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 * Hammer transaction ids are 64 bit unsigned integers and are usually
94 * synchronized with the time of day in nanoseconds.
96 * Hammer offsets are used for FIFO indexing and embed a cycle counter
97 * and volume number in addition to the offset. Most offsets are required
98 * to be 16 KB aligned.
100 typedef uint64_t hammer_tid_t;
101 typedef uint64_t hammer_off_t;
102 typedef uint32_t hammer_crc_t;
104 #define HAMMER_MIN_TID 0ULL /* unsigned */
105 #define HAMMER_MAX_TID 0xFFFFFFFFFFFFFFFFULL /* unsigned */
106 #define HAMMER_MIN_KEY -0x8000000000000000LL /* signed */
107 #define HAMMER_MAX_KEY 0x7FFFFFFFFFFFFFFFLL /* signed */
108 #define HAMMER_MIN_OBJID HAMMER_MIN_KEY /* signed */
109 #define HAMMER_MAX_OBJID HAMMER_MAX_KEY /* signed */
110 #define HAMMER_MIN_RECTYPE 0x0U /* unsigned */
111 #define HAMMER_MAX_RECTYPE 0xFFFFU /* unsigned */
112 #define HAMMER_MIN_OFFSET 0ULL /* unsigned */
113 #define HAMMER_MAX_OFFSET 0xFFFFFFFFFFFFFFFFULL /* unsigned */
116 * hammer_off_t has several different encodings. Note that not all zones
119 * zone 0: reserved for sanity
120 * zone 1 (z,v,o): raw volume relative (offset 0 is the volume header)
121 * zone 2 (z,v,o): raw buffer relative (offset 0 is the first buffer)
122 * zone 3 (z,o): undo fifo - actually zone-2 address, fixed phys array in vol hdr
123 * zone 4 (z,v,o): freemap - only real blockmap
124 * zone 8 (z,v,o): B-Tree - actually zone-2 address
125 * zone 9 (z,v,o): meta - actually zone-2 address
126 * zone 10 (z,v,o): large-data - actually zone-2 address
127 * zone 11 (z,v,o): small-data - actually zone-2 address
128 * zone 15: reserved for sanity
130 * layer1/layer2 direct map:
131 * zzzzvvvvvvvvoooo oooooooooooooooo oooooooooooooooo oooooooooooooooo
132 * ----111111111111 1111112222222222 222222222ooooooo oooooooooooooooo
135 #define HAMMER_ZONE_RAW_VOLUME 0x1000000000000000ULL
136 #define HAMMER_ZONE_RAW_BUFFER 0x2000000000000000ULL
137 #define HAMMER_ZONE_UNDO 0x3000000000000000ULL
138 #define HAMMER_ZONE_FREEMAP 0x4000000000000000ULL
139 #define HAMMER_ZONE_RESERVED05 0x5000000000000000ULL
140 #define HAMMER_ZONE_RESERVED06 0x6000000000000000ULL
141 #define HAMMER_ZONE_RESERVED07 0x7000000000000000ULL
142 #define HAMMER_ZONE_BTREE 0x8000000000000000ULL
143 #define HAMMER_ZONE_META 0x9000000000000000ULL
144 #define HAMMER_ZONE_LARGE_DATA 0xA000000000000000ULL
145 #define HAMMER_ZONE_SMALL_DATA 0xB000000000000000ULL
146 #define HAMMER_ZONE_RESERVED0C 0xC000000000000000ULL
147 #define HAMMER_ZONE_RESERVED0D 0xD000000000000000ULL
148 #define HAMMER_ZONE_RESERVED0E 0xE000000000000000ULL
149 #define HAMMER_ZONE_UNAVAIL 0xF000000000000000ULL
151 #define HAMMER_ZONE_RAW_VOLUME_INDEX 1
152 #define HAMMER_ZONE_RAW_BUFFER_INDEX 2
153 #define HAMMER_ZONE_UNDO_INDEX 3
154 #define HAMMER_ZONE_FREEMAP_INDEX 4
155 #define HAMMER_ZONE_BTREE_INDEX 8
156 #define HAMMER_ZONE_META_INDEX 9
157 #define HAMMER_ZONE_LARGE_DATA_INDEX 10
158 #define HAMMER_ZONE_SMALL_DATA_INDEX 11
159 #define HAMMER_ZONE_UNAVAIL_INDEX 15 /* unavailable */
161 #define HAMMER_MAX_ZONES 16
164 * Backend zones that are mapped to zone-2 (except for zone-3)
165 * starts from this index which is 8.
167 #define HAMMER_ZONE2_MAPPED_INDEX HAMMER_ZONE_BTREE_INDEX
169 #define HAMMER_ZONE_ENCODE(zone, ham_off) \
170 (((hammer_off_t)(zone) << 60) | (ham_off))
171 #define HAMMER_ZONE_DECODE(ham_off) \
172 (int32_t)(((hammer_off_t)(ham_off) >> 60))
174 #define HAMMER_VOL_ENCODE(vol_no) \
175 ((hammer_off_t)((vol_no) & 255) << 52)
176 #define HAMMER_VOL_DECODE(ham_off) \
177 (int32_t)(((hammer_off_t)(ham_off) >> 52) & 255)
179 #define HAMMER_OFF_SHORT_ENCODE(offset) \
180 ((hammer_off_t)(offset) & HAMMER_OFF_SHORT_MASK)
181 #define HAMMER_OFF_LONG_ENCODE(offset) \
182 ((hammer_off_t)(offset) & HAMMER_OFF_LONG_MASK)
184 #define HAMMER_ENCODE(zone, vol_no, offset) \
185 (((hammer_off_t)(zone) << 60) | \
186 HAMMER_VOL_ENCODE(vol_no) | \
187 HAMMER_OFF_SHORT_ENCODE(offset))
188 #define HAMMER_ENCODE_RAW_VOLUME(vol_no, offset) \
189 HAMMER_ENCODE(HAMMER_ZONE_RAW_VOLUME_INDEX, vol_no, offset)
190 #define HAMMER_ENCODE_RAW_BUFFER(vol_no, offset) \
191 HAMMER_ENCODE(HAMMER_ZONE_RAW_BUFFER_INDEX, vol_no, offset)
192 #define HAMMER_ENCODE_FREEMAP(vol_no, offset) \
193 HAMMER_ENCODE(HAMMER_ZONE_FREEMAP_INDEX, vol_no, offset)
196 * Translate a zone address to zone-X address.
198 #define hammer_xlate_to_zoneX(zone, offset) \
199 HAMMER_ZONE_ENCODE((zone), (offset) & ~HAMMER_OFF_ZONE_MASK)
200 #define hammer_xlate_to_zone2(offset) \
201 hammer_xlate_to_zoneX(HAMMER_ZONE_RAW_BUFFER_INDEX, (offset))
203 #define hammer_data_zone(data_len) \
204 (((data_len) >= HAMMER_BUFSIZE) ? \
205 HAMMER_ZONE_LARGE_DATA : \
206 HAMMER_ZONE_SMALL_DATA)
207 #define hammer_data_zone_index(data_len) \
208 (((data_len) >= HAMMER_BUFSIZE) ? \
209 HAMMER_ZONE_LARGE_DATA_INDEX : \
210 HAMMER_ZONE_SMALL_DATA_INDEX)
213 * Big-Block backing store
215 * A blockmap is a two-level map which translates a blockmap-backed zone
216 * offset into a raw zone 2 offset. The layer 1 handles 18 bits and the
217 * layer 2 handles 19 bits. The 8M big-block size is 23 bits so two
218 * layers gives us 18+19+23 = 60 bits of address space.
220 * When using hinting for a blockmap lookup, the hint is lost when the
221 * scan leaves the HINTBLOCK, which is typically several BIGBLOCK's.
222 * HINTBLOCK is a heuristic.
224 #define HAMMER_HINTBLOCK_SIZE (HAMMER_BIGBLOCK_SIZE * 4)
225 #define HAMMER_HINTBLOCK_MASK64 ((uint64_t)HAMMER_HINTBLOCK_SIZE - 1)
226 #define HAMMER_BIGBLOCK_SIZE (8192 * 1024)
227 #define HAMMER_BIGBLOCK_SIZE64 ((uint64_t)HAMMER_BIGBLOCK_SIZE)
228 #define HAMMER_BIGBLOCK_MASK (HAMMER_BIGBLOCK_SIZE - 1)
229 #define HAMMER_BIGBLOCK_MASK64 ((uint64_t)HAMMER_BIGBLOCK_SIZE - 1)
230 #define HAMMER_BIGBLOCK_BITS 23
232 #define HAMMER_BIGBLOCK_OVERFILL (6144 * 1024)
234 #if (1 << HAMMER_BIGBLOCK_BITS) != HAMMER_BIGBLOCK_SIZE
235 #error "HAMMER_BIGBLOCK_BITS BROKEN"
238 #define HAMMER_BUFFERS_PER_BIGBLOCK \
239 (HAMMER_BIGBLOCK_SIZE / HAMMER_BUFSIZE)
240 #define HAMMER_BUFFERS_PER_BIGBLOCK_MASK \
241 (HAMMER_BUFFERS_PER_BIGBLOCK - 1)
242 #define HAMMER_BUFFERS_PER_BIGBLOCK_MASK64 \
243 ((hammer_off_t)HAMMER_BUFFERS_PER_BIGBLOCK_MASK)
246 * Maximum number of mirrors operating in master mode (multi-master
247 * clustering and mirroring). Note that HAMMER1 does not support
248 * multi-master clustering as of 2015.
250 #define HAMMER_MAX_MASTERS 16
253 * The blockmap is somewhat of a degenerate structure. HAMMER only actually
254 * uses it in its original incarnation to implement the freemap.
256 * zone:1 raw volume (no blockmap)
257 * zone:2 raw buffer (no blockmap)
258 * zone:3 undomap (direct layer2 array in volume header)
259 * zone:4 freemap (the only real blockmap)
260 * zone:8-15 zone id used to classify big-block only, address is actually
263 struct hammer_blockmap {
264 hammer_off_t phys_offset; /* zone-2 physical offset */
265 hammer_off_t first_offset; /* zone-X logical offset (zone 3) */
266 hammer_off_t next_offset; /* zone-X logical offset */
267 hammer_off_t alloc_offset; /* zone-X logical offset */
269 hammer_crc_t entry_crc;
272 typedef struct hammer_blockmap *hammer_blockmap_t;
274 #define HAMMER_BLOCKMAP_CRCSIZE \
275 offsetof(struct hammer_blockmap, entry_crc)
278 * The blockmap is a 2-layer entity made up of big-blocks. The first layer
279 * contains 262144 32-byte entries (18 bits), the second layer contains
280 * 524288 16-byte entries (19 bits), representing 8MB (23 bit) blockmaps.
281 * 18+19+23 = 60 bits. The top four bits are the zone id.
283 * Currently only the freemap utilizes both layers in all their glory.
284 * All primary data/meta-data zones actually encode a zone-2 address
285 * requiring no real blockmap translation.
287 * The freemap uses the upper 8 bits of layer-1 to identify the volume,
288 * thus any space allocated via the freemap can be directly translated
289 * to a zone:2 (or zone:8-15) address.
291 * zone-X blockmap offset: [zone:4][layer1:18][layer2:19][big-block:23]
293 struct hammer_blockmap_layer1 {
294 hammer_off_t blocks_free; /* big-blocks free */
295 hammer_off_t phys_offset; /* UNAVAIL or zone-2 */
296 hammer_off_t reserved01;
297 hammer_crc_t layer2_crc; /* xor'd crc's of HAMMER_BLOCKSIZE */
299 hammer_crc_t layer1_crc; /* MUST BE LAST FIELD OF STRUCTURE*/
302 typedef struct hammer_blockmap_layer1 *hammer_blockmap_layer1_t;
304 #define HAMMER_LAYER1_CRCSIZE \
305 offsetof(struct hammer_blockmap_layer1, layer1_crc)
308 * layer2 entry for 8MB big-block.
310 * NOTE: bytes_free is signed and can legally go negative if/when data
311 * de-dup occurs. This field will never go higher than
312 * HAMMER_BIGBLOCK_SIZE. If exactly HAMMER_BIGBLOCK_SIZE
313 * the big-block is completely free.
315 struct hammer_blockmap_layer2 {
316 uint8_t zone; /* typed allocation zone */
319 uint32_t append_off; /* allocatable space index */
320 int32_t bytes_free; /* bytes free within this big-block */
321 hammer_crc_t entry_crc;
324 typedef struct hammer_blockmap_layer2 *hammer_blockmap_layer2_t;
326 #define HAMMER_LAYER2_CRCSIZE \
327 offsetof(struct hammer_blockmap_layer2, entry_crc)
329 #define HAMMER_BLOCKMAP_UNAVAIL ((hammer_off_t)-1LL)
331 #define HAMMER_BLOCKMAP_RADIX1 /* 262144 (18) */ \
332 (HAMMER_BIGBLOCK_SIZE / sizeof(struct hammer_blockmap_layer1))
333 #define HAMMER_BLOCKMAP_RADIX2 /* 524288 (19) */ \
334 (HAMMER_BIGBLOCK_SIZE / sizeof(struct hammer_blockmap_layer2))
336 #define HAMMER_BLOCKMAP_RADIX1_PERBUFFER \
337 (HAMMER_BLOCKMAP_RADIX1 / HAMMER_BUFFERS_PER_BIGBLOCK)
338 #define HAMMER_BLOCKMAP_RADIX2_PERBUFFER \
339 (HAMMER_BLOCKMAP_RADIX2 / HAMMER_BUFFERS_PER_BIGBLOCK)
341 #define HAMMER_BLOCKMAP_LAYER1 /* 18+19+23 - 1EB */ \
342 (HAMMER_BLOCKMAP_RADIX1 * HAMMER_BLOCKMAP_LAYER2)
343 #define HAMMER_BLOCKMAP_LAYER2 /* 19+23 - 4TB */ \
344 (HAMMER_BLOCKMAP_RADIX2 * HAMMER_BIGBLOCK_SIZE64)
346 #define HAMMER_BLOCKMAP_LAYER1_MASK (HAMMER_BLOCKMAP_LAYER1 - 1)
347 #define HAMMER_BLOCKMAP_LAYER2_MASK (HAMMER_BLOCKMAP_LAYER2 - 1)
350 * Index within layer1 or layer2 big-block for the entry representing
351 * a zone-2 physical offset.
353 #define HAMMER_BLOCKMAP_LAYER1_INDEX(zone2_offset) \
354 (((zone2_offset) & HAMMER_BLOCKMAP_LAYER1_MASK) / \
355 HAMMER_BLOCKMAP_LAYER2)
357 #define HAMMER_BLOCKMAP_LAYER2_INDEX(zone2_offset) \
358 (((zone2_offset) & HAMMER_BLOCKMAP_LAYER2_MASK) / \
359 HAMMER_BIGBLOCK_SIZE64)
362 * Byte offset within layer1 or layer2 big-block for the entry representing
363 * a zone-2 physical offset. Multiply the index by sizeof(blockmap_layer).
365 #define HAMMER_BLOCKMAP_LAYER1_OFFSET(zone2_offset) \
366 (HAMMER_BLOCKMAP_LAYER1_INDEX(zone2_offset) * \
367 sizeof(struct hammer_blockmap_layer1))
369 #define HAMMER_BLOCKMAP_LAYER2_OFFSET(zone2_offset) \
370 (HAMMER_BLOCKMAP_LAYER2_INDEX(zone2_offset) * \
371 sizeof(struct hammer_blockmap_layer2))
374 * HAMMER UNDO parameters. The UNDO fifo is mapped directly in the volume
375 * header with an array of layer2 structures. A maximum of (128x8MB) = 1GB
376 * may be reserved. The size of the undo fifo is usually set a newfs time
377 * but can be adjusted if the filesystem is taken offline.
379 #define HAMMER_UNDO_LAYER2 128 /* max layer2 undo mapping entries */
382 * All on-disk HAMMER structures which make up elements of the UNDO FIFO
383 * contain a hammer_fifo_head and hammer_fifo_tail structure. This structure
384 * contains all the information required to validate the fifo element
385 * and to scan the fifo in either direction. The head is typically embedded
386 * in higher level hammer on-disk structures while the tail is typically
387 * out-of-band. hdr_size is the size of the whole mess, including the tail.
389 * All undo structures are guaranteed to not cross a 16K filesystem
390 * buffer boundary. Most undo structures are fairly small. Data spaces
391 * are not immediately reused by HAMMER so file data is not usually recorded
392 * as part of an UNDO.
394 * PAD elements are allowed to take up only 8 bytes of space as a special
395 * case, containing only hdr_signature, hdr_type, and hdr_size fields,
396 * and with the tail overloaded onto the head structure for 8 bytes total.
398 * Every undo record has a sequence number. This number is unrelated to
399 * transaction ids and instead collects the undo transactions associated
400 * with a single atomic operation. A larger transactional operation, such
401 * as a remove(), may consist of several smaller atomic operations
402 * representing raw meta-data operations.
404 * HAMMER VERSION 4 CHANGES
406 * In HAMMER version 4 the undo structure alignment is reduced from 16384
407 * to 512 bytes in order to ensure that each 512 byte sector begins with
408 * a header. The reserved01 field in the header is now a 32 bit sequence
409 * number. This allows the recovery code to detect missing sectors
410 * without relying on the 32-bit crc and to definitively identify the current
411 * undo sequence space without having to rely on information from the volume
412 * header. In addition, new REDO entries in the undo space are used to
413 * record write, write/extend, and transaction id updates.
415 * The grand result is:
417 * (1) The volume header no longer needs to be synchronized for most
418 * flush and fsync operations.
420 * (2) Most fsync operations need only lay down REDO records
422 * (3) Data overwrite for nohistory operations covered by REDO records
423 * can be supported (instead of rolling a new block allocation),
424 * by rolling UNDO for the prior contents of the data.
426 * HAMMER VERSION 5 CHANGES
428 * Hammer version 5 contains a minor adjustment making layer2's bytes_free
429 * field signed, allowing dedup to push it into the negative domain.
431 #define HAMMER_HEAD_ALIGN 8
432 #define HAMMER_HEAD_ALIGN_MASK (HAMMER_HEAD_ALIGN - 1)
433 #define HAMMER_HEAD_DOALIGN(bytes) \
434 (((bytes) + HAMMER_HEAD_ALIGN_MASK) & ~HAMMER_HEAD_ALIGN_MASK)
436 #define HAMMER_UNDO_ALIGN 512
437 #define HAMMER_UNDO_ALIGN64 ((uint64_t)512)
438 #define HAMMER_UNDO_MASK (HAMMER_UNDO_ALIGN - 1)
439 #define HAMMER_UNDO_MASK64 (HAMMER_UNDO_ALIGN64 - 1)
441 struct hammer_fifo_head {
442 uint16_t hdr_signature;
444 uint32_t hdr_size; /* Aligned size of the whole mess */
445 uint32_t hdr_seq; /* Sequence number */
446 hammer_crc_t hdr_crc; /* XOR crc up to field w/ crc after field */
449 #define HAMMER_FIFO_HEAD_CRCOFF offsetof(struct hammer_fifo_head, hdr_crc)
451 struct hammer_fifo_tail {
452 uint16_t tail_signature;
454 uint32_t tail_size; /* aligned size of the whole mess */
457 typedef struct hammer_fifo_head *hammer_fifo_head_t;
458 typedef struct hammer_fifo_tail *hammer_fifo_tail_t;
463 #define HAMMER_HEAD_TYPE_PAD (0x0040U | 0x8000U)
464 #define HAMMER_HEAD_TYPE_DUMMY 0x0041U /* dummy entry w/seqno */
465 #define HAMMER_HEAD_TYPE_UNDO 0x0043U /* random UNDO information */
466 #define HAMMER_HEAD_TYPE_REDO 0x0044U /* data REDO / fast fsync */
468 #define HAMMER_HEAD_SIGNATURE 0xC84EU
469 #define HAMMER_TAIL_SIGNATURE 0xC74FU
472 * Misc FIFO structures.
474 * UNDO - Raw meta-data media updates.
476 struct hammer_fifo_undo {
477 struct hammer_fifo_head head;
478 hammer_off_t undo_offset; /* zone-1,2 offset */
479 int32_t undo_data_bytes;
480 int32_t undo_reserved01;
481 /* followed by data */
485 * REDO (HAMMER version 4+) - Logical file writes/truncates.
487 * REDOs contain information which will be duplicated in a later meta-data
488 * update, allowing fast write()+fsync() operations. REDOs can be ignored
489 * without harming filesystem integrity but must be processed if fsync()
490 * semantics are desired.
492 * Unlike UNDOs which are processed backwards within the recovery span,
493 * REDOs must be processed forwards starting further back (starting outside
494 * the recovery span).
496 * WRITE - Write logical file (with payload). Executed both
497 * out-of-span and in-span. Out-of-span WRITEs may be
498 * filtered out by TERMs.
500 * TRUNC - Truncate logical file (no payload). Executed both
501 * out-of-span and in-span. Out-of-span WRITEs may be
502 * filtered out by TERMs.
504 * TERM_* - Indicates meta-data was committed (if out-of-span) or
505 * will be rolled-back (in-span). Any out-of-span TERMs
506 * matching earlier WRITEs remove those WRITEs from
507 * consideration as they might conflict with a later data
508 * commit (which is not being rolled-back).
510 * SYNC - The earliest in-span SYNC (the last one when scanning
511 * backwards) tells the recovery code how far out-of-span
512 * it must go to run REDOs.
514 * NOTE: WRITEs do not always have matching TERMs even under
515 * perfect conditions because truncations might remove the
516 * buffers from consideration. I/O problems can also remove
517 * buffers from consideration.
519 * TRUNCSs do not always have matching TERMs because several
520 * truncations may be aggregated together into a single TERM.
522 struct hammer_fifo_redo {
523 struct hammer_fifo_head head;
524 int64_t redo_objid; /* file being written */
525 hammer_off_t redo_offset; /* logical offset in file */
526 int32_t redo_data_bytes;
528 uint32_t redo_localization;
529 uint32_t redo_reserved;
530 uint64_t redo_mtime; /* set mtime */
533 #define HAMMER_REDO_WRITE 0x00000001
534 #define HAMMER_REDO_TRUNC 0x00000002
535 #define HAMMER_REDO_TERM_WRITE 0x00000004
536 #define HAMMER_REDO_TERM_TRUNC 0x00000008
537 #define HAMMER_REDO_SYNC 0x00000010
539 union hammer_fifo_any {
540 struct hammer_fifo_head head;
541 struct hammer_fifo_undo undo;
542 struct hammer_fifo_redo redo;
545 typedef struct hammer_fifo_redo *hammer_fifo_redo_t;
546 typedef struct hammer_fifo_undo *hammer_fifo_undo_t;
547 typedef union hammer_fifo_any *hammer_fifo_any_t;
550 * Volume header types
552 #define HAMMER_FSBUF_VOLUME 0xC8414D4DC5523031ULL /* HAMMER01 */
553 #define HAMMER_FSBUF_VOLUME_REV 0x313052C54D4D41C8ULL /* (reverse endian) */
556 * HAMMER Volume header
558 * A HAMMER filesystem is built from any number of block devices, Each block
559 * device contains a volume header followed by however many buffers fit
562 * One of the volumes making up a HAMMER filesystem is the master, the
563 * rest are slaves. It does not have to be volume #0.
565 * The volume header takes up an entire 16K filesystem buffer and may
566 * represent up to 64KTB (65536 TB) of space.
568 * Special field notes:
570 * vol_bot_beg - offset of boot area (mem_beg - bot_beg bytes)
571 * vol_mem_beg - offset of memory log (clu_beg - mem_beg bytes)
572 * vol_buf_beg - offset of the first buffer.
574 * The memory log area allows a kernel to cache new records and data
575 * in memory without allocating space in the actual filesystem to hold
576 * the records and data. In the event that a filesystem becomes full,
577 * any records remaining in memory can be flushed to the memory log
578 * area. This allows the kernel to immediately return success.
581 #define HAMMER_BOOT_MINBYTES (32*1024)
582 #define HAMMER_BOOT_NOMBYTES (64LL*1024*1024)
583 #define HAMMER_BOOT_MAXBYTES (256LL*1024*1024)
585 #define HAMMER_MEM_MINBYTES (256*1024)
586 #define HAMMER_MEM_NOMBYTES (1LL*1024*1024*1024)
587 #define HAMMER_MEM_MAXBYTES (64LL*1024*1024*1024)
589 struct hammer_volume_ondisk {
590 uint64_t vol_signature; /* Signature */
592 int64_t vol_bot_beg; /* byte offset of boot area or 0 */
593 int64_t vol_mem_beg; /* byte offset of memory log or 0 */
594 int64_t vol_buf_beg; /* byte offset of first buffer in volume */
595 int64_t vol_buf_end; /* byte offset of volume EOF (on buf bndry) */
596 int64_t vol_locked; /* not used */
598 uuid_t vol_fsid; /* identify filesystem */
599 uuid_t vol_fstype; /* identify filesystem type */
600 char vol_name[64]; /* filesystem label */
602 int32_t vol_no; /* volume number within filesystem */
603 int32_t vol_count; /* number of volumes making up FS */
605 uint32_t vol_version; /* version control information */
606 hammer_crc_t vol_crc; /* header crc */
607 uint32_t vol_flags; /* volume flags */
608 uint32_t vol_rootvol; /* which volume is the root volume? */
610 int32_t vol_reserved04;
611 int32_t vol_reserved05;
612 uint32_t vol_reserved06;
613 uint32_t vol_reserved07;
615 int32_t vol_blocksize; /* for statfs only */
616 int32_t vol_reserved08;
617 int64_t vol_nblocks; /* total allocatable hammer bufs */
620 * These fields are initialized and space is reserved in every
621 * volume making up a HAMMER filesytem, but only the master volume
622 * contains valid data. Note that vol0_stat_bigblocks does not
623 * include big-blocks for freemap and undomap initially allocated
626 int64_t vol0_stat_bigblocks; /* total big-blocks when fs is empty */
627 int64_t vol0_stat_freebigblocks;/* number of free big-blocks */
628 int64_t vol0_stat_bytes; /* for statfs only */
629 int64_t vol0_stat_inodes; /* for statfs only */
630 int64_t vol0_stat_records; /* total records in filesystem */
631 hammer_off_t vol0_btree_root; /* B-Tree root */
632 hammer_tid_t vol0_next_tid; /* highest partially synchronized TID */
633 hammer_off_t vol0_unused03;
636 * Blockmaps for zones. Not all zones use a blockmap. Note that
637 * the entire root blockmap is cached in the hammer_mount structure.
639 struct hammer_blockmap vol0_blockmap[HAMMER_MAX_ZONES];
642 * Array of zone-2 addresses for undo FIFO.
644 hammer_off_t vol0_undo_array[HAMMER_UNDO_LAYER2];
647 typedef struct hammer_volume_ondisk *hammer_volume_ondisk_t;
649 #define HAMMER_VOLF_NEEDFLUSH 0x0004 /* volume needs flush */
651 #define HAMMER_VOL_CRCSIZE1 \
652 offsetof(struct hammer_volume_ondisk, vol_crc)
653 #define HAMMER_VOL_CRCSIZE2 \
654 (sizeof(struct hammer_volume_ondisk) - HAMMER_VOL_CRCSIZE1 - \
655 sizeof(hammer_crc_t))
657 #define HAMMER_VOL_VERSION_MIN 1 /* minimum supported version */
658 #define HAMMER_VOL_VERSION_DEFAULT 6 /* newfs default version */
659 #define HAMMER_VOL_VERSION_WIP 7 /* version >= this is WIP */
660 #define HAMMER_VOL_VERSION_MAX 6 /* maximum supported version */
662 #define HAMMER_VOL_VERSION_ONE 1
663 #define HAMMER_VOL_VERSION_TWO 2 /* new dirent layout (2.3+) */
664 #define HAMMER_VOL_VERSION_THREE 3 /* new snapshot layout (2.5+) */
665 #define HAMMER_VOL_VERSION_FOUR 4 /* new undo/flush (2.5+) */
666 #define HAMMER_VOL_VERSION_FIVE 5 /* dedup (2.9+) */
667 #define HAMMER_VOL_VERSION_SIX 6 /* DIRHASH_ALG1 */
670 * Record types are fairly straightforward. The B-Tree includes the record
671 * type in its index sort.
673 #define HAMMER_RECTYPE_UNKNOWN 0x0000
674 #define HAMMER_RECTYPE_LOWEST 0x0001 /* lowest record type avail */
675 #define HAMMER_RECTYPE_INODE 0x0001 /* inode in obj_id space */
676 #define HAMMER_RECTYPE_DATA 0x0010
677 #define HAMMER_RECTYPE_DIRENTRY 0x0011
678 #define HAMMER_RECTYPE_DB 0x0012
679 #define HAMMER_RECTYPE_EXT 0x0013 /* ext attributes */
680 #define HAMMER_RECTYPE_FIX 0x0014 /* fixed attribute */
681 #define HAMMER_RECTYPE_PFS 0x0015 /* PFS management */
682 #define HAMMER_RECTYPE_SNAPSHOT 0x0016 /* Snapshot management */
683 #define HAMMER_RECTYPE_CONFIG 0x0017 /* hammer cleanup config */
684 #define HAMMER_RECTYPE_MAX 0xFFFF
686 #define HAMMER_RECTYPE_ENTRY_START (HAMMER_RECTYPE_INODE + 1)
687 #define HAMMER_RECTYPE_CLEAN_START HAMMER_RECTYPE_EXT
689 #define HAMMER_FIXKEY_SYMLINK 1
691 #define HAMMER_OBJTYPE_UNKNOWN 0 /* never exists on-disk as unknown */
692 #define HAMMER_OBJTYPE_DIRECTORY 1
693 #define HAMMER_OBJTYPE_REGFILE 2
694 #define HAMMER_OBJTYPE_DBFILE 3
695 #define HAMMER_OBJTYPE_FIFO 4
696 #define HAMMER_OBJTYPE_CDEV 5
697 #define HAMMER_OBJTYPE_BDEV 6
698 #define HAMMER_OBJTYPE_SOFTLINK 7
699 #define HAMMER_OBJTYPE_PSEUDOFS 8 /* pseudo filesystem obj */
700 #define HAMMER_OBJTYPE_SOCKET 9
703 * HAMMER inode attribute data
705 * The data reference for a HAMMER inode points to this structure. Any
706 * modifications to the contents of this structure will result in a
707 * replacement operation.
709 * parent_obj_id is only valid for directories (which cannot be hard-linked),
710 * and specifies the parent directory obj_id. This field will also be set
711 * for non-directory inodes as a recovery aid, but can wind up holding
712 * stale information. However, since object id's are not reused, the worse
713 * that happens is that the recovery code is unable to use it.
715 * NOTE: Future note on directory hardlinks. We can implement a record type
716 * which allows us to point to multiple parent directories.
718 struct hammer_inode_data {
719 uint16_t version; /* inode data version */
720 uint16_t mode; /* basic unix permissions */
721 uint32_t uflags; /* chflags */
722 uint32_t rmajor; /* used by device nodes */
723 uint32_t rminor; /* used by device nodes */
725 int64_t parent_obj_id; /* parent directory obj_id */
730 uint8_t cap_flags; /* capability support flags (extension) */
732 uint32_t reserved03; /* RESERVED FOR POSSIBLE FUTURE BIRTHTIME */
733 uint64_t nlinks; /* hard links */
734 uint64_t size; /* filesystem object size */
738 uint32_t parent_obj_localization;
739 uint32_t integrity_crc;
741 char symlink[24]; /* HAMMER_INODE_BASESYMLEN */
743 uint64_t mtime; /* mtime must be second-to-last */
744 uint64_t atime; /* atime must be last */
748 * Neither mtime nor atime upates are CRCd by the B-Tree element.
749 * mtime updates have UNDO, atime updates do not.
751 #define HAMMER_ITIMES_BASE(ino_data) (&(ino_data)->mtime)
752 #define HAMMER_ITIMES_BYTES (sizeof(uint64_t) * 2)
754 #define HAMMER_INODE_CRCSIZE \
755 offsetof(struct hammer_inode_data, mtime)
757 #define HAMMER_INODE_DATA_VERSION 1
758 #define HAMMER_OBJID_ROOT 1 /* root inodes # */
759 #define HAMMER_INODE_BASESYMLEN 24 /* see ext.symlink */
762 * Capability & implementation flags.
764 * HAMMER_INODE_CAP_DIR_LOCAL_INO - Use inode B-Tree localization
765 * for directory entries. Also see HAMMER_DIR_INODE_LOCALIZATION().
767 #define HAMMER_INODE_CAP_DIRHASH_MASK 0x03 /* directory: hash algorithm */
768 #define HAMMER_INODE_CAP_DIRHASH_ALG0 0x00
769 #define HAMMER_INODE_CAP_DIRHASH_ALG1 0x01
770 #define HAMMER_INODE_CAP_DIRHASH_ALG2 0x02
771 #define HAMMER_INODE_CAP_DIRHASH_ALG3 0x03
772 #define HAMMER_INODE_CAP_DIR_LOCAL_INO 0x04 /* use inode localization */
775 * A HAMMER directory entry associates a HAMMER filesystem object with a
776 * namespace. It is possible to hook into a pseudo-filesystem (with its
777 * own inode numbering space) in the filesystem by setting the high
778 * 16 bits of the localization field. The low 16 bits must be 0 and
779 * are reserved for future use.
781 * Directory entries are indexed with a 128 bit namekey rather then an
782 * offset. A portion of the namekey is an iterator/randomizer to deal
785 * NOTE: leaf.base.obj_type from the related B-Tree leaf entry holds
786 * the filesystem object type of obj_id, e.g. a den_type equivalent.
787 * It is not stored in hammer_entry_data.
789 * NOTE: name field / the filename data reference is NOT terminated with \0.
791 struct hammer_entry_data {
792 int64_t obj_id; /* object being referenced */
793 uint32_t localization; /* identify pseudo-filesystem */
795 char name[16]; /* name (extended) */
798 #define HAMMER_ENTRY_NAME_OFF offsetof(struct hammer_entry_data, name[0])
799 #define HAMMER_ENTRY_SIZE(nlen) offsetof(struct hammer_entry_data, name[nlen])
802 * Symlink data which does not fit in the inode is stored in a separate
805 struct hammer_symlink_data {
806 char name[16]; /* name (extended) */
809 #define HAMMER_SYMLINK_NAME_OFF offsetof(struct hammer_symlink_data, name[0])
812 * The root inode for the primary filesystem and root inode for any
813 * pseudo-fs may be tagged with an optional data structure using
814 * HAMMER_RECTYPE_PFS and localization id. This structure allows
815 * the node to be used as a mirroring master or slave.
817 * When operating as a slave CD's into the node automatically become read-only
818 * and as-of sync_end_tid.
820 * When operating as a master the read PFSD info sets sync_end_tid to
821 * the most recently flushed TID.
823 * sync_low_tid is not yet used but will represent the highest pruning
824 * end-point, after which full history is available.
826 * We need to pack this structure making it equally sized on both 32-bit and
827 * 64-bit machines as it is part of struct hammer_ioc_mrecord_pfs which is
828 * send over the wire in hammer mirror operations. Only on 64-bit machines
829 * the size of this struct differ when packed or not. This leads us to the
830 * situation where old 64-bit systems (using the non-packed structure),
831 * which were never able to mirror to/from 32-bit systems, are now no longer
832 * able to mirror to/from newer 64-bit systems (using the packed structure).
834 struct hammer_pseudofs_data {
835 hammer_tid_t sync_low_tid; /* full history beyond this point */
836 hammer_tid_t sync_beg_tid; /* earliest tid w/ full history avail */
837 hammer_tid_t sync_end_tid; /* current synchronizatoin point */
838 uint64_t sync_beg_ts; /* real-time of last completed sync */
839 uint64_t sync_end_ts; /* initiation of current sync cycle */
840 uuid_t shared_uuid; /* shared uuid (match required) */
841 uuid_t unique_uuid; /* unique uuid of this master/slave */
842 int32_t reserved01; /* reserved for future master_id */
843 int32_t mirror_flags; /* misc flags */
844 char label[64]; /* filesystem space label */
845 char snapshots[64]; /* softlink dir for pruning */
846 int16_t prune_time; /* how long to spend pruning */
847 int16_t prune_freq; /* how often we prune */
848 int16_t reblock_time; /* how long to spend reblocking */
849 int16_t reblock_freq; /* how often we reblock */
850 int32_t snapshot_freq; /* how often we create a snapshot */
851 int32_t prune_min; /* do not prune recent history */
852 int32_t prune_max; /* do not retain history beyond here */
853 int32_t reserved[16];
856 typedef struct hammer_pseudofs_data *hammer_pseudofs_data_t;
858 #define HAMMER_PFSD_SLAVE 0x00000001
859 #define HAMMER_PFSD_DELETED 0x80000000
862 * Snapshot meta-data { Objid = HAMMER_OBJID_ROOT, Key = tid, rectype = SNAPSHOT }.
864 * Snapshot records replace the old <fs>/snapshots/<softlink> methodology. Snapshot
865 * records are mirrored but may be independantly managed once they are laid down on
868 * NOTE: The b-tree key is signed, the tid is not, so callers must still sort the
871 * NOTE: Reserved fields must be zero (as usual)
873 struct hammer_snapshot_data {
874 hammer_tid_t tid; /* the snapshot TID itself (== key) */
875 uint64_t ts; /* real-time when snapshot was made */
878 char label[64]; /* user-supplied description */
879 uint64_t reserved03[4];
883 * Config meta-data { ObjId = HAMMER_OBJID_ROOT, Key = 0, rectype = CONFIG }.
885 * Used to store the hammer cleanup config. This data is not mirrored.
887 struct hammer_config_data {
892 * Rollup various structures embedded as record data
894 union hammer_data_ondisk {
895 struct hammer_entry_data entry;
896 struct hammer_inode_data inode;
897 struct hammer_symlink_data symlink;
898 struct hammer_pseudofs_data pfsd;
899 struct hammer_snapshot_data snap;
900 struct hammer_config_data config;
903 typedef union hammer_data_ondisk *hammer_data_ondisk_t;
906 * Ondisk layout of B-Tree related structures
908 #include "hammer_btree.h"
910 #define HAMMER_DIR_INODE_LOCALIZATION(ino_data) \
911 (((ino_data)->cap_flags & HAMMER_INODE_CAP_DIR_LOCAL_INO) ? \
912 HAMMER_LOCALIZE_INODE : \
913 HAMMER_LOCALIZE_MISC)
915 #endif /* !VFS_HAMMER_DISK_H_ */