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. Most buffer-sized headers such as those
60 * used by volumes, super-clusters, clusters, and basic filesystem buffers
61 * use fixed-sized A-lists which are heavily dependant on HAMMER_BUFSIZE.
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_XDEMARC (1024 * 1024)
72 #define HAMMER_BUFMASK (HAMMER_BUFSIZE - 1)
73 #define HAMMER_XBUFMASK (HAMMER_XBUFSIZE - 1)
74 #define HAMMER_BUFFER_BITS 14
76 #if (1 << HAMMER_BUFFER_BITS) != HAMMER_BUFSIZE
77 #error "HAMMER_BUFFER_BITS BROKEN"
80 #define HAMMER_BUFSIZE64 ((u_int64_t)HAMMER_BUFSIZE)
81 #define HAMMER_BUFMASK64 ((u_int64_t)HAMMER_BUFMASK)
83 #define HAMMER_XBUFSIZE64 ((u_int64_t)HAMMER_XBUFSIZE)
84 #define HAMMER_XBUFMASK64 ((u_int64_t)HAMMER_XBUFMASK)
86 #define HAMMER_OFF_ZONE_MASK 0xF000000000000000ULL /* zone portion */
87 #define HAMMER_OFF_VOL_MASK 0x0FF0000000000000ULL /* volume portion */
88 #define HAMMER_OFF_SHORT_MASK 0x000FFFFFFFFFFFFFULL /* offset portion */
89 #define HAMMER_OFF_LONG_MASK 0x0FFFFFFFFFFFFFFFULL /* offset portion */
91 #define HAMMER_OFF_BAD ((hammer_off_t)-1)
94 * The current limit of volumes that can make up a HAMMER FS
96 #define HAMMER_MAX_VOLUMES 256
99 * Hammer transaction ids are 64 bit unsigned integers and are usually
100 * synchronized with the time of day in nanoseconds.
102 * Hammer offsets are used for FIFO indexing and embed a cycle counter
103 * and volume number in addition to the offset. Most offsets are required
104 * to be 16 KB aligned.
106 typedef u_int64_t hammer_tid_t;
107 typedef u_int64_t hammer_off_t;
108 typedef u_int32_t hammer_crc_t;
110 #define HAMMER_MIN_TID 0ULL /* unsigned */
111 #define HAMMER_MAX_TID 0xFFFFFFFFFFFFFFFFULL /* unsigned */
112 #define HAMMER_MIN_KEY -0x8000000000000000LL /* signed */
113 #define HAMMER_MAX_KEY 0x7FFFFFFFFFFFFFFFLL /* signed */
114 #define HAMMER_MIN_OBJID HAMMER_MIN_KEY /* signed */
115 #define HAMMER_MAX_OBJID HAMMER_MAX_KEY /* signed */
116 #define HAMMER_MIN_RECTYPE 0x0U /* unsigned */
117 #define HAMMER_MAX_RECTYPE 0xFFFFU /* unsigned */
118 #define HAMMER_MIN_OFFSET 0ULL /* unsigned */
119 #define HAMMER_MAX_OFFSET 0xFFFFFFFFFFFFFFFFULL /* unsigned */
122 * hammer_off_t has several different encodings. Note that not all zones
125 * zone 0: reserved for sanity
126 * zone 1 (z,v,o): raw volume relative (offset 0 is the volume header)
127 * zone 2 (z,v,o): raw buffer relative (offset 0 is the first buffer)
128 * zone 3 (z,o): undo fifo - actually zone-2 address, fixed phys array in vol hdr
129 * zone 4 (z,v,o): freemap - only real blockmap
130 * zone 8 (z,v,o): B-Tree - actually zone-2 address
131 * zone 9 (z,v,o): meta - actually zone-2 address
132 * zone 10 (z,v,o): large-data - actually zone-2 address
133 * zone 11 (z,v,o): small-data - actually zone-2 address
134 * zone 15: reserved for sanity
136 * layer1/layer2 direct map:
137 * zzzzvvvvvvvvoooo oooooooooooooooo oooooooooooooooo oooooooooooooooo
138 * ----111111111111 1111112222222222 222222222ooooooo oooooooooooooooo
141 #define HAMMER_ZONE_RAW_VOLUME 0x1000000000000000ULL
142 #define HAMMER_ZONE_RAW_BUFFER 0x2000000000000000ULL
143 #define HAMMER_ZONE_UNDO 0x3000000000000000ULL
144 #define HAMMER_ZONE_FREEMAP 0x4000000000000000ULL
145 #define HAMMER_ZONE_RESERVED05 0x5000000000000000ULL
146 #define HAMMER_ZONE_RESERVED06 0x6000000000000000ULL
147 #define HAMMER_ZONE_RESERVED07 0x7000000000000000ULL
148 #define HAMMER_ZONE_BTREE 0x8000000000000000ULL
149 #define HAMMER_ZONE_META 0x9000000000000000ULL
150 #define HAMMER_ZONE_LARGE_DATA 0xA000000000000000ULL
151 #define HAMMER_ZONE_SMALL_DATA 0xB000000000000000ULL
152 #define HAMMER_ZONE_RESERVED0C 0xC000000000000000ULL
153 #define HAMMER_ZONE_RESERVED0D 0xD000000000000000ULL
154 #define HAMMER_ZONE_RESERVED0E 0xE000000000000000ULL
155 #define HAMMER_ZONE_UNAVAIL 0xF000000000000000ULL
157 #define HAMMER_ZONE_RAW_VOLUME_INDEX 1
158 #define HAMMER_ZONE_RAW_BUFFER_INDEX 2
159 #define HAMMER_ZONE_UNDO_INDEX 3
160 #define HAMMER_ZONE_FREEMAP_INDEX 4
161 #define HAMMER_ZONE_BTREE_INDEX 8
162 #define HAMMER_ZONE_META_INDEX 9
163 #define HAMMER_ZONE_LARGE_DATA_INDEX 10
164 #define HAMMER_ZONE_SMALL_DATA_INDEX 11
165 #define HAMMER_ZONE_UNAVAIL_INDEX 15 /* unavailable */
167 #define HAMMER_MAX_ZONES 16
170 * Backend zones that are mapped to zone-2 (except for zone-3)
171 * starts from this index which is 8.
173 #define HAMMER_ZONE2_MAPPED_INDEX HAMMER_ZONE_BTREE_INDEX
175 #define HAMMER_ZONE_ENCODE(zone, ham_off) \
176 (((hammer_off_t)(zone) << 60) | (ham_off))
177 #define HAMMER_ZONE_DECODE(ham_off) \
178 (int32_t)(((hammer_off_t)(ham_off) >> 60))
180 #define HAMMER_VOL_ENCODE(vol_no) \
181 ((hammer_off_t)((vol_no) & 255) << 52)
182 #define HAMMER_VOL_DECODE(ham_off) \
183 (int32_t)(((hammer_off_t)(ham_off) >> 52) & 255)
185 #define HAMMER_OFF_SHORT_ENCODE(offset) \
186 ((hammer_off_t)(offset) & HAMMER_OFF_SHORT_MASK)
187 #define HAMMER_OFF_LONG_ENCODE(offset) \
188 ((hammer_off_t)(offset) & HAMMER_OFF_LONG_MASK)
190 #define HAMMER_ENCODE(zone, vol_no, offset) \
191 (((hammer_off_t)(zone) << 60) | \
192 HAMMER_VOL_ENCODE(vol_no) | \
193 HAMMER_OFF_SHORT_ENCODE(offset))
194 #define HAMMER_ENCODE_RAW_VOLUME(vol_no, offset) \
195 HAMMER_ENCODE(HAMMER_ZONE_RAW_VOLUME_INDEX, vol_no, offset)
196 #define HAMMER_ENCODE_RAW_BUFFER(vol_no, offset) \
197 HAMMER_ENCODE(HAMMER_ZONE_RAW_BUFFER_INDEX, vol_no, offset)
198 #define HAMMER_ENCODE_FREEMAP(vol_no, offset) \
199 HAMMER_ENCODE(HAMMER_ZONE_FREEMAP_INDEX, vol_no, offset)
202 * Translate a zone address to zone-X address.
204 #define hammer_xlate_to_zoneX(zone, offset) \
205 HAMMER_ZONE_ENCODE((zone), (offset) & ~HAMMER_OFF_ZONE_MASK)
206 #define hammer_xlate_to_zone2(offset) \
207 hammer_xlate_to_zoneX(HAMMER_ZONE_RAW_BUFFER_INDEX, (offset))
210 * Big-Block backing store
212 * A blockmap is a two-level map which translates a blockmap-backed zone
213 * offset into a raw zone 2 offset. The layer 1 handles 18 bits and the
214 * layer 2 handles 19 bits. The 8M big-block size is 23 bits so two
215 * layers gives us 18+19+23 = 60 bits of address space.
217 * When using hinting for a blockmap lookup, the hint is lost when the
218 * scan leaves the HINTBLOCK, which is typically several BIGBLOCK's.
219 * HINTBLOCK is a heuristic.
221 #define HAMMER_HINTBLOCK_SIZE (HAMMER_BIGBLOCK_SIZE * 4)
222 #define HAMMER_HINTBLOCK_MASK64 ((u_int64_t)HAMMER_HINTBLOCK_SIZE - 1)
223 #define HAMMER_BIGBLOCK_SIZE (8192 * 1024)
224 #define HAMMER_BIGBLOCK_OVERFILL (6144 * 1024)
225 #define HAMMER_BIGBLOCK_SIZE64 ((u_int64_t)HAMMER_BIGBLOCK_SIZE)
226 #define HAMMER_BIGBLOCK_MASK (HAMMER_BIGBLOCK_SIZE - 1)
227 #define HAMMER_BIGBLOCK_MASK64 ((u_int64_t)HAMMER_BIGBLOCK_SIZE - 1)
228 #define HAMMER_BIGBLOCK_BITS 23
229 #if (1 << HAMMER_BIGBLOCK_BITS) != HAMMER_BIGBLOCK_SIZE
230 #error "HAMMER_BIGBLOCK_BITS BROKEN"
233 #define HAMMER_BUFFERS_PER_BIGBLOCK \
234 (HAMMER_BIGBLOCK_SIZE / HAMMER_BUFSIZE)
235 #define HAMMER_BUFFERS_PER_BIGBLOCK_MASK \
236 (HAMMER_BUFFERS_PER_BIGBLOCK - 1)
237 #define HAMMER_BUFFERS_PER_BIGBLOCK_MASK64 \
238 ((hammer_off_t)HAMMER_BUFFERS_PER_BIGBLOCK_MASK)
241 * Maximum number of mirrors operating in master mode (multi-master
242 * clustering and mirroring). Note that HAMMER1 does not support
243 * multi-master clustering as of 2015.
245 #define HAMMER_MAX_MASTERS 16
248 * The blockmap is somewhat of a degenerate structure. HAMMER only actually
249 * uses it in its original incarnation to implement the freemap.
251 * zone:1 raw volume (no blockmap)
252 * zone:2 raw buffer (no blockmap)
253 * zone:3 undomap (direct layer2 array in volume header)
254 * zone:4 freemap (the only real blockmap)
255 * zone:8-15 zone id used to classify big-block only, address is actually
258 struct hammer_blockmap {
259 hammer_off_t phys_offset; /* zone-2 physical offset */
260 hammer_off_t first_offset; /* zone-X logical offset (zone 3) */
261 hammer_off_t next_offset; /* zone-X logical offset */
262 hammer_off_t alloc_offset; /* zone-X logical offset */
263 u_int32_t reserved01;
264 hammer_crc_t entry_crc;
267 typedef struct hammer_blockmap *hammer_blockmap_t;
269 #define HAMMER_BLOCKMAP_CRCSIZE \
270 offsetof(struct hammer_blockmap, entry_crc)
273 * The blockmap is a 2-layer entity made up of big-blocks. The first layer
274 * contains 262144 32-byte entries (18 bits), the second layer contains
275 * 524288 16-byte entries (19 bits), representing 8MB (23 bit) blockmaps.
276 * 18+19+23 = 60 bits. The top four bits are the zone id.
278 * Currently only the freemap utilizes both layers in all their glory.
279 * All primary data/meta-data zones actually encode a zone-2 address
280 * requiring no real blockmap translation.
282 * The freemap uses the upper 8 bits of layer-1 to identify the volume,
283 * thus any space allocated via the freemap can be directly translated
284 * to a zone:2 (or zone:8-15) address.
286 * zone-X blockmap offset: [zone:4][layer1:18][layer2:19][big-block:23]
288 struct hammer_blockmap_layer1 {
289 hammer_off_t blocks_free; /* big-blocks free */
290 hammer_off_t phys_offset; /* UNAVAIL or zone-2 */
291 hammer_off_t reserved01;
292 hammer_crc_t layer2_crc; /* xor'd crc's of HAMMER_BLOCKSIZE */
294 hammer_crc_t layer1_crc; /* MUST BE LAST FIELD OF STRUCTURE*/
297 typedef struct hammer_blockmap_layer1 *hammer_blockmap_layer1_t;
299 #define HAMMER_LAYER1_CRCSIZE \
300 offsetof(struct hammer_blockmap_layer1, layer1_crc)
303 * layer2 entry for 8MB big-block.
305 * NOTE: bytes_free is signed and can legally go negative if/when data
306 * de-dup occurs. This field will never go higher than
307 * HAMMER_BIGBLOCK_SIZE. If exactly HAMMER_BIGBLOCK_SIZE
308 * the big-block is completely free.
310 struct hammer_blockmap_layer2 {
311 u_int8_t zone; /* typed allocation zone */
314 u_int32_t append_off; /* allocatable space index */
315 int32_t bytes_free; /* bytes free within this big-block */
316 hammer_crc_t entry_crc;
319 typedef struct hammer_blockmap_layer2 *hammer_blockmap_layer2_t;
321 #define HAMMER_LAYER2_CRCSIZE \
322 offsetof(struct hammer_blockmap_layer2, entry_crc)
324 #define HAMMER_BLOCKMAP_UNAVAIL ((hammer_off_t)-1LL)
326 #define HAMMER_BLOCKMAP_RADIX1 /* 262144 (18) */ \
327 (HAMMER_BIGBLOCK_SIZE / sizeof(struct hammer_blockmap_layer1))
328 #define HAMMER_BLOCKMAP_RADIX2 /* 524288 (19) */ \
329 (HAMMER_BIGBLOCK_SIZE / sizeof(struct hammer_blockmap_layer2))
331 #define HAMMER_BLOCKMAP_RADIX1_PERBUFFER \
332 (HAMMER_BLOCKMAP_RADIX1 / HAMMER_BUFFERS_PER_BIGBLOCK)
333 #define HAMMER_BLOCKMAP_RADIX2_PERBUFFER \
334 (HAMMER_BLOCKMAP_RADIX2 / HAMMER_BUFFERS_PER_BIGBLOCK)
336 #define HAMMER_BLOCKMAP_LAYER1 /* 18+19+23 - 1EB */ \
337 (HAMMER_BLOCKMAP_RADIX1 * HAMMER_BLOCKMAP_LAYER2)
338 #define HAMMER_BLOCKMAP_LAYER2 /* 19+23 - 4TB */ \
339 (HAMMER_BLOCKMAP_RADIX2 * HAMMER_BIGBLOCK_SIZE64)
341 #define HAMMER_BLOCKMAP_LAYER1_MASK (HAMMER_BLOCKMAP_LAYER1 - 1)
342 #define HAMMER_BLOCKMAP_LAYER2_MASK (HAMMER_BLOCKMAP_LAYER2 - 1)
345 * Index within layer1 or layer2 big-block for the entry representing
346 * a zone-2 physical offset.
348 #define HAMMER_BLOCKMAP_LAYER1_INDEX(zone2_offset) \
349 (((zone2_offset) & HAMMER_BLOCKMAP_LAYER1_MASK) / \
350 HAMMER_BLOCKMAP_LAYER2)
352 #define HAMMER_BLOCKMAP_LAYER2_INDEX(zone2_offset) \
353 (((zone2_offset) & HAMMER_BLOCKMAP_LAYER2_MASK) / \
354 HAMMER_BIGBLOCK_SIZE64)
357 * Byte offset within layer1 or layer2 big-block for the entry representing
358 * a zone-2 physical offset. Multiply the index by sizeof(blockmap_layer).
360 #define HAMMER_BLOCKMAP_LAYER1_OFFSET(zone2_offset) \
361 (HAMMER_BLOCKMAP_LAYER1_INDEX(zone2_offset) * \
362 sizeof(struct hammer_blockmap_layer1))
364 #define HAMMER_BLOCKMAP_LAYER2_OFFSET(zone2_offset) \
365 (HAMMER_BLOCKMAP_LAYER2_INDEX(zone2_offset) * \
366 sizeof(struct hammer_blockmap_layer2))
369 * HAMMER UNDO parameters. The UNDO fifo is mapped directly in the volume
370 * header with an array of layer2 structures. A maximum of (128x8MB) = 1GB
371 * may be reserved. The size of the undo fifo is usually set a newfs time
372 * but can be adjusted if the filesystem is taken offline.
374 #define HAMMER_UNDO_LAYER2 128 /* max layer2 undo mapping entries */
377 * All on-disk HAMMER structures which make up elements of the UNDO FIFO
378 * contain a hammer_fifo_head and hammer_fifo_tail structure. This structure
379 * contains all the information required to validate the fifo element
380 * and to scan the fifo in either direction. The head is typically embedded
381 * in higher level hammer on-disk structures while the tail is typically
382 * out-of-band. hdr_size is the size of the whole mess, including the tail.
384 * All undo structures are guaranteed to not cross a 16K filesystem
385 * buffer boundary. Most undo structures are fairly small. Data spaces
386 * are not immediately reused by HAMMER so file data is not usually recorded
387 * as part of an UNDO.
389 * PAD elements are allowed to take up only 8 bytes of space as a special
390 * case, containing only hdr_signature, hdr_type, and hdr_size fields,
391 * and with the tail overloaded onto the head structure for 8 bytes total.
393 * Every undo record has a sequence number. This number is unrelated to
394 * transaction ids and instead collects the undo transactions associated
395 * with a single atomic operation. A larger transactional operation, such
396 * as a remove(), may consist of several smaller atomic operations
397 * representing raw meta-data operations.
399 * HAMMER VERSION 4 CHANGES
401 * In HAMMER version 4 the undo structure alignment is reduced from 16384
402 * to 512 bytes in order to ensure that each 512 byte sector begins with
403 * a header. The reserved01 field in the header is now a 32 bit sequence
404 * number. This allows the recovery code to detect missing sectors
405 * without relying on the 32-bit crc and to definitively identify the current
406 * undo sequence space without having to rely on information from the volume
407 * header. In addition, new REDO entries in the undo space are used to
408 * record write, write/extend, and transaction id updates.
410 * The grand result is:
412 * (1) The volume header no longer needs to be synchronized for most
413 * flush and fsync operations.
415 * (2) Most fsync operations need only lay down REDO records
417 * (3) Data overwrite for nohistory operations covered by REDO records
418 * can be supported (instead of rolling a new block allocation),
419 * by rolling UNDO for the prior contents of the data.
421 * HAMMER VERSION 5 CHANGES
423 * Hammer version 5 contains a minor adjustment making layer2's bytes_free
424 * field signed, allowing dedup to push it into the negative domain.
426 #define HAMMER_HEAD_ONDISK_SIZE 32
427 #define HAMMER_HEAD_ALIGN 8
428 #define HAMMER_HEAD_ALIGN_MASK (HAMMER_HEAD_ALIGN - 1)
429 #define HAMMER_TAIL_ONDISK_SIZE 8
430 #define HAMMER_HEAD_DOALIGN(bytes) \
431 (((bytes) + HAMMER_HEAD_ALIGN_MASK) & ~HAMMER_HEAD_ALIGN_MASK)
433 #define HAMMER_UNDO_ALIGN 512
434 #define HAMMER_UNDO_ALIGN64 ((u_int64_t)512)
435 #define HAMMER_UNDO_MASK (HAMMER_UNDO_ALIGN - 1)
436 #define HAMMER_UNDO_MASK64 (HAMMER_UNDO_ALIGN64 - 1)
438 struct hammer_fifo_head {
439 u_int16_t hdr_signature;
441 u_int32_t hdr_size; /* Aligned size of the whole mess */
442 u_int32_t hdr_seq; /* Sequence number */
443 hammer_crc_t hdr_crc; /* XOR crc up to field w/ crc after field */
446 #define HAMMER_FIFO_HEAD_CRCOFF offsetof(struct hammer_fifo_head, hdr_crc)
448 struct hammer_fifo_tail {
449 u_int16_t tail_signature;
451 u_int32_t tail_size; /* aligned size of the whole mess */
454 typedef struct hammer_fifo_head *hammer_fifo_head_t;
455 typedef struct hammer_fifo_tail *hammer_fifo_tail_t;
460 #define HAMMER_HEAD_TYPE_PAD (0x0040U|HAMMER_HEAD_FLAG_FREE)
461 #define HAMMER_HEAD_TYPE_DUMMY 0x0041U /* dummy entry w/seqno */
462 #define HAMMER_HEAD_TYPE_42 0x0042U
463 #define HAMMER_HEAD_TYPE_UNDO 0x0043U /* random UNDO information */
464 #define HAMMER_HEAD_TYPE_REDO 0x0044U /* data REDO / fast fsync */
465 #define HAMMER_HEAD_TYPE_45 0x0045U
467 #define HAMMER_HEAD_FLAG_FREE 0x8000U /* Indicates object freed */
469 #define HAMMER_HEAD_SIGNATURE 0xC84EU
470 #define HAMMER_TAIL_SIGNATURE 0xC74FU
473 * Misc FIFO structures.
475 * UNDO - Raw meta-data media updates.
477 struct hammer_fifo_undo {
478 struct hammer_fifo_head head;
479 hammer_off_t undo_offset; /* zone-1,2 offset */
480 int32_t undo_data_bytes;
481 int32_t undo_reserved01;
482 /* followed by data */
486 * REDO (HAMMER version 4+) - Logical file writes/truncates.
488 * REDOs contain information which will be duplicated in a later meta-data
489 * update, allowing fast write()+fsync() operations. REDOs can be ignored
490 * without harming filesystem integrity but must be processed if fsync()
491 * semantics are desired.
493 * Unlike UNDOs which are processed backwards within the recovery span,
494 * REDOs must be processed forwards starting further back (starting outside
495 * the recovery span).
497 * WRITE - Write logical file (with payload). Executed both
498 * out-of-span and in-span. Out-of-span WRITEs may be
499 * filtered out by TERMs.
501 * TRUNC - Truncate logical file (no payload). Executed both
502 * out-of-span and in-span. Out-of-span WRITEs may be
503 * filtered out by TERMs.
505 * TERM_* - Indicates meta-data was committed (if out-of-span) or
506 * will be rolled-back (in-span). Any out-of-span TERMs
507 * matching earlier WRITEs remove those WRITEs from
508 * consideration as they might conflict with a later data
509 * commit (which is not being rolled-back).
511 * SYNC - The earliest in-span SYNC (the last one when scanning
512 * backwards) tells the recovery code how far out-of-span
513 * it must go to run REDOs.
515 * NOTE: WRITEs do not always have matching TERMs even under
516 * perfect conditions because truncations might remove the
517 * buffers from consideration. I/O problems can also remove
518 * buffers from consideration.
520 * TRUNCSs do not always have matching TERMs because several
521 * truncations may be aggregated together into a single TERM.
523 struct hammer_fifo_redo {
524 struct hammer_fifo_head head;
525 int64_t redo_objid; /* file being written */
526 hammer_off_t redo_offset; /* logical offset in file */
527 int32_t redo_data_bytes;
528 u_int32_t redo_flags;
529 u_int32_t redo_localization;
530 u_int32_t redo_reserved;
531 u_int64_t redo_mtime; /* set mtime */
534 #define HAMMER_REDO_WRITE 0x00000001
535 #define HAMMER_REDO_TRUNC 0x00000002
536 #define HAMMER_REDO_TERM_WRITE 0x00000004
537 #define HAMMER_REDO_TERM_TRUNC 0x00000008
538 #define HAMMER_REDO_SYNC 0x00000010
540 union hammer_fifo_any {
541 struct hammer_fifo_head head;
542 struct hammer_fifo_undo undo;
543 struct hammer_fifo_redo redo;
546 typedef struct hammer_fifo_redo *hammer_fifo_redo_t;
547 typedef struct hammer_fifo_undo *hammer_fifo_undo_t;
548 typedef union hammer_fifo_any *hammer_fifo_any_t;
551 * Volume header types
553 #define HAMMER_FSBUF_VOLUME 0xC8414D4DC5523031ULL /* HAMMER01 */
554 #define HAMMER_FSBUF_VOLUME_REV 0x313052C54D4D41C8ULL /* (reverse endian) */
557 * HAMMER Volume header
559 * A HAMMER filesystem is built from any number of block devices, Each block
560 * device contains a volume header followed by however many buffers fit
563 * One of the volumes making up a HAMMER filesystem is the master, the
564 * rest are slaves. It does not have to be volume #0.
566 * The volume header takes up an entire 16K filesystem buffer and may
567 * represent up to 64KTB (65536 TB) of space.
569 * Special field notes:
571 * vol_bot_beg - offset of boot area (mem_beg - bot_beg bytes)
572 * vol_mem_beg - offset of memory log (clu_beg - mem_beg bytes)
573 * vol_buf_beg - offset of the first buffer.
575 * The memory log area allows a kernel to cache new records and data
576 * in memory without allocating space in the actual filesystem to hold
577 * the records and data. In the event that a filesystem becomes full,
578 * any records remaining in memory can be flushed to the memory log
579 * area. This allows the kernel to immediately return success.
582 #define HAMMER_BOOT_MINBYTES (32*1024)
583 #define HAMMER_BOOT_NOMBYTES (64LL*1024*1024)
584 #define HAMMER_BOOT_MAXBYTES (256LL*1024*1024)
586 #define HAMMER_MEM_MINBYTES (256*1024)
587 #define HAMMER_MEM_NOMBYTES (1LL*1024*1024*1024)
588 #define HAMMER_MEM_MAXBYTES (64LL*1024*1024*1024)
590 struct hammer_volume_ondisk {
591 u_int64_t vol_signature;/* Signature */
593 int64_t vol_bot_beg; /* byte offset of boot area or 0 */
594 int64_t vol_mem_beg; /* byte offset of memory log or 0 */
595 int64_t vol_buf_beg; /* byte offset of first buffer in volume */
596 int64_t vol_buf_end; /* byte offset of volume EOF (on buf bndry) */
597 int64_t vol_locked; /* reserved clusters are >= this offset */
599 uuid_t vol_fsid; /* identify filesystem */
600 uuid_t vol_fstype; /* identify filesystem type */
601 char vol_name[64]; /* Name of volume */
603 int32_t vol_no; /* volume number within filesystem */
604 int32_t vol_count; /* number of volumes making up FS */
606 u_int32_t vol_version; /* version control information */
607 hammer_crc_t vol_crc; /* header crc */
608 u_int32_t vol_flags; /* volume flags */
609 u_int32_t vol_rootvol; /* which volume is the root volume? */
611 int32_t vol_reserved04;
612 int32_t vol_reserved05;
613 u_int32_t vol_reserved06;
614 u_int32_t vol_reserved07;
616 int32_t vol_blocksize; /* for statfs only */
617 int32_t vol_reserved08;
618 int64_t vol_nblocks; /* total allocatable hammer bufs */
621 * These fields are initialized and space is reserved in every
622 * volume making up a HAMMER filesytem, but only the master volume
623 * contains valid data. Note that vol0_stat_bigblocks does not
624 * include big-blocks for freemap and undomap initially allocated
627 int64_t vol0_stat_bigblocks; /* total big-blocks when fs is empty */
628 int64_t vol0_stat_freebigblocks;/* number of free big-blocks */
629 int64_t vol0_stat_bytes; /* for statfs only */
630 int64_t vol0_stat_inodes; /* for statfs only */
631 int64_t vol0_stat_records; /* total records in filesystem */
632 hammer_off_t vol0_btree_root; /* B-Tree root */
633 hammer_tid_t vol0_next_tid; /* highest partially synchronized TID */
634 hammer_off_t vol0_unused03;
637 * Blockmaps for zones. Not all zones use a blockmap. Note that
638 * the entire root blockmap is cached in the hammer_mount structure.
640 struct hammer_blockmap vol0_blockmap[HAMMER_MAX_ZONES];
643 * Array of zone-2 addresses for undo FIFO.
645 hammer_off_t vol0_undo_array[HAMMER_UNDO_LAYER2];
648 typedef struct hammer_volume_ondisk *hammer_volume_ondisk_t;
650 #define HAMMER_VOLF_VALID 0x0001 /* valid entry */
651 #define HAMMER_VOLF_OPEN 0x0002 /* volume is open */
652 #define HAMMER_VOLF_NEEDFLUSH 0x0004 /* volume needs flush */
654 #define HAMMER_VOL_CRCSIZE1 \
655 offsetof(struct hammer_volume_ondisk, vol_crc)
656 #define HAMMER_VOL_CRCSIZE2 \
657 (sizeof(struct hammer_volume_ondisk) - HAMMER_VOL_CRCSIZE1 - \
658 sizeof(hammer_crc_t))
660 #define HAMMER_VOL_VERSION_MIN 1 /* minimum supported version */
661 #define HAMMER_VOL_VERSION_DEFAULT 6 /* newfs default version */
662 #define HAMMER_VOL_VERSION_WIP 7 /* version >= this is WIP */
663 #define HAMMER_VOL_VERSION_MAX 6 /* maximum supported version */
665 #define HAMMER_VOL_VERSION_ONE 1
666 #define HAMMER_VOL_VERSION_TWO 2 /* new dirent layout (2.3+) */
667 #define HAMMER_VOL_VERSION_THREE 3 /* new snapshot layout (2.5+) */
668 #define HAMMER_VOL_VERSION_FOUR 4 /* new undo/flush (2.5+) */
669 #define HAMMER_VOL_VERSION_FIVE 5 /* dedup (2.9+) */
670 #define HAMMER_VOL_VERSION_SIX 6 /* DIRHASH_ALG1 */
673 * Record types are fairly straightforward. The B-Tree includes the record
674 * type in its index sort.
676 #define HAMMER_RECTYPE_UNKNOWN 0
677 #define HAMMER_RECTYPE_LOWEST 1 /* lowest record type avail */
678 #define HAMMER_RECTYPE_INODE 1 /* inode in obj_id space */
679 #define HAMMER_RECTYPE_UNUSED02 2
680 #define HAMMER_RECTYPE_UNUSED03 3
681 #define HAMMER_RECTYPE_DATA 0x0010
682 #define HAMMER_RECTYPE_DIRENTRY 0x0011
683 #define HAMMER_RECTYPE_DB 0x0012
684 #define HAMMER_RECTYPE_EXT 0x0013 /* ext attributes */
685 #define HAMMER_RECTYPE_FIX 0x0014 /* fixed attribute */
686 #define HAMMER_RECTYPE_PFS 0x0015 /* PFS management */
687 #define HAMMER_RECTYPE_SNAPSHOT 0x0016 /* Snapshot management */
688 #define HAMMER_RECTYPE_CONFIG 0x0017 /* hammer cleanup config */
689 #define HAMMER_RECTYPE_MAX 0xFFFF
691 #define HAMMER_RECTYPE_ENTRY_START (HAMMER_RECTYPE_INODE + 1)
692 #define HAMMER_RECTYPE_CLEAN_START HAMMER_RECTYPE_EXT
694 #define HAMMER_FIXKEY_SYMLINK 1
696 #define HAMMER_OBJTYPE_UNKNOWN 0 /* never exists on-disk as unknown */
697 #define HAMMER_OBJTYPE_DIRECTORY 1
698 #define HAMMER_OBJTYPE_REGFILE 2
699 #define HAMMER_OBJTYPE_DBFILE 3
700 #define HAMMER_OBJTYPE_FIFO 4
701 #define HAMMER_OBJTYPE_CDEV 5
702 #define HAMMER_OBJTYPE_BDEV 6
703 #define HAMMER_OBJTYPE_SOFTLINK 7
704 #define HAMMER_OBJTYPE_PSEUDOFS 8 /* pseudo filesystem obj */
705 #define HAMMER_OBJTYPE_SOCKET 9
708 * HAMMER inode attribute data
710 * The data reference for a HAMMER inode points to this structure. Any
711 * modifications to the contents of this structure will result in a
712 * replacement operation.
714 * parent_obj_id is only valid for directories (which cannot be hard-linked),
715 * and specifies the parent directory obj_id. This field will also be set
716 * for non-directory inodes as a recovery aid, but can wind up holding
717 * stale information. However, since object id's are not reused, the worse
718 * that happens is that the recovery code is unable to use it.
720 * NOTE: Future note on directory hardlinks. We can implement a record type
721 * which allows us to point to multiple parent directories.
723 struct hammer_inode_data {
724 u_int16_t version; /* inode data version */
725 u_int16_t mode; /* basic unix permissions */
726 u_int32_t uflags; /* chflags */
727 u_int32_t rmajor; /* used by device nodes */
728 u_int32_t rminor; /* used by device nodes */
730 int64_t parent_obj_id; /* parent directory obj_id */
735 u_int8_t cap_flags; /* capability support flags (extension) */
736 u_int16_t reserved02;
737 u_int32_t reserved03; /* RESERVED FOR POSSIBLE FUTURE BIRTHTIME */
738 u_int64_t nlinks; /* hard links */
739 u_int64_t size; /* filesystem object size */
743 u_int32_t parent_obj_localization;
744 u_int32_t integrity_crc;
746 char symlink[24]; /* HAMMER_INODE_BASESYMLEN */
748 u_int64_t mtime; /* mtime must be second-to-last */
749 u_int64_t atime; /* atime must be last */
753 * Neither mtime nor atime upates are CRCd by the B-Tree element.
754 * mtime updates have UNDO, atime updates do not.
756 #define HAMMER_ITIMES_BASE(ino_data) (&(ino_data)->mtime)
757 #define HAMMER_ITIMES_BYTES (sizeof(u_int64_t) * 2)
759 #define HAMMER_INODE_CRCSIZE \
760 offsetof(struct hammer_inode_data, mtime)
762 #define HAMMER_INODE_DATA_VERSION 1
763 #define HAMMER_OBJID_ROOT 1 /* root inodes # */
764 #define HAMMER_INODE_BASESYMLEN 24 /* see ext.symlink */
767 * Capability & implementation flags.
769 * DIR_LOCAL_INO - Use inode B-Tree localization for directory entries.
771 #define HAMMER_INODE_CAP_DIRHASH_MASK 0x03 /* directory: hash algorithm */
772 #define HAMMER_INODE_CAP_DIRHASH_ALG0 0x00
773 #define HAMMER_INODE_CAP_DIRHASH_ALG1 0x01
774 #define HAMMER_INODE_CAP_DIRHASH_ALG2 0x02
775 #define HAMMER_INODE_CAP_DIRHASH_ALG3 0x03
776 #define HAMMER_INODE_CAP_DIR_LOCAL_INO 0x04 /* use inode localization */
779 * A HAMMER directory entry associates a HAMMER filesystem object with a
780 * namespace. It is possible to hook into a pseudo-filesystem (with its
781 * own inode numbering space) in the filesystem by setting the high
782 * 16 bits of the localization field. The low 16 bits must be 0 and
783 * are reserved for future use.
785 * Directory entries are indexed with a 128 bit namekey rather then an
786 * offset. A portion of the namekey is an iterator/randomizer to deal
789 * NOTE: leaf.base.obj_type from the related B-Tree leaf entry holds
790 * the filesystem object type of obj_id, e.g. a den_type equivalent.
791 * It is not stored in hammer_entry_data.
793 * NOTE: name field / the filename data reference is NOT terminated with \0.
795 struct hammer_entry_data {
796 int64_t obj_id; /* object being referenced */
797 u_int32_t localization; /* identify pseudo-filesystem */
798 u_int32_t reserved02;
799 char name[16]; /* name (extended) */
802 #define HAMMER_ENTRY_NAME_OFF offsetof(struct hammer_entry_data, name[0])
803 #define HAMMER_ENTRY_SIZE(nlen) offsetof(struct hammer_entry_data, name[nlen])
806 * Symlink data which does not fit in the inode is stored in a separate
809 struct hammer_symlink_data {
810 char name[16]; /* name (extended) */
813 #define HAMMER_SYMLINK_NAME_OFF offsetof(struct hammer_symlink_data, name[0])
816 * The root inode for the primary filesystem and root inode for any
817 * pseudo-fs may be tagged with an optional data structure using
818 * HAMMER_RECTYPE_PFS and localization id. This structure allows
819 * the node to be used as a mirroring master or slave.
821 * When operating as a slave CD's into the node automatically become read-only
822 * and as-of sync_end_tid.
824 * When operating as a master the read PFSD info sets sync_end_tid to
825 * the most recently flushed TID.
827 * sync_low_tid is not yet used but will represent the highest pruning
828 * end-point, after which full history is available.
830 * We need to pack this structure making it equally sized on both 32-bit and
831 * 64-bit machines as it is part of struct hammer_ioc_mrecord_pfs which is
832 * send over the wire in hammer mirror operations. Only on 64-bit machines
833 * the size of this struct differ when packed or not. This leads us to the
834 * situation where old 64-bit systems (using the non-packed structure),
835 * which were never able to mirror to/from 32-bit systems, are now no longer
836 * able to mirror to/from newer 64-bit systems (using the packed structure).
838 struct hammer_pseudofs_data {
839 hammer_tid_t sync_low_tid; /* full history beyond this point */
840 hammer_tid_t sync_beg_tid; /* earliest tid w/ full history avail */
841 hammer_tid_t sync_end_tid; /* current synchronizatoin point */
842 u_int64_t sync_beg_ts; /* real-time of last completed sync */
843 u_int64_t sync_end_ts; /* initiation of current sync cycle */
844 uuid_t shared_uuid; /* shared uuid (match required) */
845 uuid_t unique_uuid; /* unique uuid of this master/slave */
846 int32_t reserved01; /* reserved for future master_id */
847 int32_t mirror_flags; /* misc flags */
848 char label[64]; /* filesystem space label */
849 char snapshots[64]; /* softlink dir for pruning */
850 int16_t prune_time; /* how long to spend pruning */
851 int16_t prune_freq; /* how often we prune */
852 int16_t reblock_time; /* how long to spend reblocking */
853 int16_t reblock_freq; /* how often we reblock */
854 int32_t snapshot_freq; /* how often we create a snapshot */
855 int32_t prune_min; /* do not prune recent history */
856 int32_t prune_max; /* do not retain history beyond here */
857 int32_t reserved[16];
860 typedef struct hammer_pseudofs_data *hammer_pseudofs_data_t;
862 #define HAMMER_PFSD_SLAVE 0x00000001
863 #define HAMMER_PFSD_DELETED 0x80000000
866 * Snapshot meta-data { Objid = HAMMER_OBJID_ROOT, Key = tid, rectype = SNAPSHOT }.
868 * Snapshot records replace the old <fs>/snapshots/<softlink> methodology. Snapshot
869 * records are mirrored but may be independantly managed once they are laid down on
872 * NOTE: The b-tree key is signed, the tid is not, so callers must still sort the
875 * NOTE: Reserved fields must be zero (as usual)
877 struct hammer_snapshot_data {
878 hammer_tid_t tid; /* the snapshot TID itself (== key) */
879 u_int64_t ts; /* real-time when snapshot was made */
880 u_int64_t reserved01;
881 u_int64_t reserved02;
882 char label[64]; /* user-supplied description */
883 u_int64_t reserved03[4];
887 * Config meta-data { ObjId = HAMMER_OBJID_ROOT, Key = 0, rectype = CONFIG }.
889 * Used to store the hammer cleanup config. This data is not mirrored.
891 struct hammer_config_data {
896 * Rollup various structures embedded as record data
898 union hammer_data_ondisk {
899 struct hammer_entry_data entry;
900 struct hammer_inode_data inode;
901 struct hammer_symlink_data symlink;
902 struct hammer_pseudofs_data pfsd;
903 struct hammer_snapshot_data snap;
904 struct hammer_config_data config;
907 typedef union hammer_data_ondisk *hammer_data_ondisk_t;
910 * Ondisk layout of B-Tree related structures
912 #include "hammer_btree.h"