xref: /linux-6.15/fs/bcachefs/bkey_methods.c (revision bceb86be) 
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);
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(l.k, 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(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((flags & BKEY_INVALID_COMMIT) &&
175 			 !(bch2_key_types_allowed[type] & BIT_ULL(k.k->type)), c, err,
176 			 bkey_invalid_type_for_btree,
177 			 "invalid key type for btree %s (%s)",
178 			 bch2_btree_node_type_str(type),
179 			 k.k->type < KEY_TYPE_MAX
180 			 ? bch2_bkey_types[k.k->type]
181 			 : "(unknown)");
182 
183 	if (btree_node_type_is_extents(type) && !bkey_whiteout(k.k)) {
184 		bkey_fsck_err_on(k.k->size == 0, c, err,
185 				 bkey_extent_size_zero,
186 				 "size == 0");
187 
188 		bkey_fsck_err_on(k.k->size > k.k->p.offset, c, err,
189 				 bkey_extent_size_greater_than_offset,
190 				 "size greater than offset (%u > %llu)",
191 				 k.k->size, k.k->p.offset);
192 	} else {
193 		bkey_fsck_err_on(k.k->size, c, err,
194 				 bkey_size_nonzero,
195 				 "size != 0");
196 	}
197 
198 	if (type != BKEY_TYPE_btree) {
199 		enum btree_id btree = type - 1;
200 
201 		if (btree_type_has_snapshots(btree)) {
202 			bkey_fsck_err_on(!k.k->p.snapshot, c, err,
203 					 bkey_snapshot_zero,
204 					 "snapshot == 0");
205 		} else if (!btree_type_has_snapshot_field(btree)) {
206 			bkey_fsck_err_on(k.k->p.snapshot, c, err,
207 					 bkey_snapshot_nonzero,
208 					 "nonzero snapshot");
209 		} else {
210 			/*
211 			 * btree uses snapshot field but it's not required to be
212 			 * nonzero
213 			 */
214 		}
215 
216 		bkey_fsck_err_on(bkey_eq(k.k->p, POS_MAX), c, err,
217 				 bkey_at_pos_max,
218 				 "key at POS_MAX");
219 	}
220 fsck_err:
221 	return ret;
222 }
223 
224 int bch2_bkey_invalid(struct bch_fs *c, struct bkey_s_c k,
225 		      enum btree_node_type type,
226 		      enum bkey_invalid_flags flags,
227 		      struct printbuf *err)
228 {
229 	return __bch2_bkey_invalid(c, k, type, flags, err) ?:
230 		bch2_bkey_val_invalid(c, k, flags, err);
231 }
232 
233 int bch2_bkey_in_btree_node(struct bch_fs *c, struct btree *b,
234 			    struct bkey_s_c k, struct printbuf *err)
235 {
236 	int ret = 0;
237 
238 	bkey_fsck_err_on(bpos_lt(k.k->p, b->data->min_key), c, err,
239 			 bkey_before_start_of_btree_node,
240 			 "key before start of btree node");
241 
242 	bkey_fsck_err_on(bpos_gt(k.k->p, b->data->max_key), c, err,
243 			 bkey_after_end_of_btree_node,
244 			 "key past end of btree node");
245 fsck_err:
246 	return ret;
247 }
248 
249 void bch2_bpos_to_text(struct printbuf *out, struct bpos pos)
250 {
251 	if (bpos_eq(pos, POS_MIN))
252 		prt_printf(out, "POS_MIN");
253 	else if (bpos_eq(pos, POS_MAX))
254 		prt_printf(out, "POS_MAX");
255 	else if (bpos_eq(pos, SPOS_MAX))
256 		prt_printf(out, "SPOS_MAX");
257 	else {
258 		if (pos.inode == U64_MAX)
259 			prt_printf(out, "U64_MAX");
260 		else
261 			prt_printf(out, "%llu", pos.inode);
262 		prt_printf(out, ":");
263 		if (pos.offset == U64_MAX)
264 			prt_printf(out, "U64_MAX");
265 		else
266 			prt_printf(out, "%llu", pos.offset);
267 		prt_printf(out, ":");
268 		if (pos.snapshot == U32_MAX)
269 			prt_printf(out, "U32_MAX");
270 		else
271 			prt_printf(out, "%u", pos.snapshot);
272 	}
273 }
274 
275 void bch2_bkey_to_text(struct printbuf *out, const struct bkey *k)
276 {
277 	if (k) {
278 		prt_printf(out, "u64s %u type ", k->u64s);
279 
280 		if (k->type < KEY_TYPE_MAX)
281 			prt_printf(out, "%s ", bch2_bkey_types[k->type]);
282 		else
283 			prt_printf(out, "%u ", k->type);
284 
285 		bch2_bpos_to_text(out, k->p);
286 
287 		prt_printf(out, " len %u ver %llu", k->size, k->version.lo);
288 	} else {
289 		prt_printf(out, "(null)");
290 	}
291 }
292 
293 void bch2_val_to_text(struct printbuf *out, struct bch_fs *c,
294 		      struct bkey_s_c k)
295 {
296 	const struct bkey_ops *ops = bch2_bkey_type_ops(k.k->type);
297 
298 	if (likely(ops->val_to_text))
299 		ops->val_to_text(out, c, k);
300 }
301 
302 void bch2_bkey_val_to_text(struct printbuf *out, struct bch_fs *c,
303 			   struct bkey_s_c k)
304 {
305 	bch2_bkey_to_text(out, k.k);
306 
307 	if (bkey_val_bytes(k.k)) {
308 		prt_printf(out, ": ");
309 		bch2_val_to_text(out, c, k);
310 	}
311 }
312 
313 void bch2_bkey_swab_val(struct bkey_s k)
314 {
315 	const struct bkey_ops *ops = bch2_bkey_type_ops(k.k->type);
316 
317 	if (ops->swab)
318 		ops->swab(k);
319 }
320 
321 bool bch2_bkey_normalize(struct bch_fs *c, struct bkey_s k)
322 {
323 	const struct bkey_ops *ops = bch2_bkey_type_ops(k.k->type);
324 
325 	return ops->key_normalize
326 		? ops->key_normalize(c, k)
327 		: false;
328 }
329 
330 bool bch2_bkey_merge(struct bch_fs *c, struct bkey_s l, struct bkey_s_c r)
331 {
332 	const struct bkey_ops *ops = bch2_bkey_type_ops(l.k->type);
333 
334 	return ops->key_merge &&
335 		bch2_bkey_maybe_mergable(l.k, r.k) &&
336 		(u64) l.k->size + r.k->size <= KEY_SIZE_MAX &&
337 		!bch2_key_merging_disabled &&
338 		ops->key_merge(c, l, r);
339 }
340 
341 static const struct old_bkey_type {
342 	u8		btree_node_type;
343 	u8		old;
344 	u8		new;
345 } bkey_renumber_table[] = {
346 	{BKEY_TYPE_btree,	128, KEY_TYPE_btree_ptr		},
347 	{BKEY_TYPE_extents,	128, KEY_TYPE_extent		},
348 	{BKEY_TYPE_extents,	129, KEY_TYPE_extent		},
349 	{BKEY_TYPE_extents,	130, KEY_TYPE_reservation	},
350 	{BKEY_TYPE_inodes,	128, KEY_TYPE_inode		},
351 	{BKEY_TYPE_inodes,	130, KEY_TYPE_inode_generation	},
352 	{BKEY_TYPE_dirents,	128, KEY_TYPE_dirent		},
353 	{BKEY_TYPE_dirents,	129, KEY_TYPE_hash_whiteout	},
354 	{BKEY_TYPE_xattrs,	128, KEY_TYPE_xattr		},
355 	{BKEY_TYPE_xattrs,	129, KEY_TYPE_hash_whiteout	},
356 	{BKEY_TYPE_alloc,	128, KEY_TYPE_alloc		},
357 	{BKEY_TYPE_quotas,	128, KEY_TYPE_quota		},
358 };
359 
360 void bch2_bkey_renumber(enum btree_node_type btree_node_type,
361 			struct bkey_packed *k,
362 			int write)
363 {
364 	const struct old_bkey_type *i;
365 
366 	for (i = bkey_renumber_table;
367 	     i < bkey_renumber_table + ARRAY_SIZE(bkey_renumber_table);
368 	     i++)
369 		if (btree_node_type == i->btree_node_type &&
370 		    k->type == (write ? i->new : i->old)) {
371 			k->type = write ? i->old : i->new;
372 			break;
373 		}
374 }
375 
376 void __bch2_bkey_compat(unsigned level, enum btree_id btree_id,
377 			unsigned version, unsigned big_endian,
378 			int write,
379 			struct bkey_format *f,
380 			struct bkey_packed *k)
381 {
382 	const struct bkey_ops *ops;
383 	struct bkey uk;
384 	unsigned nr_compat = 5;
385 	int i;
386 
387 	/*
388 	 * Do these operations in reverse order in the write path:
389 	 */
390 
391 	for (i = 0; i < nr_compat; i++)
392 	switch (!write ? i : nr_compat - 1 - i) {
393 	case 0:
394 		if (big_endian != CPU_BIG_ENDIAN)
395 			bch2_bkey_swab_key(f, k);
396 		break;
397 	case 1:
398 		if (version < bcachefs_metadata_version_bkey_renumber)
399 			bch2_bkey_renumber(__btree_node_type(level, btree_id), k, write);
400 		break;
401 	case 2:
402 		if (version < bcachefs_metadata_version_inode_btree_change &&
403 		    btree_id == BTREE_ID_inodes) {
404 			if (!bkey_packed(k)) {
405 				struct bkey_i *u = packed_to_bkey(k);
406 
407 				swap(u->k.p.inode, u->k.p.offset);
408 			} else if (f->bits_per_field[BKEY_FIELD_INODE] &&
409 				   f->bits_per_field[BKEY_FIELD_OFFSET]) {
410 				struct bkey_format tmp = *f, *in = f, *out = &tmp;
411 
412 				swap(tmp.bits_per_field[BKEY_FIELD_INODE],
413 				     tmp.bits_per_field[BKEY_FIELD_OFFSET]);
414 				swap(tmp.field_offset[BKEY_FIELD_INODE],
415 				     tmp.field_offset[BKEY_FIELD_OFFSET]);
416 
417 				if (!write)
418 					swap(in, out);
419 
420 				uk = __bch2_bkey_unpack_key(in, k);
421 				swap(uk.p.inode, uk.p.offset);
422 				BUG_ON(!bch2_bkey_pack_key(k, &uk, out));
423 			}
424 		}
425 		break;
426 	case 3:
427 		if (version < bcachefs_metadata_version_snapshot &&
428 		    (level || btree_type_has_snapshots(btree_id))) {
429 			struct bkey_i *u = packed_to_bkey(k);
430 
431 			if (u) {
432 				u->k.p.snapshot = write
433 					? 0 : U32_MAX;
434 			} else {
435 				u64 min_packed = le64_to_cpu(f->field_offset[BKEY_FIELD_SNAPSHOT]);
436 				u64 max_packed = min_packed +
437 					~(~0ULL << f->bits_per_field[BKEY_FIELD_SNAPSHOT]);
438 
439 				uk = __bch2_bkey_unpack_key(f, k);
440 				uk.p.snapshot = write
441 					? min_packed : min_t(u64, U32_MAX, max_packed);
442 
443 				BUG_ON(!bch2_bkey_pack_key(k, &uk, f));
444 			}
445 		}
446 
447 		break;
448 	case 4: {
449 		struct bkey_s u;
450 
451 		if (!bkey_packed(k)) {
452 			u = bkey_i_to_s(packed_to_bkey(k));
453 		} else {
454 			uk = __bch2_bkey_unpack_key(f, k);
455 			u.k = &uk;
456 			u.v = bkeyp_val(f, k);
457 		}
458 
459 		if (big_endian != CPU_BIG_ENDIAN)
460 			bch2_bkey_swab_val(u);
461 
462 		ops = bch2_bkey_type_ops(k->type);
463 
464 		if (ops->compat)
465 			ops->compat(btree_id, version, big_endian, write, u);
466 		break;
467 	}
468 	default:
469 		BUG();
470 	}
471 }
472