xref: /linux-6.15/fs/bcachefs/bkey_methods.c (revision 50a38ca1) 
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(const 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(const struct bch_fs *c, struct bkey_s_c k,
44 				 enum bkey_invalid_flags flags, struct printbuf *err)
45 {
46 	if (bkey_val_bytes(k.k)) {
47 		prt_printf(err, "incorrect value size (%zu != 0)",
48 		       bkey_val_bytes(k.k));
49 		return -BCH_ERR_invalid_bkey;
50 	}
51 
52 	return 0;
53 }
54 
55 #define bch2_bkey_ops_error ((struct bkey_ops) {	\
56 	.key_invalid = empty_val_key_invalid,		\
57 })
58 
59 static int key_type_cookie_invalid(const struct bch_fs *c, struct bkey_s_c k,
60 				   enum bkey_invalid_flags flags, struct printbuf *err)
61 {
62 	return 0;
63 }
64 
65 #define bch2_bkey_ops_cookie ((struct bkey_ops) {	\
66 	.key_invalid	= key_type_cookie_invalid,	\
67 	.min_val_size	= 8,				\
68 })
69 
70 #define bch2_bkey_ops_hash_whiteout ((struct bkey_ops) {\
71 	.key_invalid = empty_val_key_invalid,		\
72 })
73 
74 static int key_type_inline_data_invalid(const struct bch_fs *c, struct bkey_s_c k,
75 					enum bkey_invalid_flags flags, struct printbuf *err)
76 {
77 	return 0;
78 }
79 
80 static void key_type_inline_data_to_text(struct printbuf *out, struct bch_fs *c,
81 					 struct bkey_s_c k)
82 {
83 	struct bkey_s_c_inline_data d = bkey_s_c_to_inline_data(k);
84 	unsigned datalen = bkey_inline_data_bytes(k.k);
85 
86 	prt_printf(out, "datalen %u: %*phN",
87 	       datalen, min(datalen, 32U), d.v->data);
88 }
89 
90 #define bch2_bkey_ops_inline_data ((struct bkey_ops) {	\
91 	.key_invalid	= key_type_inline_data_invalid,	\
92 	.val_to_text	= key_type_inline_data_to_text,	\
93 })
94 
95 static int key_type_set_invalid(const struct bch_fs *c, struct bkey_s_c k,
96 				enum bkey_invalid_flags flags, struct printbuf *err)
97 {
98 	if (bkey_val_bytes(k.k)) {
99 		prt_printf(err, "incorrect value size (%zu != %zu)",
100 		       bkey_val_bytes(k.k), sizeof(struct bch_cookie));
101 		return -BCH_ERR_invalid_bkey;
102 	}
103 
104 	return 0;
105 }
106 
107 static bool key_type_set_merge(struct bch_fs *c, struct bkey_s l, struct bkey_s_c r)
108 {
109 	bch2_key_resize(l.k, l.k->size + r.k->size);
110 	return true;
111 }
112 
113 #define bch2_bkey_ops_set ((struct bkey_ops) {		\
114 	.key_invalid	= key_type_set_invalid,		\
115 	.key_merge	= key_type_set_merge,		\
116 })
117 
118 const struct bkey_ops bch2_bkey_ops[] = {
119 #define x(name, nr) [KEY_TYPE_##name]	= bch2_bkey_ops_##name,
120 	BCH_BKEY_TYPES()
121 #undef x
122 };
123 
124 const struct bkey_ops bch2_bkey_null_ops = {
125 };
126 
127 int bch2_bkey_val_invalid(struct bch_fs *c, struct bkey_s_c k,
128 			  enum bkey_invalid_flags flags,
129 			  struct printbuf *err)
130 {
131 	const struct bkey_ops *ops = bch2_bkey_type_ops(k.k->type);
132 
133 	if (bkey_val_bytes(k.k) < ops->min_val_size) {
134 		prt_printf(err, "bad val size (%zu < %u)",
135 			   bkey_val_bytes(k.k), ops->min_val_size);
136 		return -BCH_ERR_invalid_bkey;
137 	}
138 
139 	if (!ops->key_invalid)
140 		return 0;
141 
142 	return ops->key_invalid(c, k, flags, err);
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 	if (k.k->u64s < BKEY_U64s) {
166 		prt_printf(err, "u64s too small (%u < %zu)", k.k->u64s, BKEY_U64s);
167 		return -BCH_ERR_invalid_bkey;
168 	}
169 
170 	if (type >= BKEY_TYPE_NR)
171 		return 0;
172 
173 	if (flags & BKEY_INVALID_COMMIT	 &&
174 	    !(bch2_key_types_allowed[type] & BIT_ULL(k.k->type))) {
175 		prt_printf(err, "invalid key type for btree %s (%s)",
176 			   bch2_btree_node_type_str(type), bch2_bkey_types[k.k->type]);
177 		return -BCH_ERR_invalid_bkey;
178 	}
179 
180 	if (btree_node_type_is_extents(type) && !bkey_whiteout(k.k)) {
181 		if (k.k->size == 0) {
182 			prt_printf(err, "size == 0");
183 			return -BCH_ERR_invalid_bkey;
184 		}
185 
186 		if (k.k->size > k.k->p.offset) {
187 			prt_printf(err, "size greater than offset (%u > %llu)",
188 			       k.k->size, k.k->p.offset);
189 			return -BCH_ERR_invalid_bkey;
190 		}
191 	} else {
192 		if (k.k->size) {
193 			prt_printf(err, "size != 0");
194 			return -BCH_ERR_invalid_bkey;
195 		}
196 	}
197 
198 	if (type != BKEY_TYPE_btree) {
199 		enum btree_id btree = type - 1;
200 
201 		if (!btree_type_has_snapshots(btree) &&
202 		    k.k->p.snapshot) {
203 			prt_printf(err, "nonzero snapshot");
204 			return -BCH_ERR_invalid_bkey;
205 		}
206 
207 		if (btree_type_has_snapshots(btree) &&
208 		    !k.k->p.snapshot) {
209 			prt_printf(err, "snapshot == 0");
210 			return -BCH_ERR_invalid_bkey;
211 		}
212 
213 		if (bkey_eq(k.k->p, POS_MAX)) {
214 			prt_printf(err, "key at POS_MAX");
215 			return -BCH_ERR_invalid_bkey;
216 		}
217 	}
218 
219 	return 0;
220 }
221 
222 int bch2_bkey_invalid(struct bch_fs *c, struct bkey_s_c k,
223 		      enum btree_node_type type,
224 		      enum bkey_invalid_flags flags,
225 		      struct printbuf *err)
226 {
227 	return __bch2_bkey_invalid(c, k, type, flags, err) ?:
228 		bch2_bkey_val_invalid(c, k, flags, err);
229 }
230 
231 int bch2_bkey_in_btree_node(struct btree *b, struct bkey_s_c k,
232 			    struct printbuf *err)
233 {
234 	if (bpos_lt(k.k->p, b->data->min_key)) {
235 		prt_printf(err, "key before start of btree node");
236 		return -BCH_ERR_invalid_bkey;
237 	}
238 
239 	if (bpos_gt(k.k->p, b->data->max_key)) {
240 		prt_printf(err, "key past end of btree node");
241 		return -BCH_ERR_invalid_bkey;
242 	}
243 
244 	return 0;
245 }
246 
247 void bch2_bpos_to_text(struct printbuf *out, struct bpos pos)
248 {
249 	if (bpos_eq(pos, POS_MIN))
250 		prt_printf(out, "POS_MIN");
251 	else if (bpos_eq(pos, POS_MAX))
252 		prt_printf(out, "POS_MAX");
253 	else if (bpos_eq(pos, SPOS_MAX))
254 		prt_printf(out, "SPOS_MAX");
255 	else {
256 		if (pos.inode == U64_MAX)
257 			prt_printf(out, "U64_MAX");
258 		else
259 			prt_printf(out, "%llu", pos.inode);
260 		prt_printf(out, ":");
261 		if (pos.offset == U64_MAX)
262 			prt_printf(out, "U64_MAX");
263 		else
264 			prt_printf(out, "%llu", pos.offset);
265 		prt_printf(out, ":");
266 		if (pos.snapshot == U32_MAX)
267 			prt_printf(out, "U32_MAX");
268 		else
269 			prt_printf(out, "%u", pos.snapshot);
270 	}
271 }
272 
273 void bch2_bkey_to_text(struct printbuf *out, const struct bkey *k)
274 {
275 	if (k) {
276 		prt_printf(out, "u64s %u type ", k->u64s);
277 
278 		if (k->type < KEY_TYPE_MAX)
279 			prt_printf(out, "%s ", bch2_bkey_types[k->type]);
280 		else
281 			prt_printf(out, "%u ", k->type);
282 
283 		bch2_bpos_to_text(out, k->p);
284 
285 		prt_printf(out, " len %u ver %llu", k->size, k->version.lo);
286 	} else {
287 		prt_printf(out, "(null)");
288 	}
289 }
290 
291 void bch2_val_to_text(struct printbuf *out, struct bch_fs *c,
292 		      struct bkey_s_c k)
293 {
294 	const struct bkey_ops *ops = bch2_bkey_type_ops(k.k->type);
295 
296 	if (likely(ops->val_to_text))
297 		ops->val_to_text(out, c, k);
298 }
299 
300 void bch2_bkey_val_to_text(struct printbuf *out, struct bch_fs *c,
301 			   struct bkey_s_c k)
302 {
303 	bch2_bkey_to_text(out, k.k);
304 
305 	if (bkey_val_bytes(k.k)) {
306 		prt_printf(out, ": ");
307 		bch2_val_to_text(out, c, k);
308 	}
309 }
310 
311 void bch2_bkey_swab_val(struct bkey_s k)
312 {
313 	const struct bkey_ops *ops = bch2_bkey_type_ops(k.k->type);
314 
315 	if (ops->swab)
316 		ops->swab(k);
317 }
318 
319 bool bch2_bkey_normalize(struct bch_fs *c, struct bkey_s k)
320 {
321 	const struct bkey_ops *ops = bch2_bkey_type_ops(k.k->type);
322 
323 	return ops->key_normalize
324 		? ops->key_normalize(c, k)
325 		: false;
326 }
327 
328 bool bch2_bkey_merge(struct bch_fs *c, struct bkey_s l, struct bkey_s_c r)
329 {
330 	const struct bkey_ops *ops = bch2_bkey_type_ops(l.k->type);
331 
332 	return ops->key_merge &&
333 		bch2_bkey_maybe_mergable(l.k, r.k) &&
334 		(u64) l.k->size + r.k->size <= KEY_SIZE_MAX &&
335 		!bch2_key_merging_disabled &&
336 		ops->key_merge(c, l, r);
337 }
338 
339 static const struct old_bkey_type {
340 	u8		btree_node_type;
341 	u8		old;
342 	u8		new;
343 } bkey_renumber_table[] = {
344 	{BKEY_TYPE_btree,	128, KEY_TYPE_btree_ptr		},
345 	{BKEY_TYPE_extents,	128, KEY_TYPE_extent		},
346 	{BKEY_TYPE_extents,	129, KEY_TYPE_extent		},
347 	{BKEY_TYPE_extents,	130, KEY_TYPE_reservation	},
348 	{BKEY_TYPE_inodes,	128, KEY_TYPE_inode		},
349 	{BKEY_TYPE_inodes,	130, KEY_TYPE_inode_generation	},
350 	{BKEY_TYPE_dirents,	128, KEY_TYPE_dirent		},
351 	{BKEY_TYPE_dirents,	129, KEY_TYPE_hash_whiteout	},
352 	{BKEY_TYPE_xattrs,	128, KEY_TYPE_xattr		},
353 	{BKEY_TYPE_xattrs,	129, KEY_TYPE_hash_whiteout	},
354 	{BKEY_TYPE_alloc,	128, KEY_TYPE_alloc		},
355 	{BKEY_TYPE_quotas,	128, KEY_TYPE_quota		},
356 };
357 
358 void bch2_bkey_renumber(enum btree_node_type btree_node_type,
359 			struct bkey_packed *k,
360 			int write)
361 {
362 	const struct old_bkey_type *i;
363 
364 	for (i = bkey_renumber_table;
365 	     i < bkey_renumber_table + ARRAY_SIZE(bkey_renumber_table);
366 	     i++)
367 		if (btree_node_type == i->btree_node_type &&
368 		    k->type == (write ? i->new : i->old)) {
369 			k->type = write ? i->old : i->new;
370 			break;
371 		}
372 }
373 
374 void __bch2_bkey_compat(unsigned level, enum btree_id btree_id,
375 			unsigned version, unsigned big_endian,
376 			int write,
377 			struct bkey_format *f,
378 			struct bkey_packed *k)
379 {
380 	const struct bkey_ops *ops;
381 	struct bkey uk;
382 	unsigned nr_compat = 5;
383 	int i;
384 
385 	/*
386 	 * Do these operations in reverse order in the write path:
387 	 */
388 
389 	for (i = 0; i < nr_compat; i++)
390 	switch (!write ? i : nr_compat - 1 - i) {
391 	case 0:
392 		if (big_endian != CPU_BIG_ENDIAN)
393 			bch2_bkey_swab_key(f, k);
394 		break;
395 	case 1:
396 		if (version < bcachefs_metadata_version_bkey_renumber)
397 			bch2_bkey_renumber(__btree_node_type(level, btree_id), k, write);
398 		break;
399 	case 2:
400 		if (version < bcachefs_metadata_version_inode_btree_change &&
401 		    btree_id == BTREE_ID_inodes) {
402 			if (!bkey_packed(k)) {
403 				struct bkey_i *u = packed_to_bkey(k);
404 
405 				swap(u->k.p.inode, u->k.p.offset);
406 			} else if (f->bits_per_field[BKEY_FIELD_INODE] &&
407 				   f->bits_per_field[BKEY_FIELD_OFFSET]) {
408 				struct bkey_format tmp = *f, *in = f, *out = &tmp;
409 
410 				swap(tmp.bits_per_field[BKEY_FIELD_INODE],
411 				     tmp.bits_per_field[BKEY_FIELD_OFFSET]);
412 				swap(tmp.field_offset[BKEY_FIELD_INODE],
413 				     tmp.field_offset[BKEY_FIELD_OFFSET]);
414 
415 				if (!write)
416 					swap(in, out);
417 
418 				uk = __bch2_bkey_unpack_key(in, k);
419 				swap(uk.p.inode, uk.p.offset);
420 				BUG_ON(!bch2_bkey_pack_key(k, &uk, out));
421 			}
422 		}
423 		break;
424 	case 3:
425 		if (version < bcachefs_metadata_version_snapshot &&
426 		    (level || btree_type_has_snapshots(btree_id))) {
427 			struct bkey_i *u = packed_to_bkey(k);
428 
429 			if (u) {
430 				u->k.p.snapshot = write
431 					? 0 : U32_MAX;
432 			} else {
433 				u64 min_packed = le64_to_cpu(f->field_offset[BKEY_FIELD_SNAPSHOT]);
434 				u64 max_packed = min_packed +
435 					~(~0ULL << f->bits_per_field[BKEY_FIELD_SNAPSHOT]);
436 
437 				uk = __bch2_bkey_unpack_key(f, k);
438 				uk.p.snapshot = write
439 					? min_packed : min_t(u64, U32_MAX, max_packed);
440 
441 				BUG_ON(!bch2_bkey_pack_key(k, &uk, f));
442 			}
443 		}
444 
445 		break;
446 	case 4: {
447 		struct bkey_s u;
448 
449 		if (!bkey_packed(k)) {
450 			u = bkey_i_to_s(packed_to_bkey(k));
451 		} else {
452 			uk = __bch2_bkey_unpack_key(f, k);
453 			u.k = &uk;
454 			u.v = bkeyp_val(f, k);
455 		}
456 
457 		if (big_endian != CPU_BIG_ENDIAN)
458 			bch2_bkey_swab_val(u);
459 
460 		ops = bch2_bkey_type_ops(k->type);
461 
462 		if (ops->compat)
463 			ops->compat(btree_id, version, big_endian, write, u);
464 		break;
465 	}
466 	default:
467 		BUG();
468 	}
469 }
470