Commit | Line | Data |
---|---|---|
1c6fdbd8 KO |
1 | /* SPDX-License-Identifier: GPL-2.0 */ |
2 | #ifndef _BCACHEFS_FORMAT_H | |
3 | #define _BCACHEFS_FORMAT_H | |
4 | ||
5 | /* | |
6 | * bcachefs on disk data structures | |
7 | * | |
8 | * OVERVIEW: | |
9 | * | |
10 | * There are three main types of on disk data structures in bcachefs (this is | |
11 | * reduced from 5 in bcache) | |
12 | * | |
13 | * - superblock | |
14 | * - journal | |
15 | * - btree | |
16 | * | |
17 | * The btree is the primary structure; most metadata exists as keys in the | |
18 | * various btrees. There are only a small number of btrees, they're not | |
19 | * sharded - we have one btree for extents, another for inodes, et cetera. | |
20 | * | |
21 | * SUPERBLOCK: | |
22 | * | |
23 | * The superblock contains the location of the journal, the list of devices in | |
24 | * the filesystem, and in general any metadata we need in order to decide | |
25 | * whether we can start a filesystem or prior to reading the journal/btree | |
26 | * roots. | |
27 | * | |
28 | * The superblock is extensible, and most of the contents of the superblock are | |
29 | * in variable length, type tagged fields; see struct bch_sb_field. | |
30 | * | |
31 | * Backup superblocks do not reside in a fixed location; also, superblocks do | |
32 | * not have a fixed size. To locate backup superblocks we have struct | |
33 | * bch_sb_layout; we store a copy of this inside every superblock, and also | |
34 | * before the first superblock. | |
35 | * | |
36 | * JOURNAL: | |
37 | * | |
38 | * The journal primarily records btree updates in the order they occurred; | |
39 | * journal replay consists of just iterating over all the keys in the open | |
40 | * journal entries and re-inserting them into the btrees. | |
41 | * | |
42 | * The journal also contains entry types for the btree roots, and blacklisted | |
43 | * journal sequence numbers (see journal_seq_blacklist.c). | |
44 | * | |
45 | * BTREE: | |
46 | * | |
47 | * bcachefs btrees are copy on write b+ trees, where nodes are big (typically | |
48 | * 128k-256k) and log structured. We use struct btree_node for writing the first | |
49 | * entry in a given node (offset 0), and struct btree_node_entry for all | |
50 | * subsequent writes. | |
51 | * | |
52 | * After the header, btree node entries contain a list of keys in sorted order. | |
53 | * Values are stored inline with the keys; since values are variable length (and | |
54 | * keys effectively are variable length too, due to packing) we can't do random | |
55 | * access without building up additional in memory tables in the btree node read | |
56 | * path. | |
57 | * | |
58 | * BTREE KEYS (struct bkey): | |
59 | * | |
60 | * The various btrees share a common format for the key - so as to avoid | |
61 | * switching in fastpath lookup/comparison code - but define their own | |
62 | * structures for the key values. | |
63 | * | |
64 | * The size of a key/value pair is stored as a u8 in units of u64s, so the max | |
65 | * size is just under 2k. The common part also contains a type tag for the | |
66 | * value, and a format field indicating whether the key is packed or not (and | |
67 | * also meant to allow adding new key fields in the future, if desired). | |
68 | * | |
69 | * bkeys, when stored within a btree node, may also be packed. In that case, the | |
70 | * bkey_format in that node is used to unpack it. Packed bkeys mean that we can | |
71 | * be generous with field sizes in the common part of the key format (64 bit | |
72 | * inode number, 64 bit offset, 96 bit version field, etc.) for negligible cost. | |
73 | */ | |
74 | ||
75 | #include <asm/types.h> | |
76 | #include <asm/byteorder.h> | |
77 | #include <linux/uuid.h> | |
78 | ||
79 | #ifdef __KERNEL__ | |
80 | typedef uuid_t __uuid_t; | |
81 | #endif | |
82 | ||
83 | #define LE_BITMASK(_bits, name, type, field, offset, end) \ | |
84 | static const unsigned name##_OFFSET = offset; \ | |
85 | static const unsigned name##_BITS = (end - offset); \ | |
86 | static const __u##_bits name##_MAX = (1ULL << (end - offset)) - 1; \ | |
87 | \ | |
88 | static inline __u64 name(const type *k) \ | |
89 | { \ | |
90 | return (__le##_bits##_to_cpu(k->field) >> offset) & \ | |
91 | ~(~0ULL << (end - offset)); \ | |
92 | } \ | |
93 | \ | |
94 | static inline void SET_##name(type *k, __u64 v) \ | |
95 | { \ | |
96 | __u##_bits new = __le##_bits##_to_cpu(k->field); \ | |
97 | \ | |
98 | new &= ~(~(~0ULL << (end - offset)) << offset); \ | |
99 | new |= (v & ~(~0ULL << (end - offset))) << offset; \ | |
100 | k->field = __cpu_to_le##_bits(new); \ | |
101 | } | |
102 | ||
103 | #define LE16_BITMASK(n, t, f, o, e) LE_BITMASK(16, n, t, f, o, e) | |
104 | #define LE32_BITMASK(n, t, f, o, e) LE_BITMASK(32, n, t, f, o, e) | |
105 | #define LE64_BITMASK(n, t, f, o, e) LE_BITMASK(64, n, t, f, o, e) | |
106 | ||
107 | struct bkey_format { | |
108 | __u8 key_u64s; | |
109 | __u8 nr_fields; | |
110 | /* One unused slot for now: */ | |
111 | __u8 bits_per_field[6]; | |
112 | __le64 field_offset[6]; | |
113 | }; | |
114 | ||
115 | /* Btree keys - all units are in sectors */ | |
116 | ||
117 | struct bpos { | |
118 | /* | |
119 | * Word order matches machine byte order - btree code treats a bpos as a | |
120 | * single large integer, for search/comparison purposes | |
121 | * | |
122 | * Note that wherever a bpos is embedded in another on disk data | |
123 | * structure, it has to be byte swabbed when reading in metadata that | |
124 | * wasn't written in native endian order: | |
125 | */ | |
126 | #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ | |
127 | __u32 snapshot; | |
128 | __u64 offset; | |
129 | __u64 inode; | |
130 | #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ | |
131 | __u64 inode; | |
132 | __u64 offset; /* Points to end of extent - sectors */ | |
133 | __u32 snapshot; | |
134 | #else | |
135 | #error edit for your odd byteorder. | |
136 | #endif | |
137 | } __attribute__((packed, aligned(4))); | |
138 | ||
139 | #define KEY_INODE_MAX ((__u64)~0ULL) | |
140 | #define KEY_OFFSET_MAX ((__u64)~0ULL) | |
141 | #define KEY_SNAPSHOT_MAX ((__u32)~0U) | |
142 | #define KEY_SIZE_MAX ((__u32)~0U) | |
143 | ||
144 | static inline struct bpos POS(__u64 inode, __u64 offset) | |
145 | { | |
146 | struct bpos ret; | |
147 | ||
148 | ret.inode = inode; | |
149 | ret.offset = offset; | |
150 | ret.snapshot = 0; | |
151 | ||
152 | return ret; | |
153 | } | |
154 | ||
155 | #define POS_MIN POS(0, 0) | |
156 | #define POS_MAX POS(KEY_INODE_MAX, KEY_OFFSET_MAX) | |
157 | ||
158 | /* Empty placeholder struct, for container_of() */ | |
159 | struct bch_val { | |
160 | __u64 __nothing[0]; | |
161 | }; | |
162 | ||
163 | struct bversion { | |
164 | #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ | |
165 | __u64 lo; | |
166 | __u32 hi; | |
167 | #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ | |
168 | __u32 hi; | |
169 | __u64 lo; | |
170 | #endif | |
171 | } __attribute__((packed, aligned(4))); | |
172 | ||
173 | struct bkey { | |
174 | /* Size of combined key and value, in u64s */ | |
175 | __u8 u64s; | |
176 | ||
177 | /* Format of key (0 for format local to btree node) */ | |
178 | #if defined(__LITTLE_ENDIAN_BITFIELD) | |
179 | __u8 format:7, | |
180 | needs_whiteout:1; | |
181 | #elif defined (__BIG_ENDIAN_BITFIELD) | |
182 | __u8 needs_whiteout:1, | |
183 | format:7; | |
184 | #else | |
185 | #error edit for your odd byteorder. | |
186 | #endif | |
187 | ||
188 | /* Type of the value */ | |
189 | __u8 type; | |
190 | ||
191 | #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ | |
192 | __u8 pad[1]; | |
193 | ||
194 | struct bversion version; | |
195 | __u32 size; /* extent size, in sectors */ | |
196 | struct bpos p; | |
197 | #elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ | |
198 | struct bpos p; | |
199 | __u32 size; /* extent size, in sectors */ | |
200 | struct bversion version; | |
201 | ||
202 | __u8 pad[1]; | |
203 | #endif | |
204 | } __attribute__((packed, aligned(8))); | |
205 | ||
206 | struct bkey_packed { | |
207 | __u64 _data[0]; | |
208 | ||
209 | /* Size of combined key and value, in u64s */ | |
210 | __u8 u64s; | |
211 | ||
212 | /* Format of key (0 for format local to btree node) */ | |
213 | ||
214 | /* | |
215 | * XXX: next incompat on disk format change, switch format and | |
216 | * needs_whiteout - bkey_packed() will be cheaper if format is the high | |
217 | * bits of the bitfield | |
218 | */ | |
219 | #if defined(__LITTLE_ENDIAN_BITFIELD) | |
220 | __u8 format:7, | |
221 | needs_whiteout:1; | |
222 | #elif defined (__BIG_ENDIAN_BITFIELD) | |
223 | __u8 needs_whiteout:1, | |
224 | format:7; | |
225 | #endif | |
226 | ||
227 | /* Type of the value */ | |
228 | __u8 type; | |
229 | __u8 key_start[0]; | |
230 | ||
231 | /* | |
232 | * We copy bkeys with struct assignment in various places, and while | |
233 | * that shouldn't be done with packed bkeys we can't disallow it in C, | |
234 | * and it's legal to cast a bkey to a bkey_packed - so padding it out | |
235 | * to the same size as struct bkey should hopefully be safest. | |
236 | */ | |
237 | __u8 pad[sizeof(struct bkey) - 3]; | |
238 | } __attribute__((packed, aligned(8))); | |
239 | ||
240 | #define BKEY_U64s (sizeof(struct bkey) / sizeof(__u64)) | |
241 | #define KEY_PACKED_BITS_START 24 | |
242 | ||
243 | #define KEY_FORMAT_LOCAL_BTREE 0 | |
244 | #define KEY_FORMAT_CURRENT 1 | |
245 | ||
246 | enum bch_bkey_fields { | |
247 | BKEY_FIELD_INODE, | |
248 | BKEY_FIELD_OFFSET, | |
249 | BKEY_FIELD_SNAPSHOT, | |
250 | BKEY_FIELD_SIZE, | |
251 | BKEY_FIELD_VERSION_HI, | |
252 | BKEY_FIELD_VERSION_LO, | |
253 | BKEY_NR_FIELDS, | |
254 | }; | |
255 | ||
256 | #define bkey_format_field(name, field) \ | |
257 | [BKEY_FIELD_##name] = (sizeof(((struct bkey *) NULL)->field) * 8) | |
258 | ||
259 | #define BKEY_FORMAT_CURRENT \ | |
260 | ((struct bkey_format) { \ | |
261 | .key_u64s = BKEY_U64s, \ | |
262 | .nr_fields = BKEY_NR_FIELDS, \ | |
263 | .bits_per_field = { \ | |
264 | bkey_format_field(INODE, p.inode), \ | |
265 | bkey_format_field(OFFSET, p.offset), \ | |
266 | bkey_format_field(SNAPSHOT, p.snapshot), \ | |
267 | bkey_format_field(SIZE, size), \ | |
268 | bkey_format_field(VERSION_HI, version.hi), \ | |
269 | bkey_format_field(VERSION_LO, version.lo), \ | |
270 | }, \ | |
271 | }) | |
272 | ||
273 | /* bkey with inline value */ | |
274 | struct bkey_i { | |
275 | __u64 _data[0]; | |
276 | ||
277 | union { | |
278 | struct { | |
279 | /* Size of combined key and value, in u64s */ | |
280 | __u8 u64s; | |
281 | }; | |
282 | struct { | |
283 | struct bkey k; | |
284 | struct bch_val v; | |
285 | }; | |
286 | }; | |
287 | }; | |
288 | ||
289 | #define KEY(_inode, _offset, _size) \ | |
290 | ((struct bkey) { \ | |
291 | .u64s = BKEY_U64s, \ | |
292 | .format = KEY_FORMAT_CURRENT, \ | |
293 | .p = POS(_inode, _offset), \ | |
294 | .size = _size, \ | |
295 | }) | |
296 | ||
297 | static inline void bkey_init(struct bkey *k) | |
298 | { | |
299 | *k = KEY(0, 0, 0); | |
300 | } | |
301 | ||
302 | #define bkey_bytes(_k) ((_k)->u64s * sizeof(__u64)) | |
303 | ||
304 | #define __BKEY_PADDED(key, pad) \ | |
305 | struct { struct bkey_i key; __u64 key ## _pad[pad]; } | |
306 | ||
307 | #define BKEY_VAL_TYPE(name, nr) \ | |
308 | struct bkey_i_##name { \ | |
309 | union { \ | |
310 | struct bkey k; \ | |
311 | struct bkey_i k_i; \ | |
312 | }; \ | |
313 | struct bch_##name v; \ | |
314 | } | |
315 | ||
316 | /* | |
317 | * - DELETED keys are used internally to mark keys that should be ignored but | |
318 | * override keys in composition order. Their version number is ignored. | |
319 | * | |
320 | * - DISCARDED keys indicate that the data is all 0s because it has been | |
321 | * discarded. DISCARDs may have a version; if the version is nonzero the key | |
322 | * will be persistent, otherwise the key will be dropped whenever the btree | |
323 | * node is rewritten (like DELETED keys). | |
324 | * | |
325 | * - ERROR: any read of the data returns a read error, as the data was lost due | |
326 | * to a failing device. Like DISCARDED keys, they can be removed (overridden) | |
327 | * by new writes or cluster-wide GC. Node repair can also overwrite them with | |
328 | * the same or a more recent version number, but not with an older version | |
329 | * number. | |
330 | */ | |
331 | #define KEY_TYPE_DELETED 0 | |
332 | #define KEY_TYPE_DISCARD 1 | |
333 | #define KEY_TYPE_ERROR 2 | |
334 | #define KEY_TYPE_COOKIE 3 | |
335 | #define KEY_TYPE_PERSISTENT_DISCARD 4 | |
336 | #define KEY_TYPE_GENERIC_NR 128 | |
337 | ||
338 | struct bch_cookie { | |
339 | struct bch_val v; | |
340 | __le64 cookie; | |
341 | }; | |
342 | BKEY_VAL_TYPE(cookie, KEY_TYPE_COOKIE); | |
343 | ||
344 | /* Extents */ | |
345 | ||
346 | /* | |
347 | * In extent bkeys, the value is a list of pointers (bch_extent_ptr), optionally | |
348 | * preceded by checksum/compression information (bch_extent_crc32 or | |
349 | * bch_extent_crc64). | |
350 | * | |
351 | * One major determining factor in the format of extents is how we handle and | |
352 | * represent extents that have been partially overwritten and thus trimmed: | |
353 | * | |
354 | * If an extent is not checksummed or compressed, when the extent is trimmed we | |
355 | * don't have to remember the extent we originally allocated and wrote: we can | |
356 | * merely adjust ptr->offset to point to the start of the data that is currently | |
357 | * live. The size field in struct bkey records the current (live) size of the | |
358 | * extent, and is also used to mean "size of region on disk that we point to" in | |
359 | * this case. | |
360 | * | |
361 | * Thus an extent that is not checksummed or compressed will consist only of a | |
362 | * list of bch_extent_ptrs, with none of the fields in | |
363 | * bch_extent_crc32/bch_extent_crc64. | |
364 | * | |
365 | * When an extent is checksummed or compressed, it's not possible to read only | |
366 | * the data that is currently live: we have to read the entire extent that was | |
367 | * originally written, and then return only the part of the extent that is | |
368 | * currently live. | |
369 | * | |
370 | * Thus, in addition to the current size of the extent in struct bkey, we need | |
371 | * to store the size of the originally allocated space - this is the | |
372 | * compressed_size and uncompressed_size fields in bch_extent_crc32/64. Also, | |
373 | * when the extent is trimmed, instead of modifying the offset field of the | |
374 | * pointer, we keep a second smaller offset field - "offset into the original | |
375 | * extent of the currently live region". | |
376 | * | |
377 | * The other major determining factor is replication and data migration: | |
378 | * | |
379 | * Each pointer may have its own bch_extent_crc32/64. When doing a replicated | |
380 | * write, we will initially write all the replicas in the same format, with the | |
381 | * same checksum type and compression format - however, when copygc runs later (or | |
382 | * tiering/cache promotion, anything that moves data), it is not in general | |
383 | * going to rewrite all the pointers at once - one of the replicas may be in a | |
384 | * bucket on one device that has very little fragmentation while another lives | |
385 | * in a bucket that has become heavily fragmented, and thus is being rewritten | |
386 | * sooner than the rest. | |
387 | * | |
388 | * Thus it will only move a subset of the pointers (or in the case of | |
389 | * tiering/cache promotion perhaps add a single pointer without dropping any | |
390 | * current pointers), and if the extent has been partially overwritten it must | |
391 | * write only the currently live portion (or copygc would not be able to reduce | |
392 | * fragmentation!) - which necessitates a different bch_extent_crc format for | |
393 | * the new pointer. | |
394 | * | |
395 | * But in the interests of space efficiency, we don't want to store one | |
396 | * bch_extent_crc for each pointer if we don't have to. | |
397 | * | |
398 | * Thus, a bch_extent consists of bch_extent_crc32s, bch_extent_crc64s, and | |
399 | * bch_extent_ptrs appended arbitrarily one after the other. We determine the | |
400 | * type of a given entry with a scheme similar to utf8 (except we're encoding a | |
401 | * type, not a size), encoding the type in the position of the first set bit: | |
402 | * | |
403 | * bch_extent_crc32 - 0b1 | |
404 | * bch_extent_ptr - 0b10 | |
405 | * bch_extent_crc64 - 0b100 | |
406 | * | |
407 | * We do it this way because bch_extent_crc32 is _very_ constrained on bits (and | |
408 | * bch_extent_crc64 is the least constrained). | |
409 | * | |
410 | * Then, each bch_extent_crc32/64 applies to the pointers that follow after it, | |
411 | * until the next bch_extent_crc32/64. | |
412 | * | |
413 | * If there are no bch_extent_crcs preceding a bch_extent_ptr, then that pointer | |
414 | * is neither checksummed nor compressed. | |
415 | */ | |
416 | ||
417 | /* 128 bits, sufficient for cryptographic MACs: */ | |
418 | struct bch_csum { | |
419 | __le64 lo; | |
420 | __le64 hi; | |
421 | } __attribute__((packed, aligned(8))); | |
422 | ||
423 | enum bch_csum_type { | |
424 | BCH_CSUM_NONE = 0, | |
425 | BCH_CSUM_CRC32C_NONZERO = 1, | |
426 | BCH_CSUM_CRC64_NONZERO = 2, | |
427 | BCH_CSUM_CHACHA20_POLY1305_80 = 3, | |
428 | BCH_CSUM_CHACHA20_POLY1305_128 = 4, | |
429 | BCH_CSUM_CRC32C = 5, | |
430 | BCH_CSUM_CRC64 = 6, | |
431 | BCH_CSUM_NR = 7, | |
432 | }; | |
433 | ||
434 | static const unsigned bch_crc_bytes[] = { | |
435 | [BCH_CSUM_NONE] = 0, | |
436 | [BCH_CSUM_CRC32C_NONZERO] = 4, | |
437 | [BCH_CSUM_CRC32C] = 4, | |
438 | [BCH_CSUM_CRC64_NONZERO] = 8, | |
439 | [BCH_CSUM_CRC64] = 8, | |
440 | [BCH_CSUM_CHACHA20_POLY1305_80] = 10, | |
441 | [BCH_CSUM_CHACHA20_POLY1305_128] = 16, | |
442 | }; | |
443 | ||
444 | static inline _Bool bch2_csum_type_is_encryption(enum bch_csum_type type) | |
445 | { | |
446 | switch (type) { | |
447 | case BCH_CSUM_CHACHA20_POLY1305_80: | |
448 | case BCH_CSUM_CHACHA20_POLY1305_128: | |
449 | return true; | |
450 | default: | |
451 | return false; | |
452 | } | |
453 | } | |
454 | ||
455 | enum bch_compression_type { | |
456 | BCH_COMPRESSION_NONE = 0, | |
457 | BCH_COMPRESSION_LZ4_OLD = 1, | |
458 | BCH_COMPRESSION_GZIP = 2, | |
459 | BCH_COMPRESSION_LZ4 = 3, | |
460 | BCH_COMPRESSION_ZSTD = 4, | |
461 | BCH_COMPRESSION_NR = 5, | |
462 | }; | |
463 | ||
abce30b7 KO |
464 | #define BCH_EXTENT_ENTRY_TYPES() \ |
465 | x(ptr, 0) \ | |
466 | x(crc32, 1) \ | |
467 | x(crc64, 2) \ | |
468 | x(crc128, 3) | |
469 | #define BCH_EXTENT_ENTRY_MAX 4 | |
470 | ||
1c6fdbd8 | 471 | enum bch_extent_entry_type { |
abce30b7 KO |
472 | #define x(f, n) BCH_EXTENT_ENTRY_##f = n, |
473 | BCH_EXTENT_ENTRY_TYPES() | |
474 | #undef x | |
1c6fdbd8 KO |
475 | }; |
476 | ||
1c6fdbd8 KO |
477 | /* Compressed/uncompressed size are stored biased by 1: */ |
478 | struct bch_extent_crc32 { | |
479 | #if defined(__LITTLE_ENDIAN_BITFIELD) | |
480 | __u32 type:2, | |
481 | _compressed_size:7, | |
482 | _uncompressed_size:7, | |
483 | offset:7, | |
484 | _unused:1, | |
485 | csum_type:4, | |
486 | compression_type:4; | |
487 | __u32 csum; | |
488 | #elif defined (__BIG_ENDIAN_BITFIELD) | |
489 | __u32 csum; | |
490 | __u32 compression_type:4, | |
491 | csum_type:4, | |
492 | _unused:1, | |
493 | offset:7, | |
494 | _uncompressed_size:7, | |
495 | _compressed_size:7, | |
496 | type:2; | |
497 | #endif | |
498 | } __attribute__((packed, aligned(8))); | |
499 | ||
500 | #define CRC32_SIZE_MAX (1U << 7) | |
501 | #define CRC32_NONCE_MAX 0 | |
502 | ||
503 | struct bch_extent_crc64 { | |
504 | #if defined(__LITTLE_ENDIAN_BITFIELD) | |
505 | __u64 type:3, | |
506 | _compressed_size:9, | |
507 | _uncompressed_size:9, | |
508 | offset:9, | |
509 | nonce:10, | |
510 | csum_type:4, | |
511 | compression_type:4, | |
512 | csum_hi:16; | |
513 | #elif defined (__BIG_ENDIAN_BITFIELD) | |
514 | __u64 csum_hi:16, | |
515 | compression_type:4, | |
516 | csum_type:4, | |
517 | nonce:10, | |
518 | offset:9, | |
519 | _uncompressed_size:9, | |
520 | _compressed_size:9, | |
521 | type:3; | |
522 | #endif | |
523 | __u64 csum_lo; | |
524 | } __attribute__((packed, aligned(8))); | |
525 | ||
526 | #define CRC64_SIZE_MAX (1U << 9) | |
527 | #define CRC64_NONCE_MAX ((1U << 10) - 1) | |
528 | ||
529 | struct bch_extent_crc128 { | |
530 | #if defined(__LITTLE_ENDIAN_BITFIELD) | |
531 | __u64 type:4, | |
532 | _compressed_size:13, | |
533 | _uncompressed_size:13, | |
534 | offset:13, | |
535 | nonce:13, | |
536 | csum_type:4, | |
537 | compression_type:4; | |
538 | #elif defined (__BIG_ENDIAN_BITFIELD) | |
539 | __u64 compression_type:4, | |
540 | csum_type:4, | |
541 | nonce:13, | |
542 | offset:13, | |
543 | _uncompressed_size:13, | |
544 | _compressed_size:13, | |
545 | type:4; | |
546 | #endif | |
547 | struct bch_csum csum; | |
548 | } __attribute__((packed, aligned(8))); | |
549 | ||
550 | #define CRC128_SIZE_MAX (1U << 13) | |
551 | #define CRC128_NONCE_MAX ((1U << 13) - 1) | |
552 | ||
553 | /* | |
554 | * @reservation - pointer hasn't been written to, just reserved | |
555 | */ | |
556 | struct bch_extent_ptr { | |
557 | #if defined(__LITTLE_ENDIAN_BITFIELD) | |
558 | __u64 type:1, | |
559 | cached:1, | |
560 | erasure_coded:1, | |
561 | reservation:1, | |
562 | offset:44, /* 8 petabytes */ | |
563 | dev:8, | |
564 | gen:8; | |
565 | #elif defined (__BIG_ENDIAN_BITFIELD) | |
566 | __u64 gen:8, | |
567 | dev:8, | |
568 | offset:44, | |
569 | reservation:1, | |
570 | erasure_coded:1, | |
571 | cached:1, | |
572 | type:1; | |
573 | #endif | |
574 | } __attribute__((packed, aligned(8))); | |
575 | ||
576 | struct bch_extent_reservation { | |
577 | #if defined(__LITTLE_ENDIAN_BITFIELD) | |
578 | __u64 type:5, | |
579 | unused:23, | |
580 | replicas:4, | |
581 | generation:32; | |
582 | #elif defined (__BIG_ENDIAN_BITFIELD) | |
583 | __u64 generation:32, | |
584 | replicas:4, | |
585 | unused:23, | |
586 | type:5; | |
587 | #endif | |
588 | }; | |
589 | ||
590 | union bch_extent_entry { | |
591 | #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ || __BITS_PER_LONG == 64 | |
592 | unsigned long type; | |
593 | #elif __BITS_PER_LONG == 32 | |
594 | struct { | |
595 | unsigned long pad; | |
596 | unsigned long type; | |
597 | }; | |
598 | #else | |
599 | #error edit for your odd byteorder. | |
600 | #endif | |
abce30b7 KO |
601 | |
602 | #define x(f, n) struct bch_extent_##f f; | |
603 | BCH_EXTENT_ENTRY_TYPES() | |
604 | #undef x | |
1c6fdbd8 KO |
605 | }; |
606 | ||
607 | enum { | |
608 | BCH_EXTENT = 128, | |
609 | ||
610 | /* | |
611 | * This is kind of a hack, we're overloading the type for a boolean that | |
612 | * really should be part of the value - BCH_EXTENT and BCH_EXTENT_CACHED | |
613 | * have the same value type: | |
614 | */ | |
615 | BCH_EXTENT_CACHED = 129, | |
616 | ||
617 | /* | |
618 | * Persistent reservation: | |
619 | */ | |
620 | BCH_RESERVATION = 130, | |
621 | }; | |
622 | ||
623 | struct bch_extent { | |
624 | struct bch_val v; | |
625 | ||
626 | __u64 _data[0]; | |
627 | union bch_extent_entry start[]; | |
628 | } __attribute__((packed, aligned(8))); | |
629 | BKEY_VAL_TYPE(extent, BCH_EXTENT); | |
630 | ||
631 | struct bch_reservation { | |
632 | struct bch_val v; | |
633 | ||
634 | __le32 generation; | |
635 | __u8 nr_replicas; | |
636 | __u8 pad[3]; | |
637 | } __attribute__((packed, aligned(8))); | |
638 | BKEY_VAL_TYPE(reservation, BCH_RESERVATION); | |
639 | ||
640 | /* Maximum size (in u64s) a single pointer could be: */ | |
641 | #define BKEY_EXTENT_PTR_U64s_MAX\ | |
642 | ((sizeof(struct bch_extent_crc128) + \ | |
643 | sizeof(struct bch_extent_ptr)) / sizeof(u64)) | |
644 | ||
645 | /* Maximum possible size of an entire extent value: */ | |
646 | #define BKEY_EXTENT_VAL_U64s_MAX \ | |
647 | (BKEY_EXTENT_PTR_U64s_MAX * (BCH_REPLICAS_MAX + 1)) | |
648 | ||
649 | #define BKEY_PADDED(key) __BKEY_PADDED(key, BKEY_EXTENT_VAL_U64s_MAX) | |
650 | ||
651 | /* * Maximum possible size of an entire extent, key + value: */ | |
652 | #define BKEY_EXTENT_U64s_MAX (BKEY_U64s + BKEY_EXTENT_VAL_U64s_MAX) | |
653 | ||
654 | /* Btree pointers don't carry around checksums: */ | |
655 | #define BKEY_BTREE_PTR_VAL_U64s_MAX \ | |
656 | ((sizeof(struct bch_extent_ptr)) / sizeof(u64) * BCH_REPLICAS_MAX) | |
657 | #define BKEY_BTREE_PTR_U64s_MAX \ | |
658 | (BKEY_U64s + BKEY_BTREE_PTR_VAL_U64s_MAX) | |
659 | ||
660 | /* Inodes */ | |
661 | ||
662 | #define BLOCKDEV_INODE_MAX 4096 | |
663 | ||
664 | #define BCACHEFS_ROOT_INO 4096 | |
665 | ||
666 | enum bch_inode_types { | |
667 | BCH_INODE_FS = 128, | |
668 | BCH_INODE_BLOCKDEV = 129, | |
669 | BCH_INODE_GENERATION = 130, | |
670 | }; | |
671 | ||
672 | struct bch_inode { | |
673 | struct bch_val v; | |
674 | ||
675 | __le64 bi_hash_seed; | |
676 | __le32 bi_flags; | |
677 | __le16 bi_mode; | |
678 | __u8 fields[0]; | |
679 | } __attribute__((packed, aligned(8))); | |
680 | BKEY_VAL_TYPE(inode, BCH_INODE_FS); | |
681 | ||
682 | struct bch_inode_generation { | |
683 | struct bch_val v; | |
684 | ||
685 | __le32 bi_generation; | |
686 | __le32 pad; | |
687 | } __attribute__((packed, aligned(8))); | |
688 | BKEY_VAL_TYPE(inode_generation, BCH_INODE_GENERATION); | |
689 | ||
690 | #define BCH_INODE_FIELDS() \ | |
691 | BCH_INODE_FIELD(bi_atime, 64) \ | |
692 | BCH_INODE_FIELD(bi_ctime, 64) \ | |
693 | BCH_INODE_FIELD(bi_mtime, 64) \ | |
694 | BCH_INODE_FIELD(bi_otime, 64) \ | |
695 | BCH_INODE_FIELD(bi_size, 64) \ | |
696 | BCH_INODE_FIELD(bi_sectors, 64) \ | |
697 | BCH_INODE_FIELD(bi_uid, 32) \ | |
698 | BCH_INODE_FIELD(bi_gid, 32) \ | |
699 | BCH_INODE_FIELD(bi_nlink, 32) \ | |
700 | BCH_INODE_FIELD(bi_generation, 32) \ | |
701 | BCH_INODE_FIELD(bi_dev, 32) \ | |
702 | BCH_INODE_FIELD(bi_data_checksum, 8) \ | |
703 | BCH_INODE_FIELD(bi_compression, 8) \ | |
704 | BCH_INODE_FIELD(bi_project, 32) \ | |
705 | BCH_INODE_FIELD(bi_background_compression, 8) \ | |
706 | BCH_INODE_FIELD(bi_data_replicas, 8) \ | |
707 | BCH_INODE_FIELD(bi_promote_target, 16) \ | |
708 | BCH_INODE_FIELD(bi_foreground_target, 16) \ | |
709 | BCH_INODE_FIELD(bi_background_target, 16) | |
710 | ||
711 | #define BCH_INODE_FIELDS_INHERIT() \ | |
712 | BCH_INODE_FIELD(bi_data_checksum) \ | |
713 | BCH_INODE_FIELD(bi_compression) \ | |
714 | BCH_INODE_FIELD(bi_project) \ | |
715 | BCH_INODE_FIELD(bi_background_compression) \ | |
716 | BCH_INODE_FIELD(bi_data_replicas) \ | |
717 | BCH_INODE_FIELD(bi_promote_target) \ | |
718 | BCH_INODE_FIELD(bi_foreground_target) \ | |
719 | BCH_INODE_FIELD(bi_background_target) | |
720 | ||
721 | enum { | |
722 | /* | |
723 | * User flags (get/settable with FS_IOC_*FLAGS, correspond to FS_*_FL | |
724 | * flags) | |
725 | */ | |
726 | __BCH_INODE_SYNC = 0, | |
727 | __BCH_INODE_IMMUTABLE = 1, | |
728 | __BCH_INODE_APPEND = 2, | |
729 | __BCH_INODE_NODUMP = 3, | |
730 | __BCH_INODE_NOATIME = 4, | |
731 | ||
732 | __BCH_INODE_I_SIZE_DIRTY= 5, | |
733 | __BCH_INODE_I_SECTORS_DIRTY= 6, | |
734 | __BCH_INODE_UNLINKED = 7, | |
735 | ||
736 | /* bits 20+ reserved for packed fields below: */ | |
737 | }; | |
738 | ||
739 | #define BCH_INODE_SYNC (1 << __BCH_INODE_SYNC) | |
740 | #define BCH_INODE_IMMUTABLE (1 << __BCH_INODE_IMMUTABLE) | |
741 | #define BCH_INODE_APPEND (1 << __BCH_INODE_APPEND) | |
742 | #define BCH_INODE_NODUMP (1 << __BCH_INODE_NODUMP) | |
743 | #define BCH_INODE_NOATIME (1 << __BCH_INODE_NOATIME) | |
744 | #define BCH_INODE_I_SIZE_DIRTY (1 << __BCH_INODE_I_SIZE_DIRTY) | |
745 | #define BCH_INODE_I_SECTORS_DIRTY (1 << __BCH_INODE_I_SECTORS_DIRTY) | |
746 | #define BCH_INODE_UNLINKED (1 << __BCH_INODE_UNLINKED) | |
747 | ||
748 | LE32_BITMASK(INODE_STR_HASH, struct bch_inode, bi_flags, 20, 24); | |
749 | LE32_BITMASK(INODE_NR_FIELDS, struct bch_inode, bi_flags, 24, 32); | |
750 | ||
751 | struct bch_inode_blockdev { | |
752 | struct bch_val v; | |
753 | ||
754 | __le64 i_size; | |
755 | __le64 i_flags; | |
756 | ||
757 | /* Seconds: */ | |
758 | __le64 i_ctime; | |
759 | __le64 i_mtime; | |
760 | ||
761 | __uuid_t i_uuid; | |
762 | __u8 i_label[32]; | |
763 | } __attribute__((packed, aligned(8))); | |
764 | BKEY_VAL_TYPE(inode_blockdev, BCH_INODE_BLOCKDEV); | |
765 | ||
766 | /* Thin provisioned volume, or cache for another block device? */ | |
767 | LE64_BITMASK(CACHED_DEV, struct bch_inode_blockdev, i_flags, 0, 1) | |
768 | ||
769 | /* Dirents */ | |
770 | ||
771 | /* | |
772 | * Dirents (and xattrs) have to implement string lookups; since our b-tree | |
773 | * doesn't support arbitrary length strings for the key, we instead index by a | |
774 | * 64 bit hash (currently truncated sha1) of the string, stored in the offset | |
775 | * field of the key - using linear probing to resolve hash collisions. This also | |
776 | * provides us with the readdir cookie posix requires. | |
777 | * | |
778 | * Linear probing requires us to use whiteouts for deletions, in the event of a | |
779 | * collision: | |
780 | */ | |
781 | ||
782 | enum { | |
783 | BCH_DIRENT = 128, | |
784 | BCH_DIRENT_WHITEOUT = 129, | |
785 | }; | |
786 | ||
787 | struct bch_dirent { | |
788 | struct bch_val v; | |
789 | ||
790 | /* Target inode number: */ | |
791 | __le64 d_inum; | |
792 | ||
793 | /* | |
794 | * Copy of mode bits 12-15 from the target inode - so userspace can get | |
795 | * the filetype without having to do a stat() | |
796 | */ | |
797 | __u8 d_type; | |
798 | ||
799 | __u8 d_name[]; | |
800 | } __attribute__((packed, aligned(8))); | |
801 | BKEY_VAL_TYPE(dirent, BCH_DIRENT); | |
802 | ||
803 | #define BCH_NAME_MAX (U8_MAX * sizeof(u64) - \ | |
804 | sizeof(struct bkey) - \ | |
805 | offsetof(struct bch_dirent, d_name)) | |
806 | ||
807 | ||
808 | /* Xattrs */ | |
809 | ||
810 | enum { | |
811 | BCH_XATTR = 128, | |
812 | BCH_XATTR_WHITEOUT = 129, | |
813 | }; | |
814 | ||
815 | #define BCH_XATTR_INDEX_USER 0 | |
816 | #define BCH_XATTR_INDEX_POSIX_ACL_ACCESS 1 | |
817 | #define BCH_XATTR_INDEX_POSIX_ACL_DEFAULT 2 | |
818 | #define BCH_XATTR_INDEX_TRUSTED 3 | |
819 | #define BCH_XATTR_INDEX_SECURITY 4 | |
820 | ||
821 | struct bch_xattr { | |
822 | struct bch_val v; | |
823 | __u8 x_type; | |
824 | __u8 x_name_len; | |
825 | __le16 x_val_len; | |
826 | __u8 x_name[]; | |
827 | } __attribute__((packed, aligned(8))); | |
828 | BKEY_VAL_TYPE(xattr, BCH_XATTR); | |
829 | ||
830 | /* Bucket/allocation information: */ | |
831 | ||
832 | enum { | |
833 | BCH_ALLOC = 128, | |
834 | }; | |
835 | ||
836 | enum { | |
837 | BCH_ALLOC_FIELD_READ_TIME = 0, | |
838 | BCH_ALLOC_FIELD_WRITE_TIME = 1, | |
839 | }; | |
840 | ||
841 | struct bch_alloc { | |
842 | struct bch_val v; | |
843 | __u8 fields; | |
844 | __u8 gen; | |
845 | __u8 data[]; | |
846 | } __attribute__((packed, aligned(8))); | |
847 | BKEY_VAL_TYPE(alloc, BCH_ALLOC); | |
848 | ||
849 | /* Quotas: */ | |
850 | ||
851 | enum { | |
852 | BCH_QUOTA = 128, | |
853 | }; | |
854 | ||
855 | enum quota_types { | |
856 | QTYP_USR = 0, | |
857 | QTYP_GRP = 1, | |
858 | QTYP_PRJ = 2, | |
859 | QTYP_NR = 3, | |
860 | }; | |
861 | ||
862 | enum quota_counters { | |
863 | Q_SPC = 0, | |
864 | Q_INO = 1, | |
865 | Q_COUNTERS = 2, | |
866 | }; | |
867 | ||
868 | struct bch_quota_counter { | |
869 | __le64 hardlimit; | |
870 | __le64 softlimit; | |
871 | }; | |
872 | ||
873 | struct bch_quota { | |
874 | struct bch_val v; | |
875 | struct bch_quota_counter c[Q_COUNTERS]; | |
876 | } __attribute__((packed, aligned(8))); | |
877 | BKEY_VAL_TYPE(quota, BCH_QUOTA); | |
878 | ||
879 | /* Optional/variable size superblock sections: */ | |
880 | ||
881 | struct bch_sb_field { | |
882 | __u64 _data[0]; | |
883 | __le32 u64s; | |
884 | __le32 type; | |
885 | }; | |
886 | ||
887 | #define BCH_SB_FIELDS() \ | |
888 | x(journal, 0) \ | |
889 | x(members, 1) \ | |
890 | x(crypt, 2) \ | |
af9d3bc2 | 891 | x(replicas_v0, 3) \ |
1c6fdbd8 KO |
892 | x(quota, 4) \ |
893 | x(disk_groups, 5) \ | |
af9d3bc2 KO |
894 | x(clean, 6) \ |
895 | x(replicas, 7) | |
1c6fdbd8 KO |
896 | |
897 | enum bch_sb_field_type { | |
898 | #define x(f, nr) BCH_SB_FIELD_##f = nr, | |
899 | BCH_SB_FIELDS() | |
900 | #undef x | |
901 | BCH_SB_FIELD_NR | |
902 | }; | |
903 | ||
904 | /* BCH_SB_FIELD_journal: */ | |
905 | ||
906 | struct bch_sb_field_journal { | |
907 | struct bch_sb_field field; | |
908 | __le64 buckets[0]; | |
909 | }; | |
910 | ||
911 | /* BCH_SB_FIELD_members: */ | |
912 | ||
8b335bae KO |
913 | #define BCH_MIN_NR_NBUCKETS (1 << 6) |
914 | ||
1c6fdbd8 KO |
915 | struct bch_member { |
916 | __uuid_t uuid; | |
917 | __le64 nbuckets; /* device size */ | |
918 | __le16 first_bucket; /* index of first bucket used */ | |
919 | __le16 bucket_size; /* sectors */ | |
920 | __le32 pad; | |
921 | __le64 last_mount; /* time_t */ | |
922 | ||
923 | __le64 flags[2]; | |
924 | }; | |
925 | ||
926 | LE64_BITMASK(BCH_MEMBER_STATE, struct bch_member, flags[0], 0, 4) | |
927 | /* 4-10 unused, was TIER, HAS_(META)DATA */ | |
928 | LE64_BITMASK(BCH_MEMBER_REPLACEMENT, struct bch_member, flags[0], 10, 14) | |
929 | LE64_BITMASK(BCH_MEMBER_DISCARD, struct bch_member, flags[0], 14, 15) | |
930 | LE64_BITMASK(BCH_MEMBER_DATA_ALLOWED, struct bch_member, flags[0], 15, 20) | |
931 | LE64_BITMASK(BCH_MEMBER_GROUP, struct bch_member, flags[0], 20, 28) | |
932 | LE64_BITMASK(BCH_MEMBER_DURABILITY, struct bch_member, flags[0], 28, 30) | |
933 | ||
934 | #define BCH_TIER_MAX 4U | |
935 | ||
936 | #if 0 | |
937 | LE64_BITMASK(BCH_MEMBER_NR_READ_ERRORS, struct bch_member, flags[1], 0, 20); | |
938 | LE64_BITMASK(BCH_MEMBER_NR_WRITE_ERRORS,struct bch_member, flags[1], 20, 40); | |
939 | #endif | |
940 | ||
941 | enum bch_member_state { | |
942 | BCH_MEMBER_STATE_RW = 0, | |
943 | BCH_MEMBER_STATE_RO = 1, | |
944 | BCH_MEMBER_STATE_FAILED = 2, | |
945 | BCH_MEMBER_STATE_SPARE = 3, | |
946 | BCH_MEMBER_STATE_NR = 4, | |
947 | }; | |
948 | ||
949 | enum cache_replacement { | |
950 | CACHE_REPLACEMENT_LRU = 0, | |
951 | CACHE_REPLACEMENT_FIFO = 1, | |
952 | CACHE_REPLACEMENT_RANDOM = 2, | |
953 | CACHE_REPLACEMENT_NR = 3, | |
954 | }; | |
955 | ||
956 | struct bch_sb_field_members { | |
957 | struct bch_sb_field field; | |
958 | struct bch_member members[0]; | |
959 | }; | |
960 | ||
961 | /* BCH_SB_FIELD_crypt: */ | |
962 | ||
963 | struct nonce { | |
964 | __le32 d[4]; | |
965 | }; | |
966 | ||
967 | struct bch_key { | |
968 | __le64 key[4]; | |
969 | }; | |
970 | ||
971 | #define BCH_KEY_MAGIC \ | |
972 | (((u64) 'b' << 0)|((u64) 'c' << 8)| \ | |
973 | ((u64) 'h' << 16)|((u64) '*' << 24)| \ | |
974 | ((u64) '*' << 32)|((u64) 'k' << 40)| \ | |
975 | ((u64) 'e' << 48)|((u64) 'y' << 56)) | |
976 | ||
977 | struct bch_encrypted_key { | |
978 | __le64 magic; | |
979 | struct bch_key key; | |
980 | }; | |
981 | ||
982 | /* | |
983 | * If this field is present in the superblock, it stores an encryption key which | |
984 | * is used encrypt all other data/metadata. The key will normally be encrypted | |
985 | * with the key userspace provides, but if encryption has been turned off we'll | |
986 | * just store the master key unencrypted in the superblock so we can access the | |
987 | * previously encrypted data. | |
988 | */ | |
989 | struct bch_sb_field_crypt { | |
990 | struct bch_sb_field field; | |
991 | ||
992 | __le64 flags; | |
993 | __le64 kdf_flags; | |
994 | struct bch_encrypted_key key; | |
995 | }; | |
996 | ||
997 | LE64_BITMASK(BCH_CRYPT_KDF_TYPE, struct bch_sb_field_crypt, flags, 0, 4); | |
998 | ||
999 | enum bch_kdf_types { | |
1000 | BCH_KDF_SCRYPT = 0, | |
1001 | BCH_KDF_NR = 1, | |
1002 | }; | |
1003 | ||
1004 | /* stored as base 2 log of scrypt params: */ | |
1005 | LE64_BITMASK(BCH_KDF_SCRYPT_N, struct bch_sb_field_crypt, kdf_flags, 0, 16); | |
1006 | LE64_BITMASK(BCH_KDF_SCRYPT_R, struct bch_sb_field_crypt, kdf_flags, 16, 32); | |
1007 | LE64_BITMASK(BCH_KDF_SCRYPT_P, struct bch_sb_field_crypt, kdf_flags, 32, 48); | |
1008 | ||
1009 | /* BCH_SB_FIELD_replicas: */ | |
1010 | ||
1011 | enum bch_data_type { | |
1012 | BCH_DATA_NONE = 0, | |
1013 | BCH_DATA_SB = 1, | |
1014 | BCH_DATA_JOURNAL = 2, | |
1015 | BCH_DATA_BTREE = 3, | |
1016 | BCH_DATA_USER = 4, | |
1017 | BCH_DATA_CACHED = 5, | |
1018 | BCH_DATA_NR = 6, | |
1019 | }; | |
1020 | ||
af9d3bc2 KO |
1021 | struct bch_replicas_entry_v0 { |
1022 | __u8 data_type; | |
1023 | __u8 nr_devs; | |
1024 | __u8 devs[]; | |
1025 | } __attribute__((packed)); | |
1026 | ||
1027 | struct bch_sb_field_replicas_v0 { | |
1028 | struct bch_sb_field field; | |
1029 | struct bch_replicas_entry_v0 entries[]; | |
1030 | } __attribute__((packed, aligned(8))); | |
1031 | ||
1c6fdbd8 | 1032 | struct bch_replicas_entry { |
7a920560 KO |
1033 | __u8 data_type; |
1034 | __u8 nr_devs; | |
af9d3bc2 | 1035 | __u8 nr_required; |
7a920560 | 1036 | __u8 devs[]; |
af9d3bc2 | 1037 | } __attribute__((packed)); |
1c6fdbd8 KO |
1038 | |
1039 | struct bch_sb_field_replicas { | |
1040 | struct bch_sb_field field; | |
1041 | struct bch_replicas_entry entries[]; | |
af9d3bc2 | 1042 | } __attribute__((packed, aligned(8))); |
1c6fdbd8 KO |
1043 | |
1044 | /* BCH_SB_FIELD_quota: */ | |
1045 | ||
1046 | struct bch_sb_quota_counter { | |
1047 | __le32 timelimit; | |
1048 | __le32 warnlimit; | |
1049 | }; | |
1050 | ||
1051 | struct bch_sb_quota_type { | |
1052 | __le64 flags; | |
1053 | struct bch_sb_quota_counter c[Q_COUNTERS]; | |
1054 | }; | |
1055 | ||
1056 | struct bch_sb_field_quota { | |
1057 | struct bch_sb_field field; | |
1058 | struct bch_sb_quota_type q[QTYP_NR]; | |
1059 | } __attribute__((packed, aligned(8))); | |
1060 | ||
1061 | /* BCH_SB_FIELD_disk_groups: */ | |
1062 | ||
1063 | #define BCH_SB_LABEL_SIZE 32 | |
1064 | ||
1065 | struct bch_disk_group { | |
1066 | __u8 label[BCH_SB_LABEL_SIZE]; | |
1067 | __le64 flags[2]; | |
1068 | }; | |
1069 | ||
1070 | LE64_BITMASK(BCH_GROUP_DELETED, struct bch_disk_group, flags[0], 0, 1) | |
1071 | LE64_BITMASK(BCH_GROUP_DATA_ALLOWED, struct bch_disk_group, flags[0], 1, 6) | |
1072 | LE64_BITMASK(BCH_GROUP_PARENT, struct bch_disk_group, flags[0], 6, 24) | |
1073 | ||
1074 | struct bch_sb_field_disk_groups { | |
1075 | struct bch_sb_field field; | |
1076 | struct bch_disk_group entries[0]; | |
1077 | }; | |
1078 | ||
1079 | /* | |
1080 | * On clean shutdown, store btree roots and current journal sequence number in | |
1081 | * the superblock: | |
1082 | */ | |
1083 | struct jset_entry { | |
1084 | __le16 u64s; | |
1085 | __u8 btree_id; | |
1086 | __u8 level; | |
1087 | __u8 type; /* designates what this jset holds */ | |
1088 | __u8 pad[3]; | |
1089 | ||
1090 | union { | |
1091 | struct bkey_i start[0]; | |
1092 | __u64 _data[0]; | |
1093 | }; | |
1094 | }; | |
1095 | ||
1096 | struct bch_sb_field_clean { | |
1097 | struct bch_sb_field field; | |
1098 | ||
1099 | __le32 flags; | |
1100 | __le16 read_clock; | |
1101 | __le16 write_clock; | |
1102 | __le64 journal_seq; | |
1103 | ||
1104 | union { | |
1105 | struct jset_entry start[0]; | |
1106 | __u64 _data[0]; | |
1107 | }; | |
1108 | }; | |
1109 | ||
1110 | /* Superblock: */ | |
1111 | ||
1112 | /* | |
1113 | * Version 8: BCH_SB_ENCODED_EXTENT_MAX_BITS | |
1114 | * BCH_MEMBER_DATA_ALLOWED | |
1115 | * Version 9: incompatible extent nonce change | |
1116 | */ | |
1117 | ||
1118 | #define BCH_SB_VERSION_MIN 7 | |
1119 | #define BCH_SB_VERSION_EXTENT_MAX 8 | |
1120 | #define BCH_SB_VERSION_EXTENT_NONCE_V1 9 | |
1121 | #define BCH_SB_VERSION_MAX 9 | |
1122 | ||
1123 | #define BCH_SB_SECTOR 8 | |
1124 | #define BCH_SB_MEMBERS_MAX 64 /* XXX kill */ | |
1125 | ||
1126 | struct bch_sb_layout { | |
1127 | __uuid_t magic; /* bcachefs superblock UUID */ | |
1128 | __u8 layout_type; | |
1129 | __u8 sb_max_size_bits; /* base 2 of 512 byte sectors */ | |
1130 | __u8 nr_superblocks; | |
1131 | __u8 pad[5]; | |
1132 | __le64 sb_offset[61]; | |
1133 | } __attribute__((packed, aligned(8))); | |
1134 | ||
1135 | #define BCH_SB_LAYOUT_SECTOR 7 | |
1136 | ||
1137 | /* | |
1138 | * @offset - sector where this sb was written | |
1139 | * @version - on disk format version | |
1140 | * @magic - identifies as a bcachefs superblock (BCACHE_MAGIC) | |
1141 | * @seq - incremented each time superblock is written | |
1142 | * @uuid - used for generating various magic numbers and identifying | |
1143 | * member devices, never changes | |
1144 | * @user_uuid - user visible UUID, may be changed | |
1145 | * @label - filesystem label | |
1146 | * @seq - identifies most recent superblock, incremented each time | |
1147 | * superblock is written | |
1148 | * @features - enabled incompatible features | |
1149 | */ | |
1150 | struct bch_sb { | |
1151 | struct bch_csum csum; | |
1152 | __le16 version; | |
1153 | __le16 version_min; | |
1154 | __le16 pad[2]; | |
1155 | __uuid_t magic; | |
1156 | __uuid_t uuid; | |
1157 | __uuid_t user_uuid; | |
1158 | __u8 label[BCH_SB_LABEL_SIZE]; | |
1159 | __le64 offset; | |
1160 | __le64 seq; | |
1161 | ||
1162 | __le16 block_size; | |
1163 | __u8 dev_idx; | |
1164 | __u8 nr_devices; | |
1165 | __le32 u64s; | |
1166 | ||
1167 | __le64 time_base_lo; | |
1168 | __le32 time_base_hi; | |
1169 | __le32 time_precision; | |
1170 | ||
1171 | __le64 flags[8]; | |
1172 | __le64 features[2]; | |
1173 | __le64 compat[2]; | |
1174 | ||
1175 | struct bch_sb_layout layout; | |
1176 | ||
1177 | union { | |
1178 | struct bch_sb_field start[0]; | |
1179 | __le64 _data[0]; | |
1180 | }; | |
1181 | } __attribute__((packed, aligned(8))); | |
1182 | ||
1183 | /* | |
1184 | * Flags: | |
1185 | * BCH_SB_INITALIZED - set on first mount | |
1186 | * BCH_SB_CLEAN - did we shut down cleanly? Just a hint, doesn't affect | |
1187 | * behaviour of mount/recovery path: | |
1188 | * BCH_SB_INODE_32BIT - limit inode numbers to 32 bits | |
1189 | * BCH_SB_128_BIT_MACS - 128 bit macs instead of 80 | |
1190 | * BCH_SB_ENCRYPTION_TYPE - if nonzero encryption is enabled; overrides | |
1191 | * DATA/META_CSUM_TYPE. Also indicates encryption | |
1192 | * algorithm in use, if/when we get more than one | |
1193 | */ | |
1194 | ||
1195 | LE16_BITMASK(BCH_SB_BLOCK_SIZE, struct bch_sb, block_size, 0, 16); | |
1196 | ||
1197 | LE64_BITMASK(BCH_SB_INITIALIZED, struct bch_sb, flags[0], 0, 1); | |
1198 | LE64_BITMASK(BCH_SB_CLEAN, struct bch_sb, flags[0], 1, 2); | |
1199 | LE64_BITMASK(BCH_SB_CSUM_TYPE, struct bch_sb, flags[0], 2, 8); | |
1200 | LE64_BITMASK(BCH_SB_ERROR_ACTION, struct bch_sb, flags[0], 8, 12); | |
1201 | ||
1202 | LE64_BITMASK(BCH_SB_BTREE_NODE_SIZE, struct bch_sb, flags[0], 12, 28); | |
1203 | ||
1204 | LE64_BITMASK(BCH_SB_GC_RESERVE, struct bch_sb, flags[0], 28, 33); | |
1205 | LE64_BITMASK(BCH_SB_ROOT_RESERVE, struct bch_sb, flags[0], 33, 40); | |
1206 | ||
1207 | LE64_BITMASK(BCH_SB_META_CSUM_TYPE, struct bch_sb, flags[0], 40, 44); | |
1208 | LE64_BITMASK(BCH_SB_DATA_CSUM_TYPE, struct bch_sb, flags[0], 44, 48); | |
1209 | ||
1210 | LE64_BITMASK(BCH_SB_META_REPLICAS_WANT, struct bch_sb, flags[0], 48, 52); | |
1211 | LE64_BITMASK(BCH_SB_DATA_REPLICAS_WANT, struct bch_sb, flags[0], 52, 56); | |
1212 | ||
1213 | LE64_BITMASK(BCH_SB_POSIX_ACL, struct bch_sb, flags[0], 56, 57); | |
1214 | LE64_BITMASK(BCH_SB_USRQUOTA, struct bch_sb, flags[0], 57, 58); | |
1215 | LE64_BITMASK(BCH_SB_GRPQUOTA, struct bch_sb, flags[0], 58, 59); | |
1216 | LE64_BITMASK(BCH_SB_PRJQUOTA, struct bch_sb, flags[0], 59, 60); | |
1217 | ||
1218 | /* 60-64 unused */ | |
1219 | ||
1220 | LE64_BITMASK(BCH_SB_STR_HASH_TYPE, struct bch_sb, flags[1], 0, 4); | |
1221 | LE64_BITMASK(BCH_SB_COMPRESSION_TYPE, struct bch_sb, flags[1], 4, 8); | |
1222 | LE64_BITMASK(BCH_SB_INODE_32BIT, struct bch_sb, flags[1], 8, 9); | |
1223 | ||
1224 | LE64_BITMASK(BCH_SB_128_BIT_MACS, struct bch_sb, flags[1], 9, 10); | |
1225 | LE64_BITMASK(BCH_SB_ENCRYPTION_TYPE, struct bch_sb, flags[1], 10, 14); | |
1226 | ||
1227 | /* | |
1228 | * Max size of an extent that may require bouncing to read or write | |
1229 | * (checksummed, compressed): 64k | |
1230 | */ | |
1231 | LE64_BITMASK(BCH_SB_ENCODED_EXTENT_MAX_BITS, | |
1232 | struct bch_sb, flags[1], 14, 20); | |
1233 | ||
1234 | LE64_BITMASK(BCH_SB_META_REPLICAS_REQ, struct bch_sb, flags[1], 20, 24); | |
1235 | LE64_BITMASK(BCH_SB_DATA_REPLICAS_REQ, struct bch_sb, flags[1], 24, 28); | |
1236 | ||
1237 | LE64_BITMASK(BCH_SB_PROMOTE_TARGET, struct bch_sb, flags[1], 28, 40); | |
1238 | LE64_BITMASK(BCH_SB_FOREGROUND_TARGET, struct bch_sb, flags[1], 40, 52); | |
1239 | LE64_BITMASK(BCH_SB_BACKGROUND_TARGET, struct bch_sb, flags[1], 52, 64); | |
1240 | ||
1241 | LE64_BITMASK(BCH_SB_BACKGROUND_COMPRESSION_TYPE, | |
1242 | struct bch_sb, flags[2], 0, 4); | |
a50ed7c8 | 1243 | LE64_BITMASK(BCH_SB_GC_RESERVE_BYTES, struct bch_sb, flags[2], 4, 64); |
1c6fdbd8 KO |
1244 | |
1245 | /* Features: */ | |
1246 | enum bch_sb_features { | |
1247 | BCH_FEATURE_LZ4 = 0, | |
1248 | BCH_FEATURE_GZIP = 1, | |
1249 | BCH_FEATURE_ZSTD = 2, | |
c258f28e KO |
1250 | BCH_FEATURE_ATOMIC_NLINK = 3, /* should have gone under compat */ |
1251 | BCH_FEATURE_NR, | |
1c6fdbd8 KO |
1252 | }; |
1253 | ||
1254 | /* options: */ | |
1255 | ||
1256 | #define BCH_REPLICAS_MAX 4U | |
1257 | ||
1258 | enum bch_error_actions { | |
1259 | BCH_ON_ERROR_CONTINUE = 0, | |
1260 | BCH_ON_ERROR_RO = 1, | |
1261 | BCH_ON_ERROR_PANIC = 2, | |
1262 | BCH_NR_ERROR_ACTIONS = 3, | |
1263 | }; | |
1264 | ||
1265 | enum bch_csum_opts { | |
1266 | BCH_CSUM_OPT_NONE = 0, | |
1267 | BCH_CSUM_OPT_CRC32C = 1, | |
1268 | BCH_CSUM_OPT_CRC64 = 2, | |
1269 | BCH_CSUM_OPT_NR = 3, | |
1270 | }; | |
1271 | ||
1272 | enum bch_str_hash_opts { | |
1273 | BCH_STR_HASH_CRC32C = 0, | |
1274 | BCH_STR_HASH_CRC64 = 1, | |
1275 | BCH_STR_HASH_SIPHASH = 2, | |
1276 | BCH_STR_HASH_NR = 3, | |
1277 | }; | |
1278 | ||
1279 | #define BCH_COMPRESSION_TYPES() \ | |
1280 | x(NONE) \ | |
1281 | x(LZ4) \ | |
1282 | x(GZIP) \ | |
1283 | x(ZSTD) | |
1284 | ||
1285 | enum bch_compression_opts { | |
1286 | #define x(t) BCH_COMPRESSION_OPT_##t, | |
1287 | BCH_COMPRESSION_TYPES() | |
1288 | #undef x | |
1289 | BCH_COMPRESSION_OPT_NR | |
1290 | }; | |
1291 | ||
1292 | /* | |
1293 | * Magic numbers | |
1294 | * | |
1295 | * The various other data structures have their own magic numbers, which are | |
1296 | * xored with the first part of the cache set's UUID | |
1297 | */ | |
1298 | ||
1299 | #define BCACHE_MAGIC \ | |
1300 | UUID_INIT(0xc68573f6, 0x4e1a, 0x45ca, \ | |
1301 | 0x82, 0x65, 0xf5, 0x7f, 0x48, 0xba, 0x6d, 0x81) | |
1302 | #define BCHFS_MAGIC \ | |
1303 | UUID_INIT(0xc68573f6, 0x66ce, 0x90a9, \ | |
1304 | 0xd9, 0x6a, 0x60, 0xcf, 0x80, 0x3d, 0xf7, 0xef) | |
1305 | ||
1306 | #define BCACHEFS_STATFS_MAGIC 0xca451a4e | |
1307 | ||
1308 | #define JSET_MAGIC __cpu_to_le64(0x245235c1a3625032ULL) | |
1309 | #define BSET_MAGIC __cpu_to_le64(0x90135c78b99e07f5ULL) | |
1310 | ||
1311 | static inline __le64 __bch2_sb_magic(struct bch_sb *sb) | |
1312 | { | |
1313 | __le64 ret; | |
1314 | memcpy(&ret, &sb->uuid, sizeof(ret)); | |
1315 | return ret; | |
1316 | } | |
1317 | ||
1318 | static inline __u64 __jset_magic(struct bch_sb *sb) | |
1319 | { | |
1320 | return __le64_to_cpu(__bch2_sb_magic(sb) ^ JSET_MAGIC); | |
1321 | } | |
1322 | ||
1323 | static inline __u64 __bset_magic(struct bch_sb *sb) | |
1324 | { | |
1325 | return __le64_to_cpu(__bch2_sb_magic(sb) ^ BSET_MAGIC); | |
1326 | } | |
1327 | ||
1328 | /* Journal */ | |
1329 | ||
1330 | #define BCACHE_JSET_VERSION_UUIDv1 1 | |
1331 | #define BCACHE_JSET_VERSION_UUID 1 /* Always latest UUID format */ | |
1332 | #define BCACHE_JSET_VERSION_JKEYS 2 | |
1333 | #define BCACHE_JSET_VERSION 2 | |
1334 | ||
1335 | #define JSET_KEYS_U64s (sizeof(struct jset_entry) / sizeof(__u64)) | |
1336 | ||
1337 | #define BCH_JSET_ENTRY_TYPES() \ | |
1338 | x(btree_keys, 0) \ | |
1339 | x(btree_root, 1) \ | |
1340 | x(prio_ptrs, 2) \ | |
1341 | x(blacklist, 3) \ | |
1342 | x(blacklist_v2, 4) | |
1343 | ||
1344 | enum { | |
1345 | #define x(f, nr) BCH_JSET_ENTRY_##f = nr, | |
1346 | BCH_JSET_ENTRY_TYPES() | |
1347 | #undef x | |
1348 | BCH_JSET_ENTRY_NR | |
1349 | }; | |
1350 | ||
1351 | /* | |
1352 | * Journal sequence numbers can be blacklisted: bsets record the max sequence | |
1353 | * number of all the journal entries they contain updates for, so that on | |
1354 | * recovery we can ignore those bsets that contain index updates newer that what | |
1355 | * made it into the journal. | |
1356 | * | |
1357 | * This means that we can't reuse that journal_seq - we have to skip it, and | |
1358 | * then record that we skipped it so that the next time we crash and recover we | |
1359 | * don't think there was a missing journal entry. | |
1360 | */ | |
1361 | struct jset_entry_blacklist { | |
1362 | struct jset_entry entry; | |
1363 | __le64 seq; | |
1364 | }; | |
1365 | ||
1366 | struct jset_entry_blacklist_v2 { | |
1367 | struct jset_entry entry; | |
1368 | __le64 start; | |
1369 | __le64 end; | |
1370 | }; | |
1371 | ||
1372 | /* | |
1373 | * On disk format for a journal entry: | |
1374 | * seq is monotonically increasing; every journal entry has its own unique | |
1375 | * sequence number. | |
1376 | * | |
1377 | * last_seq is the oldest journal entry that still has keys the btree hasn't | |
1378 | * flushed to disk yet. | |
1379 | * | |
1380 | * version is for on disk format changes. | |
1381 | */ | |
1382 | struct jset { | |
1383 | struct bch_csum csum; | |
1384 | ||
1385 | __le64 magic; | |
1386 | __le64 seq; | |
1387 | __le32 version; | |
1388 | __le32 flags; | |
1389 | ||
1390 | __le32 u64s; /* size of d[] in u64s */ | |
1391 | ||
1392 | __u8 encrypted_start[0]; | |
1393 | ||
1394 | __le16 read_clock; | |
1395 | __le16 write_clock; | |
1396 | ||
1397 | /* Sequence number of oldest dirty journal entry */ | |
1398 | __le64 last_seq; | |
1399 | ||
1400 | ||
1401 | union { | |
1402 | struct jset_entry start[0]; | |
1403 | __u64 _data[0]; | |
1404 | }; | |
1405 | } __attribute__((packed, aligned(8))); | |
1406 | ||
1407 | LE32_BITMASK(JSET_CSUM_TYPE, struct jset, flags, 0, 4); | |
1408 | LE32_BITMASK(JSET_BIG_ENDIAN, struct jset, flags, 4, 5); | |
1409 | ||
8b335bae | 1410 | #define BCH_JOURNAL_BUCKETS_MIN 8 |
1c6fdbd8 KO |
1411 | |
1412 | /* Btree: */ | |
1413 | ||
1414 | #define DEFINE_BCH_BTREE_IDS() \ | |
1415 | DEF_BTREE_ID(EXTENTS, 0, "extents") \ | |
1416 | DEF_BTREE_ID(INODES, 1, "inodes") \ | |
1417 | DEF_BTREE_ID(DIRENTS, 2, "dirents") \ | |
1418 | DEF_BTREE_ID(XATTRS, 3, "xattrs") \ | |
1419 | DEF_BTREE_ID(ALLOC, 4, "alloc") \ | |
1420 | DEF_BTREE_ID(QUOTAS, 5, "quotas") | |
1421 | ||
1422 | #define DEF_BTREE_ID(kwd, val, name) BTREE_ID_##kwd = val, | |
1423 | ||
1424 | enum btree_id { | |
1425 | DEFINE_BCH_BTREE_IDS() | |
1426 | BTREE_ID_NR | |
1427 | }; | |
1428 | ||
1429 | #undef DEF_BTREE_ID | |
1430 | ||
1431 | #define BTREE_MAX_DEPTH 4U | |
1432 | ||
1433 | /* Btree nodes */ | |
1434 | ||
1435 | /* Version 1: Seed pointer into btree node checksum | |
1436 | */ | |
1437 | #define BCACHE_BSET_CSUM 1 | |
1438 | #define BCACHE_BSET_KEY_v1 2 | |
1439 | #define BCACHE_BSET_JOURNAL_SEQ 3 | |
1440 | #define BCACHE_BSET_VERSION 3 | |
1441 | ||
1442 | /* | |
1443 | * Btree nodes | |
1444 | * | |
1445 | * On disk a btree node is a list/log of these; within each set the keys are | |
1446 | * sorted | |
1447 | */ | |
1448 | struct bset { | |
1449 | __le64 seq; | |
1450 | ||
1451 | /* | |
1452 | * Highest journal entry this bset contains keys for. | |
1453 | * If on recovery we don't see that journal entry, this bset is ignored: | |
1454 | * this allows us to preserve the order of all index updates after a | |
1455 | * crash, since the journal records a total order of all index updates | |
1456 | * and anything that didn't make it to the journal doesn't get used. | |
1457 | */ | |
1458 | __le64 journal_seq; | |
1459 | ||
1460 | __le32 flags; | |
1461 | __le16 version; | |
1462 | __le16 u64s; /* count of d[] in u64s */ | |
1463 | ||
1464 | union { | |
1465 | struct bkey_packed start[0]; | |
1466 | __u64 _data[0]; | |
1467 | }; | |
1468 | } __attribute__((packed, aligned(8))); | |
1469 | ||
1470 | LE32_BITMASK(BSET_CSUM_TYPE, struct bset, flags, 0, 4); | |
1471 | ||
1472 | LE32_BITMASK(BSET_BIG_ENDIAN, struct bset, flags, 4, 5); | |
1473 | LE32_BITMASK(BSET_SEPARATE_WHITEOUTS, | |
1474 | struct bset, flags, 5, 6); | |
1475 | ||
1476 | struct btree_node { | |
1477 | struct bch_csum csum; | |
1478 | __le64 magic; | |
1479 | ||
1480 | /* this flags field is encrypted, unlike bset->flags: */ | |
1481 | __le64 flags; | |
1482 | ||
1483 | /* Closed interval: */ | |
1484 | struct bpos min_key; | |
1485 | struct bpos max_key; | |
1486 | struct bch_extent_ptr ptr; | |
1487 | struct bkey_format format; | |
1488 | ||
1489 | union { | |
1490 | struct bset keys; | |
1491 | struct { | |
1492 | __u8 pad[22]; | |
1493 | __le16 u64s; | |
1494 | __u64 _data[0]; | |
1495 | ||
1496 | }; | |
1497 | }; | |
1498 | } __attribute__((packed, aligned(8))); | |
1499 | ||
1500 | LE64_BITMASK(BTREE_NODE_ID, struct btree_node, flags, 0, 4); | |
1501 | LE64_BITMASK(BTREE_NODE_LEVEL, struct btree_node, flags, 4, 8); | |
1502 | /* 8-32 unused */ | |
1503 | LE64_BITMASK(BTREE_NODE_SEQ, struct btree_node, flags, 32, 64); | |
1504 | ||
1505 | struct btree_node_entry { | |
1506 | struct bch_csum csum; | |
1507 | ||
1508 | union { | |
1509 | struct bset keys; | |
1510 | struct { | |
1511 | __u8 pad[22]; | |
1512 | __le16 u64s; | |
1513 | __u64 _data[0]; | |
1514 | ||
1515 | }; | |
1516 | }; | |
1517 | } __attribute__((packed, aligned(8))); | |
1518 | ||
1519 | #endif /* _BCACHEFS_FORMAT_H */ |