2 * Copyright (c) 2007 The DragonFly Project. All rights reserved.
4 * This code is derived from software contributed to The DragonFly Project
5 * by Matthew Dillon <dillon@backplane.com>
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
17 * 3. Neither the name of The DragonFly Project nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific, prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
25 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * $DragonFly: src/sys/vfs/hammer/hammer_disk.h,v 1.55 2008/11/13 02:18:43 dillon Exp $
37 #ifndef VFS_HAMMER_DISK_H_
38 #define VFS_HAMMER_DISK_H_
40 #include <sys/endian.h>
47 * The structures below represent the on-disk format for a HAMMER
48 * filesystem. Note that all fields for on-disk structures are naturally
49 * aligned. HAMMER uses little endian for fields in on-disk structures.
50 * HAMMER doesn't support big endian arch, but is planned.
52 * Most of HAMMER revolves around the concept of an object identifier. An
53 * obj_id is a 64 bit quantity which uniquely identifies a filesystem object
54 * FOR THE ENTIRE LIFE OF THE FILESYSTEM. This uniqueness allows backups
55 * and mirrors to retain varying amounts of filesystem history by removing
56 * any possibility of conflict through identifier reuse.
58 * A HAMMER filesystem may span multiple volumes.
60 * A HAMMER filesystem uses a 16K filesystem buffer size. All filesystem
61 * I/O is done in multiples of 16K.
63 * 64K X-bufs are used for blocks >= a file's 1MB mark.
65 * Per-volume storage limit: 52 bits 4096 TB
66 * Per-Zone storage limit: 60 bits 1 MTB
67 * Per-filesystem storage limit: 60 bits 1 MTB
69 #define HAMMER_BUFSIZE 16384
70 #define HAMMER_XBUFSIZE 65536
71 #define HAMMER_HBUFSIZE (HAMMER_BUFSIZE / 2)
72 #define HAMMER_XDEMARC (1024 * 1024)
73 #define HAMMER_BUFMASK (HAMMER_BUFSIZE - 1)
74 #define HAMMER_XBUFMASK (HAMMER_XBUFSIZE - 1)
76 #define HAMMER_BUFSIZE64 ((uint64_t)HAMMER_BUFSIZE)
77 #define HAMMER_BUFMASK64 ((uint64_t)HAMMER_BUFMASK)
79 #define HAMMER_XBUFSIZE64 ((uint64_t)HAMMER_XBUFSIZE)
80 #define HAMMER_XBUFMASK64 ((uint64_t)HAMMER_XBUFMASK)
82 #define HAMMER_OFF_ZONE_MASK 0xF000000000000000ULL /* zone portion */
83 #define HAMMER_OFF_VOL_MASK 0x0FF0000000000000ULL /* volume portion */
84 #define HAMMER_OFF_SHORT_MASK 0x000FFFFFFFFFFFFFULL /* offset portion */
85 #define HAMMER_OFF_LONG_MASK 0x0FFFFFFFFFFFFFFFULL /* offset portion */
87 #define HAMMER_OFF_BAD ((hammer_off_t)-1)
90 * The current limit of volumes that can make up a HAMMER FS
92 #define HAMMER_MAX_VOLUMES 256
95 * Reserved space for (future) header junk after the volume header.
97 #define HAMMER_MIN_VOL_JUNK (HAMMER_BUFSIZE * 16) /* 256 KB */
98 #define HAMMER_MAX_VOL_JUNK HAMMER_MIN_VOL_JUNK
99 #define HAMMER_VOL_JUNK_SIZE HAMMER_MIN_VOL_JUNK
102 * Hammer transaction ids are 64 bit unsigned integers and are usually
103 * synchronized with the time of day in nanoseconds.
105 * Hammer offsets are used for FIFO indexing and embed a cycle counter
106 * and volume number in addition to the offset. Most offsets are required
107 * to be 16 KB aligned.
109 typedef uint64_t hammer_tid_t;
110 typedef uint64_t hammer_off_t;
111 typedef uint32_t hammer_crc_t;
113 #define HAMMER_MIN_TID 0ULL /* unsigned */
114 #define HAMMER_MAX_TID 0xFFFFFFFFFFFFFFFFULL /* unsigned */
115 #define HAMMER_MIN_KEY -0x8000000000000000LL /* signed */
116 #define HAMMER_MAX_KEY 0x7FFFFFFFFFFFFFFFLL /* signed */
117 #define HAMMER_MIN_OBJID HAMMER_MIN_KEY /* signed */
118 #define HAMMER_MAX_OBJID HAMMER_MAX_KEY /* signed */
119 #define HAMMER_MIN_RECTYPE 0x0U /* unsigned */
120 #define HAMMER_MAX_RECTYPE 0xFFFFU /* unsigned */
121 #define HAMMER_MIN_OFFSET 0ULL /* unsigned */
122 #define HAMMER_MAX_OFFSET 0xFFFFFFFFFFFFFFFFULL /* unsigned */
125 * hammer_off_t has several different encodings. Note that not all zones
126 * encode a vol_no. Zone bits are not a part of filesystem capacity as
127 * the zone bits aren't directly or indirectly mapped to physical volumes.
129 * In other words, HAMMER's logical filesystem offset consists of 64 bits,
130 * but the filesystem is considered 60 bits filesystem, not 64 bits.
131 * The maximum filesystem capacity is 1EB, not 16EB.
133 * zone 0: available, a big-block that contains the offset is unused
134 * zone 1 (z,v,o): raw volume relative (offset 0 is the volume header)
135 * zone 2 (z,v,o): raw buffer relative (offset 0 is the first buffer)
136 * zone 3 (z,o): undo fifo - actually zone-2 address, fixed phys array in vol hdr
137 * zone 4 (z,v,o): freemap - only real blockmap
138 * zone 8 (z,v,o): B-Tree - actually zone-2 address
139 * zone 9 (z,v,o): meta - actually zone-2 address
140 * zone 10 (z,v,o): large-data - actually zone-2 address
141 * zone 11 (z,v,o): small-data - actually zone-2 address
142 * zone 15: unavailable, usually the offset is beyond volume size
144 * layer1/layer2 direct map:
145 * Maximum HAMMER filesystem capacity from volume aspect
146 * 2^8(max volumes) * 2^52(max volume size) = 2^60 = 1EB
147 * <------------------------------------------------------------->
149 * <------><----------------------------------------------------->
150 * zzzzvvvvvvvvoooo oooooooooooooooo oooooooooooooooo oooooooooooooooo
151 * ----111111111111 1111112222222222 222222222ooooooo oooooooooooooooo
152 * <-----------------><------------------><---------------------->
153 * 18bits 19bits 23bits
154 * <------------------------------------------------------------->
155 * 2^18(layer1) * 2^19(layer2) * 2^23(big-block) = 2^60 = 1EB
156 * Maximum HAMMER filesystem capacity from blockmap aspect
159 #define HAMMER_ZONE_RAW_VOLUME 0x1000000000000000ULL
160 #define HAMMER_ZONE_RAW_BUFFER 0x2000000000000000ULL
161 #define HAMMER_ZONE_UNDO 0x3000000000000000ULL
162 #define HAMMER_ZONE_FREEMAP 0x4000000000000000ULL
163 #define HAMMER_ZONE_RESERVED05 0x5000000000000000ULL /* not used */
164 #define HAMMER_ZONE_RESERVED06 0x6000000000000000ULL /* not used */
165 #define HAMMER_ZONE_RESERVED07 0x7000000000000000ULL /* not used */
166 #define HAMMER_ZONE_BTREE 0x8000000000000000ULL
167 #define HAMMER_ZONE_META 0x9000000000000000ULL
168 #define HAMMER_ZONE_LARGE_DATA 0xA000000000000000ULL
169 #define HAMMER_ZONE_SMALL_DATA 0xB000000000000000ULL
170 #define HAMMER_ZONE_RESERVED0C 0xC000000000000000ULL /* not used */
171 #define HAMMER_ZONE_RESERVED0D 0xD000000000000000ULL /* not used */
172 #define HAMMER_ZONE_RESERVED0E 0xE000000000000000ULL /* not used */
173 #define HAMMER_ZONE_UNAVAIL 0xF000000000000000ULL
175 #define HAMMER_ZONE_RAW_VOLUME_INDEX 1
176 #define HAMMER_ZONE_RAW_BUFFER_INDEX 2
177 #define HAMMER_ZONE_UNDO_INDEX 3
178 #define HAMMER_ZONE_FREEMAP_INDEX 4
179 #define HAMMER_ZONE_BTREE_INDEX 8
180 #define HAMMER_ZONE_META_INDEX 9
181 #define HAMMER_ZONE_LARGE_DATA_INDEX 10
182 #define HAMMER_ZONE_SMALL_DATA_INDEX 11
183 #define HAMMER_ZONE_UNAVAIL_INDEX 15
185 #define HAMMER_MAX_ZONES 16
187 #define HAMMER_ZONE(offset) ((offset) & HAMMER_OFF_ZONE_MASK)
189 #define hammer_is_zone_raw_volume(offset) \
190 (HAMMER_ZONE(offset) == HAMMER_ZONE_RAW_VOLUME)
191 #define hammer_is_zone_raw_buffer(offset) \
192 (HAMMER_ZONE(offset) == HAMMER_ZONE_RAW_BUFFER)
193 #define hammer_is_zone_undo(offset) \
194 (HAMMER_ZONE(offset) == HAMMER_ZONE_UNDO)
195 #define hammer_is_zone_freemap(offset) \
196 (HAMMER_ZONE(offset) == HAMMER_ZONE_FREEMAP)
197 #define hammer_is_zone_btree(offset) \
198 (HAMMER_ZONE(offset) == HAMMER_ZONE_BTREE)
199 #define hammer_is_zone_meta(offset) \
200 (HAMMER_ZONE(offset) == HAMMER_ZONE_META)
201 #define hammer_is_zone_large_data(offset) \
202 (HAMMER_ZONE(offset) == HAMMER_ZONE_LARGE_DATA)
203 #define hammer_is_zone_small_data(offset) \
204 (HAMMER_ZONE(offset) == HAMMER_ZONE_SMALL_DATA)
205 #define hammer_is_zone_unavail(offset) \
206 (HAMMER_ZONE(offset) == HAMMER_ZONE_UNAVAIL)
207 #define hammer_is_zone_data(offset) \
208 (hammer_is_zone_large_data(offset) || hammer_is_zone_small_data(offset))
211 * Test if the zone is directly mapped to zone-2 offset via freemap.
213 #define hammer_is_zone2_mapped_index(zone) \
214 ((zone) >= HAMMER_ZONE_BTREE_INDEX && \
215 (zone) < HAMMER_MAX_ZONES)
217 * Test if the zone is directly mapped to zone-2 offset. The word
218 * directly here means the zone is neither RAW_VOLUME nor UNDO zone.
220 #define hammer_is_direct_mapped_index(zone) \
221 (((zone) == HAMMER_ZONE_RAW_BUFFER_INDEX) || \
222 ((zone) == HAMMER_ZONE_FREEMAP_INDEX) || \
223 hammer_is_zone2_mapped_index(zone))
225 #define HAMMER_ZONE_ENCODE(zone, ham_off) \
226 (((hammer_off_t)(zone) << 60) | (ham_off))
227 #define HAMMER_ZONE_DECODE(ham_off) \
228 ((int)(((hammer_off_t)(ham_off) >> 60)))
230 #define HAMMER_VOL_ENCODE(vol_no) \
231 ((hammer_off_t)((vol_no) & 255) << 52)
232 #define HAMMER_VOL_DECODE(ham_off) \
233 ((int)(((hammer_off_t)(ham_off) >> 52) & 255))
235 #define HAMMER_OFF_SHORT_ENCODE(offset) \
236 ((hammer_off_t)(offset) & HAMMER_OFF_SHORT_MASK)
237 #define HAMMER_OFF_LONG_ENCODE(offset) \
238 ((hammer_off_t)(offset) & HAMMER_OFF_LONG_MASK)
240 #define HAMMER_ENCODE(zone, vol_no, offset) \
241 (((hammer_off_t)(zone) << 60) | \
242 HAMMER_VOL_ENCODE(vol_no) | \
243 HAMMER_OFF_SHORT_ENCODE(offset))
244 #define HAMMER_ENCODE_RAW_VOLUME(vol_no, offset) \
245 HAMMER_ENCODE(HAMMER_ZONE_RAW_VOLUME_INDEX, vol_no, offset)
246 #define HAMMER_ENCODE_RAW_BUFFER(vol_no, offset) \
247 HAMMER_ENCODE(HAMMER_ZONE_RAW_BUFFER_INDEX, vol_no, offset)
248 #define HAMMER_ENCODE_FREEMAP(vol_no, offset) \
249 HAMMER_ENCODE(HAMMER_ZONE_FREEMAP_INDEX, vol_no, offset)
252 * Translate a zone address to zone-X address.
254 #define hammer_xlate_to_zoneX(zone, offset) \
255 HAMMER_ZONE_ENCODE((zone), (offset) & ~HAMMER_OFF_ZONE_MASK)
256 #define hammer_xlate_to_zone2(offset) \
257 hammer_xlate_to_zoneX(HAMMER_ZONE_RAW_BUFFER_INDEX, (offset))
259 #define hammer_data_zone(data_len) \
260 (((data_len) >= HAMMER_BUFSIZE) ? \
261 HAMMER_ZONE_LARGE_DATA : \
262 HAMMER_ZONE_SMALL_DATA)
263 #define hammer_data_zone_index(data_len) \
264 (((data_len) >= HAMMER_BUFSIZE) ? \
265 HAMMER_ZONE_LARGE_DATA_INDEX : \
266 HAMMER_ZONE_SMALL_DATA_INDEX)
269 * Big-Block backing store
271 * A blockmap is a two-level map which translates a blockmap-backed zone
272 * offset into a raw zone 2 offset. The layer 1 handles 18 bits and the
273 * layer 2 handles 19 bits. The 8M big-block size is 23 bits so two
274 * layers gives us 18+19+23 = 60 bits of address space.
276 * When using hinting for a blockmap lookup, the hint is lost when the
277 * scan leaves the HINTBLOCK, which is typically several BIGBLOCK's.
278 * HINTBLOCK is a heuristic.
280 #define HAMMER_HINTBLOCK_SIZE (HAMMER_BIGBLOCK_SIZE * 4)
281 #define HAMMER_HINTBLOCK_MASK64 ((uint64_t)HAMMER_HINTBLOCK_SIZE - 1)
282 #define HAMMER_BIGBLOCK_SIZE (8192 * 1024)
283 #define HAMMER_BIGBLOCK_SIZE64 ((uint64_t)HAMMER_BIGBLOCK_SIZE)
284 #define HAMMER_BIGBLOCK_MASK (HAMMER_BIGBLOCK_SIZE - 1)
285 #define HAMMER_BIGBLOCK_MASK64 ((uint64_t)HAMMER_BIGBLOCK_SIZE - 1)
286 #define HAMMER_BIGBLOCK_BITS 23
288 #define HAMMER_BIGBLOCK_OVERFILL (6144 * 1024)
290 #if (1 << HAMMER_BIGBLOCK_BITS) != HAMMER_BIGBLOCK_SIZE
291 #error "HAMMER_BIGBLOCK_BITS BROKEN"
294 #define HAMMER_BUFFERS_PER_BIGBLOCK \
295 (HAMMER_BIGBLOCK_SIZE / HAMMER_BUFSIZE)
296 #define HAMMER_BUFFERS_PER_BIGBLOCK_MASK \
297 (HAMMER_BUFFERS_PER_BIGBLOCK - 1)
298 #define HAMMER_BUFFERS_PER_BIGBLOCK_MASK64 \
299 ((hammer_off_t)HAMMER_BUFFERS_PER_BIGBLOCK_MASK)
302 * Maximum number of mirrors operating in master mode (multi-master
303 * clustering and mirroring). Note that HAMMER1 does not support
304 * multi-master clustering as of 2015.
306 #define HAMMER_MAX_MASTERS 16
309 * The blockmap is somewhat of a degenerate structure. HAMMER only actually
310 * uses it in its original incarnation to implement the freemap.
312 * zone:1 raw volume (no blockmap)
313 * zone:2 raw buffer (no blockmap)
314 * zone:3 undomap (direct layer2 array in volume header)
315 * zone:4 freemap (the only real blockmap)
316 * zone:8-15 zone id used to classify big-block only, address is actually
319 typedef struct hammer_blockmap {
320 hammer_off_t phys_offset; /* zone-2 physical offset */
321 hammer_off_t first_offset; /* zone-X logical offset (zone 3) */
322 hammer_off_t next_offset; /* zone-X logical offset */
323 hammer_off_t alloc_offset; /* zone-X logical offset */
325 hammer_crc_t entry_crc;
326 } *hammer_blockmap_t;
328 #define HAMMER_BLOCKMAP_CRCSIZE \
329 offsetof(struct hammer_blockmap, entry_crc)
332 * The blockmap is a 2-layer entity made up of big-blocks. The first layer
333 * contains 262144 32-byte entries (18 bits), the second layer contains
334 * 524288 16-byte entries (19 bits), representing 8MB (23 bit) blockmaps.
335 * 18+19+23 = 60 bits. The top four bits are the zone id.
337 * Currently only the freemap utilizes both layers in all their glory.
338 * All primary data/meta-data zones actually encode a zone-2 address
339 * requiring no real blockmap translation.
341 * The freemap uses the upper 8 bits of layer-1 to identify the volume,
342 * thus any space allocated via the freemap can be directly translated
343 * to a zone:2 (or zone:8-15) address.
345 * zone-X blockmap offset: [zone:4][layer1:18][layer2:19][big-block:23]
349 * 32 bytes layer1 entry for 8MB big-block.
350 * A big-block can hold 2^23 / 2^5 = 2^18 layer1 entries,
351 * which equals bits assigned for layer1 in zone-2 address.
353 typedef struct hammer_blockmap_layer1 {
354 hammer_off_t blocks_free; /* big-blocks free */
355 hammer_off_t phys_offset; /* UNAVAIL or zone-2 */
356 hammer_off_t reserved01;
357 hammer_crc_t layer2_crc; /* xor'd crc's of HAMMER_BLOCKSIZE */
359 hammer_crc_t layer1_crc; /* MUST BE LAST FIELD OF STRUCTURE*/
360 } *hammer_blockmap_layer1_t;
362 #define HAMMER_LAYER1_CRCSIZE \
363 offsetof(struct hammer_blockmap_layer1, layer1_crc)
366 * 16 bytes layer2 entry for 8MB big-blocks.
367 * A big-block can hold 2^23 / 2^4 = 2^19 layer2 entries,
368 * which equals bits assigned for layer2 in zone-2 address.
370 * NOTE: bytes_free is signed and can legally go negative if/when data
371 * de-dup occurs. This field will never go higher than
372 * HAMMER_BIGBLOCK_SIZE. If exactly HAMMER_BIGBLOCK_SIZE
373 * the big-block is completely free.
375 typedef struct hammer_blockmap_layer2 {
376 uint8_t zone; /* typed allocation zone */
379 uint32_t append_off; /* allocatable space index */
380 int32_t bytes_free; /* bytes free within this big-block */
381 hammer_crc_t entry_crc;
382 } *hammer_blockmap_layer2_t;
384 #define HAMMER_LAYER2_CRCSIZE \
385 offsetof(struct hammer_blockmap_layer2, entry_crc)
387 #define HAMMER_BLOCKMAP_UNAVAIL ((hammer_off_t)-1LL)
389 #define HAMMER_BLOCKMAP_RADIX1 /* 2^18 = 262144 */ \
390 ((int)(HAMMER_BIGBLOCK_SIZE / sizeof(struct hammer_blockmap_layer1)))
391 #define HAMMER_BLOCKMAP_RADIX2 /* 2^19 = 524288 */ \
392 ((int)(HAMMER_BIGBLOCK_SIZE / sizeof(struct hammer_blockmap_layer2)))
394 #define HAMMER_BLOCKMAP_LAYER1 /* 2^(18+19+23) = 1EB */ \
395 (HAMMER_BLOCKMAP_RADIX1 * HAMMER_BLOCKMAP_LAYER2)
396 #define HAMMER_BLOCKMAP_LAYER2 /* 2^(19+23) = 4TB */ \
397 (HAMMER_BLOCKMAP_RADIX2 * HAMMER_BIGBLOCK_SIZE64)
399 #define HAMMER_BLOCKMAP_LAYER1_MASK (HAMMER_BLOCKMAP_LAYER1 - 1)
400 #define HAMMER_BLOCKMAP_LAYER2_MASK (HAMMER_BLOCKMAP_LAYER2 - 1)
403 * Index within layer1 or layer2 big-block for the entry representing
404 * a zone-2 physical offset.
406 #define HAMMER_BLOCKMAP_LAYER1_INDEX(zone2_offset) \
407 ((int)(((zone2_offset) & HAMMER_BLOCKMAP_LAYER1_MASK) / \
408 HAMMER_BLOCKMAP_LAYER2))
410 #define HAMMER_BLOCKMAP_LAYER2_INDEX(zone2_offset) \
411 ((int)(((zone2_offset) & HAMMER_BLOCKMAP_LAYER2_MASK) / \
412 HAMMER_BIGBLOCK_SIZE64))
415 * Byte offset within layer1 or layer2 big-block for the entry representing
416 * a zone-2 physical offset. Multiply the index by sizeof(blockmap_layer).
418 #define HAMMER_BLOCKMAP_LAYER1_OFFSET(zone2_offset) \
419 (HAMMER_BLOCKMAP_LAYER1_INDEX(zone2_offset) * \
420 sizeof(struct hammer_blockmap_layer1))
422 #define HAMMER_BLOCKMAP_LAYER2_OFFSET(zone2_offset) \
423 (HAMMER_BLOCKMAP_LAYER2_INDEX(zone2_offset) * \
424 sizeof(struct hammer_blockmap_layer2))
427 * HAMMER UNDO parameters. The UNDO fifo is mapped directly in the volume
428 * header with an array of zone-2 offsets. A maximum of (128x8MB) = 1GB,
429 * and minimum of (64x8MB) = 512MB may be reserved. The size of the undo
430 * fifo is usually set a newfs time.
432 #define HAMMER_MIN_UNDO_BIGBLOCKS 64
433 #define HAMMER_MAX_UNDO_BIGBLOCKS 128
436 * All on-disk HAMMER structures which make up elements of the UNDO FIFO
437 * contain a hammer_fifo_head and hammer_fifo_tail structure. This structure
438 * contains all the information required to validate the fifo element
439 * and to scan the fifo in either direction. The head is typically embedded
440 * in higher level hammer on-disk structures while the tail is typically
441 * out-of-band. hdr_size is the size of the whole mess, including the tail.
443 * All undo structures are guaranteed to not cross a 16K filesystem
444 * buffer boundary. Most undo structures are fairly small. Data spaces
445 * are not immediately reused by HAMMER so file data is not usually recorded
446 * as part of an UNDO.
448 * PAD elements are allowed to take up only 8 bytes of space as a special
449 * case, containing only hdr_signature, hdr_type, and hdr_size fields,
450 * and with the tail overloaded onto the head structure for 8 bytes total.
452 * Every undo record has a sequence number. This number is unrelated to
453 * transaction ids and instead collects the undo transactions associated
454 * with a single atomic operation. A larger transactional operation, such
455 * as a remove(), may consist of several smaller atomic operations
456 * representing raw meta-data operations.
458 * HAMMER VERSION 4 CHANGES
460 * In HAMMER version 4 the undo structure alignment is reduced from 16384
461 * to 512 bytes in order to ensure that each 512 byte sector begins with
462 * a header. The hdr_seq field in the header is a 32 bit sequence number
463 * which allows the recovery code to detect missing sectors
464 * without relying on the 32-bit crc and to definitively identify the current
465 * undo sequence space without having to rely on information from the volume
466 * header. In addition, new REDO entries in the undo space are used to
467 * record write, write/extend, and transaction id updates.
469 * The grand result is:
471 * (1) The volume header no longer needs to be synchronized for most
472 * flush and fsync operations.
474 * (2) Most fsync operations need only lay down REDO records
476 * (3) Data overwrite for nohistory operations covered by REDO records
477 * can be supported (instead of rolling a new block allocation),
478 * by rolling UNDO for the prior contents of the data.
480 * HAMMER VERSION 5 CHANGES
482 * Hammer version 5 contains a minor adjustment making layer2's bytes_free
483 * field signed, allowing dedup to push it into the negative domain.
485 #define HAMMER_HEAD_ALIGN 8
486 #define HAMMER_HEAD_ALIGN_MASK (HAMMER_HEAD_ALIGN - 1)
487 #define HAMMER_HEAD_DOALIGN(bytes) \
488 (((bytes) + HAMMER_HEAD_ALIGN_MASK) & ~HAMMER_HEAD_ALIGN_MASK)
490 #define HAMMER_UNDO_ALIGN 512
491 #define HAMMER_UNDO_ALIGN64 ((uint64_t)512)
492 #define HAMMER_UNDO_MASK (HAMMER_UNDO_ALIGN - 1)
493 #define HAMMER_UNDO_MASK64 (HAMMER_UNDO_ALIGN64 - 1)
495 typedef struct hammer_fifo_head {
496 uint16_t hdr_signature;
498 uint32_t hdr_size; /* Aligned size of the whole mess */
499 uint32_t hdr_seq; /* Sequence number */
500 hammer_crc_t hdr_crc; /* XOR crc up to field w/ crc after field */
501 } *hammer_fifo_head_t;
503 #define HAMMER_FIFO_HEAD_CRCOFF offsetof(struct hammer_fifo_head, hdr_crc)
505 typedef struct hammer_fifo_tail {
506 uint16_t tail_signature;
508 uint32_t tail_size; /* aligned size of the whole mess */
509 } *hammer_fifo_tail_t;
514 * NOTE: 0x8000U part of HAMMER_HEAD_TYPE_PAD can be removed if the HAMMER
515 * version ever gets bumped again. It exists only to keep compatibility with
518 #define HAMMER_HEAD_TYPE_PAD (0x0040U | 0x8000U)
519 #define HAMMER_HEAD_TYPE_DUMMY 0x0041U /* dummy entry w/seqno */
520 #define HAMMER_HEAD_TYPE_UNDO 0x0043U /* random UNDO information */
521 #define HAMMER_HEAD_TYPE_REDO 0x0044U /* data REDO / fast fsync */
523 #define HAMMER_HEAD_SIGNATURE 0xC84EU
524 #define HAMMER_TAIL_SIGNATURE 0xC74FU
527 * Misc FIFO structures.
529 * UNDO - Raw meta-data media updates.
531 typedef struct hammer_fifo_undo {
532 struct hammer_fifo_head head;
533 hammer_off_t undo_offset; /* zone-1,2 offset */
534 int32_t undo_data_bytes;
535 int32_t undo_reserved01;
536 /* followed by data */
537 } *hammer_fifo_undo_t;
540 * REDO (HAMMER version 4+) - Logical file writes/truncates.
542 * REDOs contain information which will be duplicated in a later meta-data
543 * update, allowing fast write()+fsync() operations. REDOs can be ignored
544 * without harming filesystem integrity but must be processed if fsync()
545 * semantics are desired.
547 * Unlike UNDOs which are processed backwards within the recovery span,
548 * REDOs must be processed forwards starting further back (starting outside
549 * the recovery span).
551 * WRITE - Write logical file (with payload). Executed both
552 * out-of-span and in-span. Out-of-span WRITEs may be
553 * filtered out by TERMs.
555 * TRUNC - Truncate logical file (no payload). Executed both
556 * out-of-span and in-span. Out-of-span WRITEs may be
557 * filtered out by TERMs.
559 * TERM_* - Indicates meta-data was committed (if out-of-span) or
560 * will be rolled-back (in-span). Any out-of-span TERMs
561 * matching earlier WRITEs remove those WRITEs from
562 * consideration as they might conflict with a later data
563 * commit (which is not being rolled-back).
565 * SYNC - The earliest in-span SYNC (the last one when scanning
566 * backwards) tells the recovery code how far out-of-span
567 * it must go to run REDOs.
569 * NOTE: WRITEs do not always have matching TERMs even under
570 * perfect conditions because truncations might remove the
571 * buffers from consideration. I/O problems can also remove
572 * buffers from consideration.
574 * TRUNCSs do not always have matching TERMs because several
575 * truncations may be aggregated together into a single TERM.
577 typedef struct hammer_fifo_redo {
578 struct hammer_fifo_head head;
579 int64_t redo_objid; /* file being written */
580 hammer_off_t redo_offset; /* logical offset in file */
581 int32_t redo_data_bytes;
583 uint32_t redo_localization;
584 uint32_t redo_reserved;
585 uint64_t redo_mtime; /* set mtime */
586 } *hammer_fifo_redo_t;
588 #define HAMMER_REDO_WRITE 0x00000001
589 #define HAMMER_REDO_TRUNC 0x00000002
590 #define HAMMER_REDO_TERM_WRITE 0x00000004
591 #define HAMMER_REDO_TERM_TRUNC 0x00000008
592 #define HAMMER_REDO_SYNC 0x00000010
594 typedef union hammer_fifo_any {
595 struct hammer_fifo_head head;
596 struct hammer_fifo_undo undo;
597 struct hammer_fifo_redo redo;
598 } *hammer_fifo_any_t;
601 * Volume header types
603 #define HAMMER_FSBUF_VOLUME 0xC8414D4DC5523031ULL /* HAMMER01 */
604 #define HAMMER_FSBUF_VOLUME_REV 0x313052C54D4D41C8ULL /* (reverse endian) */
607 * HAMMER Volume header
609 * A HAMMER filesystem can be built from 1-256 block devices, each block
610 * device contains a volume header followed by however many buffers fit
613 * One of the volumes making up a HAMMER filesystem is the root volume.
614 * The root volume is always volume #0 which is the first block device path
615 * specified by newfs_hammer(8). All HAMMER volumes have a volume header,
616 * however the root volume may be the only volume that has valid values for
617 * some fields in the header.
619 * Special field notes:
621 * vol_bot_beg - offset of boot area (mem_beg - bot_beg bytes)
622 * vol_mem_beg - offset of memory log (buf_beg - mem_beg bytes)
623 * vol_buf_beg - offset of the first buffer in volume
624 * vol_buf_end - offset of volume EOF (on buffer boundary)
626 * The memory log area allows a kernel to cache new records and data
627 * in memory without allocating space in the actual filesystem to hold
628 * the records and data. In the event that a filesystem becomes full,
629 * any records remaining in memory can be flushed to the memory log
630 * area. This allows the kernel to immediately return success.
632 * The buffer offset is a physical offset of zone-2 offset. The lower
633 * 52 bits of the zone-2 offset is added to the buffer offset of each
634 * volume to generate an actual I/O offset within the block device.
636 * NOTE: boot area and memory log are currently not used.
640 * Filesystem type string
642 #define HAMMER_FSTYPE_STRING "DragonFly HAMMER"
645 * These macros are only used by userspace when userspace commands either
646 * initialize or add a new HAMMER volume.
648 #define HAMMER_BOOT_MINBYTES (32*1024)
649 #define HAMMER_BOOT_NOMBYTES (64LL*1024*1024)
650 #define HAMMER_BOOT_MAXBYTES (256LL*1024*1024)
652 #define HAMMER_MEM_MINBYTES (256*1024)
653 #define HAMMER_MEM_NOMBYTES (1LL*1024*1024*1024)
654 #define HAMMER_MEM_MAXBYTES (64LL*1024*1024*1024)
656 typedef struct hammer_volume_ondisk {
657 uint64_t vol_signature; /* HAMMER_FSBUF_VOLUME for a valid header */
660 * These are relative to block device offset, not zone offsets.
662 int64_t vol_bot_beg; /* offset of boot area */
663 int64_t vol_mem_beg; /* offset of memory log */
664 int64_t vol_buf_beg; /* offset of the first buffer in volume */
665 int64_t vol_buf_end; /* offset of volume EOF (on buffer boundary) */
666 int64_t vol_reserved01;
668 uuid_t vol_fsid; /* identify filesystem */
669 uuid_t vol_fstype; /* identify filesystem type */
670 char vol_label[64]; /* filesystem label */
672 int32_t vol_no; /* volume number within filesystem */
673 int32_t vol_count; /* number of volumes making up filesystem */
675 uint32_t vol_version; /* version control information */
676 hammer_crc_t vol_crc; /* header crc */
677 uint32_t vol_flags; /* volume flags */
678 uint32_t vol_rootvol; /* the root volume number (must be 0) */
680 uint32_t vol_reserved[8];
683 * These fields are initialized and space is reserved in every
684 * volume making up a HAMMER filesytem, but only the root volume
685 * contains valid data. Note that vol0_stat_bigblocks does not
686 * include big-blocks for freemap and undomap initially allocated
687 * by newfs_hammer(8).
689 int64_t vol0_stat_bigblocks; /* total big-blocks when fs is empty */
690 int64_t vol0_stat_freebigblocks;/* number of free big-blocks */
691 int64_t vol0_reserved01;
692 int64_t vol0_stat_inodes; /* for statfs only */
693 int64_t vol0_reserved02;
694 hammer_off_t vol0_btree_root; /* B-Tree root offset in zone-8 */
695 hammer_tid_t vol0_next_tid; /* highest partially synchronized TID */
696 hammer_off_t vol0_reserved03;
699 * Blockmaps for zones. Not all zones use a blockmap. Note that
700 * the entire root blockmap is cached in the hammer_mount structure.
702 struct hammer_blockmap vol0_blockmap[HAMMER_MAX_ZONES];
705 * Array of zone-2 addresses for undo FIFO.
707 hammer_off_t vol0_undo_array[HAMMER_MAX_UNDO_BIGBLOCKS];
708 } *hammer_volume_ondisk_t;
710 #define HAMMER_ROOT_VOLNO 0
712 #define HAMMER_VOLF_NEEDFLUSH 0x0004 /* volume needs flush */
714 #define HAMMER_VOL_CRCSIZE1 \
715 offsetof(struct hammer_volume_ondisk, vol_crc)
716 #define HAMMER_VOL_CRCSIZE2 \
717 (sizeof(struct hammer_volume_ondisk) - HAMMER_VOL_CRCSIZE1 - \
718 sizeof(hammer_crc_t))
720 #define HAMMER_VOL_VERSION_MIN 1 /* minimum supported version */
721 #define HAMMER_VOL_VERSION_DEFAULT 6 /* newfs default version */
722 #define HAMMER_VOL_VERSION_WIP 7 /* version >= this is WIP */
723 #define HAMMER_VOL_VERSION_MAX 6 /* maximum supported version */
725 #define HAMMER_VOL_VERSION_ONE 1
726 #define HAMMER_VOL_VERSION_TWO 2 /* new dirent layout (2.3+) */
727 #define HAMMER_VOL_VERSION_THREE 3 /* new snapshot layout (2.5+) */
728 #define HAMMER_VOL_VERSION_FOUR 4 /* new undo/flush (2.5+) */
729 #define HAMMER_VOL_VERSION_FIVE 5 /* dedup (2.9+) */
730 #define HAMMER_VOL_VERSION_SIX 6 /* DIRHASH_ALG1 */
733 * Translate a zone-2 address to physical address
735 #define hammer_xlate_to_phys(volume, zone2_offset) \
736 ((volume)->vol_buf_beg + HAMMER_OFF_SHORT_ENCODE(zone2_offset))
739 * Effective per-volume filesystem capacity including big-blocks for layer1/2
741 #define HAMMER_VOL_BUF_SIZE(volume) \
742 ((volume)->vol_buf_end - (volume)->vol_buf_beg)
745 * Record types are fairly straightforward. The B-Tree includes the record
746 * type in its index sort.
748 #define HAMMER_RECTYPE_UNKNOWN 0x0000
749 #define HAMMER_RECTYPE_INODE 0x0001 /* inode in obj_id space */
750 #define HAMMER_RECTYPE_DATA 0x0010
751 #define HAMMER_RECTYPE_DIRENTRY 0x0011
752 #define HAMMER_RECTYPE_DB 0x0012
753 #define HAMMER_RECTYPE_EXT 0x0013 /* ext attributes */
754 #define HAMMER_RECTYPE_FIX 0x0014 /* fixed attribute */
755 #define HAMMER_RECTYPE_PFS 0x0015 /* PFS management */
756 #define HAMMER_RECTYPE_SNAPSHOT 0x0016 /* Snapshot management */
757 #define HAMMER_RECTYPE_CONFIG 0x0017 /* hammer cleanup config */
758 #define HAMMER_RECTYPE_MAX 0xFFFF
760 #define HAMMER_RECTYPE_ENTRY_START (HAMMER_RECTYPE_INODE + 1)
761 #define HAMMER_RECTYPE_CLEAN_START HAMMER_RECTYPE_EXT
763 #define HAMMER_FIXKEY_SYMLINK 1
765 #define HAMMER_OBJTYPE_UNKNOWN 0 /* never exists on-disk as unknown */
766 #define HAMMER_OBJTYPE_DIRECTORY 1
767 #define HAMMER_OBJTYPE_REGFILE 2
768 #define HAMMER_OBJTYPE_DBFILE 3
769 #define HAMMER_OBJTYPE_FIFO 4
770 #define HAMMER_OBJTYPE_CDEV 5
771 #define HAMMER_OBJTYPE_BDEV 6
772 #define HAMMER_OBJTYPE_SOFTLINK 7
773 #define HAMMER_OBJTYPE_PSEUDOFS 8 /* pseudo filesystem obj */
774 #define HAMMER_OBJTYPE_SOCKET 9
777 * HAMMER inode attribute data
779 * The data reference for a HAMMER inode points to this structure. Any
780 * modifications to the contents of this structure will result in a
781 * replacement operation.
783 * parent_obj_id is only valid for directories (which cannot be hard-linked),
784 * and specifies the parent directory obj_id. This field will also be set
785 * for non-directory inodes as a recovery aid, but can wind up holding
786 * stale information. However, since object id's are not reused, the worse
787 * that happens is that the recovery code is unable to use it.
788 * A parent_obj_id of 0 means it's a root inode of root or non-root PFS.
790 * NOTE: Future note on directory hardlinks. We can implement a record type
791 * which allows us to point to multiple parent directories.
793 typedef struct hammer_inode_data {
794 uint16_t version; /* inode data version */
795 uint16_t mode; /* basic unix permissions */
796 uint32_t uflags; /* chflags */
797 uint32_t rmajor; /* used by device nodes */
798 uint32_t rminor; /* used by device nodes */
800 int64_t parent_obj_id; /* parent directory obj_id */
805 uint8_t cap_flags; /* capability support flags (extension) */
808 uint64_t nlinks; /* hard links */
809 uint64_t size; /* filesystem object size */
811 char symlink[24]; /* HAMMER_INODE_BASESYMLEN */
813 uint64_t mtime; /* mtime must be second-to-last */
814 uint64_t atime; /* atime must be last */
815 } *hammer_inode_data_t;
818 * Neither mtime nor atime upates are CRCd by the B-Tree element.
819 * mtime updates have UNDO, atime updates do not.
821 #define HAMMER_INODE_CRCSIZE \
822 offsetof(struct hammer_inode_data, mtime)
824 #define HAMMER_INODE_DATA_VERSION 1
825 #define HAMMER_OBJID_ROOT 1 /* root inodes # */
826 #define HAMMER_INODE_BASESYMLEN 24 /* see ext.symlink */
829 * Capability & implementation flags.
831 * HAMMER_INODE_CAP_DIR_LOCAL_INO - Use inode B-Tree localization
832 * for directory entries. Also see HAMMER_DIR_INODE_LOCALIZATION().
834 #define HAMMER_INODE_CAP_DIRHASH_MASK 0x03 /* directory: hash algorithm */
835 #define HAMMER_INODE_CAP_DIRHASH_ALG0 0x00
836 #define HAMMER_INODE_CAP_DIRHASH_ALG1 0x01
837 #define HAMMER_INODE_CAP_DIRHASH_ALG2 0x02
838 #define HAMMER_INODE_CAP_DIRHASH_ALG3 0x03
839 #define HAMMER_INODE_CAP_DIR_LOCAL_INO 0x04 /* use inode localization */
842 * A HAMMER directory entry associates a HAMMER filesystem object with a
843 * namespace. It is hooked into a pseudo-filesystem (with its own inode
844 * numbering space) in the filesystem by setting the high 16 bits of the
845 * localization field. The low 16 bits must be 0 and are reserved for
848 * Directory entries are indexed with a 128 bit namekey rather then an
849 * offset. A portion of the namekey is an iterator/randomizer to deal
852 * NOTE: leaf.base.obj_type from the related B-Tree leaf entry holds
853 * the filesystem object type of obj_id, e.g. a den_type equivalent.
854 * It is not stored in hammer_direntry_data.
856 * NOTE: name field / the filename data reference is NOT terminated with \0.
858 typedef struct hammer_direntry_data {
859 int64_t obj_id; /* object being referenced */
860 uint32_t localization; /* identify pseudo-filesystem */
862 char name[16]; /* name (extended) */
863 } *hammer_direntry_data_t;
865 #define HAMMER_ENTRY_NAME_OFF offsetof(struct hammer_direntry_data, name[0])
866 #define HAMMER_ENTRY_SIZE(nlen) offsetof(struct hammer_direntry_data, name[nlen])
869 * Symlink data which does not fit in the inode is stored in a separate
872 typedef struct hammer_symlink_data {
873 char name[16]; /* name (extended) */
874 } *hammer_symlink_data_t;
876 #define HAMMER_SYMLINK_NAME_OFF offsetof(struct hammer_symlink_data, name[0])
879 * The root inode for the primary filesystem and root inode for any
880 * pseudo-fs may be tagged with an optional data structure using
881 * HAMMER_RECTYPE_PFS and localization id. This structure allows
882 * the node to be used as a mirroring master or slave.
884 * When operating as a slave CD's into the node automatically become read-only
885 * and as-of sync_end_tid.
887 * When operating as a master the read PFSD info sets sync_end_tid to
888 * the most recently flushed TID.
890 * sync_low_tid is not yet used but will represent the highest pruning
891 * end-point, after which full history is available.
893 * We need to pack this structure making it equally sized on both 32-bit and
894 * 64-bit machines as it is part of struct hammer_ioc_mrecord_pfs which is
895 * send over the wire in hammer mirror operations. Only on 64-bit machines
896 * the size of this struct differ when packed or not. This leads us to the
897 * situation where old 64-bit systems (using the non-packed structure),
898 * which were never able to mirror to/from 32-bit systems, are now no longer
899 * able to mirror to/from newer 64-bit systems (using the packed structure).
901 struct hammer_pseudofs_data {
902 hammer_tid_t sync_low_tid; /* full history beyond this point */
903 hammer_tid_t sync_beg_tid; /* earliest tid w/ full history avail */
904 hammer_tid_t sync_end_tid; /* current synchronizatoin point */
905 uint64_t sync_beg_ts; /* real-time of last completed sync */
906 uint64_t sync_end_ts; /* initiation of current sync cycle */
907 uuid_t shared_uuid; /* shared uuid (match required) */
908 uuid_t unique_uuid; /* unique uuid of this master/slave */
909 int32_t reserved01; /* reserved for future master_id */
910 int32_t mirror_flags; /* misc flags */
911 char label[64]; /* filesystem space label */
912 char snapshots[64]; /* softlink dir for pruning */
913 int32_t reserved02; /* was prune_{time,freq} */
914 int32_t reserved03; /* was reblock_{time,freq} */
915 int32_t reserved04; /* was snapshot_freq */
916 int32_t prune_min; /* do not prune recent history */
917 int32_t prune_max; /* do not retain history beyond here */
918 int32_t reserved[16];
921 typedef struct hammer_pseudofs_data *hammer_pseudofs_data_t;
923 #define HAMMER_PFSD_SLAVE 0x00000001
924 #define HAMMER_PFSD_DELETED 0x80000000
926 #define hammer_is_pfs_slave(pfsd) \
927 (((pfsd)->mirror_flags & HAMMER_PFSD_SLAVE) != 0)
928 #define hammer_is_pfs_master(pfsd) \
929 (!hammer_is_pfs_slave(pfsd))
930 #define hammer_is_pfs_deleted(pfsd) \
931 (((pfsd)->mirror_flags & HAMMER_PFSD_DELETED) != 0)
933 #define HAMMER_MAX_PFS 65536
934 #define HAMMER_MAX_PFSID (HAMMER_MAX_PFS - 1)
935 #define HAMMER_ROOT_PFSID 0
938 * Snapshot meta-data { Objid = HAMMER_OBJID_ROOT, Key = tid, rectype = SNAPSHOT }.
940 * Snapshot records replace the old <fs>/snapshots/<softlink> methodology. Snapshot
941 * records are mirrored but may be independantly managed once they are laid down on
944 * NOTE: The b-tree key is signed, the tid is not, so callers must still sort the
947 * NOTE: Reserved fields must be zero (as usual)
949 typedef struct hammer_snapshot_data {
950 hammer_tid_t tid; /* the snapshot TID itself (== key) */
951 uint64_t ts; /* real-time when snapshot was made */
954 char label[64]; /* user-supplied description */
955 uint64_t reserved03[4];
956 } *hammer_snapshot_data_t;
959 * Config meta-data { ObjId = HAMMER_OBJID_ROOT, Key = 0, rectype = CONFIG }.
961 * Used to store the hammer cleanup config. This data is not mirrored.
963 typedef struct hammer_config_data {
965 } *hammer_config_data_t;
968 * Rollup various structures embedded as record data
970 typedef union hammer_data_ondisk {
971 struct hammer_direntry_data entry;
972 struct hammer_inode_data inode;
973 struct hammer_symlink_data symlink;
974 struct hammer_pseudofs_data pfsd;
975 struct hammer_snapshot_data snap;
976 struct hammer_config_data config;
977 } *hammer_data_ondisk_t;
980 * Ondisk layout of B-Tree related structures
982 #include "hammer_btree.h"
984 #define HAMMER_DIR_INODE_LOCALIZATION(ino_data) \
985 (((ino_data)->cap_flags & HAMMER_INODE_CAP_DIR_LOCAL_INO) ? \
986 HAMMER_LOCALIZE_INODE : \
987 HAMMER_LOCALIZE_MISC)
989 #endif /* !VFS_HAMMER_DISK_H_ */