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: 59 bits 512 KTB (due to blockmap)
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 */
90 #define HAMMER_OFF_SHORT_REC_MASK 0x000FFFFFFF000000ULL /* recovery boundary */
91 #define HAMMER_OFF_LONG_REC_MASK 0x0FFFFFFFFF000000ULL /* recovery boundary */
92 #define HAMMER_RECOVERY_BND 0x0000000001000000ULL
94 #define HAMMER_OFF_BAD ((hammer_off_t)-1)
97 * The current limit of volumes that can make up a HAMMER FS
99 #define HAMMER_MAX_VOLUMES 256
102 * Hammer transction ids are 64 bit unsigned integers and are usually
103 * synchronized with the time of day in nanoseconds.
105 * Hammer offsets are used for FIFO indexing and embed a cycle counter
106 * and volume number in addition to the offset. Most offsets are required
107 * to be 64-byte aligned.
109 typedef u_int64_t hammer_tid_t;
110 typedef u_int64_t hammer_off_t;
111 typedef u_int32_t hammer_seq_t;
112 typedef u_int32_t hammer_crc_t;
114 #define HAMMER_MIN_TID 0ULL /* unsigned */
115 #define HAMMER_MAX_TID 0xFFFFFFFFFFFFFFFFULL /* unsigned */
116 #define HAMMER_MIN_KEY -0x8000000000000000LL /* signed */
117 #define HAMMER_MAX_KEY 0x7FFFFFFFFFFFFFFFLL /* signed */
118 #define HAMMER_MIN_OBJID HAMMER_MIN_KEY /* signed */
119 #define HAMMER_MAX_OBJID HAMMER_MAX_KEY /* signed */
120 #define HAMMER_MIN_RECTYPE 0x0U /* unsigned */
121 #define HAMMER_MAX_RECTYPE 0xFFFFU /* unsigned */
122 #define HAMMER_MIN_OFFSET 0ULL /* unsigned */
123 #define HAMMER_MAX_OFFSET 0xFFFFFFFFFFFFFFFFULL /* unsigned */
126 * hammer_off_t has several different encodings. Note that not all zones
129 * zone 0: reserved for sanity
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 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): Record - actually zone-2 address
136 * zone 10 (z,v,o): Large-data - actually zone-2 address
137 * zone 15: reserved for sanity
139 * layer1/layer2 direct map:
140 * zzzzvvvvvvvvoooo oooooooooooooooo oooooooooooooooo oooooooooooooooo
141 * ----111111111111 1111112222222222 222222222ooooooo oooooooooooooooo
144 #define HAMMER_ZONE_RAW_VOLUME 0x1000000000000000ULL
145 #define HAMMER_ZONE_RAW_BUFFER 0x2000000000000000ULL
146 #define HAMMER_ZONE_UNDO 0x3000000000000000ULL
147 #define HAMMER_ZONE_FREEMAP 0x4000000000000000ULL
148 #define HAMMER_ZONE_RESERVED05 0x5000000000000000ULL
149 #define HAMMER_ZONE_RESERVED06 0x6000000000000000ULL
150 #define HAMMER_ZONE_RESERVED07 0x7000000000000000ULL
151 #define HAMMER_ZONE_BTREE 0x8000000000000000ULL
152 #define HAMMER_ZONE_META 0x9000000000000000ULL
153 #define HAMMER_ZONE_LARGE_DATA 0xA000000000000000ULL
154 #define HAMMER_ZONE_SMALL_DATA 0xB000000000000000ULL
155 #define HAMMER_ZONE_RESERVED0C 0xC000000000000000ULL
156 #define HAMMER_ZONE_RESERVED0D 0xD000000000000000ULL
157 #define HAMMER_ZONE_RESERVED0E 0xE000000000000000ULL
158 #define HAMMER_ZONE_UNAVAIL 0xF000000000000000ULL
160 #define HAMMER_ZONE_RAW_VOLUME_INDEX 1
161 #define HAMMER_ZONE_RAW_BUFFER_INDEX 2
162 #define HAMMER_ZONE_UNDO_INDEX 3
163 #define HAMMER_ZONE_FREEMAP_INDEX 4
164 #define HAMMER_ZONE_BTREE_INDEX 8
165 #define HAMMER_ZONE_META_INDEX 9
166 #define HAMMER_ZONE_LARGE_DATA_INDEX 10
167 #define HAMMER_ZONE_SMALL_DATA_INDEX 11
168 #define HAMMER_ZONE_UNAVAIL_INDEX 15 /* unavailable */
170 #define HAMMER_MAX_ZONES 16
172 #define HAMMER_VOL_ENCODE(vol_no) \
173 ((hammer_off_t)((vol_no) & 255) << 52)
174 #define HAMMER_VOL_DECODE(ham_off) \
175 (int32_t)(((hammer_off_t)(ham_off) >> 52) & 255)
176 #define HAMMER_ZONE_DECODE(ham_off) \
177 (int32_t)(((hammer_off_t)(ham_off) >> 60))
178 #define HAMMER_ZONE_ENCODE(zone, ham_off) \
179 (((hammer_off_t)(zone) << 60) | (ham_off))
180 #define HAMMER_SHORT_OFF_ENCODE(offset) \
181 ((hammer_off_t)(offset) & HAMMER_OFF_SHORT_MASK)
182 #define HAMMER_LONG_OFF_ENCODE(offset) \
183 ((hammer_off_t)(offset) & HAMMER_OFF_LONG_MASK)
185 #define HAMMER_ENCODE_RAW_VOLUME(vol_no, offset) \
186 (HAMMER_ZONE_RAW_VOLUME | \
187 HAMMER_VOL_ENCODE(vol_no) | \
188 HAMMER_SHORT_OFF_ENCODE(offset))
190 #define HAMMER_ENCODE_RAW_BUFFER(vol_no, offset) \
191 (HAMMER_ZONE_RAW_BUFFER | \
192 HAMMER_VOL_ENCODE(vol_no) | \
193 HAMMER_SHORT_OFF_ENCODE(offset))
195 #define HAMMER_ENCODE_FREEMAP(vol_no, offset) \
196 (HAMMER_ZONE_FREEMAP | \
197 HAMMER_VOL_ENCODE(vol_no) | \
198 HAMMER_SHORT_OFF_ENCODE(offset))
201 * Large-Block backing store
203 * A blockmap is a two-level map which translates a blockmap-backed zone
204 * offset into a raw zone 2 offset. Each layer handles 18 bits. The 8M
205 * large-block size is 23 bits so two layers gives us 23+18+18 = 59 bits
208 * When using hinting for a blockmap lookup, the hint is lost when the
209 * scan leaves the HINTBLOCK, which is typically several LARGEBLOCK's.
210 * HINTBLOCK is a heuristic.
212 #define HAMMER_HINTBLOCK_SIZE (HAMMER_LARGEBLOCK_SIZE * 4)
213 #define HAMMER_HINTBLOCK_MASK64 ((u_int64_t)HAMMER_HINTBLOCK_SIZE - 1)
214 #define HAMMER_LARGEBLOCK_SIZE (8192 * 1024)
215 #define HAMMER_LARGEBLOCK_OVERFILL (6144 * 1024)
216 #define HAMMER_LARGEBLOCK_SIZE64 ((u_int64_t)HAMMER_LARGEBLOCK_SIZE)
217 #define HAMMER_LARGEBLOCK_MASK (HAMMER_LARGEBLOCK_SIZE - 1)
218 #define HAMMER_LARGEBLOCK_MASK64 ((u_int64_t)HAMMER_LARGEBLOCK_SIZE - 1)
219 #define HAMMER_LARGEBLOCK_BITS 23
220 #if (1 << HAMMER_LARGEBLOCK_BITS) != HAMMER_LARGEBLOCK_SIZE
221 #error "HAMMER_LARGEBLOCK_BITS BROKEN"
224 #define HAMMER_BUFFERS_PER_LARGEBLOCK \
225 (HAMMER_LARGEBLOCK_SIZE / HAMMER_BUFSIZE)
226 #define HAMMER_BUFFERS_PER_LARGEBLOCK_MASK \
227 (HAMMER_BUFFERS_PER_LARGEBLOCK - 1)
228 #define HAMMER_BUFFERS_PER_LARGEBLOCK_MASK64 \
229 ((hammer_off_t)HAMMER_BUFFERS_PER_LARGEBLOCK_MASK)
232 * Maximum number of mirrors operating in master mode (multi-master
233 * clustering and mirroring).
235 #define HAMMER_MAX_MASTERS 16
238 * The blockmap is somewhat of a degenerate structure. HAMMER only actually
239 * uses it in its original incarnation to implement the free-map.
241 * zone:1 raw volume (no blockmap)
242 * zone:2 raw buffer (no blockmap)
243 * zone:3 undo-map (direct layer2 array in volume header)
244 * zone:4 free-map (the only real blockmap)
245 * zone:8-15 zone id used to classify big-block only, address is actually
248 struct hammer_blockmap {
249 hammer_off_t phys_offset; /* zone-2 physical offset */
250 hammer_off_t first_offset; /* zone-X logical offset (zone 3) */
251 hammer_off_t next_offset; /* zone-X logical offset */
252 hammer_off_t alloc_offset; /* zone-X logical offset */
253 u_int32_t reserved01;
254 hammer_crc_t entry_crc;
257 typedef struct hammer_blockmap *hammer_blockmap_t;
259 #define HAMMER_BLOCKMAP_CRCSIZE \
260 offsetof(struct hammer_blockmap, entry_crc)
263 * The blockmap is a 2-layer entity made up of big-blocks. The first layer
264 * contains 262144 32-byte entries (18 bits), the second layer contains
265 * 524288 16-byte entries (19 bits), representing 8MB (23 bit) blockmaps.
266 * 18+19+23 = 60 bits. The top four bits are the zone id.
268 * Currently only the freemap utilizes both layers in all their glory.
269 * All primary data/meta-data zones actually encode a zone-2 address
270 * requiring no real blockmap translation.
272 * The freemap uses the upper 8 bits of layer-1 to identify the volume,
273 * thus any space allocated via the freemap can be directly translated
274 * to a zone:2 (or zone:8-15) address.
276 * zone-X blockmap offset: [z:4][layer1:18][layer2:19][bigblock:23]
278 struct hammer_blockmap_layer1 {
279 hammer_off_t blocks_free; /* big-blocks free */
280 hammer_off_t phys_offset; /* UNAVAIL or zone-2 */
281 hammer_off_t reserved01;
282 hammer_crc_t layer2_crc; /* xor'd crc's of HAMMER_BLOCKSIZE */
284 hammer_crc_t layer1_crc; /* MUST BE LAST FIELD OF STRUCTURE*/
287 typedef struct hammer_blockmap_layer1 *hammer_blockmap_layer1_t;
289 #define HAMMER_LAYER1_CRCSIZE \
290 offsetof(struct hammer_blockmap_layer1, layer1_crc)
293 * layer2 entry for 8MB bigblock.
295 * NOTE: bytes_free is signed and can legally go negative if/when data
296 * de-dup occurs. This field will never go higher than
297 * HAMMER_LARGEBLOCK_SIZE. If exactly HAMMER_LARGEBLOCK_SIZE
298 * the big-block is completely free.
300 struct hammer_blockmap_layer2 {
301 u_int8_t zone; /* typed allocation zone */
304 u_int32_t append_off; /* allocatable space index */
305 int32_t bytes_free; /* bytes free within this bigblock */
306 hammer_crc_t entry_crc;
309 typedef struct hammer_blockmap_layer2 *hammer_blockmap_layer2_t;
311 #define HAMMER_LAYER2_CRCSIZE \
312 offsetof(struct hammer_blockmap_layer2, entry_crc)
314 #define HAMMER_BLOCKMAP_FREE 0ULL
315 #define HAMMER_BLOCKMAP_UNAVAIL ((hammer_off_t)-1LL)
317 #define HAMMER_BLOCKMAP_RADIX1 /* 262144 (18) */ \
318 (HAMMER_LARGEBLOCK_SIZE / sizeof(struct hammer_blockmap_layer1))
319 #define HAMMER_BLOCKMAP_RADIX2 /* 524288 (19) */ \
320 (HAMMER_LARGEBLOCK_SIZE / sizeof(struct hammer_blockmap_layer2))
322 #define HAMMER_BLOCKMAP_RADIX1_PERBUFFER \
323 (HAMMER_BLOCKMAP_RADIX1 / (HAMMER_LARGEBLOCK_SIZE / HAMMER_BUFSIZE))
324 #define HAMMER_BLOCKMAP_RADIX2_PERBUFFER \
325 (HAMMER_BLOCKMAP_RADIX2 / (HAMMER_LARGEBLOCK_SIZE / HAMMER_BUFSIZE))
327 #define HAMMER_BLOCKMAP_LAYER1 /* 18+19+23 */ \
328 (HAMMER_BLOCKMAP_RADIX1 * HAMMER_BLOCKMAP_LAYER2)
329 #define HAMMER_BLOCKMAP_LAYER2 /* 19+23 - 4TB */ \
330 (HAMMER_BLOCKMAP_RADIX2 * HAMMER_LARGEBLOCK_SIZE64)
332 #define HAMMER_BLOCKMAP_LAYER1_MASK (HAMMER_BLOCKMAP_LAYER1 - 1)
333 #define HAMMER_BLOCKMAP_LAYER2_MASK (HAMMER_BLOCKMAP_LAYER2 - 1)
336 * byte offset within layer1 or layer2 big-block for the entry representing
337 * a zone-2 physical offset.
339 #define HAMMER_BLOCKMAP_LAYER1_OFFSET(zone2_offset) \
340 (((zone2_offset) & HAMMER_BLOCKMAP_LAYER1_MASK) / \
341 HAMMER_BLOCKMAP_LAYER2 * sizeof(struct hammer_blockmap_layer1))
343 #define HAMMER_BLOCKMAP_LAYER2_OFFSET(zone2_offset) \
344 (((zone2_offset) & HAMMER_BLOCKMAP_LAYER2_MASK) / \
345 HAMMER_LARGEBLOCK_SIZE64 * sizeof(struct hammer_blockmap_layer2))
348 * HAMMER UNDO parameters. The UNDO fifo is mapped directly in the volume
349 * header with an array of layer2 structures. A maximum of (128x8MB) = 1GB
350 * may be reserved. The size of the undo fifo is usually set a newfs time
351 * but can be adjusted if the filesystem is taken offline.
353 #define HAMMER_UNDO_LAYER2 128 /* max layer2 undo mapping entries */
356 * All on-disk HAMMER structures which make up elements of the UNDO FIFO
357 * contain a hammer_fifo_head and hammer_fifo_tail structure. This structure
358 * contains all the information required to validate the fifo element
359 * and to scan the fifo in either direction. The head is typically embedded
360 * in higher level hammer on-disk structures while the tail is typically
361 * out-of-band. hdr_size is the size of the whole mess, including the tail.
363 * All undo structures are guaranteed to not cross a 16K filesystem
364 * buffer boundary. Most undo structures are fairly small. Data spaces
365 * are not immediately reused by HAMMER so file data is not usually recorded
366 * as part of an UNDO.
368 * PAD elements are allowed to take up only 8 bytes of space as a special
369 * case, containing only hdr_signature, hdr_type, and hdr_size fields,
370 * and with the tail overloaded onto the head structure for 8 bytes total.
372 * Every undo record has a sequence number. This number is unrelated to
373 * transaction ids and instead collects the undo transactions associated
374 * with a single atomic operation. A larger transactional operation, such
375 * as a remove(), may consist of several smaller atomic operations
376 * representing raw meta-data operations.
378 * HAMMER VERSION 4 CHANGES
380 * In HAMMER version 4 the undo structure alignment is reduced from 16384
381 * to 512 bytes in order to ensure that each 512 byte sector begins with
382 * a header. The reserved01 field in the header is now a 32 bit sequence
383 * number. This allows the recovery code to detect missing sectors
384 * without relying on the 32-bit crc and to definitively identify the current
385 * undo sequence space without having to rely on information from the volume
386 * header. In addition, new REDO entries in the undo space are used to
387 * record write, write/extend, and transaction id updates.
389 * The grand result is:
391 * (1) The volume header no longer needs to be synchronized for most
392 * flush and fsync operations.
394 * (2) Most fsync operations need only lay down REDO records
396 * (3) Data overwrite for nohistory operations covered by REDO records
397 * can be supported (instead of rolling a new block allocation),
398 * by rolling UNDO for the prior contents of the data.
400 #define HAMMER_HEAD_ONDISK_SIZE 32
401 #define HAMMER_HEAD_ALIGN 8
402 #define HAMMER_HEAD_ALIGN_MASK (HAMMER_HEAD_ALIGN - 1)
403 #define HAMMER_TAIL_ONDISK_SIZE 8
404 #define HAMMER_HEAD_DOALIGN(bytes) \
405 (((bytes) + HAMMER_HEAD_ALIGN_MASK) & ~HAMMER_HEAD_ALIGN_MASK)
407 #define HAMMER_UNDO_ALIGN 512
408 #define HAMMER_UNDO_ALIGN64 ((u_int64_t)512)
409 #define HAMMER_UNDO_MASK (HAMMER_UNDO_ALIGN - 1)
410 #define HAMMER_UNDO_MASK64 (HAMMER_UNDO_ALIGN64 - 1)
412 struct hammer_fifo_head {
413 u_int16_t hdr_signature;
415 u_int32_t hdr_size; /* Aligned size of the whole mess */
416 u_int32_t hdr_seq; /* Sequence number */
417 hammer_crc_t hdr_crc; /* XOR crc up to field w/ crc after field */
420 #define HAMMER_FIFO_HEAD_CRCOFF offsetof(struct hammer_fifo_head, hdr_crc)
422 struct hammer_fifo_tail {
423 u_int16_t tail_signature;
425 u_int32_t tail_size; /* aligned size of the whole mess */
428 typedef struct hammer_fifo_head *hammer_fifo_head_t;
429 typedef struct hammer_fifo_tail *hammer_fifo_tail_t;
434 #define HAMMER_HEAD_TYPE_PAD (0x0040U|HAMMER_HEAD_FLAG_FREE)
435 #define HAMMER_HEAD_TYPE_DUMMY 0x0041U /* dummy entry w/seqno */
436 #define HAMMER_HEAD_TYPE_42 0x0042U
437 #define HAMMER_HEAD_TYPE_UNDO 0x0043U /* random UNDO information */
438 #define HAMMER_HEAD_TYPE_REDO 0x0044U /* data REDO / fast fsync */
439 #define HAMMER_HEAD_TYPE_45 0x0045U
441 #define HAMMER_HEAD_FLAG_FREE 0x8000U /* Indicates object freed */
443 #define HAMMER_HEAD_SIGNATURE 0xC84EU
444 #define HAMMER_TAIL_SIGNATURE 0xC74FU
446 #define HAMMER_HEAD_SEQ_BEG 0x80000000U
447 #define HAMMER_HEAD_SEQ_END 0x40000000U
448 #define HAMMER_HEAD_SEQ_MASK 0x3FFFFFFFU
451 * Misc FIFO structures.
453 * UNDO - Raw meta-data media updates.
455 struct hammer_fifo_undo {
456 struct hammer_fifo_head head;
457 hammer_off_t undo_offset; /* zone-1 offset */
458 int32_t undo_data_bytes;
459 int32_t undo_reserved01;
460 /* followed by data */
464 * REDO (HAMMER version 4+) - Logical file writes/truncates.
466 * REDOs contain information which will be duplicated in a later meta-data
467 * update, allowing fast write()+fsync() operations. REDOs can be ignored
468 * without harming filesystem integrity but must be processed if fsync()
469 * semantics are desired.
471 * Unlike UNDOs which are processed backwards within the recovery span,
472 * REDOs must be processed forwards starting further back (starting outside
473 * the recovery span).
475 * WRITE - Write logical file (with payload). Executed both
476 * out-of-span and in-span. Out-of-span WRITEs may be
477 * filtered out by TERMs.
479 * TRUNC - Truncate logical file (no payload). Executed both
480 * out-of-span and in-span. Out-of-span WRITEs may be
481 * filtered out by TERMs.
483 * TERM_* - Indicates meta-data was committed (if out-of-span) or
484 * will be rolled-back (in-span). Any out-of-span TERMs
485 * matching earlier WRITEs remove those WRITEs from
486 * consideration as they might conflict with a later data
487 * commit (which is not being rolled-back).
489 * SYNC - The earliest in-span SYNC (the last one when scanning
490 * backwards) tells the recovery code how far out-of-span
491 * it must go to run REDOs.
493 * NOTE: WRITEs do not always have matching TERMs even under
494 * perfect conditions because truncations might remove the
495 * buffers from consideration. I/O problems can also remove
496 * buffers from consideration.
498 * TRUNCSs do not always have matching TERMs because several
499 * truncations may be aggregated together into a single TERM.
501 struct hammer_fifo_redo {
502 struct hammer_fifo_head head;
503 int64_t redo_objid; /* file being written */
504 hammer_off_t redo_offset; /* logical offset in file */
505 int32_t redo_data_bytes;
506 u_int32_t redo_flags;
507 u_int32_t redo_localization;
508 u_int32_t redo_reserved;
509 u_int64_t redo_mtime; /* set mtime */
512 #define HAMMER_REDO_WRITE 0x00000001
513 #define HAMMER_REDO_TRUNC 0x00000002
514 #define HAMMER_REDO_TERM_WRITE 0x00000004
515 #define HAMMER_REDO_TERM_TRUNC 0x00000008
516 #define HAMMER_REDO_SYNC 0x00000010
518 union hammer_fifo_any {
519 struct hammer_fifo_head head;
520 struct hammer_fifo_undo undo;
521 struct hammer_fifo_redo redo;
524 typedef struct hammer_fifo_redo *hammer_fifo_redo_t;
525 typedef struct hammer_fifo_undo *hammer_fifo_undo_t;
526 typedef union hammer_fifo_any *hammer_fifo_any_t;
529 * Volume header types
531 #define HAMMER_FSBUF_VOLUME 0xC8414D4DC5523031ULL /* HAMMER01 */
532 #define HAMMER_FSBUF_VOLUME_REV 0x313052C54D4D41C8ULL /* (reverse endian) */
535 * The B-Tree structures need hammer_fsbuf_head.
537 #include "hammer_btree.h"
540 * HAMMER Volume header
542 * A HAMMER filesystem is built from any number of block devices, Each block
543 * device contains a volume header followed by however many buffers fit
546 * One of the volumes making up a HAMMER filesystem is the master, the
547 * rest are slaves. It does not have to be volume #0.
549 * The volume header takes up an entire 16K filesystem buffer and may
550 * represent up to 64KTB (65536 TB) of space.
552 * Special field notes:
554 * vol_bot_beg - offset of boot area (mem_beg - bot_beg bytes)
555 * vol_mem_beg - offset of memory log (clu_beg - mem_beg bytes)
556 * vol_buf_beg - offset of the first buffer.
558 * The memory log area allows a kernel to cache new records and data
559 * in memory without allocating space in the actual filesystem to hold
560 * the records and data. In the event that a filesystem becomes full,
561 * any records remaining in memory can be flushed to the memory log
562 * area. This allows the kernel to immediately return success.
565 #define HAMMER_BOOT_MINBYTES (32*1024)
566 #define HAMMER_BOOT_NOMBYTES (64LL*1024*1024)
567 #define HAMMER_BOOT_MAXBYTES (256LL*1024*1024)
569 #define HAMMER_MEM_MINBYTES (256*1024)
570 #define HAMMER_MEM_NOMBYTES (1LL*1024*1024*1024)
571 #define HAMMER_MEM_MAXBYTES (64LL*1024*1024*1024)
573 struct hammer_volume_ondisk {
574 u_int64_t vol_signature;/* Signature */
576 int64_t vol_bot_beg; /* byte offset of boot area or 0 */
577 int64_t vol_mem_beg; /* byte offset of memory log or 0 */
578 int64_t vol_buf_beg; /* byte offset of first buffer in volume */
579 int64_t vol_buf_end; /* byte offset of volume EOF (on buf bndry) */
580 int64_t vol_locked; /* reserved clusters are >= this offset */
582 uuid_t vol_fsid; /* identify filesystem */
583 uuid_t vol_fstype; /* identify filesystem type */
584 char vol_name[64]; /* Name of volume */
586 int32_t vol_no; /* volume number within filesystem */
587 int32_t vol_count; /* number of volumes making up FS */
589 u_int32_t vol_version; /* version control information */
590 hammer_crc_t vol_crc; /* header crc */
591 u_int32_t vol_flags; /* volume flags */
592 u_int32_t vol_rootvol; /* which volume is the root volume? */
594 int32_t vol_reserved04;
595 int32_t vol_reserved05;
596 u_int32_t vol_reserved06;
597 u_int32_t vol_reserved07;
599 int32_t vol_blocksize; /* for statfs only */
600 int32_t vol_reserved08;
601 int64_t vol_nblocks; /* total allocatable hammer bufs */
604 * These fields are initialized and space is reserved in every
605 * volume making up a HAMMER filesytem, but only the master volume
606 * contains valid data.
608 int64_t vol0_stat_bigblocks; /* total bigblocks when fs is empty */
609 int64_t vol0_stat_freebigblocks;/* number of free bigblocks */
610 int64_t vol0_stat_bytes; /* for statfs only */
611 int64_t vol0_stat_inodes; /* for statfs only */
612 int64_t vol0_stat_records; /* total records in filesystem */
613 hammer_off_t vol0_btree_root; /* B-Tree root */
614 hammer_tid_t vol0_next_tid; /* highest partially synchronized TID */
615 hammer_off_t vol0_unused03;
618 * Blockmaps for zones. Not all zones use a blockmap. Note that
619 * the entire root blockmap is cached in the hammer_mount structure.
621 struct hammer_blockmap vol0_blockmap[HAMMER_MAX_ZONES];
624 * Array of zone-2 addresses for undo FIFO.
626 hammer_off_t vol0_undo_array[HAMMER_UNDO_LAYER2];
630 typedef struct hammer_volume_ondisk *hammer_volume_ondisk_t;
632 #define HAMMER_VOLF_VALID 0x0001 /* valid entry */
633 #define HAMMER_VOLF_OPEN 0x0002 /* volume is open */
634 #define HAMMER_VOLF_NEEDFLUSH 0x0004 /* volume needs flush */
636 #define HAMMER_VOL_CRCSIZE1 \
637 offsetof(struct hammer_volume_ondisk, vol_crc)
638 #define HAMMER_VOL_CRCSIZE2 \
639 (sizeof(struct hammer_volume_ondisk) - HAMMER_VOL_CRCSIZE1 - \
640 sizeof(hammer_crc_t))
642 #define HAMMER_VOL_VERSION_MIN 1 /* minimum supported version */
643 #define HAMMER_VOL_VERSION_DEFAULT 4 /* newfs default version */
644 #define HAMMER_VOL_VERSION_WIP 5 /* version >= this is WIP */
645 #define HAMMER_VOL_VERSION_MAX 4 /* maximum supported version */
647 #define HAMMER_VOL_VERSION_ONE 1
648 #define HAMMER_VOL_VERSION_TWO 2 /* new dirent layout (2.3+) */
649 #define HAMMER_VOL_VERSION_THREE 3 /* new snapshot layout (2.5+) */
650 #define HAMMER_VOL_VERSION_FOUR 4 /* new undo/flush (2.5+) */
653 * Record types are fairly straightforward. The B-Tree includes the record
654 * type in its index sort.
656 #define HAMMER_RECTYPE_UNKNOWN 0
657 #define HAMMER_RECTYPE_LOWEST 1 /* lowest record type avail */
658 #define HAMMER_RECTYPE_INODE 1 /* inode in obj_id space */
659 #define HAMMER_RECTYPE_UNUSED02 2
660 #define HAMMER_RECTYPE_UNUSED03 3
661 #define HAMMER_RECTYPE_DATA 0x0010
662 #define HAMMER_RECTYPE_DIRENTRY 0x0011
663 #define HAMMER_RECTYPE_DB 0x0012
664 #define HAMMER_RECTYPE_EXT 0x0013 /* ext attributes */
665 #define HAMMER_RECTYPE_FIX 0x0014 /* fixed attribute */
666 #define HAMMER_RECTYPE_PFS 0x0015 /* PFS management */
667 #define HAMMER_RECTYPE_SNAPSHOT 0x0016 /* Snapshot management */
668 #define HAMMER_RECTYPE_CONFIG 0x0017 /* hammer cleanup config */
669 #define HAMMER_RECTYPE_MOVED 0x8000 /* special recovery flag */
670 #define HAMMER_RECTYPE_MAX 0xFFFF
672 #define HAMMER_RECTYPE_CLEAN_START HAMMER_RECTYPE_EXT
674 #define HAMMER_FIXKEY_SYMLINK 1
676 #define HAMMER_OBJTYPE_UNKNOWN 0 /* (never exists on-disk) */
677 #define HAMMER_OBJTYPE_DIRECTORY 1
678 #define HAMMER_OBJTYPE_REGFILE 2
679 #define HAMMER_OBJTYPE_DBFILE 3
680 #define HAMMER_OBJTYPE_FIFO 4
681 #define HAMMER_OBJTYPE_CDEV 5
682 #define HAMMER_OBJTYPE_BDEV 6
683 #define HAMMER_OBJTYPE_SOFTLINK 7
684 #define HAMMER_OBJTYPE_PSEUDOFS 8 /* pseudo filesystem obj */
685 #define HAMMER_OBJTYPE_SOCKET 9
688 * HAMMER inode attribute data
690 * The data reference for a HAMMER inode points to this structure. Any
691 * modifications to the contents of this structure will result in a
692 * replacement operation.
694 * parent_obj_id is only valid for directories (which cannot be hard-linked),
695 * and specifies the parent directory obj_id. This field will also be set
696 * for non-directory inodes as a recovery aid, but can wind up holding
697 * stale information. However, since object id's are not reused, the worse
698 * that happens is that the recovery code is unable to use it.
700 * NOTE: Future note on directory hardlinks. We can implement a record type
701 * which allows us to point to multiple parent directories.
703 * NOTE: atime is stored in the inode's B-Tree element and not in the inode
704 * data. This allows the atime to be updated without having to lay down a
707 struct hammer_inode_data {
708 u_int16_t version; /* inode data version */
709 u_int16_t mode; /* basic unix permissions */
710 u_int32_t uflags; /* chflags */
711 u_int32_t rmajor; /* used by device nodes */
712 u_int32_t rminor; /* used by device nodes */
714 int64_t parent_obj_id; /* parent directory obj_id */
719 u_int8_t cap_flags; /* capability support flags (extension) */
720 u_int16_t reserved02;
721 u_int32_t reserved03; /* RESERVED FOR POSSIBLE FUTURE BIRTHTIME */
722 u_int64_t nlinks; /* hard links */
723 u_int64_t size; /* filesystem object size */
727 u_int32_t parent_obj_localization;
728 u_int32_t integrity_crc;
730 char symlink[24]; /* HAMMER_INODE_BASESYMLEN */
732 u_int64_t mtime; /* mtime must be second-to-last */
733 u_int64_t atime; /* atime must be last */
737 * Neither mtime nor atime upates are CRCd by the B-Tree element.
738 * mtime updates have UNDO, atime updates do not.
740 #define HAMMER_ITIMES_BASE(ino_data) (&(ino_data)->mtime)
741 #define HAMMER_ITIMES_BYTES (sizeof(u_int64_t) * 2)
743 #define HAMMER_INODE_CRCSIZE \
744 offsetof(struct hammer_inode_data, mtime)
746 #define HAMMER_INODE_DATA_VERSION 1
747 #define HAMMER_OBJID_ROOT 1
748 #define HAMMER_INODE_BASESYMLEN 24 /* see ext.symlink */
751 * Capability & implementation flags.
753 * DIR_LOCAL_INO - Use inode B-Tree localization for directory entries.
755 #define HAMMER_INODE_CAP_DIRHASH_MASK 0x03 /* directory: hash algorithm */
756 #define HAMMER_INODE_CAP_DIRHASH_ALG0 0x00
757 #define HAMMER_INODE_CAP_DIRHASH_ALG1 0x01
758 #define HAMMER_INODE_CAP_DIRHASH_ALG2 0x02
759 #define HAMMER_INODE_CAP_DIRHASH_ALG3 0x03
760 #define HAMMER_INODE_CAP_DIR_LOCAL_INO 0x04 /* use inode localization */
763 * A HAMMER directory entry associates a HAMMER filesystem object with a
764 * namespace. It is possible to hook into a pseudo-filesystem (with its
765 * own inode numbering space) in the filesystem by setting the high
766 * 16 bits of the localization field. The low 16 bits must be 0 and
767 * are reserved for future use.
769 * Directory entries are indexed with a 128 bit namekey rather then an
770 * offset. A portion of the namekey is an iterator/randomizer to deal
773 * NOTE: base.base.obj_type from the related B-Tree leaf entry holds
774 * the filesystem object type of obj_id, e.g. a den_type equivalent.
775 * It is not stored in hammer_entry_data.
777 * NOTE: den_name / the filename data reference is NOT terminated with \0.
779 struct hammer_entry_data {
780 int64_t obj_id; /* object being referenced */
781 u_int32_t localization; /* identify pseudo-filesystem */
782 u_int32_t reserved02;
783 char name[16]; /* name (extended) */
786 #define HAMMER_ENTRY_NAME_OFF offsetof(struct hammer_entry_data, name[0])
787 #define HAMMER_ENTRY_SIZE(nlen) offsetof(struct hammer_entry_data, name[nlen])
790 * Symlink data which does not fit in the inode is stored in a separte
793 struct hammer_symlink_data {
797 #define HAMMER_SYMLINK_NAME_OFF offsetof(struct hammer_symlink_data, name[0])
800 * The root inode for the primary filesystem and root inode for any
801 * pseudo-fs may be tagged with an optional data structure using
802 * HAMMER_RECTYPE_FIX/HAMMER_FIXKEY_PSEUDOFS. This structure allows
803 * the node to be used as a mirroring master or slave.
805 * When operating as a slave CD's into the node automatically become read-only
806 * and as-of sync_end_tid.
808 * When operating as a master the read PFSD info sets sync_end_tid to
809 * the most recently flushed TID.
811 * sync_low_tid is not yet used but will represent the highest pruning
812 * end-point, after which full history is available.
814 * We need to pack this structure making it equally sized on both 32-bit and
815 * 64-bit machines as it is part of struct hammer_ioc_mrecord_pfs which is
816 * send over the wire in hammer mirror operations. Only on 64-bit machines
817 * the size of this struct differ when packed or not. This leads us to the
818 * situation where old 64-bit systems (using the non-packed structure),
819 * which were never able to mirror to/from 32-bit systems, are now no longer
820 * able to mirror to/from newer 64-bit systems (using the packed structure).
822 struct hammer_pseudofs_data {
823 hammer_tid_t sync_low_tid; /* full history beyond this point */
824 hammer_tid_t sync_beg_tid; /* earliest tid w/ full history avail */
825 hammer_tid_t sync_end_tid; /* current synchronizatoin point */
826 u_int64_t sync_beg_ts; /* real-time of last completed sync */
827 u_int64_t sync_end_ts; /* initiation of current sync cycle */
828 uuid_t shared_uuid; /* shared uuid (match required) */
829 uuid_t unique_uuid; /* unique uuid of this master/slave */
830 int32_t reserved01; /* reserved for future master_id */
831 int32_t mirror_flags; /* misc flags */
832 char label[64]; /* filesystem space label */
833 char snapshots[64]; /* softlink dir for pruning */
834 int16_t prune_time; /* how long to spend pruning */
835 int16_t prune_freq; /* how often we prune */
836 int16_t reblock_time; /* how long to spend reblocking */
837 int16_t reblock_freq; /* how often we reblock */
838 int32_t snapshot_freq; /* how often we create a snapshot */
839 int32_t prune_min; /* do not prune recent history */
840 int32_t prune_max; /* do not retain history beyond here */
841 int32_t reserved[16];
844 typedef struct hammer_pseudofs_data *hammer_pseudofs_data_t;
846 #define HAMMER_PFSD_SLAVE 0x00000001
847 #define HAMMER_PFSD_DELETED 0x80000000
850 * Snapshot meta-data { Objid = HAMMER_OBJID_ROOT, Key = tid, rectype = SNAPSHOT }.
852 * Snapshot records replace the old <fs>/snapshots/<softlink> methodology. Snapshot
853 * records are mirrored but may be independantly managed once they are laid down on
856 * NOTE: The b-tree key is signed, the tid is not, so callers must still sort the
859 * NOTE: Reserved fields must be zero (as usual)
861 struct hammer_snapshot_data {
862 hammer_tid_t tid; /* the snapshot TID itself (== key) */
863 u_int64_t ts; /* real-time when snapshot was made */
864 u_int64_t reserved01;
865 u_int64_t reserved02;
866 char label[64]; /* user-supplied description */
867 u_int64_t reserved03[4];
871 * Config meta-data { ObjId = HAMMER_OBJID_ROOT, Key = 0, rectype = CONFIG }.
873 * Used to store the hammer cleanup config. This data is not mirrored.
875 struct hammer_config_data {
880 * Rollup various structures embedded as record data
882 union hammer_data_ondisk {
883 struct hammer_entry_data entry;
884 struct hammer_inode_data inode;
885 struct hammer_symlink_data symlink;
886 struct hammer_pseudofs_data pfsd;
887 struct hammer_snapshot_data snap;
888 struct hammer_config_data config;
891 typedef union hammer_data_ondisk *hammer_data_ondisk_t;