HAMMER 56C/Many: Performance tuning - MEDIA STRUCTURES CHANGED!
[dragonfly.git] / sys / vfs / hammer / hammer_disk.h
CommitLineData
8750964d
MD
1/*
2 * Copyright (c) 2007 The DragonFly Project. All rights reserved.
3 *
4 * This code is derived from software contributed to The DragonFly Project
5 * by Matthew Dillon <dillon@backplane.com>
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 *
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
16 * distribution.
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.
20 *
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
32 * SUCH DAMAGE.
33 *
4a2796f3 34 * $DragonFly: src/sys/vfs/hammer/hammer_disk.h,v 1.39 2008/06/20 05:38:26 dillon Exp $
8750964d
MD
35 */
36
7dc57964
MD
37#ifndef VFS_HAMMER_DISK_H_
38#define VFS_HAMMER_DISK_H_
39
8750964d
MD
40#ifndef _SYS_UUID_H_
41#include <sys/uuid.h>
42#endif
43
44/*
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.
49 *
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.
55 *
00e31441 56 * A HAMMER filesystem may span multiple volumes.
8750964d
MD
57 *
58 * A HAMMER filesystem uses a 16K filesystem buffer size. All filesystem
c60bb2c5
MD
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.
40043e7f 62 *
4a2796f3
MD
63 * 64K X-bufs are used for blocks >= a file's 1MB mark.
64 *
40043e7f
MD
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
8750964d 68 */
40043e7f 69#define HAMMER_BUFSIZE 16384
4a2796f3
MD
70#define HAMMER_XBUFSIZE 65536
71#define HAMMER_XDEMARC (1024 * 1024)
40043e7f 72#define HAMMER_BUFMASK (HAMMER_BUFSIZE - 1)
4a2796f3 73#define HAMMER_XBUFMASK (HAMMER_XBUFSIZE - 1)
40043e7f
MD
74#define HAMMER_BUFFER_BITS 14
75
76#if (1 << HAMMER_BUFFER_BITS) != HAMMER_BUFSIZE
77#error "HAMMER_BUFFER_BITS BROKEN"
78#endif
8750964d 79
47197d71
MD
80#define HAMMER_BUFSIZE64 ((u_int64_t)HAMMER_BUFSIZE)
81#define HAMMER_BUFMASK64 ((u_int64_t)HAMMER_BUFMASK)
82
4a2796f3
MD
83#define HAMMER_XBUFSIZE64 ((u_int64_t)HAMMER_XBUFSIZE)
84#define HAMMER_XBUFMASK64 ((u_int64_t)HAMMER_XBUFMASK)
85
47197d71
MD
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
93
8750964d
MD
94/*
95 * Hammer transction ids are 64 bit unsigned integers and are usually
96 * synchronized with the time of day in nanoseconds.
47197d71
MD
97 *
98 * Hammer offsets are used for FIFO indexing and embed a cycle counter
99 * and volume number in addition to the offset. Most offsets are required
100 * to be 64-byte aligned.
8750964d
MD
101 */
102typedef u_int64_t hammer_tid_t;
47197d71 103typedef u_int64_t hammer_off_t;
059819e3
MD
104typedef u_int32_t hammer_seq_t;
105typedef u_int32_t hammer_crc_t;
47197d71
MD
106
107#define HAMMER_MIN_TID 0ULL /* unsigned */
108#define HAMMER_MAX_TID 0xFFFFFFFFFFFFFFFFULL /* unsigned */
109#define HAMMER_MIN_KEY -0x8000000000000000LL /* signed */
110#define HAMMER_MAX_KEY 0x7FFFFFFFFFFFFFFFLL /* signed */
111#define HAMMER_MIN_OBJID HAMMER_MIN_KEY /* signed */
112#define HAMMER_MAX_OBJID HAMMER_MAX_KEY /* signed */
113#define HAMMER_MIN_RECTYPE 0x0U /* unsigned */
114#define HAMMER_MAX_RECTYPE 0xFFFFU /* unsigned */
115#define HAMMER_MIN_OFFSET 0ULL /* unsigned */
116#define HAMMER_MAX_OFFSET 0xFFFFFFFFFFFFFFFFULL /* unsigned */
117
118/*
119 * hammer_off_t has several different encodings. Note that not all zones
120 * encode a vol_no.
121 *
cb51be26 122 * zone 0: reserved for sanity
47197d71
MD
123 * zone 1 (z,v,o): raw volume relative (offset 0 is the volume header)
124 * zone 2 (z,v,o): raw buffer relative (offset 0 is the first buffer)
cb51be26
MD
125 * zone 3 (z,o): undo fifo - actually fixed phys array in vol hdr
126 * zone 4 (z,v,o): freemap - only real blockmap
127 * zone 8 (z,v,o): B-Tree - actually zone-2 address
128 * zone 9 (z,v,o): Record - actually zone-2 address
129 * zone 10 (z,v,o): Large-data - actually zone-2 address
130 * zone 15: reserved for sanity
47197d71
MD
131 */
132
133#define HAMMER_ZONE_RAW_VOLUME 0x1000000000000000ULL
134#define HAMMER_ZONE_RAW_BUFFER 0x2000000000000000ULL
40043e7f 135#define HAMMER_ZONE_UNDO 0x3000000000000000ULL
c3be93f2 136#define HAMMER_ZONE_FREEMAP 0x4000000000000000ULL
40043e7f
MD
137#define HAMMER_ZONE_RESERVED05 0x5000000000000000ULL
138#define HAMMER_ZONE_RESERVED06 0x6000000000000000ULL
139#define HAMMER_ZONE_RESERVED07 0x7000000000000000ULL
140#define HAMMER_ZONE_BTREE 0x8000000000000000ULL
bf3b416b 141#define HAMMER_ZONE_META 0x9000000000000000ULL
40043e7f
MD
142#define HAMMER_ZONE_LARGE_DATA 0xA000000000000000ULL
143#define HAMMER_ZONE_SMALL_DATA 0xB000000000000000ULL
144#define HAMMER_ZONE_RESERVED0C 0xC000000000000000ULL
145#define HAMMER_ZONE_RESERVED0D 0xD000000000000000ULL
146#define HAMMER_ZONE_RESERVED0E 0xE000000000000000ULL
cb51be26 147#define HAMMER_ZONE_UNAVAIL 0xF000000000000000ULL
40043e7f
MD
148
149#define HAMMER_ZONE_RAW_VOLUME_INDEX 1
150#define HAMMER_ZONE_RAW_BUFFER_INDEX 2
151#define HAMMER_ZONE_UNDO_INDEX 3
c3be93f2 152#define HAMMER_ZONE_FREEMAP_INDEX 4
40043e7f 153#define HAMMER_ZONE_BTREE_INDEX 8
bf3b416b 154#define HAMMER_ZONE_META_INDEX 9
40043e7f
MD
155#define HAMMER_ZONE_LARGE_DATA_INDEX 10
156#define HAMMER_ZONE_SMALL_DATA_INDEX 11
cb51be26 157#define HAMMER_ZONE_UNAVAIL_INDEX 15 /* unavailable */
f03c9cf4 158
40043e7f 159#define HAMMER_MAX_ZONES 16
47197d71
MD
160
161#define HAMMER_VOL_ENCODE(vol_no) \
162 ((hammer_off_t)((vol_no) & 255) << 52)
163#define HAMMER_VOL_DECODE(ham_off) \
164 (int32_t)(((hammer_off_t)(ham_off) >> 52) & 255)
40043e7f
MD
165#define HAMMER_ZONE_DECODE(ham_off) \
166 (int32_t)(((hammer_off_t)(ham_off) >> 60))
f03c9cf4
MD
167#define HAMMER_ZONE_ENCODE(zone, ham_off) \
168 (((hammer_off_t)(zone) << 60) | (ham_off))
47197d71
MD
169#define HAMMER_SHORT_OFF_ENCODE(offset) \
170 ((hammer_off_t)(offset) & HAMMER_OFF_SHORT_MASK)
171#define HAMMER_LONG_OFF_ENCODE(offset) \
172 ((hammer_off_t)(offset) & HAMMER_OFF_LONG_MASK)
173
174#define HAMMER_ENCODE_RAW_VOLUME(vol_no, offset) \
175 (HAMMER_ZONE_RAW_VOLUME | \
176 HAMMER_VOL_ENCODE(vol_no) | \
177 HAMMER_SHORT_OFF_ENCODE(offset))
178
179#define HAMMER_ENCODE_RAW_BUFFER(vol_no, offset) \
180 (HAMMER_ZONE_RAW_BUFFER | \
181 HAMMER_VOL_ENCODE(vol_no) | \
182 HAMMER_SHORT_OFF_ENCODE(offset))
8750964d 183
c3be93f2
MD
184#define HAMMER_ENCODE_FREEMAP(vol_no, offset) \
185 (HAMMER_ZONE_FREEMAP | \
186 HAMMER_VOL_ENCODE(vol_no) | \
187 HAMMER_SHORT_OFF_ENCODE(offset))
188
40043e7f
MD
189/*
190 * Large-Block backing store
191 *
192 * A blockmap is a two-level map which translates a blockmap-backed zone
193 * offset into a raw zone 2 offset. Each layer handles 18 bits. The 8M
194 * large-block size is 23 bits so two layers gives us 23+18+18 = 59 bits
195 * of address space.
196 */
197#define HAMMER_LARGEBLOCK_SIZE (8192 * 1024)
c3be93f2 198#define HAMMER_LARGEBLOCK_SIZE64 ((u_int64_t)HAMMER_LARGEBLOCK_SIZE)
40043e7f
MD
199#define HAMMER_LARGEBLOCK_MASK (HAMMER_LARGEBLOCK_SIZE - 1)
200#define HAMMER_LARGEBLOCK_MASK64 ((u_int64_t)HAMMER_LARGEBLOCK_SIZE - 1)
201#define HAMMER_LARGEBLOCK_BITS 23
202#if (1 << HAMMER_LARGEBLOCK_BITS) != HAMMER_LARGEBLOCK_SIZE
203#error "HAMMER_LARGEBLOCK_BITS BROKEN"
204#endif
205
206#define HAMMER_BUFFERS_PER_LARGEBLOCK \
207 (HAMMER_LARGEBLOCK_SIZE / HAMMER_BUFSIZE)
208#define HAMMER_BUFFERS_PER_LARGEBLOCK_MASK \
209 (HAMMER_BUFFERS_PER_LARGEBLOCK - 1)
210#define HAMMER_BUFFERS_PER_LARGEBLOCK_MASK64 \
211 ((hammer_off_t)HAMMER_BUFFERS_PER_LARGEBLOCK_MASK)
212
c3be93f2 213/*
cb51be26
MD
214 * The blockmap is somewhat of a degenerate structure. HAMMER only actually
215 * uses it in its original incarnation to implement the free-map.
bf686dbe 216 *
cb51be26
MD
217 * zone:1 raw volume (no blockmap)
218 * zone:2 raw buffer (no blockmap)
219 * zone:3 undo-map (direct layer2 array in volume header)
220 * zone:4 free-map (the only real blockmap)
221 * zone:8-15 zone id used to classify big-block only, address is actually
222 * a zone-2 address.
c3be93f2
MD
223 */
224struct hammer_blockmap {
225 hammer_off_t phys_offset; /* zone-2 physical offset */
bf686dbe 226 hammer_off_t first_offset; /* zone-X logical offset (zone 3) */
c3be93f2
MD
227 hammer_off_t next_offset; /* zone-X logical offset */
228 hammer_off_t alloc_offset; /* zone-X logical offset */
c3be93f2 229 u_int32_t reserved01;
19619882 230 hammer_crc_t entry_crc;
c3be93f2 231};
40043e7f 232
c3be93f2 233typedef struct hammer_blockmap *hammer_blockmap_t;
40043e7f 234
19619882
MD
235#define HAMMER_BLOCKMAP_CRCSIZE \
236 offsetof(struct hammer_blockmap, entry_crc)
237
c3be93f2
MD
238/*
239 * The blockmap is a 2-layer entity made up of big-blocks. The first layer
240 * contains 262144 32-byte entries (18 bits), the second layer contains
241 * 524288 16-byte entries (19 bits), representing 8MB (23 bit) blockmaps.
242 * 18+19+23 = 60 bits. The top four bits are the zone id.
243 *
cb51be26
MD
244 * Currently only the freemap utilizes both layers in all their glory.
245 * All primary data/meta-data zones actually encode a zone-2 address
246 * requiring no real blockmap translation.
c3be93f2 247 *
cb51be26
MD
248 * The freemap uses the upper 8 bits of layer-1 to identify the volume,
249 * thus any space allocated via the freemap can be directly translated
250 * to a zone:2 (or zone:8-15) address.
c3be93f2 251 *
cb51be26 252 * zone-X blockmap offset: [z:4][layer1:18][layer2:19][bigblock:23]
c3be93f2
MD
253 */
254struct hammer_blockmap_layer1 {
255 hammer_off_t blocks_free; /* big-blocks free */
256 hammer_off_t phys_offset; /* UNAVAIL or zone-2 */
c3be93f2 257 hammer_off_t reserved01;
19619882
MD
258 hammer_crc_t layer2_crc; /* xor'd crc's of HAMMER_BLOCKSIZE */
259 /* (not yet used) */
260 hammer_crc_t layer1_crc; /* MUST BE LAST FIELD OF STRUCTURE*/
c3be93f2 261};
40043e7f 262
47637bff
MD
263typedef struct hammer_blockmap_layer1 *hammer_blockmap_layer1_t;
264
19619882
MD
265#define HAMMER_LAYER1_CRCSIZE \
266 offsetof(struct hammer_blockmap_layer1, layer1_crc)
267
c3be93f2 268struct hammer_blockmap_layer2 {
cb51be26
MD
269 u_int8_t zone; /* typed allocation zone */
270 u_int8_t unused01;
271 u_int16_t unused02;
272 u_int32_t append_off; /* allocatable space index */
19619882
MD
273 u_int32_t bytes_free; /* bytes free within this bigblock */
274 hammer_crc_t entry_crc;
40043e7f
MD
275};
276
47637bff
MD
277typedef struct hammer_blockmap_layer2 *hammer_blockmap_layer2_t;
278
19619882
MD
279#define HAMMER_LAYER2_CRCSIZE \
280 offsetof(struct hammer_blockmap_layer2, entry_crc)
281
c3be93f2
MD
282#define HAMMER_BLOCKMAP_FREE 0ULL
283#define HAMMER_BLOCKMAP_UNAVAIL ((hammer_off_t)-1LL)
284
285#define HAMMER_BLOCKMAP_RADIX1 /* 262144 (18) */ \
286 (HAMMER_LARGEBLOCK_SIZE / sizeof(struct hammer_blockmap_layer1))
287#define HAMMER_BLOCKMAP_RADIX2 /* 524288 (19) */ \
288 (HAMMER_LARGEBLOCK_SIZE / sizeof(struct hammer_blockmap_layer2))
289
290#define HAMMER_BLOCKMAP_RADIX1_PERBUFFER \
291 (HAMMER_BLOCKMAP_RADIX1 / (HAMMER_LARGEBLOCK_SIZE / HAMMER_BUFSIZE))
292#define HAMMER_BLOCKMAP_RADIX2_PERBUFFER \
293 (HAMMER_BLOCKMAP_RADIX2 / (HAMMER_LARGEBLOCK_SIZE / HAMMER_BUFSIZE))
294
295#define HAMMER_BLOCKMAP_LAYER1 /* 18+19+23 */ \
296 (HAMMER_BLOCKMAP_RADIX1 * HAMMER_BLOCKMAP_LAYER2)
6f97fce3 297#define HAMMER_BLOCKMAP_LAYER2 /* 19+23 - 4TB */ \
c3be93f2
MD
298 (HAMMER_BLOCKMAP_RADIX2 * HAMMER_LARGEBLOCK_SIZE64)
299
300#define HAMMER_BLOCKMAP_LAYER1_MASK (HAMMER_BLOCKMAP_LAYER1 - 1)
301#define HAMMER_BLOCKMAP_LAYER2_MASK (HAMMER_BLOCKMAP_LAYER2 - 1)
302
303/*
304 * byte offset within layer1 or layer2 big-block for the entry representing
305 * a zone-2 physical offset.
306 */
307#define HAMMER_BLOCKMAP_LAYER1_OFFSET(zone2_offset) \
308 (((zone2_offset) & HAMMER_BLOCKMAP_LAYER1_MASK) / \
309 HAMMER_BLOCKMAP_LAYER2 * sizeof(struct hammer_blockmap_layer1))
310
311#define HAMMER_BLOCKMAP_LAYER2_OFFSET(zone2_offset) \
312 (((zone2_offset) & HAMMER_BLOCKMAP_LAYER2_MASK) / \
313 HAMMER_LARGEBLOCK_SIZE64 * sizeof(struct hammer_blockmap_layer2))
66325755 314
bf686dbe
MD
315/*
316 * HAMMER UNDO parameters. The UNDO fifo is mapped directly in the volume
cb51be26 317 * header with an array of layer2 structures. A maximum of (128x8MB) = 1GB
bf686dbe
MD
318 * may be reserved. The size of the undo fifo is usually set a newfs time
319 * but can be adjusted if the filesystem is taken offline.
320 */
321
cb51be26 322#define HAMMER_UNDO_LAYER2 128 /* max layer2 undo mapping entries */
bf686dbe 323
9380c752 324/*
36f82b23
MD
325 * All on-disk HAMMER structures which make up elements of the UNDO FIFO
326 * contain a hammer_fifo_head and hammer_fifo_tail structure. This structure
40043e7f
MD
327 * contains all the information required to validate the fifo element
328 * and to scan the fifo in either direction. The head is typically embedded
329 * in higher level hammer on-disk structures while the tail is typically
330 * out-of-band. hdr_size is the size of the whole mess, including the tail.
47197d71 331 *
36f82b23
MD
332 * All undo structures are guaranteed to not cross a 16K filesystem
333 * buffer boundary. Most undo structures are fairly small. Data spaces
334 * are not immediately reused by HAMMER so file data is not usually recorded
335 * as part of an UNDO.
40043e7f
MD
336 *
337 * PAD elements are allowed to take up only 8 bytes of space as a special
338 * case, containing only hdr_signature, hdr_type, and hdr_size fields,
339 * and with the tail overloaded onto the head structure for 8 bytes total.
36f82b23
MD
340 *
341 * Every undo record has a sequence number. This number is unrelated to
342 * transaction ids and instead collects the undo transactions associated
343 * with a single atomic operation. A larger transactional operation, such
344 * as a remove(), may consist of several smaller atomic operations
345 * representing raw meta-data operations.
9380c752 346 */
36f82b23 347#define HAMMER_HEAD_ONDISK_SIZE 32
40043e7f 348#define HAMMER_HEAD_ALIGN 8
47197d71 349#define HAMMER_HEAD_ALIGN_MASK (HAMMER_HEAD_ALIGN - 1)
40043e7f 350#define HAMMER_TAIL_ONDISK_SIZE 8
47197d71
MD
351
352struct hammer_fifo_head {
353 u_int16_t hdr_signature;
354 u_int16_t hdr_type;
40043e7f 355 u_int32_t hdr_size; /* aligned size of the whole mess */
059819e3 356 u_int32_t reserved01; /* (0) reserved for future use */
19619882 357 hammer_crc_t hdr_crc; /* XOR crc up to field w/ crc after field */
40043e7f
MD
358};
359
19619882
MD
360#define HAMMER_FIFO_HEAD_CRCOFF offsetof(struct hammer_fifo_head, hdr_crc)
361
40043e7f
MD
362struct hammer_fifo_tail {
363 u_int16_t tail_signature;
364 u_int16_t tail_type;
365 u_int32_t tail_size; /* aligned size of the whole mess */
47197d71
MD
366};
367
368typedef struct hammer_fifo_head *hammer_fifo_head_t;
40043e7f 369typedef struct hammer_fifo_tail *hammer_fifo_tail_t;
9380c752 370
8750964d 371/*
47197d71 372 * Fifo header types.
8750964d 373 */
40043e7f
MD
374#define HAMMER_HEAD_TYPE_PAD (0x0040U|HAMMER_HEAD_FLAG_FREE)
375#define HAMMER_HEAD_TYPE_VOL 0x0041U /* Volume (dummy header) */
376#define HAMMER_HEAD_TYPE_BTREE 0x0042U /* B-Tree node */
377#define HAMMER_HEAD_TYPE_UNDO 0x0043U /* random UNDO information */
378#define HAMMER_HEAD_TYPE_DELETE 0x0044U /* record deletion */
379#define HAMMER_HEAD_TYPE_RECORD 0x0045U /* Filesystem record */
47197d71 380
40043e7f 381#define HAMMER_HEAD_FLAG_FREE 0x8000U /* Indicates object freed */
8750964d 382
47197d71 383#define HAMMER_HEAD_SIGNATURE 0xC84EU
40043e7f 384#define HAMMER_TAIL_SIGNATURE 0xC74FU
47197d71 385
36f82b23
MD
386#define HAMMER_HEAD_SEQ_BEG 0x80000000U
387#define HAMMER_HEAD_SEQ_END 0x40000000U
388#define HAMMER_HEAD_SEQ_MASK 0x3FFFFFFFU
389
47197d71 390/*
bf686dbe 391 * Misc FIFO structures.
47197d71
MD
392 */
393struct hammer_fifo_undo {
394 struct hammer_fifo_head head;
bf686dbe
MD
395 hammer_off_t undo_offset; /* zone-1 offset */
396 int32_t undo_data_bytes;
397 int32_t undo_reserved01;
47197d71 398 /* followed by data */
8750964d
MD
399};
400
47197d71 401typedef struct hammer_fifo_undo *hammer_fifo_undo_t;
8750964d 402
bf686dbe
MD
403struct hammer_fifo_buf_commit {
404 hammer_off_t undo_offset;
405};
406
c60bb2c5 407/*
47197d71 408 * Volume header types
c60bb2c5 409 */
8750964d 410#define HAMMER_FSBUF_VOLUME 0xC8414D4DC5523031ULL /* HAMMER01 */
8750964d
MD
411#define HAMMER_FSBUF_VOLUME_REV 0x313052C54D4D41C8ULL /* (reverse endian) */
412
413/*
414 * The B-Tree structures need hammer_fsbuf_head.
415 */
416#include "hammer_btree.h"
417
418/*
419 * HAMMER Volume header
420 *
421 * A HAMMER filesystem is built from any number of block devices, Each block
47197d71
MD
422 * device contains a volume header followed by however many buffers fit
423 * into the volume.
8750964d 424 *
47197d71
MD
425 * One of the volumes making up a HAMMER filesystem is the master, the
426 * rest are slaves. It does not have to be volume #0.
c60bb2c5 427 *
47197d71
MD
428 * The volume header takes up an entire 16K filesystem buffer and may
429 * represent up to 64KTB (65536 TB) of space.
a89aec1b
MD
430 *
431 * Special field notes:
432 *
433 * vol_bot_beg - offset of boot area (mem_beg - bot_beg bytes)
434 * vol_mem_beg - offset of memory log (clu_beg - mem_beg bytes)
47197d71 435 * vol_buf_beg - offset of the first buffer.
a89aec1b
MD
436 *
437 * The memory log area allows a kernel to cache new records and data
438 * in memory without allocating space in the actual filesystem to hold
439 * the records and data. In the event that a filesystem becomes full,
440 * any records remaining in memory can be flushed to the memory log
441 * area. This allows the kernel to immediately return success.
8750964d 442 */
8750964d 443
a89aec1b
MD
444#define HAMMER_BOOT_MINBYTES (32*1024)
445#define HAMMER_BOOT_NOMBYTES (64LL*1024*1024)
446#define HAMMER_BOOT_MAXBYTES (256LL*1024*1024)
447
448#define HAMMER_MEM_MINBYTES (256*1024)
449#define HAMMER_MEM_NOMBYTES (1LL*1024*1024*1024)
450#define HAMMER_MEM_MAXBYTES (64LL*1024*1024*1024)
451
8750964d 452struct hammer_volume_ondisk {
40043e7f
MD
453 u_int64_t vol_signature;/* Signature */
454
a89aec1b
MD
455 int64_t vol_bot_beg; /* byte offset of boot area or 0 */
456 int64_t vol_mem_beg; /* byte offset of memory log or 0 */
47197d71
MD
457 int64_t vol_buf_beg; /* byte offset of first buffer in volume */
458 int64_t vol_buf_end; /* byte offset of volume EOF (on buf bndry) */
8750964d
MD
459 int64_t vol_locked; /* reserved clusters are >= this offset */
460
461 uuid_t vol_fsid; /* identify filesystem */
462 uuid_t vol_fstype; /* identify filesystem type */
463 char vol_name[64]; /* Name of volume */
464
465 int32_t vol_no; /* volume number within filesystem */
466 int32_t vol_count; /* number of volumes making up FS */
467
468 u_int32_t vol_version; /* version control information */
19619882 469 hammer_crc_t vol_crc; /* header crc */
8750964d
MD
470 u_int32_t vol_flags; /* volume flags */
471 u_int32_t vol_rootvol; /* which volume is the root volume? */
472
40043e7f 473 int32_t vol_reserved04;
47197d71 474 int32_t vol_reserved05;
8750964d
MD
475 u_int32_t vol_reserved06;
476 u_int32_t vol_reserved07;
477
fbc6e32a 478 int32_t vol_blocksize; /* for statfs only */
47197d71 479 int32_t vol_reserved08;
fbc6e32a
MD
480 int64_t vol_nblocks; /* total allocatable hammer bufs */
481
8750964d
MD
482 /*
483 * These fields are initialized and space is reserved in every
484 * volume making up a HAMMER filesytem, but only the master volume
485 * contains valid data.
486 */
c3be93f2
MD
487 int64_t vol0_stat_bigblocks; /* total bigblocks when fs is empty */
488 int64_t vol0_stat_freebigblocks;/* number of free bigblocks */
fbc6e32a
MD
489 int64_t vol0_stat_bytes; /* for statfs only */
490 int64_t vol0_stat_inodes; /* for statfs only */
b33e2cc0 491 int64_t vol0_stat_records; /* total records in filesystem */
47197d71
MD
492 hammer_off_t vol0_btree_root; /* B-Tree root */
493 hammer_tid_t vol0_next_tid; /* highest synchronized TID */
cb51be26 494 hammer_off_t vol0_unused03; /* limit the zone size */
40043e7f
MD
495
496 /*
0729c8c8
MD
497 * Blockmaps for zones. Not all zones use a blockmap. Note that
498 * the entire root blockmap is cached in the hammer_mount structure.
40043e7f 499 */
c3be93f2 500 struct hammer_blockmap vol0_blockmap[HAMMER_MAX_ZONES];
40043e7f 501
bf686dbe 502 /*
cb51be26 503 * Array of zone-2 addresses for undo FIFO.
bf686dbe 504 */
cb51be26 505 hammer_off_t vol0_undo_array[HAMMER_UNDO_LAYER2];
bf686dbe 506
8750964d
MD
507};
508
8cd0a023
MD
509typedef struct hammer_volume_ondisk *hammer_volume_ondisk_t;
510
c60bb2c5
MD
511#define HAMMER_VOLF_VALID 0x0001 /* valid entry */
512#define HAMMER_VOLF_OPEN 0x0002 /* volume is open */
8cd0a023 513
19619882
MD
514#define HAMMER_VOL_CRCSIZE1 \
515 offsetof(struct hammer_volume_ondisk, vol_crc)
516#define HAMMER_VOL_CRCSIZE2 \
517 (sizeof(struct hammer_volume_ondisk) - HAMMER_VOL_CRCSIZE1 - \
518 sizeof(hammer_crc_t))
519
c60bb2c5
MD
520/*
521 * Record types are fairly straightforward. The B-Tree includes the record
522 * type in its index sort.
523 *
524 * In particular please note that it is possible to create a pseudo-
525 * filesystem within a HAMMER filesystem by creating a special object
526 * type within a directory. Pseudo-filesystems are used as replication
527 * targets and even though they are built within a HAMMER filesystem they
528 * get their own obj_id space (and thus can serve as a replication target)
529 * and look like a mount point to the system.
8cd0a023 530 *
b3deaf57
MD
531 * NOTE: hammer_ip_delete_range_all() deletes all record types greater
532 * then HAMMER_RECTYPE_INODE.
c60bb2c5 533 */
8750964d 534#define HAMMER_RECTYPE_UNKNOWN 0
66325755 535#define HAMMER_RECTYPE_LOWEST 1 /* lowest record type avail */
8750964d 536#define HAMMER_RECTYPE_INODE 1 /* inode in obj_id space */
c60bb2c5 537#define HAMMER_RECTYPE_PSEUDO_INODE 2 /* pseudo filesysem */
bf3b416b 538#define HAMMER_RECTYPE_UNUSED03 3 /* inter-cluster reference */
bf686dbe
MD
539#define HAMMER_RECTYPE_DATA 0x0010
540#define HAMMER_RECTYPE_DIRENTRY 0x0011
541#define HAMMER_RECTYPE_DB 0x0012
542#define HAMMER_RECTYPE_EXT 0x0013 /* ext attributes */
543#define HAMMER_RECTYPE_FIX 0x0014 /* fixed attribute */
544#define HAMMER_RECTYPE_MOVED 0x8000 /* special recovery flag */
7a04d74f
MD
545
546#define HAMMER_FIXKEY_SYMLINK 1
8750964d 547
66325755 548#define HAMMER_OBJTYPE_UNKNOWN 0 /* (never exists on-disk) */
8750964d
MD
549#define HAMMER_OBJTYPE_DIRECTORY 1
550#define HAMMER_OBJTYPE_REGFILE 2
551#define HAMMER_OBJTYPE_DBFILE 3
552#define HAMMER_OBJTYPE_FIFO 4
c60bb2c5
MD
553#define HAMMER_OBJTYPE_CDEV 5
554#define HAMMER_OBJTYPE_BDEV 6
555#define HAMMER_OBJTYPE_SOFTLINK 7
556#define HAMMER_OBJTYPE_PSEUDOFS 8 /* pseudo filesystem obj */
557
8750964d 558/*
11ad5ade 559 * HAMMER inode attribute data
8750964d 560 *
11ad5ade
MD
561 * The data reference for a HAMMER inode points to this structure. Any
562 * modifications to the contents of this structure will result in a
563 * replacement operation.
8750964d 564 *
11ad5ade
MD
565 * parent_obj_id is only valid for directories (which cannot be hard-linked),
566 * and specifies the parent directory obj_id. This field will also be set
567 * for non-directory inodes as a recovery aid, but can wind up specifying
568 * stale information. However, since object id's are not reused, the worse
569 * that happens is that the recovery code is unable to use it.
8750964d 570 *
11ad5ade
MD
571 * NOTE: atime is stored in the inode's B-Tree element and not in the inode
572 * data. This allows the atime to be updated without having to lay down a
573 * new record.
8750964d 574 */
11ad5ade
MD
575struct hammer_inode_data {
576 u_int16_t version; /* inode data version */
577 u_int16_t mode; /* basic unix permissions */
578 u_int32_t uflags; /* chflags */
579 u_int32_t rmajor; /* used by device nodes */
580 u_int32_t rminor; /* used by device nodes */
581 u_int64_t ctime;
582 u_int64_t parent_obj_id;/* parent directory obj_id */
583 uuid_t uid;
584 uuid_t gid;
585
586 u_int8_t obj_type;
587 u_int8_t reserved01;
588 u_int16_t reserved02;
589 u_int32_t reserved03;
11ad5ade 590 u_int64_t nlinks; /* hard links */
bcac4bbb
MD
591 u_int64_t size; /* filesystem object size */
592 u_int64_t mtime;
593 u_int64_t atime; /* atime must be just after mtime */
2f85fa4d
MD
594 union {
595 char reserved06[24];
596 char symlink[24]; /* HAMMER_INODE_BASESYMLEN */
597 } ext;
8750964d
MD
598};
599
bcac4bbb
MD
600#define HAMMER_ITIMES_BASE(ino_data) (&(ino_data)->mtime)
601#define HAMMER_ITIMES_BYTES (sizeof(u_int64_t) * 2)
602
11ad5ade
MD
603#define HAMMER_INODE_DATA_VERSION 1
604#define HAMMER_OBJID_ROOT 1
2f85fa4d 605#define HAMMER_INODE_BASESYMLEN 24
8750964d
MD
606
607/*
608 * A directory entry specifies the HAMMER filesystem object id, a copy of
609 * the file type, and file name (either embedded or as out-of-band data).
610 * If the file name is short enough to fit into den_name[] (including a
611 * terminating nul) then it will be embedded in the record, otherwise it
612 * is stored out-of-band. The base record's data reference always points
613 * to the nul-terminated filename regardless.
614 *
615 * Directory entries are indexed with a 128 bit namekey rather then an
616 * offset. A portion of the namekey is an iterator or randomizer to deal
617 * with collisions.
66325755 618 *
6b4f890b
MD
619 * NOTE: base.base.obj_type holds the filesystem object type of obj_id,
620 * e.g. a den_type equivalent.
621 *
622 * NOTE: den_name / the filename data reference is NOT terminated with \0.
66325755 623 *
8750964d 624 */
11ad5ade 625struct hammer_entry_data {
8750964d
MD
626 u_int64_t obj_id; /* object being referenced */
627 u_int64_t reserved01;
11ad5ade 628 char name[16]; /* name (extended) */
8750964d
MD
629};
630
11ad5ade
MD
631#define HAMMER_ENTRY_NAME_OFF offsetof(struct hammer_entry_data, name[0])
632#define HAMMER_ENTRY_SIZE(nlen) offsetof(struct hammer_entry_data, name[nlen])
19619882 633
11ad5ade
MD
634struct hammer_symlink_data {
635 char name[16];
8750964d
MD
636};
637
11ad5ade 638#define HAMMER_SYMLINK_NAME_OFF offsetof(struct hammer_symlink_data, name[0])
7a04d74f 639
c60bb2c5
MD
640/*
641 * Rollup various structures embedded as record data
642 */
427e5fc6 643union hammer_data_ondisk {
11ad5ade 644 struct hammer_entry_data entry;
c60bb2c5 645 struct hammer_inode_data inode;
11ad5ade 646 struct hammer_symlink_data symlink;
c60bb2c5
MD
647};
648
11ad5ade
MD
649typedef union hammer_data_ondisk *hammer_data_ondisk_t;
650
7dc57964 651#endif