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_
45 * The structures below represent the on-disk format for a HAMMER
46 * filesystem. Note that all fields for on-disk structures are naturally
47 * aligned. The host endian format is used - compatibility is possible
48 * if the implementation detects reversed endian and adjusts data accordingly.
50 * Most of HAMMER revolves around the concept of an object identifier. An
51 * obj_id is a 64 bit quantity which uniquely identifies a filesystem object
52 * FOR THE ENTIRE LIFE OF THE FILESYSTEM. This uniqueness allows backups
53 * and mirrors to retain varying amounts of filesystem history by removing
54 * any possibility of conflict through identifier reuse.
56 * A HAMMER filesystem may span multiple volumes.
58 * A HAMMER filesystem uses a 16K filesystem buffer size. All filesystem
59 * I/O is done in multiples of 16K.
61 * 64K X-bufs are used for blocks >= a file's 1MB mark.
63 * Per-volume storage limit: 52 bits 4096 TB
64 * Per-Zone storage limit: 60 bits 1 MTB
65 * Per-filesystem storage limit: 60 bits 1 MTB
67 #define HAMMER_BUFSIZE 16384
68 #define HAMMER_XBUFSIZE 65536
69 #define HAMMER_HBUFSIZE (HAMMER_BUFSIZE / 2)
70 #define HAMMER_XDEMARC (1024 * 1024)
71 #define HAMMER_BUFMASK (HAMMER_BUFSIZE - 1)
72 #define HAMMER_XBUFMASK (HAMMER_XBUFSIZE - 1)
74 #define HAMMER_BUFSIZE64 ((uint64_t)HAMMER_BUFSIZE)
75 #define HAMMER_BUFMASK64 ((uint64_t)HAMMER_BUFMASK)
77 #define HAMMER_XBUFSIZE64 ((uint64_t)HAMMER_XBUFSIZE)
78 #define HAMMER_XBUFMASK64 ((uint64_t)HAMMER_XBUFMASK)
80 #define HAMMER_OFF_ZONE_MASK 0xF000000000000000ULL /* zone portion */
81 #define HAMMER_OFF_VOL_MASK 0x0FF0000000000000ULL /* volume portion */
82 #define HAMMER_OFF_SHORT_MASK 0x000FFFFFFFFFFFFFULL /* offset portion */
83 #define HAMMER_OFF_LONG_MASK 0x0FFFFFFFFFFFFFFFULL /* offset portion */
85 #define HAMMER_OFF_BAD ((hammer_off_t)-1)
88 * The current limit of volumes that can make up a HAMMER FS
90 #define HAMMER_MAX_VOLUMES 256
93 * Reserved space for (future) header junk after the volume header.
95 #define HAMMER_VOL_ALLOC (HAMMER_BUFSIZE * 16)
98 * Hammer transaction ids are 64 bit unsigned integers and are usually
99 * synchronized with the time of day in nanoseconds.
101 * Hammer offsets are used for FIFO indexing and embed a cycle counter
102 * and volume number in addition to the offset. Most offsets are required
103 * to be 16 KB aligned.
105 typedef uint64_t hammer_tid_t;
106 typedef uint64_t hammer_off_t;
107 typedef uint32_t hammer_crc_t;
109 #define HAMMER_MIN_TID 0ULL /* unsigned */
110 #define HAMMER_MAX_TID 0xFFFFFFFFFFFFFFFFULL /* unsigned */
111 #define HAMMER_MIN_KEY -0x8000000000000000LL /* signed */
112 #define HAMMER_MAX_KEY 0x7FFFFFFFFFFFFFFFLL /* signed */
113 #define HAMMER_MIN_OBJID HAMMER_MIN_KEY /* signed */
114 #define HAMMER_MAX_OBJID HAMMER_MAX_KEY /* signed */
115 #define HAMMER_MIN_RECTYPE 0x0U /* unsigned */
116 #define HAMMER_MAX_RECTYPE 0xFFFFU /* unsigned */
117 #define HAMMER_MIN_OFFSET 0ULL /* unsigned */
118 #define HAMMER_MAX_OFFSET 0xFFFFFFFFFFFFFFFFULL /* unsigned */
121 * hammer_off_t has several different encodings. Note that not all zones
122 * encode a vol_no. Zone bits are not a part of filesystem capacity as
123 * the zone bits aren't directly or indirectly mapped to physical volumes.
125 * In other words, HAMMER's logical filesystem offset consists of 64 bits,
126 * but the filesystem is considered 60 bits filesystem, not 64 bits.
127 * The maximum filesystem capacity is 1EB, not 16EB.
129 * zone 0: available, a big-block that contains the offset is unused
130 * zone 1 (z,v,o): raw volume relative (offset 0 is the volume header)
131 * zone 2 (z,v,o): raw buffer relative (offset 0 is the first buffer)
132 * zone 3 (z,o): undo fifo - actually zone-2 address, fixed phys array in vol hdr
133 * zone 4 (z,v,o): freemap - only real blockmap
134 * zone 8 (z,v,o): B-Tree - actually zone-2 address
135 * zone 9 (z,v,o): meta - actually zone-2 address
136 * zone 10 (z,v,o): large-data - actually zone-2 address
137 * zone 11 (z,v,o): small-data - actually zone-2 address
138 * zone 15: unavailable, usually the offset is beyond volume size
140 * layer1/layer2 direct map:
141 * Maximum HAMMER filesystem capacity from volume aspect
142 * 2^8(max volumes) * 2^52(max volume size) = 2^60 = 1EB
143 * <------------------------------------------------------------->
145 * <------><----------------------------------------------------->
146 * zzzzvvvvvvvvoooo oooooooooooooooo oooooooooooooooo oooooooooooooooo
147 * ----111111111111 1111112222222222 222222222ooooooo oooooooooooooooo
148 * <-----------------><------------------><---------------------->
149 * 18bits 19bits 23bits
150 * <------------------------------------------------------------->
151 * 2^18(layer1) * 2^19(layer2) * 2^23(big-block) = 2^60 = 1EB
152 * Maximum HAMMER filesystem capacity from blockmap aspect
155 #define HAMMER_ZONE_RAW_VOLUME 0x1000000000000000ULL
156 #define HAMMER_ZONE_RAW_BUFFER 0x2000000000000000ULL
157 #define HAMMER_ZONE_UNDO 0x3000000000000000ULL
158 #define HAMMER_ZONE_FREEMAP 0x4000000000000000ULL
159 #define HAMMER_ZONE_RESERVED05 0x5000000000000000ULL /* not used */
160 #define HAMMER_ZONE_RESERVED06 0x6000000000000000ULL /* not used */
161 #define HAMMER_ZONE_RESERVED07 0x7000000000000000ULL /* not used */
162 #define HAMMER_ZONE_BTREE 0x8000000000000000ULL
163 #define HAMMER_ZONE_META 0x9000000000000000ULL
164 #define HAMMER_ZONE_LARGE_DATA 0xA000000000000000ULL
165 #define HAMMER_ZONE_SMALL_DATA 0xB000000000000000ULL
166 #define HAMMER_ZONE_RESERVED0C 0xC000000000000000ULL /* not used */
167 #define HAMMER_ZONE_RESERVED0D 0xD000000000000000ULL /* not used */
168 #define HAMMER_ZONE_RESERVED0E 0xE000000000000000ULL /* not used */
169 #define HAMMER_ZONE_UNAVAIL 0xF000000000000000ULL
171 #define HAMMER_ZONE_RAW_VOLUME_INDEX 1
172 #define HAMMER_ZONE_RAW_BUFFER_INDEX 2
173 #define HAMMER_ZONE_UNDO_INDEX 3
174 #define HAMMER_ZONE_FREEMAP_INDEX 4
175 #define HAMMER_ZONE_BTREE_INDEX 8
176 #define HAMMER_ZONE_META_INDEX 9
177 #define HAMMER_ZONE_LARGE_DATA_INDEX 10
178 #define HAMMER_ZONE_SMALL_DATA_INDEX 11
179 #define HAMMER_ZONE_UNAVAIL_INDEX 15
181 #define HAMMER_MAX_ZONES 16
183 #define HAMMER_ZONE(offset) ((offset) & HAMMER_OFF_ZONE_MASK)
185 #define hammer_is_zone_raw_volume(offset) \
186 (HAMMER_ZONE(offset) == HAMMER_ZONE_RAW_VOLUME)
187 #define hammer_is_zone_raw_buffer(offset) \
188 (HAMMER_ZONE(offset) == HAMMER_ZONE_RAW_BUFFER)
189 #define hammer_is_zone_undo(offset) \
190 (HAMMER_ZONE(offset) == HAMMER_ZONE_UNDO)
191 #define hammer_is_zone_freemap(offset) \
192 (HAMMER_ZONE(offset) == HAMMER_ZONE_FREEMAP)
193 #define hammer_is_zone_btree(offset) \
194 (HAMMER_ZONE(offset) == HAMMER_ZONE_BTREE)
195 #define hammer_is_zone_meta(offset) \
196 (HAMMER_ZONE(offset) == HAMMER_ZONE_META)
197 #define hammer_is_zone_large_data(offset) \
198 (HAMMER_ZONE(offset) == HAMMER_ZONE_LARGE_DATA)
199 #define hammer_is_zone_small_data(offset) \
200 (HAMMER_ZONE(offset) == HAMMER_ZONE_SMALL_DATA)
201 #define hammer_is_zone_unavail(offset) \
202 (HAMMER_ZONE(offset) == HAMMER_ZONE_UNAVAIL)
203 #define hammer_is_zone_data(offset) \
204 (hammer_is_zone_large_data(offset) || hammer_is_zone_small_data(offset))
207 * Test if the zone is directly mapped to zone-2 offset via freemap.
209 #define hammer_is_zone2_mapped_index(zone) \
210 ((zone) >= HAMMER_ZONE_BTREE_INDEX && \
211 (zone) < HAMMER_MAX_ZONES)
213 * Test if the zone is directly mapped to zone-2 offset. The word
214 * directly here means the zone is neither RAW_VOLUME nor UNDO zone.
216 #define hammer_is_direct_mapped_index(zone) \
217 (((zone) == HAMMER_ZONE_RAW_BUFFER_INDEX) || \
218 ((zone) == HAMMER_ZONE_FREEMAP_INDEX) || \
219 hammer_is_zone2_mapped_index(zone))
221 #define HAMMER_ZONE_ENCODE(zone, ham_off) \
222 (((hammer_off_t)(zone) << 60) | (ham_off))
223 #define HAMMER_ZONE_DECODE(ham_off) \
224 ((int)(((hammer_off_t)(ham_off) >> 60)))
226 #define HAMMER_VOL_ENCODE(vol_no) \
227 ((hammer_off_t)((vol_no) & 255) << 52)
228 #define HAMMER_VOL_DECODE(ham_off) \
229 ((int)(((hammer_off_t)(ham_off) >> 52) & 255))
231 #define HAMMER_OFF_SHORT_ENCODE(offset) \
232 ((hammer_off_t)(offset) & HAMMER_OFF_SHORT_MASK)
233 #define HAMMER_OFF_LONG_ENCODE(offset) \
234 ((hammer_off_t)(offset) & HAMMER_OFF_LONG_MASK)
236 #define HAMMER_ENCODE(zone, vol_no, offset) \
237 (((hammer_off_t)(zone) << 60) | \
238 HAMMER_VOL_ENCODE(vol_no) | \
239 HAMMER_OFF_SHORT_ENCODE(offset))
240 #define HAMMER_ENCODE_RAW_VOLUME(vol_no, offset) \
241 HAMMER_ENCODE(HAMMER_ZONE_RAW_VOLUME_INDEX, vol_no, offset)
242 #define HAMMER_ENCODE_RAW_BUFFER(vol_no, offset) \
243 HAMMER_ENCODE(HAMMER_ZONE_RAW_BUFFER_INDEX, vol_no, offset)
244 #define HAMMER_ENCODE_FREEMAP(vol_no, offset) \
245 HAMMER_ENCODE(HAMMER_ZONE_FREEMAP_INDEX, vol_no, offset)
248 * Translate a zone address to zone-X address.
250 #define hammer_xlate_to_zoneX(zone, offset) \
251 HAMMER_ZONE_ENCODE((zone), (offset) & ~HAMMER_OFF_ZONE_MASK)
252 #define hammer_xlate_to_zone2(offset) \
253 hammer_xlate_to_zoneX(HAMMER_ZONE_RAW_BUFFER_INDEX, (offset))
255 #define hammer_data_zone(data_len) \
256 (((data_len) >= HAMMER_BUFSIZE) ? \
257 HAMMER_ZONE_LARGE_DATA : \
258 HAMMER_ZONE_SMALL_DATA)
259 #define hammer_data_zone_index(data_len) \
260 (((data_len) >= HAMMER_BUFSIZE) ? \
261 HAMMER_ZONE_LARGE_DATA_INDEX : \
262 HAMMER_ZONE_SMALL_DATA_INDEX)
265 * Big-Block backing store
267 * A blockmap is a two-level map which translates a blockmap-backed zone
268 * offset into a raw zone 2 offset. The layer 1 handles 18 bits and the
269 * layer 2 handles 19 bits. The 8M big-block size is 23 bits so two
270 * layers gives us 18+19+23 = 60 bits of address space.
272 * When using hinting for a blockmap lookup, the hint is lost when the
273 * scan leaves the HINTBLOCK, which is typically several BIGBLOCK's.
274 * HINTBLOCK is a heuristic.
276 #define HAMMER_HINTBLOCK_SIZE (HAMMER_BIGBLOCK_SIZE * 4)
277 #define HAMMER_HINTBLOCK_MASK64 ((uint64_t)HAMMER_HINTBLOCK_SIZE - 1)
278 #define HAMMER_BIGBLOCK_SIZE (8192 * 1024)
279 #define HAMMER_BIGBLOCK_SIZE64 ((uint64_t)HAMMER_BIGBLOCK_SIZE)
280 #define HAMMER_BIGBLOCK_MASK (HAMMER_BIGBLOCK_SIZE - 1)
281 #define HAMMER_BIGBLOCK_MASK64 ((uint64_t)HAMMER_BIGBLOCK_SIZE - 1)
282 #define HAMMER_BIGBLOCK_BITS 23
284 #define HAMMER_BIGBLOCK_OVERFILL (6144 * 1024)
286 #if (1 << HAMMER_BIGBLOCK_BITS) != HAMMER_BIGBLOCK_SIZE
287 #error "HAMMER_BIGBLOCK_BITS BROKEN"
290 #define HAMMER_BUFFERS_PER_BIGBLOCK \
291 (HAMMER_BIGBLOCK_SIZE / HAMMER_BUFSIZE)
292 #define HAMMER_BUFFERS_PER_BIGBLOCK_MASK \
293 (HAMMER_BUFFERS_PER_BIGBLOCK - 1)
294 #define HAMMER_BUFFERS_PER_BIGBLOCK_MASK64 \
295 ((hammer_off_t)HAMMER_BUFFERS_PER_BIGBLOCK_MASK)
298 * Maximum number of mirrors operating in master mode (multi-master
299 * clustering and mirroring). Note that HAMMER1 does not support
300 * multi-master clustering as of 2015.
302 #define HAMMER_MAX_MASTERS 16
305 * The blockmap is somewhat of a degenerate structure. HAMMER only actually
306 * uses it in its original incarnation to implement the freemap.
308 * zone:1 raw volume (no blockmap)
309 * zone:2 raw buffer (no blockmap)
310 * zone:3 undomap (direct layer2 array in volume header)
311 * zone:4 freemap (the only real blockmap)
312 * zone:8-15 zone id used to classify big-block only, address is actually
315 typedef struct hammer_blockmap {
316 hammer_off_t phys_offset; /* zone-2 physical offset */
317 hammer_off_t first_offset; /* zone-X logical offset (zone 3) */
318 hammer_off_t next_offset; /* zone-X logical offset */
319 hammer_off_t alloc_offset; /* zone-X logical offset */
321 hammer_crc_t entry_crc;
322 } *hammer_blockmap_t;
324 #define HAMMER_BLOCKMAP_CRCSIZE \
325 offsetof(struct hammer_blockmap, entry_crc)
328 * The blockmap is a 2-layer entity made up of big-blocks. The first layer
329 * contains 262144 32-byte entries (18 bits), the second layer contains
330 * 524288 16-byte entries (19 bits), representing 8MB (23 bit) blockmaps.
331 * 18+19+23 = 60 bits. The top four bits are the zone id.
333 * Currently only the freemap utilizes both layers in all their glory.
334 * All primary data/meta-data zones actually encode a zone-2 address
335 * requiring no real blockmap translation.
337 * The freemap uses the upper 8 bits of layer-1 to identify the volume,
338 * thus any space allocated via the freemap can be directly translated
339 * to a zone:2 (or zone:8-15) address.
341 * zone-X blockmap offset: [zone:4][layer1:18][layer2:19][big-block:23]
345 * 32 bytes layer1 entry for 8MB big-block.
346 * A big-block can hold 2^23 / 2^5 = 2^18 layer1 entries,
347 * which equals bits assigned for layer1 in zone-2 address.
349 typedef struct hammer_blockmap_layer1 {
350 hammer_off_t blocks_free; /* big-blocks free */
351 hammer_off_t phys_offset; /* UNAVAIL or zone-2 */
352 hammer_off_t reserved01;
353 hammer_crc_t layer2_crc; /* xor'd crc's of HAMMER_BLOCKSIZE */
355 hammer_crc_t layer1_crc; /* MUST BE LAST FIELD OF STRUCTURE*/
356 } *hammer_blockmap_layer1_t;
358 #define HAMMER_LAYER1_CRCSIZE \
359 offsetof(struct hammer_blockmap_layer1, layer1_crc)
362 * 16 bytes layer2 entry for 8MB big-blocks.
363 * A big-block can hold 2^23 / 2^4 = 2^19 layer2 entries,
364 * which equals bits assigned for layer2 in zone-2 address.
366 * NOTE: bytes_free is signed and can legally go negative if/when data
367 * de-dup occurs. This field will never go higher than
368 * HAMMER_BIGBLOCK_SIZE. If exactly HAMMER_BIGBLOCK_SIZE
369 * the big-block is completely free.
371 typedef struct hammer_blockmap_layer2 {
372 uint8_t zone; /* typed allocation zone */
375 uint32_t append_off; /* allocatable space index */
376 int32_t bytes_free; /* bytes free within this big-block */
377 hammer_crc_t entry_crc;
378 } *hammer_blockmap_layer2_t;
380 #define HAMMER_LAYER2_CRCSIZE \
381 offsetof(struct hammer_blockmap_layer2, entry_crc)
383 #define HAMMER_BLOCKMAP_UNAVAIL ((hammer_off_t)-1LL)
385 #define HAMMER_BLOCKMAP_RADIX1 /* 2^18 = 262144 */ \
386 ((int)(HAMMER_BIGBLOCK_SIZE / sizeof(struct hammer_blockmap_layer1)))
387 #define HAMMER_BLOCKMAP_RADIX2 /* 2^19 = 524288 */ \
388 ((int)(HAMMER_BIGBLOCK_SIZE / sizeof(struct hammer_blockmap_layer2)))
390 #define HAMMER_BLOCKMAP_LAYER1 /* 2^(18+19+23) = 1EB */ \
391 (HAMMER_BLOCKMAP_RADIX1 * HAMMER_BLOCKMAP_LAYER2)
392 #define HAMMER_BLOCKMAP_LAYER2 /* 2^(19+23) = 4TB */ \
393 (HAMMER_BLOCKMAP_RADIX2 * HAMMER_BIGBLOCK_SIZE64)
395 #define HAMMER_BLOCKMAP_LAYER1_MASK (HAMMER_BLOCKMAP_LAYER1 - 1)
396 #define HAMMER_BLOCKMAP_LAYER2_MASK (HAMMER_BLOCKMAP_LAYER2 - 1)
399 * Index within layer1 or layer2 big-block for the entry representing
400 * a zone-2 physical offset.
402 #define HAMMER_BLOCKMAP_LAYER1_INDEX(zone2_offset) \
403 ((int)(((zone2_offset) & HAMMER_BLOCKMAP_LAYER1_MASK) / \
404 HAMMER_BLOCKMAP_LAYER2))
406 #define HAMMER_BLOCKMAP_LAYER2_INDEX(zone2_offset) \
407 ((int)(((zone2_offset) & HAMMER_BLOCKMAP_LAYER2_MASK) / \
408 HAMMER_BIGBLOCK_SIZE64))
411 * Byte offset within layer1 or layer2 big-block for the entry representing
412 * a zone-2 physical offset. Multiply the index by sizeof(blockmap_layer).
414 #define HAMMER_BLOCKMAP_LAYER1_OFFSET(zone2_offset) \
415 (HAMMER_BLOCKMAP_LAYER1_INDEX(zone2_offset) * \
416 sizeof(struct hammer_blockmap_layer1))
418 #define HAMMER_BLOCKMAP_LAYER2_OFFSET(zone2_offset) \
419 (HAMMER_BLOCKMAP_LAYER2_INDEX(zone2_offset) * \
420 sizeof(struct hammer_blockmap_layer2))
423 * HAMMER UNDO parameters. The UNDO fifo is mapped directly in the volume
424 * header with an array of layer2 structures. A maximum of (128x8MB) = 1GB
425 * may be reserved. The size of the undo fifo is usually set a newfs time
426 * but can be adjusted if the filesystem is taken offline.
428 #define HAMMER_UNDO_LAYER2 128 /* max layer2 undo mapping entries */
431 * All on-disk HAMMER structures which make up elements of the UNDO FIFO
432 * contain a hammer_fifo_head and hammer_fifo_tail structure. This structure
433 * contains all the information required to validate the fifo element
434 * and to scan the fifo in either direction. The head is typically embedded
435 * in higher level hammer on-disk structures while the tail is typically
436 * out-of-band. hdr_size is the size of the whole mess, including the tail.
438 * All undo structures are guaranteed to not cross a 16K filesystem
439 * buffer boundary. Most undo structures are fairly small. Data spaces
440 * are not immediately reused by HAMMER so file data is not usually recorded
441 * as part of an UNDO.
443 * PAD elements are allowed to take up only 8 bytes of space as a special
444 * case, containing only hdr_signature, hdr_type, and hdr_size fields,
445 * and with the tail overloaded onto the head structure for 8 bytes total.
447 * Every undo record has a sequence number. This number is unrelated to
448 * transaction ids and instead collects the undo transactions associated
449 * with a single atomic operation. A larger transactional operation, such
450 * as a remove(), may consist of several smaller atomic operations
451 * representing raw meta-data operations.
453 * HAMMER VERSION 4 CHANGES
455 * In HAMMER version 4 the undo structure alignment is reduced from 16384
456 * to 512 bytes in order to ensure that each 512 byte sector begins with
457 * a header. The hdr_seq field in the header is a 32 bit sequence number
458 * which allows the recovery code to detect missing sectors
459 * without relying on the 32-bit crc and to definitively identify the current
460 * undo sequence space without having to rely on information from the volume
461 * header. In addition, new REDO entries in the undo space are used to
462 * record write, write/extend, and transaction id updates.
464 * The grand result is:
466 * (1) The volume header no longer needs to be synchronized for most
467 * flush and fsync operations.
469 * (2) Most fsync operations need only lay down REDO records
471 * (3) Data overwrite for nohistory operations covered by REDO records
472 * can be supported (instead of rolling a new block allocation),
473 * by rolling UNDO for the prior contents of the data.
475 * HAMMER VERSION 5 CHANGES
477 * Hammer version 5 contains a minor adjustment making layer2's bytes_free
478 * field signed, allowing dedup to push it into the negative domain.
480 #define HAMMER_HEAD_ALIGN 8
481 #define HAMMER_HEAD_ALIGN_MASK (HAMMER_HEAD_ALIGN - 1)
482 #define HAMMER_HEAD_DOALIGN(bytes) \
483 (((bytes) + HAMMER_HEAD_ALIGN_MASK) & ~HAMMER_HEAD_ALIGN_MASK)
485 #define HAMMER_UNDO_ALIGN 512
486 #define HAMMER_UNDO_ALIGN64 ((uint64_t)512)
487 #define HAMMER_UNDO_MASK (HAMMER_UNDO_ALIGN - 1)
488 #define HAMMER_UNDO_MASK64 (HAMMER_UNDO_ALIGN64 - 1)
490 typedef struct hammer_fifo_head {
491 uint16_t hdr_signature;
493 uint32_t hdr_size; /* Aligned size of the whole mess */
494 uint32_t hdr_seq; /* Sequence number */
495 hammer_crc_t hdr_crc; /* XOR crc up to field w/ crc after field */
496 } *hammer_fifo_head_t;
498 #define HAMMER_FIFO_HEAD_CRCOFF offsetof(struct hammer_fifo_head, hdr_crc)
500 typedef struct hammer_fifo_tail {
501 uint16_t tail_signature;
503 uint32_t tail_size; /* aligned size of the whole mess */
504 } *hammer_fifo_tail_t;
509 * NOTE: 0x8000U part of HAMMER_HEAD_TYPE_PAD can be removed if the HAMMER
510 * version ever gets bumped again. It exists only to keep compatibility with
513 #define HAMMER_HEAD_TYPE_PAD (0x0040U | 0x8000U)
514 #define HAMMER_HEAD_TYPE_DUMMY 0x0041U /* dummy entry w/seqno */
515 #define HAMMER_HEAD_TYPE_UNDO 0x0043U /* random UNDO information */
516 #define HAMMER_HEAD_TYPE_REDO 0x0044U /* data REDO / fast fsync */
518 #define HAMMER_HEAD_SIGNATURE 0xC84EU
519 #define HAMMER_TAIL_SIGNATURE 0xC74FU
522 * Misc FIFO structures.
524 * UNDO - Raw meta-data media updates.
526 typedef struct hammer_fifo_undo {
527 struct hammer_fifo_head head;
528 hammer_off_t undo_offset; /* zone-1,2 offset */
529 int32_t undo_data_bytes;
530 int32_t undo_reserved01;
531 /* followed by data */
532 } *hammer_fifo_undo_t;
535 * REDO (HAMMER version 4+) - Logical file writes/truncates.
537 * REDOs contain information which will be duplicated in a later meta-data
538 * update, allowing fast write()+fsync() operations. REDOs can be ignored
539 * without harming filesystem integrity but must be processed if fsync()
540 * semantics are desired.
542 * Unlike UNDOs which are processed backwards within the recovery span,
543 * REDOs must be processed forwards starting further back (starting outside
544 * the recovery span).
546 * WRITE - Write logical file (with payload). Executed both
547 * out-of-span and in-span. Out-of-span WRITEs may be
548 * filtered out by TERMs.
550 * TRUNC - Truncate logical file (no payload). Executed both
551 * out-of-span and in-span. Out-of-span WRITEs may be
552 * filtered out by TERMs.
554 * TERM_* - Indicates meta-data was committed (if out-of-span) or
555 * will be rolled-back (in-span). Any out-of-span TERMs
556 * matching earlier WRITEs remove those WRITEs from
557 * consideration as they might conflict with a later data
558 * commit (which is not being rolled-back).
560 * SYNC - The earliest in-span SYNC (the last one when scanning
561 * backwards) tells the recovery code how far out-of-span
562 * it must go to run REDOs.
564 * NOTE: WRITEs do not always have matching TERMs even under
565 * perfect conditions because truncations might remove the
566 * buffers from consideration. I/O problems can also remove
567 * buffers from consideration.
569 * TRUNCSs do not always have matching TERMs because several
570 * truncations may be aggregated together into a single TERM.
572 typedef struct hammer_fifo_redo {
573 struct hammer_fifo_head head;
574 int64_t redo_objid; /* file being written */
575 hammer_off_t redo_offset; /* logical offset in file */
576 int32_t redo_data_bytes;
578 uint32_t redo_localization;
579 uint32_t redo_reserved;
580 uint64_t redo_mtime; /* set mtime */
581 } *hammer_fifo_redo_t;
583 #define HAMMER_REDO_WRITE 0x00000001
584 #define HAMMER_REDO_TRUNC 0x00000002
585 #define HAMMER_REDO_TERM_WRITE 0x00000004
586 #define HAMMER_REDO_TERM_TRUNC 0x00000008
587 #define HAMMER_REDO_SYNC 0x00000010
589 typedef union hammer_fifo_any {
590 struct hammer_fifo_head head;
591 struct hammer_fifo_undo undo;
592 struct hammer_fifo_redo redo;
593 } *hammer_fifo_any_t;
596 * Volume header types
598 #define HAMMER_FSBUF_VOLUME 0xC8414D4DC5523031ULL /* HAMMER01 */
599 #define HAMMER_FSBUF_VOLUME_REV 0x313052C54D4D41C8ULL /* (reverse endian) */
602 * HAMMER Volume header
604 * A HAMMER filesystem can be built from 1-256 block devices, each block
605 * device contains a volume header followed by however many buffers fit
608 * One of the volumes making up a HAMMER filesystem is the root volume.
609 * The root volume is always volume #0 which is the first block device path
610 * specified by newfs_hammer(8). All HAMMER volumes have a volume header,
611 * however the root volume may be the only volume that has valid values for
612 * some fields in the header.
614 * Special field notes:
616 * vol_bot_beg - offset of boot area (mem_beg - bot_beg bytes)
617 * vol_mem_beg - offset of memory log (buf_beg - mem_beg bytes)
618 * vol_buf_beg - offset of the first buffer in volume
619 * vol_buf_end - offset of volume EOF (on buffer boundary)
621 * The memory log area allows a kernel to cache new records and data
622 * in memory without allocating space in the actual filesystem to hold
623 * the records and data. In the event that a filesystem becomes full,
624 * any records remaining in memory can be flushed to the memory log
625 * area. This allows the kernel to immediately return success.
627 * The buffer offset is a physical offset of zone-2 offset. The lower
628 * 52 bits of the zone-2 offset is added to the buffer offset of each
629 * volume to generate an actual I/O offset within the block device.
631 * NOTE: boot area and memory log are currently not used.
635 * These macros are only used by userspace when userspace commands either
636 * initialize or add a new HAMMER volume.
638 #define HAMMER_BOOT_MINBYTES (32*1024)
639 #define HAMMER_BOOT_NOMBYTES (64LL*1024*1024)
640 #define HAMMER_BOOT_MAXBYTES (256LL*1024*1024)
642 #define HAMMER_MEM_MINBYTES (256*1024)
643 #define HAMMER_MEM_NOMBYTES (1LL*1024*1024*1024)
644 #define HAMMER_MEM_MAXBYTES (64LL*1024*1024*1024)
646 typedef struct hammer_volume_ondisk {
647 uint64_t vol_signature; /* HAMMER_FSBUF_VOLUME for a valid header */
650 * These are relative to block device offset, not zone offsets.
652 int64_t vol_bot_beg; /* offset of boot area */
653 int64_t vol_mem_beg; /* offset of memory log */
654 int64_t vol_buf_beg; /* offset of the first buffer in volume */
655 int64_t vol_buf_end; /* offset of volume EOF (on buffer boundary) */
656 int64_t vol_reserved01;
658 uuid_t vol_fsid; /* identify filesystem */
659 uuid_t vol_fstype; /* identify filesystem type */
660 char vol_label[64]; /* filesystem label */
662 int32_t vol_no; /* volume number within filesystem */
663 int32_t vol_count; /* number of volumes making up filesystem */
665 uint32_t vol_version; /* version control information */
666 hammer_crc_t vol_crc; /* header crc */
667 uint32_t vol_flags; /* volume flags */
668 uint32_t vol_rootvol; /* the root volume number (must be 0) */
670 uint32_t vol_reserved[8];
673 * These fields are initialized and space is reserved in every
674 * volume making up a HAMMER filesytem, but only the root volume
675 * contains valid data. Note that vol0_stat_bigblocks does not
676 * include big-blocks for freemap and undomap initially allocated
677 * by newfs_hammer(8).
679 int64_t vol0_stat_bigblocks; /* total big-blocks when fs is empty */
680 int64_t vol0_stat_freebigblocks;/* number of free big-blocks */
681 int64_t vol0_reserved01;
682 int64_t vol0_stat_inodes; /* for statfs only */
683 int64_t vol0_reserved02;
684 hammer_off_t vol0_btree_root; /* B-Tree root offset in zone-8 */
685 hammer_tid_t vol0_next_tid; /* highest partially synchronized TID */
686 hammer_off_t vol0_reserved03;
689 * Blockmaps for zones. Not all zones use a blockmap. Note that
690 * the entire root blockmap is cached in the hammer_mount structure.
692 struct hammer_blockmap vol0_blockmap[HAMMER_MAX_ZONES];
695 * Array of zone-2 addresses for undo FIFO.
697 hammer_off_t vol0_undo_array[HAMMER_UNDO_LAYER2];
698 } *hammer_volume_ondisk_t;
700 #define HAMMER_ROOT_VOLNO 0
702 #define HAMMER_VOLF_NEEDFLUSH 0x0004 /* volume needs flush */
704 #define HAMMER_VOL_CRCSIZE1 \
705 offsetof(struct hammer_volume_ondisk, vol_crc)
706 #define HAMMER_VOL_CRCSIZE2 \
707 (sizeof(struct hammer_volume_ondisk) - HAMMER_VOL_CRCSIZE1 - \
708 sizeof(hammer_crc_t))
710 #define HAMMER_VOL_VERSION_MIN 1 /* minimum supported version */
711 #define HAMMER_VOL_VERSION_DEFAULT 6 /* newfs default version */
712 #define HAMMER_VOL_VERSION_WIP 7 /* version >= this is WIP */
713 #define HAMMER_VOL_VERSION_MAX 6 /* maximum supported version */
715 #define HAMMER_VOL_VERSION_ONE 1
716 #define HAMMER_VOL_VERSION_TWO 2 /* new dirent layout (2.3+) */
717 #define HAMMER_VOL_VERSION_THREE 3 /* new snapshot layout (2.5+) */
718 #define HAMMER_VOL_VERSION_FOUR 4 /* new undo/flush (2.5+) */
719 #define HAMMER_VOL_VERSION_FIVE 5 /* dedup (2.9+) */
720 #define HAMMER_VOL_VERSION_SIX 6 /* DIRHASH_ALG1 */
723 * Translate a zone-2 address to physical address
725 #define hammer_xlate_to_phys(volume, zone2_offset) \
726 ((volume)->vol_buf_beg + \
727 ((zone2_offset) & HAMMER_OFF_SHORT_MASK))
730 * Effective per-volume filesystem capacity including big-blocks for layer1/2
732 #define HAMMER_VOL_BUF_SIZE(volume) \
733 ((volume)->vol_buf_end - (volume)->vol_buf_beg)
736 * Record types are fairly straightforward. The B-Tree includes the record
737 * type in its index sort.
739 #define HAMMER_RECTYPE_UNKNOWN 0x0000
740 #define HAMMER_RECTYPE_INODE 0x0001 /* inode in obj_id space */
741 #define HAMMER_RECTYPE_DATA 0x0010
742 #define HAMMER_RECTYPE_DIRENTRY 0x0011
743 #define HAMMER_RECTYPE_DB 0x0012
744 #define HAMMER_RECTYPE_EXT 0x0013 /* ext attributes */
745 #define HAMMER_RECTYPE_FIX 0x0014 /* fixed attribute */
746 #define HAMMER_RECTYPE_PFS 0x0015 /* PFS management */
747 #define HAMMER_RECTYPE_SNAPSHOT 0x0016 /* Snapshot management */
748 #define HAMMER_RECTYPE_CONFIG 0x0017 /* hammer cleanup config */
749 #define HAMMER_RECTYPE_MAX 0xFFFF
751 #define HAMMER_RECTYPE_ENTRY_START (HAMMER_RECTYPE_INODE + 1)
752 #define HAMMER_RECTYPE_CLEAN_START HAMMER_RECTYPE_EXT
754 #define HAMMER_FIXKEY_SYMLINK 1
756 #define HAMMER_OBJTYPE_UNKNOWN 0 /* never exists on-disk as unknown */
757 #define HAMMER_OBJTYPE_DIRECTORY 1
758 #define HAMMER_OBJTYPE_REGFILE 2
759 #define HAMMER_OBJTYPE_DBFILE 3
760 #define HAMMER_OBJTYPE_FIFO 4
761 #define HAMMER_OBJTYPE_CDEV 5
762 #define HAMMER_OBJTYPE_BDEV 6
763 #define HAMMER_OBJTYPE_SOFTLINK 7
764 #define HAMMER_OBJTYPE_PSEUDOFS 8 /* pseudo filesystem obj */
765 #define HAMMER_OBJTYPE_SOCKET 9
768 * HAMMER inode attribute data
770 * The data reference for a HAMMER inode points to this structure. Any
771 * modifications to the contents of this structure will result in a
772 * replacement operation.
774 * parent_obj_id is only valid for directories (which cannot be hard-linked),
775 * and specifies the parent directory obj_id. This field will also be set
776 * for non-directory inodes as a recovery aid, but can wind up holding
777 * stale information. However, since object id's are not reused, the worse
778 * that happens is that the recovery code is unable to use it.
779 * A parent_obj_id of 0 means it's a root inode of root or non-root PFS.
781 * NOTE: Future note on directory hardlinks. We can implement a record type
782 * which allows us to point to multiple parent directories.
784 typedef struct hammer_inode_data {
785 uint16_t version; /* inode data version */
786 uint16_t mode; /* basic unix permissions */
787 uint32_t uflags; /* chflags */
788 uint32_t rmajor; /* used by device nodes */
789 uint32_t rminor; /* used by device nodes */
791 int64_t parent_obj_id; /* parent directory obj_id */
796 uint8_t cap_flags; /* capability support flags (extension) */
799 uint64_t nlinks; /* hard links */
800 uint64_t size; /* filesystem object size */
802 char symlink[24]; /* HAMMER_INODE_BASESYMLEN */
804 uint64_t mtime; /* mtime must be second-to-last */
805 uint64_t atime; /* atime must be last */
806 } *hammer_inode_data_t;
809 * Neither mtime nor atime upates are CRCd by the B-Tree element.
810 * mtime updates have UNDO, atime updates do not.
812 #define HAMMER_INODE_CRCSIZE \
813 offsetof(struct hammer_inode_data, mtime)
815 #define HAMMER_INODE_DATA_VERSION 1
816 #define HAMMER_OBJID_ROOT 1 /* root inodes # */
817 #define HAMMER_INODE_BASESYMLEN 24 /* see ext.symlink */
820 * Capability & implementation flags.
822 * HAMMER_INODE_CAP_DIR_LOCAL_INO - Use inode B-Tree localization
823 * for directory entries. Also see HAMMER_DIR_INODE_LOCALIZATION().
825 #define HAMMER_INODE_CAP_DIRHASH_MASK 0x03 /* directory: hash algorithm */
826 #define HAMMER_INODE_CAP_DIRHASH_ALG0 0x00
827 #define HAMMER_INODE_CAP_DIRHASH_ALG1 0x01
828 #define HAMMER_INODE_CAP_DIRHASH_ALG2 0x02
829 #define HAMMER_INODE_CAP_DIRHASH_ALG3 0x03
830 #define HAMMER_INODE_CAP_DIR_LOCAL_INO 0x04 /* use inode localization */
833 * A HAMMER directory entry associates a HAMMER filesystem object with a
834 * namespace. It is hooked into a pseudo-filesystem (with its own inode
835 * numbering space) in the filesystem by setting the high 16 bits of the
836 * localization field. The low 16 bits must be 0 and are reserved for
839 * Directory entries are indexed with a 128 bit namekey rather then an
840 * offset. A portion of the namekey is an iterator/randomizer to deal
843 * NOTE: leaf.base.obj_type from the related B-Tree leaf entry holds
844 * the filesystem object type of obj_id, e.g. a den_type equivalent.
845 * It is not stored in hammer_direntry_data.
847 * NOTE: name field / the filename data reference is NOT terminated with \0.
849 typedef struct hammer_direntry_data {
850 int64_t obj_id; /* object being referenced */
851 uint32_t localization; /* identify pseudo-filesystem */
853 char name[16]; /* name (extended) */
854 } *hammer_direntry_data_t;
856 #define HAMMER_ENTRY_NAME_OFF offsetof(struct hammer_direntry_data, name[0])
857 #define HAMMER_ENTRY_SIZE(nlen) offsetof(struct hammer_direntry_data, name[nlen])
860 * Symlink data which does not fit in the inode is stored in a separate
863 typedef struct hammer_symlink_data {
864 char name[16]; /* name (extended) */
865 } *hammer_symlink_data_t;
867 #define HAMMER_SYMLINK_NAME_OFF offsetof(struct hammer_symlink_data, name[0])
870 * The root inode for the primary filesystem and root inode for any
871 * pseudo-fs may be tagged with an optional data structure using
872 * HAMMER_RECTYPE_PFS and localization id. This structure allows
873 * the node to be used as a mirroring master or slave.
875 * When operating as a slave CD's into the node automatically become read-only
876 * and as-of sync_end_tid.
878 * When operating as a master the read PFSD info sets sync_end_tid to
879 * the most recently flushed TID.
881 * sync_low_tid is not yet used but will represent the highest pruning
882 * end-point, after which full history is available.
884 * We need to pack this structure making it equally sized on both 32-bit and
885 * 64-bit machines as it is part of struct hammer_ioc_mrecord_pfs which is
886 * send over the wire in hammer mirror operations. Only on 64-bit machines
887 * the size of this struct differ when packed or not. This leads us to the
888 * situation where old 64-bit systems (using the non-packed structure),
889 * which were never able to mirror to/from 32-bit systems, are now no longer
890 * able to mirror to/from newer 64-bit systems (using the packed structure).
892 struct hammer_pseudofs_data {
893 hammer_tid_t sync_low_tid; /* full history beyond this point */
894 hammer_tid_t sync_beg_tid; /* earliest tid w/ full history avail */
895 hammer_tid_t sync_end_tid; /* current synchronizatoin point */
896 uint64_t sync_beg_ts; /* real-time of last completed sync */
897 uint64_t sync_end_ts; /* initiation of current sync cycle */
898 uuid_t shared_uuid; /* shared uuid (match required) */
899 uuid_t unique_uuid; /* unique uuid of this master/slave */
900 int32_t reserved01; /* reserved for future master_id */
901 int32_t mirror_flags; /* misc flags */
902 char label[64]; /* filesystem space label */
903 char snapshots[64]; /* softlink dir for pruning */
904 int32_t reserved02; /* was prune_{time,freq} */
905 int32_t reserved03; /* was reblock_{time,freq} */
906 int32_t reserved04; /* was snapshot_freq */
907 int32_t prune_min; /* do not prune recent history */
908 int32_t prune_max; /* do not retain history beyond here */
909 int32_t reserved[16];
912 typedef struct hammer_pseudofs_data *hammer_pseudofs_data_t;
914 #define HAMMER_PFSD_SLAVE 0x00000001
915 #define HAMMER_PFSD_DELETED 0x80000000
917 #define hammer_is_pfs_slave(pfsd) \
918 (((pfsd)->mirror_flags & HAMMER_PFSD_SLAVE) != 0)
919 #define hammer_is_pfs_master(pfsd) \
920 (!hammer_is_pfs_slave(pfsd))
921 #define hammer_is_pfs_deleted(pfsd) \
922 (((pfsd)->mirror_flags & HAMMER_PFSD_DELETED) != 0)
924 #define HAMMER_MAX_PFS 65536
925 #define HAMMER_MAX_PFSID (HAMMER_MAX_PFS - 1)
926 #define HAMMER_ROOT_PFSID 0
929 * Snapshot meta-data { Objid = HAMMER_OBJID_ROOT, Key = tid, rectype = SNAPSHOT }.
931 * Snapshot records replace the old <fs>/snapshots/<softlink> methodology. Snapshot
932 * records are mirrored but may be independantly managed once they are laid down on
935 * NOTE: The b-tree key is signed, the tid is not, so callers must still sort the
938 * NOTE: Reserved fields must be zero (as usual)
940 typedef struct hammer_snapshot_data {
941 hammer_tid_t tid; /* the snapshot TID itself (== key) */
942 uint64_t ts; /* real-time when snapshot was made */
945 char label[64]; /* user-supplied description */
946 uint64_t reserved03[4];
947 } *hammer_snapshot_data_t;
950 * Config meta-data { ObjId = HAMMER_OBJID_ROOT, Key = 0, rectype = CONFIG }.
952 * Used to store the hammer cleanup config. This data is not mirrored.
954 typedef struct hammer_config_data {
956 } *hammer_config_data_t;
959 * Rollup various structures embedded as record data
961 typedef union hammer_data_ondisk {
962 struct hammer_direntry_data entry;
963 struct hammer_inode_data inode;
964 struct hammer_symlink_data symlink;
965 struct hammer_pseudofs_data pfsd;
966 struct hammer_snapshot_data snap;
967 struct hammer_config_data config;
968 } *hammer_data_ondisk_t;
971 * Ondisk layout of B-Tree related structures
973 #include "hammer_btree.h"
975 #define HAMMER_DIR_INODE_LOCALIZATION(ino_data) \
976 (((ino_data)->cap_flags & HAMMER_INODE_CAP_DIR_LOCAL_INO) ? \
977 HAMMER_LOCALIZE_INODE : \
978 HAMMER_LOCALIZE_MISC)
980 #endif /* !VFS_HAMMER_DISK_H_ */