1 // SPDX-License-Identifier: GPL-2.0
4 #include "backpointers.h"
5 #include "bkey_methods.h"
6 #include "btree_cache.h"
7 #include "btree_types.h"
8 #include "alloc_background.h"
19 #include "subvolume.h"
22 const char * const bch2_bkey_types[] = {
23 #define x(name, nr) #name,
29 static int deleted_key_invalid(struct bch_fs *c, struct bkey_s_c k,
30 enum bkey_invalid_flags flags, struct printbuf *err)
35 #define bch2_bkey_ops_deleted ((struct bkey_ops) { \
36 .key_invalid = deleted_key_invalid, \
39 #define bch2_bkey_ops_whiteout ((struct bkey_ops) { \
40 .key_invalid = deleted_key_invalid, \
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)
48 bkey_fsck_err_on(bkey_val_bytes(k.k), c, err,
49 bkey_val_size_nonzero,
50 "incorrect value size (%zu != 0)",
56 #define bch2_bkey_ops_error ((struct bkey_ops) { \
57 .key_invalid = empty_val_key_invalid, \
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)
66 static void key_type_cookie_to_text(struct printbuf *out, struct bch_fs *c,
69 struct bkey_s_c_cookie ck = bkey_s_c_to_cookie(k);
71 prt_printf(out, "%llu", le64_to_cpu(ck.v->cookie));
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, \
80 #define bch2_bkey_ops_hash_whiteout ((struct bkey_ops) {\
81 .key_invalid = empty_val_key_invalid, \
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)
90 static void key_type_inline_data_to_text(struct printbuf *out, struct bch_fs *c,
93 struct bkey_s_c_inline_data d = bkey_s_c_to_inline_data(k);
94 unsigned datalen = bkey_inline_data_bytes(k.k);
96 prt_printf(out, "datalen %u: %*phN",
97 datalen, min(datalen, 32U), d.v->data);
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, \
105 static bool key_type_set_merge(struct bch_fs *c, struct bkey_s l, struct bkey_s_c r)
107 bch2_key_resize(l.k, l.k->size + r.k->size);
111 #define bch2_bkey_ops_set ((struct bkey_ops) { \
112 .key_invalid = empty_val_key_invalid, \
113 .key_merge = key_type_set_merge, \
116 const struct bkey_ops bch2_bkey_ops[] = {
117 #define x(name, nr) [KEY_TYPE_##name] = bch2_bkey_ops_##name,
122 const struct bkey_ops bch2_bkey_null_ops = {
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)
129 const struct bkey_ops *ops = bch2_bkey_type_ops(k.k->type);
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);
137 if (!ops->key_invalid)
140 ret = ops->key_invalid(c, k, flags, err);
145 static u64 bch2_key_types_allowed[] = {
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,
155 const char *bch2_btree_node_type_str(enum btree_node_type type)
157 return type == BKEY_TYPE_btree ? "internal btree node" : bch2_btree_id_str(type - 1);
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)
167 bkey_fsck_err_on(k.k->u64s < BKEY_U64s, c, err,
169 "u64s too small (%u < %zu)", k.k->u64s, BKEY_U64s);
171 if (type >= BKEY_TYPE_NR)
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), bch2_bkey_types[k.k->type]);
180 if (btree_node_type_is_extents(type) && !bkey_whiteout(k.k)) {
181 bkey_fsck_err_on(k.k->size == 0, c, err,
182 bkey_extent_size_zero,
185 bkey_fsck_err_on(k.k->size > k.k->p.offset, c, err,
186 bkey_extent_size_greater_than_offset,
187 "size greater than offset (%u > %llu)",
188 k.k->size, k.k->p.offset);
190 bkey_fsck_err_on(k.k->size, c, err,
195 if (type != BKEY_TYPE_btree) {
196 enum btree_id btree = type - 1;
198 if (btree_type_has_snapshots(btree)) {
199 bkey_fsck_err_on(!k.k->p.snapshot, c, err,
202 } else if (!btree_type_has_snapshot_field(btree)) {
203 bkey_fsck_err_on(k.k->p.snapshot, c, err,
204 bkey_snapshot_nonzero,
208 * btree uses snapshot field but it's not required to be
213 bkey_fsck_err_on(bkey_eq(k.k->p, POS_MAX), c, err,
221 int bch2_bkey_invalid(struct bch_fs *c, struct bkey_s_c k,
222 enum btree_node_type type,
223 enum bkey_invalid_flags flags,
224 struct printbuf *err)
226 return __bch2_bkey_invalid(c, k, type, flags, err) ?:
227 bch2_bkey_val_invalid(c, k, flags, err);
230 int bch2_bkey_in_btree_node(struct bch_fs *c, struct btree *b,
231 struct bkey_s_c k, struct printbuf *err)
235 bkey_fsck_err_on(bpos_lt(k.k->p, b->data->min_key), c, err,
236 bkey_before_start_of_btree_node,
237 "key before start of btree node");
239 bkey_fsck_err_on(bpos_gt(k.k->p, b->data->max_key), c, err,
240 bkey_after_end_of_btree_node,
241 "key past end of btree node");
246 void bch2_bpos_to_text(struct printbuf *out, struct bpos pos)
248 if (bpos_eq(pos, POS_MIN))
249 prt_printf(out, "POS_MIN");
250 else if (bpos_eq(pos, POS_MAX))
251 prt_printf(out, "POS_MAX");
252 else if (bpos_eq(pos, SPOS_MAX))
253 prt_printf(out, "SPOS_MAX");
255 if (pos.inode == U64_MAX)
256 prt_printf(out, "U64_MAX");
258 prt_printf(out, "%llu", pos.inode);
259 prt_printf(out, ":");
260 if (pos.offset == U64_MAX)
261 prt_printf(out, "U64_MAX");
263 prt_printf(out, "%llu", pos.offset);
264 prt_printf(out, ":");
265 if (pos.snapshot == U32_MAX)
266 prt_printf(out, "U32_MAX");
268 prt_printf(out, "%u", pos.snapshot);
272 void bch2_bkey_to_text(struct printbuf *out, const struct bkey *k)
275 prt_printf(out, "u64s %u type ", k->u64s);
277 if (k->type < KEY_TYPE_MAX)
278 prt_printf(out, "%s ", bch2_bkey_types[k->type]);
280 prt_printf(out, "%u ", k->type);
282 bch2_bpos_to_text(out, k->p);
284 prt_printf(out, " len %u ver %llu", k->size, k->version.lo);
286 prt_printf(out, "(null)");
290 void bch2_val_to_text(struct printbuf *out, struct bch_fs *c,
293 const struct bkey_ops *ops = bch2_bkey_type_ops(k.k->type);
295 if (likely(ops->val_to_text))
296 ops->val_to_text(out, c, k);
299 void bch2_bkey_val_to_text(struct printbuf *out, struct bch_fs *c,
302 bch2_bkey_to_text(out, k.k);
304 if (bkey_val_bytes(k.k)) {
305 prt_printf(out, ": ");
306 bch2_val_to_text(out, c, k);
310 void bch2_bkey_swab_val(struct bkey_s k)
312 const struct bkey_ops *ops = bch2_bkey_type_ops(k.k->type);
318 bool bch2_bkey_normalize(struct bch_fs *c, struct bkey_s k)
320 const struct bkey_ops *ops = bch2_bkey_type_ops(k.k->type);
322 return ops->key_normalize
323 ? ops->key_normalize(c, k)
327 bool bch2_bkey_merge(struct bch_fs *c, struct bkey_s l, struct bkey_s_c r)
329 const struct bkey_ops *ops = bch2_bkey_type_ops(l.k->type);
331 return ops->key_merge &&
332 bch2_bkey_maybe_mergable(l.k, r.k) &&
333 (u64) l.k->size + r.k->size <= KEY_SIZE_MAX &&
334 !bch2_key_merging_disabled &&
335 ops->key_merge(c, l, r);
338 static const struct old_bkey_type {
342 } bkey_renumber_table[] = {
343 {BKEY_TYPE_btree, 128, KEY_TYPE_btree_ptr },
344 {BKEY_TYPE_extents, 128, KEY_TYPE_extent },
345 {BKEY_TYPE_extents, 129, KEY_TYPE_extent },
346 {BKEY_TYPE_extents, 130, KEY_TYPE_reservation },
347 {BKEY_TYPE_inodes, 128, KEY_TYPE_inode },
348 {BKEY_TYPE_inodes, 130, KEY_TYPE_inode_generation },
349 {BKEY_TYPE_dirents, 128, KEY_TYPE_dirent },
350 {BKEY_TYPE_dirents, 129, KEY_TYPE_hash_whiteout },
351 {BKEY_TYPE_xattrs, 128, KEY_TYPE_xattr },
352 {BKEY_TYPE_xattrs, 129, KEY_TYPE_hash_whiteout },
353 {BKEY_TYPE_alloc, 128, KEY_TYPE_alloc },
354 {BKEY_TYPE_quotas, 128, KEY_TYPE_quota },
357 void bch2_bkey_renumber(enum btree_node_type btree_node_type,
358 struct bkey_packed *k,
361 const struct old_bkey_type *i;
363 for (i = bkey_renumber_table;
364 i < bkey_renumber_table + ARRAY_SIZE(bkey_renumber_table);
366 if (btree_node_type == i->btree_node_type &&
367 k->type == (write ? i->new : i->old)) {
368 k->type = write ? i->old : i->new;
373 void __bch2_bkey_compat(unsigned level, enum btree_id btree_id,
374 unsigned version, unsigned big_endian,
376 struct bkey_format *f,
377 struct bkey_packed *k)
379 const struct bkey_ops *ops;
381 unsigned nr_compat = 5;
385 * Do these operations in reverse order in the write path:
388 for (i = 0; i < nr_compat; i++)
389 switch (!write ? i : nr_compat - 1 - i) {
391 if (big_endian != CPU_BIG_ENDIAN)
392 bch2_bkey_swab_key(f, k);
395 if (version < bcachefs_metadata_version_bkey_renumber)
396 bch2_bkey_renumber(__btree_node_type(level, btree_id), k, write);
399 if (version < bcachefs_metadata_version_inode_btree_change &&
400 btree_id == BTREE_ID_inodes) {
401 if (!bkey_packed(k)) {
402 struct bkey_i *u = packed_to_bkey(k);
404 swap(u->k.p.inode, u->k.p.offset);
405 } else if (f->bits_per_field[BKEY_FIELD_INODE] &&
406 f->bits_per_field[BKEY_FIELD_OFFSET]) {
407 struct bkey_format tmp = *f, *in = f, *out = &tmp;
409 swap(tmp.bits_per_field[BKEY_FIELD_INODE],
410 tmp.bits_per_field[BKEY_FIELD_OFFSET]);
411 swap(tmp.field_offset[BKEY_FIELD_INODE],
412 tmp.field_offset[BKEY_FIELD_OFFSET]);
417 uk = __bch2_bkey_unpack_key(in, k);
418 swap(uk.p.inode, uk.p.offset);
419 BUG_ON(!bch2_bkey_pack_key(k, &uk, out));
424 if (version < bcachefs_metadata_version_snapshot &&
425 (level || btree_type_has_snapshots(btree_id))) {
426 struct bkey_i *u = packed_to_bkey(k);
429 u->k.p.snapshot = write
432 u64 min_packed = le64_to_cpu(f->field_offset[BKEY_FIELD_SNAPSHOT]);
433 u64 max_packed = min_packed +
434 ~(~0ULL << f->bits_per_field[BKEY_FIELD_SNAPSHOT]);
436 uk = __bch2_bkey_unpack_key(f, k);
437 uk.p.snapshot = write
438 ? min_packed : min_t(u64, U32_MAX, max_packed);
440 BUG_ON(!bch2_bkey_pack_key(k, &uk, f));
448 if (!bkey_packed(k)) {
449 u = bkey_i_to_s(packed_to_bkey(k));
451 uk = __bch2_bkey_unpack_key(f, k);
453 u.v = bkeyp_val(f, k);
456 if (big_endian != CPU_BIG_ENDIAN)
457 bch2_bkey_swab_val(u);
459 ops = bch2_bkey_type_ops(k->type);
462 ops->compat(btree_id, version, big_endian, write, u);