1 | // SPDX-License-Identifier: GPL-2.0 |
2 | |
3 | #include "bcachefs.h" |
4 | #include "backpointers.h" |
5 | #include "bkey_methods.h" |
6 | #include "btree_cache.h" |
7 | #include "btree_types.h" |
8 | #include "alloc_background.h" |
9 | #include "dirent.h" |
10 | #include "ec.h" |
11 | #include "error.h" |
12 | #include "extents.h" |
13 | #include "inode.h" |
14 | #include "io_misc.h" |
15 | #include "lru.h" |
16 | #include "quota.h" |
17 | #include "reflink.h" |
18 | #include "snapshot.h" |
19 | #include "subvolume.h" |
20 | #include "xattr.h" |
21 | |
22 | const char * const bch2_bkey_types[] = { |
23 | #define x(name, nr) #name, |
24 | BCH_BKEY_TYPES() |
25 | #undef x |
26 | NULL |
27 | }; |
28 | |
29 | static int deleted_key_invalid(struct bch_fs *c, struct bkey_s_c k, |
30 | enum bkey_invalid_flags flags, struct printbuf *err) |
31 | { |
32 | return 0; |
33 | } |
34 | |
35 | #define bch2_bkey_ops_deleted ((struct bkey_ops) { \ |
36 | .key_invalid = deleted_key_invalid, \ |
37 | }) |
38 | |
39 | #define bch2_bkey_ops_whiteout ((struct bkey_ops) { \ |
40 | .key_invalid = deleted_key_invalid, \ |
41 | }) |
42 | |
43 | static int empty_val_key_invalid(struct bch_fs *c, struct bkey_s_c k, |
44 | enum bkey_invalid_flags flags, struct printbuf *err) |
45 | { |
46 | int ret = 0; |
47 | |
48 | bkey_fsck_err_on(bkey_val_bytes(k.k), c, err, |
49 | bkey_val_size_nonzero, |
50 | "incorrect value size (%zu != 0)" , |
51 | bkey_val_bytes(k.k)); |
52 | fsck_err: |
53 | return ret; |
54 | } |
55 | |
56 | #define bch2_bkey_ops_error ((struct bkey_ops) { \ |
57 | .key_invalid = empty_val_key_invalid, \ |
58 | }) |
59 | |
60 | static int key_type_cookie_invalid(struct bch_fs *c, struct bkey_s_c k, |
61 | enum bkey_invalid_flags flags, struct printbuf *err) |
62 | { |
63 | return 0; |
64 | } |
65 | |
66 | static void key_type_cookie_to_text(struct printbuf *out, struct bch_fs *c, |
67 | struct bkey_s_c k) |
68 | { |
69 | struct bkey_s_c_cookie ck = bkey_s_c_to_cookie(k); |
70 | |
71 | prt_printf(out, "%llu" , le64_to_cpu(ck.v->cookie)); |
72 | } |
73 | |
74 | #define bch2_bkey_ops_cookie ((struct bkey_ops) { \ |
75 | .key_invalid = key_type_cookie_invalid, \ |
76 | .val_to_text = key_type_cookie_to_text, \ |
77 | .min_val_size = 8, \ |
78 | }) |
79 | |
80 | #define bch2_bkey_ops_hash_whiteout ((struct bkey_ops) {\ |
81 | .key_invalid = empty_val_key_invalid, \ |
82 | }) |
83 | |
84 | static int key_type_inline_data_invalid(struct bch_fs *c, struct bkey_s_c k, |
85 | enum bkey_invalid_flags flags, struct printbuf *err) |
86 | { |
87 | return 0; |
88 | } |
89 | |
90 | static void key_type_inline_data_to_text(struct printbuf *out, struct bch_fs *c, |
91 | struct bkey_s_c k) |
92 | { |
93 | struct bkey_s_c_inline_data d = bkey_s_c_to_inline_data(k); |
94 | unsigned datalen = bkey_inline_data_bytes(k: k.k); |
95 | |
96 | prt_printf(out, "datalen %u: %*phN" , |
97 | datalen, min(datalen, 32U), d.v->data); |
98 | } |
99 | |
100 | #define bch2_bkey_ops_inline_data ((struct bkey_ops) { \ |
101 | .key_invalid = key_type_inline_data_invalid, \ |
102 | .val_to_text = key_type_inline_data_to_text, \ |
103 | }) |
104 | |
105 | static bool key_type_set_merge(struct bch_fs *c, struct bkey_s l, struct bkey_s_c r) |
106 | { |
107 | bch2_key_resize(k: l.k, new_size: l.k->size + r.k->size); |
108 | return true; |
109 | } |
110 | |
111 | #define bch2_bkey_ops_set ((struct bkey_ops) { \ |
112 | .key_invalid = empty_val_key_invalid, \ |
113 | .key_merge = key_type_set_merge, \ |
114 | }) |
115 | |
116 | const struct bkey_ops bch2_bkey_ops[] = { |
117 | #define x(name, nr) [KEY_TYPE_##name] = bch2_bkey_ops_##name, |
118 | BCH_BKEY_TYPES() |
119 | #undef x |
120 | }; |
121 | |
122 | const struct bkey_ops bch2_bkey_null_ops = { |
123 | }; |
124 | |
125 | int bch2_bkey_val_invalid(struct bch_fs *c, struct bkey_s_c k, |
126 | enum bkey_invalid_flags flags, |
127 | struct printbuf *err) |
128 | { |
129 | const struct bkey_ops *ops = bch2_bkey_type_ops(type: k.k->type); |
130 | int ret = 0; |
131 | |
132 | bkey_fsck_err_on(bkey_val_bytes(k.k) < ops->min_val_size, c, err, |
133 | bkey_val_size_too_small, |
134 | "bad val size (%zu < %u)" , |
135 | bkey_val_bytes(k.k), ops->min_val_size); |
136 | |
137 | if (!ops->key_invalid) |
138 | return 0; |
139 | |
140 | ret = ops->key_invalid(c, k, flags, err); |
141 | fsck_err: |
142 | return ret; |
143 | } |
144 | |
145 | static u64 bch2_key_types_allowed[] = { |
146 | [BKEY_TYPE_btree] = |
147 | BIT_ULL(KEY_TYPE_deleted)| |
148 | BIT_ULL(KEY_TYPE_btree_ptr)| |
149 | BIT_ULL(KEY_TYPE_btree_ptr_v2), |
150 | #define x(name, nr, flags, keys) [BKEY_TYPE_##name] = BIT_ULL(KEY_TYPE_deleted)|keys, |
151 | BCH_BTREE_IDS() |
152 | #undef x |
153 | }; |
154 | |
155 | const char *bch2_btree_node_type_str(enum btree_node_type type) |
156 | { |
157 | return type == BKEY_TYPE_btree ? "internal btree node" : bch2_btree_id_str(type - 1); |
158 | } |
159 | |
160 | int __bch2_bkey_invalid(struct bch_fs *c, struct bkey_s_c k, |
161 | enum btree_node_type type, |
162 | enum bkey_invalid_flags flags, |
163 | struct printbuf *err) |
164 | { |
165 | int ret = 0; |
166 | |
167 | bkey_fsck_err_on(k.k->u64s < BKEY_U64s, c, err, |
168 | bkey_u64s_too_small, |
169 | "u64s too small (%u < %zu)" , k.k->u64s, BKEY_U64s); |
170 | |
171 | if (type >= BKEY_TYPE_NR) |
172 | return 0; |
173 | |
174 | bkey_fsck_err_on((type == BKEY_TYPE_btree || |
175 | (flags & BKEY_INVALID_COMMIT)) && |
176 | !(bch2_key_types_allowed[type] & BIT_ULL(k.k->type)), c, err, |
177 | bkey_invalid_type_for_btree, |
178 | "invalid key type for btree %s (%s)" , |
179 | bch2_btree_node_type_str(type), |
180 | k.k->type < KEY_TYPE_MAX |
181 | ? bch2_bkey_types[k.k->type] |
182 | : "(unknown)" ); |
183 | |
184 | if (btree_node_type_is_extents(type) && !bkey_whiteout(k.k)) { |
185 | bkey_fsck_err_on(k.k->size == 0, c, err, |
186 | bkey_extent_size_zero, |
187 | "size == 0" ); |
188 | |
189 | bkey_fsck_err_on(k.k->size > k.k->p.offset, c, err, |
190 | bkey_extent_size_greater_than_offset, |
191 | "size greater than offset (%u > %llu)" , |
192 | k.k->size, k.k->p.offset); |
193 | } else { |
194 | bkey_fsck_err_on(k.k->size, c, err, |
195 | bkey_size_nonzero, |
196 | "size != 0" ); |
197 | } |
198 | |
199 | if (type != BKEY_TYPE_btree) { |
200 | enum btree_id btree = type - 1; |
201 | |
202 | if (btree_type_has_snapshots(id: btree)) { |
203 | bkey_fsck_err_on(!k.k->p.snapshot, c, err, |
204 | bkey_snapshot_zero, |
205 | "snapshot == 0" ); |
206 | } else if (!btree_type_has_snapshot_field(id: btree)) { |
207 | bkey_fsck_err_on(k.k->p.snapshot, c, err, |
208 | bkey_snapshot_nonzero, |
209 | "nonzero snapshot" ); |
210 | } else { |
211 | /* |
212 | * btree uses snapshot field but it's not required to be |
213 | * nonzero |
214 | */ |
215 | } |
216 | |
217 | bkey_fsck_err_on(bkey_eq(k.k->p, POS_MAX), c, err, |
218 | bkey_at_pos_max, |
219 | "key at POS_MAX" ); |
220 | } |
221 | fsck_err: |
222 | return ret; |
223 | } |
224 | |
225 | int bch2_bkey_invalid(struct bch_fs *c, struct bkey_s_c k, |
226 | enum btree_node_type type, |
227 | enum bkey_invalid_flags flags, |
228 | struct printbuf *err) |
229 | { |
230 | return __bch2_bkey_invalid(c, k, type, flags, err) ?: |
231 | bch2_bkey_val_invalid(c, k, flags, err); |
232 | } |
233 | |
234 | int bch2_bkey_in_btree_node(struct bch_fs *c, struct btree *b, |
235 | struct bkey_s_c k, struct printbuf *err) |
236 | { |
237 | int ret = 0; |
238 | |
239 | bkey_fsck_err_on(bpos_lt(k.k->p, b->data->min_key), c, err, |
240 | bkey_before_start_of_btree_node, |
241 | "key before start of btree node" ); |
242 | |
243 | bkey_fsck_err_on(bpos_gt(k.k->p, b->data->max_key), c, err, |
244 | bkey_after_end_of_btree_node, |
245 | "key past end of btree node" ); |
246 | fsck_err: |
247 | return ret; |
248 | } |
249 | |
250 | void bch2_bpos_to_text(struct printbuf *out, struct bpos pos) |
251 | { |
252 | if (bpos_eq(l: pos, POS_MIN)) |
253 | prt_printf(out, "POS_MIN" ); |
254 | else if (bpos_eq(l: pos, POS_MAX)) |
255 | prt_printf(out, "POS_MAX" ); |
256 | else if (bpos_eq(l: pos, SPOS_MAX)) |
257 | prt_printf(out, "SPOS_MAX" ); |
258 | else { |
259 | if (pos.inode == U64_MAX) |
260 | prt_printf(out, "U64_MAX" ); |
261 | else |
262 | prt_printf(out, "%llu" , pos.inode); |
263 | prt_printf(out, ":" ); |
264 | if (pos.offset == U64_MAX) |
265 | prt_printf(out, "U64_MAX" ); |
266 | else |
267 | prt_printf(out, "%llu" , pos.offset); |
268 | prt_printf(out, ":" ); |
269 | if (pos.snapshot == U32_MAX) |
270 | prt_printf(out, "U32_MAX" ); |
271 | else |
272 | prt_printf(out, "%u" , pos.snapshot); |
273 | } |
274 | } |
275 | |
276 | void bch2_bkey_to_text(struct printbuf *out, const struct bkey *k) |
277 | { |
278 | if (k) { |
279 | prt_printf(out, "u64s %u type " , k->u64s); |
280 | |
281 | if (k->type < KEY_TYPE_MAX) |
282 | prt_printf(out, "%s " , bch2_bkey_types[k->type]); |
283 | else |
284 | prt_printf(out, "%u " , k->type); |
285 | |
286 | bch2_bpos_to_text(out, pos: k->p); |
287 | |
288 | prt_printf(out, " len %u ver %llu" , k->size, k->version.lo); |
289 | } else { |
290 | prt_printf(out, "(null)" ); |
291 | } |
292 | } |
293 | |
294 | void bch2_val_to_text(struct printbuf *out, struct bch_fs *c, |
295 | struct bkey_s_c k) |
296 | { |
297 | const struct bkey_ops *ops = bch2_bkey_type_ops(type: k.k->type); |
298 | |
299 | if (likely(ops->val_to_text)) |
300 | ops->val_to_text(out, c, k); |
301 | } |
302 | |
303 | void bch2_bkey_val_to_text(struct printbuf *out, struct bch_fs *c, |
304 | struct bkey_s_c k) |
305 | { |
306 | bch2_bkey_to_text(out, k: k.k); |
307 | |
308 | if (bkey_val_bytes(k: k.k)) { |
309 | prt_printf(out, ": " ); |
310 | bch2_val_to_text(out, c, k); |
311 | } |
312 | } |
313 | |
314 | void bch2_bkey_swab_val(struct bkey_s k) |
315 | { |
316 | const struct bkey_ops *ops = bch2_bkey_type_ops(type: k.k->type); |
317 | |
318 | if (ops->swab) |
319 | ops->swab(k); |
320 | } |
321 | |
322 | bool bch2_bkey_normalize(struct bch_fs *c, struct bkey_s k) |
323 | { |
324 | const struct bkey_ops *ops = bch2_bkey_type_ops(type: k.k->type); |
325 | |
326 | return ops->key_normalize |
327 | ? ops->key_normalize(c, k) |
328 | : false; |
329 | } |
330 | |
331 | bool bch2_bkey_merge(struct bch_fs *c, struct bkey_s l, struct bkey_s_c r) |
332 | { |
333 | const struct bkey_ops *ops = bch2_bkey_type_ops(type: l.k->type); |
334 | |
335 | return ops->key_merge && |
336 | bch2_bkey_maybe_mergable(l: l.k, r: r.k) && |
337 | (u64) l.k->size + r.k->size <= KEY_SIZE_MAX && |
338 | !bch2_key_merging_disabled && |
339 | ops->key_merge(c, l, r); |
340 | } |
341 | |
342 | static const struct old_bkey_type { |
343 | u8 btree_node_type; |
344 | u8 old; |
345 | u8 new; |
346 | } bkey_renumber_table[] = { |
347 | {BKEY_TYPE_btree, 128, KEY_TYPE_btree_ptr }, |
348 | {BKEY_TYPE_extents, 128, KEY_TYPE_extent }, |
349 | {BKEY_TYPE_extents, 129, KEY_TYPE_extent }, |
350 | {BKEY_TYPE_extents, 130, KEY_TYPE_reservation }, |
351 | {BKEY_TYPE_inodes, 128, KEY_TYPE_inode }, |
352 | {BKEY_TYPE_inodes, 130, KEY_TYPE_inode_generation }, |
353 | {BKEY_TYPE_dirents, 128, KEY_TYPE_dirent }, |
354 | {BKEY_TYPE_dirents, 129, KEY_TYPE_hash_whiteout }, |
355 | {BKEY_TYPE_xattrs, 128, KEY_TYPE_xattr }, |
356 | {BKEY_TYPE_xattrs, 129, KEY_TYPE_hash_whiteout }, |
357 | {BKEY_TYPE_alloc, 128, KEY_TYPE_alloc }, |
358 | {BKEY_TYPE_quotas, 128, KEY_TYPE_quota }, |
359 | }; |
360 | |
361 | void bch2_bkey_renumber(enum btree_node_type btree_node_type, |
362 | struct bkey_packed *k, |
363 | int write) |
364 | { |
365 | const struct old_bkey_type *i; |
366 | |
367 | for (i = bkey_renumber_table; |
368 | i < bkey_renumber_table + ARRAY_SIZE(bkey_renumber_table); |
369 | i++) |
370 | if (btree_node_type == i->btree_node_type && |
371 | k->type == (write ? i->new : i->old)) { |
372 | k->type = write ? i->old : i->new; |
373 | break; |
374 | } |
375 | } |
376 | |
377 | void __bch2_bkey_compat(unsigned level, enum btree_id btree_id, |
378 | unsigned version, unsigned big_endian, |
379 | int write, |
380 | struct bkey_format *f, |
381 | struct bkey_packed *k) |
382 | { |
383 | const struct bkey_ops *ops; |
384 | struct bkey uk; |
385 | unsigned nr_compat = 5; |
386 | int i; |
387 | |
388 | /* |
389 | * Do these operations in reverse order in the write path: |
390 | */ |
391 | |
392 | for (i = 0; i < nr_compat; i++) |
393 | switch (!write ? i : nr_compat - 1 - i) { |
394 | case 0: |
395 | if (big_endian != CPU_BIG_ENDIAN) |
396 | bch2_bkey_swab_key(f, k); |
397 | break; |
398 | case 1: |
399 | if (version < bcachefs_metadata_version_bkey_renumber) |
400 | bch2_bkey_renumber(btree_node_type: __btree_node_type(level, id: btree_id), k, write); |
401 | break; |
402 | case 2: |
403 | if (version < bcachefs_metadata_version_inode_btree_change && |
404 | btree_id == BTREE_ID_inodes) { |
405 | if (!bkey_packed(k)) { |
406 | struct bkey_i *u = packed_to_bkey(k); |
407 | |
408 | swap(u->k.p.inode, u->k.p.offset); |
409 | } else if (f->bits_per_field[BKEY_FIELD_INODE] && |
410 | f->bits_per_field[BKEY_FIELD_OFFSET]) { |
411 | struct bkey_format tmp = *f, *in = f, *out = &tmp; |
412 | |
413 | swap(tmp.bits_per_field[BKEY_FIELD_INODE], |
414 | tmp.bits_per_field[BKEY_FIELD_OFFSET]); |
415 | swap(tmp.field_offset[BKEY_FIELD_INODE], |
416 | tmp.field_offset[BKEY_FIELD_OFFSET]); |
417 | |
418 | if (!write) |
419 | swap(in, out); |
420 | |
421 | uk = __bch2_bkey_unpack_key(in, k); |
422 | swap(uk.p.inode, uk.p.offset); |
423 | BUG_ON(!bch2_bkey_pack_key(k, &uk, out)); |
424 | } |
425 | } |
426 | break; |
427 | case 3: |
428 | if (version < bcachefs_metadata_version_snapshot && |
429 | (level || btree_type_has_snapshots(id: btree_id))) { |
430 | struct bkey_i *u = packed_to_bkey(k); |
431 | |
432 | if (u) { |
433 | u->k.p.snapshot = write |
434 | ? 0 : U32_MAX; |
435 | } else { |
436 | u64 min_packed = le64_to_cpu(f->field_offset[BKEY_FIELD_SNAPSHOT]); |
437 | u64 max_packed = min_packed + |
438 | ~(~0ULL << f->bits_per_field[BKEY_FIELD_SNAPSHOT]); |
439 | |
440 | uk = __bch2_bkey_unpack_key(f, k); |
441 | uk.p.snapshot = write |
442 | ? min_packed : min_t(u64, U32_MAX, max_packed); |
443 | |
444 | BUG_ON(!bch2_bkey_pack_key(k, &uk, f)); |
445 | } |
446 | } |
447 | |
448 | break; |
449 | case 4: { |
450 | struct bkey_s u; |
451 | |
452 | if (!bkey_packed(k)) { |
453 | u = bkey_i_to_s(k: packed_to_bkey(k)); |
454 | } else { |
455 | uk = __bch2_bkey_unpack_key(f, k); |
456 | u.k = &uk; |
457 | u.v = bkeyp_val(f, k); |
458 | } |
459 | |
460 | if (big_endian != CPU_BIG_ENDIAN) |
461 | bch2_bkey_swab_val(k: u); |
462 | |
463 | ops = bch2_bkey_type_ops(type: k->type); |
464 | |
465 | if (ops->compat) |
466 | ops->compat(btree_id, version, big_endian, write, u); |
467 | break; |
468 | } |
469 | default: |
470 | BUG(); |
471 | } |
472 | } |
473 | |