xref: /linux-6.15/fs/bcachefs/bkey_methods.c (revision b030e262) 
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 "io_misc.h"
14 #include "lru.h"
15 #include "quota.h"
16 #include "reflink.h"
17 #include "snapshot.h"
18 #include "subvolume.h"
19 #include "xattr.h"
20 
21 const char * const bch2_bkey_types[] = {
22 #define x(name, nr) #name,
23 	BCH_BKEY_TYPES()
24 #undef x
25 	NULL
26 };
27 
28 static int deleted_key_invalid(const struct bch_fs *c, struct bkey_s_c k,
29 			       unsigned flags, struct printbuf *err)
30 {
31 	return 0;
32 }
33 
34 #define bch2_bkey_ops_deleted ((struct bkey_ops) {	\
35 	.key_invalid = deleted_key_invalid,		\
36 })
37 
38 #define bch2_bkey_ops_whiteout ((struct bkey_ops) {	\
39 	.key_invalid = deleted_key_invalid,		\
40 })
41 
42 static int empty_val_key_invalid(const struct bch_fs *c, struct bkey_s_c k,
43 				 unsigned flags, struct printbuf *err)
44 {
45 	if (bkey_val_bytes(k.k)) {
46 		prt_printf(err, "incorrect value size (%zu != 0)",
47 		       bkey_val_bytes(k.k));
48 		return -BCH_ERR_invalid_bkey;
49 	}
50 
51 	return 0;
52 }
53 
54 #define bch2_bkey_ops_error ((struct bkey_ops) {	\
55 	.key_invalid = empty_val_key_invalid,		\
56 })
57 
58 static int key_type_cookie_invalid(const struct bch_fs *c, struct bkey_s_c k,
59 				   unsigned flags, struct printbuf *err)
60 {
61 	return 0;
62 }
63 
64 #define bch2_bkey_ops_cookie ((struct bkey_ops) {	\
65 	.key_invalid	= key_type_cookie_invalid,	\
66 	.min_val_size	= 8,				\
67 })
68 
69 #define bch2_bkey_ops_hash_whiteout ((struct bkey_ops) {\
70 	.key_invalid = empty_val_key_invalid,		\
71 })
72 
73 static int key_type_inline_data_invalid(const struct bch_fs *c, struct bkey_s_c k,
74 					unsigned flags, struct printbuf *err)
75 {
76 	return 0;
77 }
78 
79 static void key_type_inline_data_to_text(struct printbuf *out, struct bch_fs *c,
80 					 struct bkey_s_c k)
81 {
82 	struct bkey_s_c_inline_data d = bkey_s_c_to_inline_data(k);
83 	unsigned datalen = bkey_inline_data_bytes(k.k);
84 
85 	prt_printf(out, "datalen %u: %*phN",
86 	       datalen, min(datalen, 32U), d.v->data);
87 }
88 
89 #define bch2_bkey_ops_inline_data ((struct bkey_ops) {	\
90 	.key_invalid	= key_type_inline_data_invalid,	\
91 	.val_to_text	= key_type_inline_data_to_text,	\
92 })
93 
94 static int key_type_set_invalid(const struct bch_fs *c, struct bkey_s_c k,
95 				unsigned flags, struct printbuf *err)
96 {
97 	if (bkey_val_bytes(k.k)) {
98 		prt_printf(err, "incorrect value size (%zu != %zu)",
99 		       bkey_val_bytes(k.k), sizeof(struct bch_cookie));
100 		return -BCH_ERR_invalid_bkey;
101 	}
102 
103 	return 0;
104 }
105 
106 static bool key_type_set_merge(struct bch_fs *c, struct bkey_s l, struct bkey_s_c r)
107 {
108 	bch2_key_resize(l.k, l.k->size + r.k->size);
109 	return true;
110 }
111 
112 #define bch2_bkey_ops_set ((struct bkey_ops) {		\
113 	.key_invalid	= key_type_set_invalid,		\
114 	.key_merge	= key_type_set_merge,		\
115 })
116 
117 const struct bkey_ops bch2_bkey_ops[] = {
118 #define x(name, nr) [KEY_TYPE_##name]	= bch2_bkey_ops_##name,
119 	BCH_BKEY_TYPES()
120 #undef x
121 };
122 
123 const struct bkey_ops bch2_bkey_null_ops = {
124 	.min_val_size = U8_MAX,
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 #define x(name, nr, flags, keys)	[BKEY_TYPE_##name] = BIT_ULL(KEY_TYPE_deleted)|keys,
147 	BCH_BTREE_IDS()
148 #undef x
149 	[BKEY_TYPE_btree] =
150 		BIT_ULL(KEY_TYPE_deleted)|
151 		BIT_ULL(KEY_TYPE_btree_ptr)|
152 		BIT_ULL(KEY_TYPE_btree_ptr_v2),
153 };
154 
155 int __bch2_bkey_invalid(struct bch_fs *c, struct bkey_s_c k,
156 			enum btree_node_type type,
157 			enum bkey_invalid_flags flags,
158 			struct printbuf *err)
159 {
160 	if (k.k->u64s < BKEY_U64s) {
161 		prt_printf(err, "u64s too small (%u < %zu)", k.k->u64s, BKEY_U64s);
162 		return -BCH_ERR_invalid_bkey;
163 	}
164 
165 	if (flags & BKEY_INVALID_COMMIT	 &&
166 	    !(bch2_key_types_allowed[type] & BIT_ULL(k.k->type))) {
167 		prt_printf(err, "invalid key type for btree %s (%s)",
168 			   bch2_btree_ids[type], bch2_bkey_types[k.k->type]);
169 		return -BCH_ERR_invalid_bkey;
170 	}
171 
172 	if (btree_node_type_is_extents(type) && !bkey_whiteout(k.k)) {
173 		if (k.k->size == 0) {
174 			prt_printf(err, "size == 0");
175 			return -BCH_ERR_invalid_bkey;
176 		}
177 
178 		if (k.k->size > k.k->p.offset) {
179 			prt_printf(err, "size greater than offset (%u > %llu)",
180 			       k.k->size, k.k->p.offset);
181 			return -BCH_ERR_invalid_bkey;
182 		}
183 	} else {
184 		if (k.k->size) {
185 			prt_printf(err, "size != 0");
186 			return -BCH_ERR_invalid_bkey;
187 		}
188 	}
189 
190 	if (type != BKEY_TYPE_btree) {
191 		if (!btree_type_has_snapshots((enum btree_id) type) &&
192 		    k.k->p.snapshot) {
193 			prt_printf(err, "nonzero snapshot");
194 			return -BCH_ERR_invalid_bkey;
195 		}
196 
197 		if (btree_type_has_snapshots((enum btree_id) type) &&
198 		    !k.k->p.snapshot) {
199 			prt_printf(err, "snapshot == 0");
200 			return -BCH_ERR_invalid_bkey;
201 		}
202 
203 		if (bkey_eq(k.k->p, POS_MAX)) {
204 			prt_printf(err, "key at POS_MAX");
205 			return -BCH_ERR_invalid_bkey;
206 		}
207 	}
208 
209 	return 0;
210 }
211 
212 int bch2_bkey_invalid(struct bch_fs *c, struct bkey_s_c k,
213 		      enum btree_node_type type,
214 		      enum bkey_invalid_flags flags,
215 		      struct printbuf *err)
216 {
217 	return __bch2_bkey_invalid(c, k, type, flags, err) ?:
218 		bch2_bkey_val_invalid(c, k, flags, err);
219 }
220 
221 int bch2_bkey_in_btree_node(struct btree *b, struct bkey_s_c k,
222 			    struct printbuf *err)
223 {
224 	if (bpos_lt(k.k->p, b->data->min_key)) {
225 		prt_printf(err, "key before start of btree node");
226 		return -BCH_ERR_invalid_bkey;
227 	}
228 
229 	if (bpos_gt(k.k->p, b->data->max_key)) {
230 		prt_printf(err, "key past end of btree node");
231 		return -BCH_ERR_invalid_bkey;
232 	}
233 
234 	return 0;
235 }
236 
237 void bch2_bpos_to_text(struct printbuf *out, struct bpos pos)
238 {
239 	if (bpos_eq(pos, POS_MIN))
240 		prt_printf(out, "POS_MIN");
241 	else if (bpos_eq(pos, POS_MAX))
242 		prt_printf(out, "POS_MAX");
243 	else if (bpos_eq(pos, SPOS_MAX))
244 		prt_printf(out, "SPOS_MAX");
245 	else {
246 		if (pos.inode == U64_MAX)
247 			prt_printf(out, "U64_MAX");
248 		else
249 			prt_printf(out, "%llu", pos.inode);
250 		prt_printf(out, ":");
251 		if (pos.offset == U64_MAX)
252 			prt_printf(out, "U64_MAX");
253 		else
254 			prt_printf(out, "%llu", pos.offset);
255 		prt_printf(out, ":");
256 		if (pos.snapshot == U32_MAX)
257 			prt_printf(out, "U32_MAX");
258 		else
259 			prt_printf(out, "%u", pos.snapshot);
260 	}
261 }
262 
263 void bch2_bkey_to_text(struct printbuf *out, const struct bkey *k)
264 {
265 	if (k) {
266 		prt_printf(out, "u64s %u type ", k->u64s);
267 
268 		if (k->type < KEY_TYPE_MAX)
269 			prt_printf(out, "%s ", bch2_bkey_types[k->type]);
270 		else
271 			prt_printf(out, "%u ", k->type);
272 
273 		bch2_bpos_to_text(out, k->p);
274 
275 		prt_printf(out, " len %u ver %llu", k->size, k->version.lo);
276 	} else {
277 		prt_printf(out, "(null)");
278 	}
279 }
280 
281 void bch2_val_to_text(struct printbuf *out, struct bch_fs *c,
282 		      struct bkey_s_c k)
283 {
284 	const struct bkey_ops *ops = bch2_bkey_type_ops(k.k->type);
285 
286 	if (likely(ops->val_to_text))
287 		ops->val_to_text(out, c, k);
288 }
289 
290 void bch2_bkey_val_to_text(struct printbuf *out, struct bch_fs *c,
291 			   struct bkey_s_c k)
292 {
293 	bch2_bkey_to_text(out, k.k);
294 
295 	if (bkey_val_bytes(k.k)) {
296 		prt_printf(out, ": ");
297 		bch2_val_to_text(out, c, k);
298 	}
299 }
300 
301 void bch2_bkey_swab_val(struct bkey_s k)
302 {
303 	const struct bkey_ops *ops = bch2_bkey_type_ops(k.k->type);
304 
305 	if (ops->swab)
306 		ops->swab(k);
307 }
308 
309 bool bch2_bkey_normalize(struct bch_fs *c, struct bkey_s k)
310 {
311 	const struct bkey_ops *ops = bch2_bkey_type_ops(k.k->type);
312 
313 	return ops->key_normalize
314 		? ops->key_normalize(c, k)
315 		: false;
316 }
317 
318 bool bch2_bkey_merge(struct bch_fs *c, struct bkey_s l, struct bkey_s_c r)
319 {
320 	const struct bkey_ops *ops = bch2_bkey_type_ops(l.k->type);
321 
322 	return ops->key_merge &&
323 		bch2_bkey_maybe_mergable(l.k, r.k) &&
324 		(u64) l.k->size + r.k->size <= KEY_SIZE_MAX &&
325 		!bch2_key_merging_disabled &&
326 		ops->key_merge(c, l, r);
327 }
328 
329 static const struct old_bkey_type {
330 	u8		btree_node_type;
331 	u8		old;
332 	u8		new;
333 } bkey_renumber_table[] = {
334 	{BKEY_TYPE_btree,	128, KEY_TYPE_btree_ptr		},
335 	{BKEY_TYPE_extents,	128, KEY_TYPE_extent		},
336 	{BKEY_TYPE_extents,	129, KEY_TYPE_extent		},
337 	{BKEY_TYPE_extents,	130, KEY_TYPE_reservation	},
338 	{BKEY_TYPE_inodes,	128, KEY_TYPE_inode		},
339 	{BKEY_TYPE_inodes,	130, KEY_TYPE_inode_generation	},
340 	{BKEY_TYPE_dirents,	128, KEY_TYPE_dirent		},
341 	{BKEY_TYPE_dirents,	129, KEY_TYPE_hash_whiteout	},
342 	{BKEY_TYPE_xattrs,	128, KEY_TYPE_xattr		},
343 	{BKEY_TYPE_xattrs,	129, KEY_TYPE_hash_whiteout	},
344 	{BKEY_TYPE_alloc,	128, KEY_TYPE_alloc		},
345 	{BKEY_TYPE_quotas,	128, KEY_TYPE_quota		},
346 };
347 
348 void bch2_bkey_renumber(enum btree_node_type btree_node_type,
349 			struct bkey_packed *k,
350 			int write)
351 {
352 	const struct old_bkey_type *i;
353 
354 	for (i = bkey_renumber_table;
355 	     i < bkey_renumber_table + ARRAY_SIZE(bkey_renumber_table);
356 	     i++)
357 		if (btree_node_type == i->btree_node_type &&
358 		    k->type == (write ? i->new : i->old)) {
359 			k->type = write ? i->old : i->new;
360 			break;
361 		}
362 }
363 
364 void __bch2_bkey_compat(unsigned level, enum btree_id btree_id,
365 			unsigned version, unsigned big_endian,
366 			int write,
367 			struct bkey_format *f,
368 			struct bkey_packed *k)
369 {
370 	const struct bkey_ops *ops;
371 	struct bkey uk;
372 	struct bkey_s u;
373 	unsigned nr_compat = 5;
374 	int i;
375 
376 	/*
377 	 * Do these operations in reverse order in the write path:
378 	 */
379 
380 	for (i = 0; i < nr_compat; i++)
381 	switch (!write ? i : nr_compat - 1 - i) {
382 	case 0:
383 		if (big_endian != CPU_BIG_ENDIAN)
384 			bch2_bkey_swab_key(f, k);
385 		break;
386 	case 1:
387 		if (version < bcachefs_metadata_version_bkey_renumber)
388 			bch2_bkey_renumber(__btree_node_type(level, btree_id), k, write);
389 		break;
390 	case 2:
391 		if (version < bcachefs_metadata_version_inode_btree_change &&
392 		    btree_id == BTREE_ID_inodes) {
393 			if (!bkey_packed(k)) {
394 				struct bkey_i *u = packed_to_bkey(k);
395 
396 				swap(u->k.p.inode, u->k.p.offset);
397 			} else if (f->bits_per_field[BKEY_FIELD_INODE] &&
398 				   f->bits_per_field[BKEY_FIELD_OFFSET]) {
399 				struct bkey_format tmp = *f, *in = f, *out = &tmp;
400 
401 				swap(tmp.bits_per_field[BKEY_FIELD_INODE],
402 				     tmp.bits_per_field[BKEY_FIELD_OFFSET]);
403 				swap(tmp.field_offset[BKEY_FIELD_INODE],
404 				     tmp.field_offset[BKEY_FIELD_OFFSET]);
405 
406 				if (!write)
407 					swap(in, out);
408 
409 				uk = __bch2_bkey_unpack_key(in, k);
410 				swap(uk.p.inode, uk.p.offset);
411 				BUG_ON(!bch2_bkey_pack_key(k, &uk, out));
412 			}
413 		}
414 		break;
415 	case 3:
416 		if (version < bcachefs_metadata_version_snapshot &&
417 		    (level || btree_type_has_snapshots(btree_id))) {
418 			struct bkey_i *u = packed_to_bkey(k);
419 
420 			if (u) {
421 				u->k.p.snapshot = write
422 					? 0 : U32_MAX;
423 			} else {
424 				u64 min_packed = le64_to_cpu(f->field_offset[BKEY_FIELD_SNAPSHOT]);
425 				u64 max_packed = min_packed +
426 					~(~0ULL << f->bits_per_field[BKEY_FIELD_SNAPSHOT]);
427 
428 				uk = __bch2_bkey_unpack_key(f, k);
429 				uk.p.snapshot = write
430 					? min_packed : min_t(u64, U32_MAX, max_packed);
431 
432 				BUG_ON(!bch2_bkey_pack_key(k, &uk, f));
433 			}
434 		}
435 
436 		break;
437 	case 4:
438 		if (!bkey_packed(k)) {
439 			u = bkey_i_to_s(packed_to_bkey(k));
440 		} else {
441 			uk = __bch2_bkey_unpack_key(f, k);
442 			u.k = &uk;
443 			u.v = bkeyp_val(f, k);
444 		}
445 
446 		if (big_endian != CPU_BIG_ENDIAN)
447 			bch2_bkey_swab_val(u);
448 
449 		ops = bch2_bkey_type_ops(k->type);
450 
451 		if (ops->compat)
452 			ops->compat(btree_id, version, big_endian, write, u);
453 		break;
454 	default:
455 		BUG();
456 	}
457 }
458