xref: /linux-6.15/fs/bcachefs/bkey_methods.c (revision 8e877caa) 
1 // SPDX-License-Identifier: GPL-2.0
2 
3 #include "bcachefs.h"
4 #include "backpointers.h"
5 #include "bkey_methods.h"
6 #include "btree_types.h"
7 #include "alloc_background.h"
8 #include "dirent.h"
9 #include "ec.h"
10 #include "error.h"
11 #include "extents.h"
12 #include "inode.h"
13 #include "lru.h"
14 #include "quota.h"
15 #include "reflink.h"
16 #include "snapshot.h"
17 #include "subvolume.h"
18 #include "xattr.h"
19 
20 const char * const bch2_bkey_types[] = {
21 #define x(name, nr) #name,
22 	BCH_BKEY_TYPES()
23 #undef x
24 	NULL
25 };
26 
27 static int deleted_key_invalid(const struct bch_fs *c, struct bkey_s_c k,
28 			       unsigned flags, struct printbuf *err)
29 {
30 	return 0;
31 }
32 
33 #define bch2_bkey_ops_deleted ((struct bkey_ops) {	\
34 	.key_invalid = deleted_key_invalid,		\
35 })
36 
37 #define bch2_bkey_ops_whiteout ((struct bkey_ops) {	\
38 	.key_invalid = deleted_key_invalid,		\
39 })
40 
41 static int empty_val_key_invalid(const struct bch_fs *c, struct bkey_s_c k,
42 				 unsigned flags, struct printbuf *err)
43 {
44 	if (bkey_val_bytes(k.k)) {
45 		prt_printf(err, "incorrect value size (%zu != 0)",
46 		       bkey_val_bytes(k.k));
47 		return -BCH_ERR_invalid_bkey;
48 	}
49 
50 	return 0;
51 }
52 
53 #define bch2_bkey_ops_error ((struct bkey_ops) {	\
54 	.key_invalid = empty_val_key_invalid,		\
55 })
56 
57 static int key_type_cookie_invalid(const struct bch_fs *c, struct bkey_s_c k,
58 				   unsigned flags, struct printbuf *err)
59 {
60 	return 0;
61 }
62 
63 #define bch2_bkey_ops_cookie ((struct bkey_ops) {	\
64 	.key_invalid	= key_type_cookie_invalid,	\
65 	.min_val_size	= 8,				\
66 })
67 
68 #define bch2_bkey_ops_hash_whiteout ((struct bkey_ops) {\
69 	.key_invalid = empty_val_key_invalid,		\
70 })
71 
72 static int key_type_inline_data_invalid(const struct bch_fs *c, struct bkey_s_c k,
73 					unsigned flags, struct printbuf *err)
74 {
75 	return 0;
76 }
77 
78 static void key_type_inline_data_to_text(struct printbuf *out, struct bch_fs *c,
79 					 struct bkey_s_c k)
80 {
81 	struct bkey_s_c_inline_data d = bkey_s_c_to_inline_data(k);
82 	unsigned datalen = bkey_inline_data_bytes(k.k);
83 
84 	prt_printf(out, "datalen %u: %*phN",
85 	       datalen, min(datalen, 32U), d.v->data);
86 }
87 
88 #define bch2_bkey_ops_inline_data ((struct bkey_ops) {	\
89 	.key_invalid	= key_type_inline_data_invalid,	\
90 	.val_to_text	= key_type_inline_data_to_text,	\
91 })
92 
93 static int key_type_set_invalid(const struct bch_fs *c, struct bkey_s_c k,
94 				unsigned flags, struct printbuf *err)
95 {
96 	if (bkey_val_bytes(k.k)) {
97 		prt_printf(err, "incorrect value size (%zu != %zu)",
98 		       bkey_val_bytes(k.k), sizeof(struct bch_cookie));
99 		return -BCH_ERR_invalid_bkey;
100 	}
101 
102 	return 0;
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	= key_type_set_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 	.min_val_size = U8_MAX,
124 };
125 
126 int bch2_bkey_val_invalid(struct bch_fs *c, struct bkey_s_c k,
127 			  enum bkey_invalid_flags flags,
128 			  struct printbuf *err)
129 {
130 	const struct bkey_ops *ops = bch2_bkey_type_ops(k.k->type);
131 
132 	if (bkey_val_bytes(k.k) < ops->min_val_size) {
133 		prt_printf(err, "bad val size (%zu < %u)",
134 			   bkey_val_bytes(k.k), ops->min_val_size);
135 		return -BCH_ERR_invalid_bkey;
136 	}
137 
138 	if (!ops->key_invalid)
139 		return 0;
140 
141 	return ops->key_invalid(c, k, flags, err);
142 }
143 
144 static u64 bch2_key_types_allowed[] = {
145 #define x(name, nr, flags, keys)	[BKEY_TYPE_##name] = BIT_ULL(KEY_TYPE_deleted)|keys,
146 	BCH_BTREE_IDS()
147 #undef x
148 	[BKEY_TYPE_btree] =
149 		BIT_ULL(KEY_TYPE_deleted)|
150 		BIT_ULL(KEY_TYPE_btree_ptr)|
151 		BIT_ULL(KEY_TYPE_btree_ptr_v2),
152 };
153 
154 int __bch2_bkey_invalid(struct bch_fs *c, struct bkey_s_c k,
155 			enum btree_node_type type,
156 			enum bkey_invalid_flags flags,
157 			struct printbuf *err)
158 {
159 	if (k.k->u64s < BKEY_U64s) {
160 		prt_printf(err, "u64s too small (%u < %zu)", k.k->u64s, BKEY_U64s);
161 		return -BCH_ERR_invalid_bkey;
162 	}
163 
164 	if (flags & BKEY_INVALID_COMMIT	 &&
165 	    !(bch2_key_types_allowed[type] & BIT_ULL(k.k->type))) {
166 		prt_printf(err, "invalid key type for btree %s (%s)",
167 			   bch2_btree_ids[type], bch2_bkey_types[k.k->type]);
168 		return -BCH_ERR_invalid_bkey;
169 	}
170 
171 	if (btree_node_type_is_extents(type) && !bkey_whiteout(k.k)) {
172 		if (k.k->size == 0) {
173 			prt_printf(err, "size == 0");
174 			return -BCH_ERR_invalid_bkey;
175 		}
176 
177 		if (k.k->size > k.k->p.offset) {
178 			prt_printf(err, "size greater than offset (%u > %llu)",
179 			       k.k->size, k.k->p.offset);
180 			return -BCH_ERR_invalid_bkey;
181 		}
182 	} else {
183 		if (k.k->size) {
184 			prt_printf(err, "size != 0");
185 			return -BCH_ERR_invalid_bkey;
186 		}
187 	}
188 
189 	if (type != BKEY_TYPE_btree) {
190 		if (!btree_type_has_snapshots((enum btree_id) type) &&
191 		    k.k->p.snapshot) {
192 			prt_printf(err, "nonzero snapshot");
193 			return -BCH_ERR_invalid_bkey;
194 		}
195 
196 		if (btree_type_has_snapshots((enum btree_id) type) &&
197 		    !k.k->p.snapshot) {
198 			prt_printf(err, "snapshot == 0");
199 			return -BCH_ERR_invalid_bkey;
200 		}
201 
202 		if (bkey_eq(k.k->p, POS_MAX)) {
203 			prt_printf(err, "key at POS_MAX");
204 			return -BCH_ERR_invalid_bkey;
205 		}
206 	}
207 
208 	return 0;
209 }
210 
211 int bch2_bkey_invalid(struct bch_fs *c, struct bkey_s_c k,
212 		      enum btree_node_type type,
213 		      enum bkey_invalid_flags flags,
214 		      struct printbuf *err)
215 {
216 	return __bch2_bkey_invalid(c, k, type, flags, err) ?:
217 		bch2_bkey_val_invalid(c, k, flags, err);
218 }
219 
220 int bch2_bkey_in_btree_node(struct btree *b, struct bkey_s_c k,
221 			    struct printbuf *err)
222 {
223 	if (bpos_lt(k.k->p, b->data->min_key)) {
224 		prt_printf(err, "key before start of btree node");
225 		return -BCH_ERR_invalid_bkey;
226 	}
227 
228 	if (bpos_gt(k.k->p, b->data->max_key)) {
229 		prt_printf(err, "key past end of btree node");
230 		return -BCH_ERR_invalid_bkey;
231 	}
232 
233 	return 0;
234 }
235 
236 void bch2_bpos_to_text(struct printbuf *out, struct bpos pos)
237 {
238 	if (bpos_eq(pos, POS_MIN))
239 		prt_printf(out, "POS_MIN");
240 	else if (bpos_eq(pos, POS_MAX))
241 		prt_printf(out, "POS_MAX");
242 	else if (bpos_eq(pos, SPOS_MAX))
243 		prt_printf(out, "SPOS_MAX");
244 	else {
245 		if (pos.inode == U64_MAX)
246 			prt_printf(out, "U64_MAX");
247 		else
248 			prt_printf(out, "%llu", pos.inode);
249 		prt_printf(out, ":");
250 		if (pos.offset == U64_MAX)
251 			prt_printf(out, "U64_MAX");
252 		else
253 			prt_printf(out, "%llu", pos.offset);
254 		prt_printf(out, ":");
255 		if (pos.snapshot == U32_MAX)
256 			prt_printf(out, "U32_MAX");
257 		else
258 			prt_printf(out, "%u", pos.snapshot);
259 	}
260 }
261 
262 void bch2_bkey_to_text(struct printbuf *out, const struct bkey *k)
263 {
264 	if (k) {
265 		prt_printf(out, "u64s %u type ", k->u64s);
266 
267 		if (k->type < KEY_TYPE_MAX)
268 			prt_printf(out, "%s ", bch2_bkey_types[k->type]);
269 		else
270 			prt_printf(out, "%u ", k->type);
271 
272 		bch2_bpos_to_text(out, k->p);
273 
274 		prt_printf(out, " len %u ver %llu", k->size, k->version.lo);
275 	} else {
276 		prt_printf(out, "(null)");
277 	}
278 }
279 
280 void bch2_val_to_text(struct printbuf *out, struct bch_fs *c,
281 		      struct bkey_s_c k)
282 {
283 	const struct bkey_ops *ops = bch2_bkey_type_ops(k.k->type);
284 
285 	if (likely(ops->val_to_text))
286 		ops->val_to_text(out, c, k);
287 }
288 
289 void bch2_bkey_val_to_text(struct printbuf *out, struct bch_fs *c,
290 			   struct bkey_s_c k)
291 {
292 	bch2_bkey_to_text(out, k.k);
293 
294 	if (bkey_val_bytes(k.k)) {
295 		prt_printf(out, ": ");
296 		bch2_val_to_text(out, c, k);
297 	}
298 }
299 
300 void bch2_bkey_swab_val(struct bkey_s k)
301 {
302 	const struct bkey_ops *ops = bch2_bkey_type_ops(k.k->type);
303 
304 	if (ops->swab)
305 		ops->swab(k);
306 }
307 
308 bool bch2_bkey_normalize(struct bch_fs *c, struct bkey_s k)
309 {
310 	const struct bkey_ops *ops = bch2_bkey_type_ops(k.k->type);
311 
312 	return ops->key_normalize
313 		? ops->key_normalize(c, k)
314 		: false;
315 }
316 
317 bool bch2_bkey_merge(struct bch_fs *c, struct bkey_s l, struct bkey_s_c r)
318 {
319 	const struct bkey_ops *ops = bch2_bkey_type_ops(l.k->type);
320 
321 	return ops->key_merge &&
322 		bch2_bkey_maybe_mergable(l.k, r.k) &&
323 		(u64) l.k->size + r.k->size <= KEY_SIZE_MAX &&
324 		!bch2_key_merging_disabled &&
325 		ops->key_merge(c, l, r);
326 }
327 
328 static const struct old_bkey_type {
329 	u8		btree_node_type;
330 	u8		old;
331 	u8		new;
332 } bkey_renumber_table[] = {
333 	{BKEY_TYPE_btree,	128, KEY_TYPE_btree_ptr		},
334 	{BKEY_TYPE_extents,	128, KEY_TYPE_extent		},
335 	{BKEY_TYPE_extents,	129, KEY_TYPE_extent		},
336 	{BKEY_TYPE_extents,	130, KEY_TYPE_reservation	},
337 	{BKEY_TYPE_inodes,	128, KEY_TYPE_inode		},
338 	{BKEY_TYPE_inodes,	130, KEY_TYPE_inode_generation	},
339 	{BKEY_TYPE_dirents,	128, KEY_TYPE_dirent		},
340 	{BKEY_TYPE_dirents,	129, KEY_TYPE_hash_whiteout	},
341 	{BKEY_TYPE_xattrs,	128, KEY_TYPE_xattr		},
342 	{BKEY_TYPE_xattrs,	129, KEY_TYPE_hash_whiteout	},
343 	{BKEY_TYPE_alloc,	128, KEY_TYPE_alloc		},
344 	{BKEY_TYPE_quotas,	128, KEY_TYPE_quota		},
345 };
346 
347 void bch2_bkey_renumber(enum btree_node_type btree_node_type,
348 			struct bkey_packed *k,
349 			int write)
350 {
351 	const struct old_bkey_type *i;
352 
353 	for (i = bkey_renumber_table;
354 	     i < bkey_renumber_table + ARRAY_SIZE(bkey_renumber_table);
355 	     i++)
356 		if (btree_node_type == i->btree_node_type &&
357 		    k->type == (write ? i->new : i->old)) {
358 			k->type = write ? i->old : i->new;
359 			break;
360 		}
361 }
362 
363 void __bch2_bkey_compat(unsigned level, enum btree_id btree_id,
364 			unsigned version, unsigned big_endian,
365 			int write,
366 			struct bkey_format *f,
367 			struct bkey_packed *k)
368 {
369 	const struct bkey_ops *ops;
370 	struct bkey uk;
371 	struct bkey_s u;
372 	unsigned nr_compat = 5;
373 	int i;
374 
375 	/*
376 	 * Do these operations in reverse order in the write path:
377 	 */
378 
379 	for (i = 0; i < nr_compat; i++)
380 	switch (!write ? i : nr_compat - 1 - i) {
381 	case 0:
382 		if (big_endian != CPU_BIG_ENDIAN)
383 			bch2_bkey_swab_key(f, k);
384 		break;
385 	case 1:
386 		if (version < bcachefs_metadata_version_bkey_renumber)
387 			bch2_bkey_renumber(__btree_node_type(level, btree_id), k, write);
388 		break;
389 	case 2:
390 		if (version < bcachefs_metadata_version_inode_btree_change &&
391 		    btree_id == BTREE_ID_inodes) {
392 			if (!bkey_packed(k)) {
393 				struct bkey_i *u = packed_to_bkey(k);
394 
395 				swap(u->k.p.inode, u->k.p.offset);
396 			} else if (f->bits_per_field[BKEY_FIELD_INODE] &&
397 				   f->bits_per_field[BKEY_FIELD_OFFSET]) {
398 				struct bkey_format tmp = *f, *in = f, *out = &tmp;
399 
400 				swap(tmp.bits_per_field[BKEY_FIELD_INODE],
401 				     tmp.bits_per_field[BKEY_FIELD_OFFSET]);
402 				swap(tmp.field_offset[BKEY_FIELD_INODE],
403 				     tmp.field_offset[BKEY_FIELD_OFFSET]);
404 
405 				if (!write)
406 					swap(in, out);
407 
408 				uk = __bch2_bkey_unpack_key(in, k);
409 				swap(uk.p.inode, uk.p.offset);
410 				BUG_ON(!bch2_bkey_pack_key(k, &uk, out));
411 			}
412 		}
413 		break;
414 	case 3:
415 		if (version < bcachefs_metadata_version_snapshot &&
416 		    (level || btree_type_has_snapshots(btree_id))) {
417 			struct bkey_i *u = packed_to_bkey(k);
418 
419 			if (u) {
420 				u->k.p.snapshot = write
421 					? 0 : U32_MAX;
422 			} else {
423 				u64 min_packed = le64_to_cpu(f->field_offset[BKEY_FIELD_SNAPSHOT]);
424 				u64 max_packed = min_packed +
425 					~(~0ULL << f->bits_per_field[BKEY_FIELD_SNAPSHOT]);
426 
427 				uk = __bch2_bkey_unpack_key(f, k);
428 				uk.p.snapshot = write
429 					? min_packed : min_t(u64, U32_MAX, max_packed);
430 
431 				BUG_ON(!bch2_bkey_pack_key(k, &uk, f));
432 			}
433 		}
434 
435 		break;
436 	case 4:
437 		if (!bkey_packed(k)) {
438 			u = bkey_i_to_s(packed_to_bkey(k));
439 		} else {
440 			uk = __bch2_bkey_unpack_key(f, k);
441 			u.k = &uk;
442 			u.v = bkeyp_val(f, k);
443 		}
444 
445 		if (big_endian != CPU_BIG_ENDIAN)
446 			bch2_bkey_swab_val(u);
447 
448 		ops = bch2_bkey_type_ops(k->type);
449 
450 		if (ops->compat)
451 			ops->compat(btree_id, version, big_endian, write, u);
452 		break;
453 	default:
454 		BUG();
455 	}
456 }
457