1 // SPDX-License-Identifier: GPL-2.0
4 #include "btree_key_cache.h"
5 #include "bkey_methods.h"
6 #include "btree_update.h"
10 #include "extent_update.h"
13 #include "subvolume.h"
16 #include <linux/random.h>
18 #include <asm/unaligned.h>
20 const char * const bch2_inode_opts[] = {
21 #define x(name, ...) #name,
27 static const u8 byte_table[8] = { 1, 2, 3, 4, 6, 8, 10, 13 };
29 static int inode_decode_field(const u8 *in, const u8 *end,
30 u64 out[2], unsigned *out_bits)
32 __be64 be[2] = { 0, 0 };
33 unsigned bytes, shift;
43 * position of highest set bit indicates number of bytes:
44 * shift = number of bits to remove in high byte:
46 shift = 8 - __fls(*in); /* 1 <= shift <= 8 */
47 bytes = byte_table[shift - 1];
52 p = (u8 *) be + 16 - bytes;
54 *p ^= (1 << 8) >> shift;
56 out[0] = be64_to_cpu(be[0]);
57 out[1] = be64_to_cpu(be[1]);
58 *out_bits = out[0] ? 64 + fls64(out[0]) : fls64(out[1]);
63 void bch2_inode_pack(struct bch_fs *c,
64 struct bkey_inode_buf *packed,
65 const struct bch_inode_unpacked *inode)
67 struct bkey_i_inode_v2 *k = &packed->inode;
68 u8 *out = k->v.fields;
69 u8 *end = (void *) &packed[1];
70 u8 *last_nonzero_field = out;
71 unsigned nr_fields = 0, last_nonzero_fieldnr = 0;
75 bkey_inode_v2_init(&packed->inode.k_i);
76 packed->inode.k.p.offset = inode->bi_inum;
77 packed->inode.v.bi_journal_seq = cpu_to_le64(inode->bi_journal_seq);
78 packed->inode.v.bi_hash_seed = inode->bi_hash_seed;
79 packed->inode.v.bi_flags = cpu_to_le64(inode->bi_flags);
80 packed->inode.v.bi_flags = cpu_to_le64(inode->bi_flags);
81 packed->inode.v.bi_mode = cpu_to_le16(inode->bi_mode);
83 #define x(_name, _bits) \
87 ret = bch2_varint_encode_fast(out, inode->_name); \
93 last_nonzero_field = out; \
94 last_nonzero_fieldnr = nr_fields; \
106 out = last_nonzero_field;
107 nr_fields = last_nonzero_fieldnr;
109 bytes = out - (u8 *) &packed->inode.v;
110 set_bkey_val_bytes(&packed->inode.k, bytes);
111 memset_u64s_tail(&packed->inode.v, 0, bytes);
113 SET_INODEv2_NR_FIELDS(&k->v, nr_fields);
115 if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG)) {
116 struct bch_inode_unpacked unpacked;
118 int ret = bch2_inode_unpack(bkey_i_to_s_c(&packed->inode.k_i),
121 BUG_ON(unpacked.bi_inum != inode->bi_inum);
122 BUG_ON(unpacked.bi_hash_seed != inode->bi_hash_seed);
123 BUG_ON(unpacked.bi_mode != inode->bi_mode);
125 #define x(_name, _bits) if (unpacked._name != inode->_name) \
126 panic("unpacked %llu should be %llu", \
127 (u64) unpacked._name, (u64) inode->_name);
133 static noinline int bch2_inode_unpack_v1(struct bkey_s_c_inode inode,
134 struct bch_inode_unpacked *unpacked)
136 const u8 *in = inode.v->fields;
137 const u8 *end = bkey_val_end(inode);
139 unsigned fieldnr = 0, field_bits;
142 #define x(_name, _bits) \
143 if (fieldnr++ == INODE_NR_FIELDS(inode.v)) { \
144 memset(&unpacked->_name, 0, \
145 sizeof(*unpacked) - \
146 offsetof(struct bch_inode_unpacked, _name)); \
150 ret = inode_decode_field(in, end, field, &field_bits); \
154 if (field_bits > sizeof(unpacked->_name) * 8) \
157 unpacked->_name = field[1]; \
163 /* XXX: signal if there were more fields than expected? */
167 static int bch2_inode_unpack_v2(struct bch_inode_unpacked *unpacked,
168 const u8 *in, const u8 *end,
171 unsigned fieldnr = 0;
175 #define x(_name, _bits) \
176 if (fieldnr < nr_fields) { \
177 ret = bch2_varint_decode_fast(in, end, &v[0]); \
183 ret = bch2_varint_decode_fast(in, end, &v[1]); \
194 unpacked->_name = v[0]; \
195 if (v[1] || v[0] != unpacked->_name) \
202 /* XXX: signal if there were more fields than expected? */
206 int bch2_inode_unpack(struct bkey_s_c k,
207 struct bch_inode_unpacked *unpacked)
210 case KEY_TYPE_inode: {
211 struct bkey_s_c_inode inode = bkey_s_c_to_inode(k);
213 unpacked->bi_inum = inode.k->p.offset;
214 unpacked->bi_journal_seq= 0;
215 unpacked->bi_hash_seed = inode.v->bi_hash_seed;
216 unpacked->bi_flags = le32_to_cpu(inode.v->bi_flags);
217 unpacked->bi_mode = le16_to_cpu(inode.v->bi_mode);
219 if (INODE_NEW_VARINT(inode.v)) {
220 return bch2_inode_unpack_v2(unpacked, inode.v->fields,
222 INODE_NR_FIELDS(inode.v));
224 return bch2_inode_unpack_v1(inode, unpacked);
228 case KEY_TYPE_inode_v2: {
229 struct bkey_s_c_inode_v2 inode = bkey_s_c_to_inode_v2(k);
231 unpacked->bi_inum = inode.k->p.offset;
232 unpacked->bi_journal_seq= le64_to_cpu(inode.v->bi_journal_seq);
233 unpacked->bi_hash_seed = inode.v->bi_hash_seed;
234 unpacked->bi_flags = le64_to_cpu(inode.v->bi_flags);
235 unpacked->bi_mode = le16_to_cpu(inode.v->bi_mode);
237 return bch2_inode_unpack_v2(unpacked, inode.v->fields,
239 INODEv2_NR_FIELDS(inode.v));
246 int bch2_inode_peek(struct btree_trans *trans,
247 struct btree_iter *iter,
248 struct bch_inode_unpacked *inode,
249 subvol_inum inum, unsigned flags)
255 ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot);
259 bch2_trans_iter_init(trans, iter, BTREE_ID_inodes,
260 SPOS(0, inum.inum, snapshot),
261 flags|BTREE_ITER_CACHED);
262 k = bch2_btree_iter_peek_slot(iter);
267 ret = bkey_is_inode(k.k) ? 0 : -ENOENT;
271 ret = bch2_inode_unpack(k, inode);
277 bch2_trans_iter_exit(trans, iter);
281 int bch2_inode_write(struct btree_trans *trans,
282 struct btree_iter *iter,
283 struct bch_inode_unpacked *inode)
285 struct bkey_inode_buf *inode_p;
287 inode_p = bch2_trans_kmalloc(trans, sizeof(*inode_p));
289 return PTR_ERR(inode_p);
291 bch2_inode_pack(trans->c, inode_p, inode);
292 inode_p->inode.k.p.snapshot = iter->snapshot;
293 return bch2_trans_update(trans, iter, &inode_p->inode.k_i, 0);
296 const char *bch2_inode_invalid(const struct bch_fs *c, struct bkey_s_c k)
298 struct bkey_s_c_inode inode = bkey_s_c_to_inode(k);
299 struct bch_inode_unpacked unpacked;
302 return "nonzero k.p.inode";
304 if (bkey_val_bytes(k.k) < sizeof(struct bch_inode))
305 return "incorrect value size";
307 if (k.k->p.offset < BLOCKDEV_INODE_MAX)
308 return "fs inode in blockdev range";
310 if (INODE_STR_HASH(inode.v) >= BCH_STR_HASH_NR)
311 return "invalid str hash type";
313 if (bch2_inode_unpack(k, &unpacked))
314 return "invalid variable length fields";
316 if (unpacked.bi_data_checksum >= BCH_CSUM_OPT_NR + 1)
317 return "invalid data checksum type";
319 if (unpacked.bi_compression >= BCH_COMPRESSION_OPT_NR + 1)
320 return "invalid data checksum type";
322 if ((unpacked.bi_flags & BCH_INODE_UNLINKED) &&
323 unpacked.bi_nlink != 0)
324 return "flagged as unlinked but bi_nlink != 0";
326 if (unpacked.bi_subvol && !S_ISDIR(unpacked.bi_mode))
327 return "subvolume root but not a directory";
332 const char *bch2_inode_v2_invalid(const struct bch_fs *c, struct bkey_s_c k)
334 struct bkey_s_c_inode_v2 inode = bkey_s_c_to_inode_v2(k);
335 struct bch_inode_unpacked unpacked;
338 return "nonzero k.p.inode";
340 if (bkey_val_bytes(k.k) < sizeof(struct bch_inode))
341 return "incorrect value size";
343 if (k.k->p.offset < BLOCKDEV_INODE_MAX)
344 return "fs inode in blockdev range";
346 if (INODEv2_STR_HASH(inode.v) >= BCH_STR_HASH_NR)
347 return "invalid str hash type";
349 if (bch2_inode_unpack(k, &unpacked))
350 return "invalid variable length fields";
352 if (unpacked.bi_data_checksum >= BCH_CSUM_OPT_NR + 1)
353 return "invalid data checksum type";
355 if (unpacked.bi_compression >= BCH_COMPRESSION_OPT_NR + 1)
356 return "invalid data checksum type";
358 if ((unpacked.bi_flags & BCH_INODE_UNLINKED) &&
359 unpacked.bi_nlink != 0)
360 return "flagged as unlinked but bi_nlink != 0";
362 if (unpacked.bi_subvol && !S_ISDIR(unpacked.bi_mode))
363 return "subvolume root but not a directory";
368 static void __bch2_inode_unpacked_to_text(struct printbuf *out, struct bch_inode_unpacked *inode)
370 pr_buf(out, "mode %o flags %x journal_seq %llu",
371 inode->bi_mode, inode->bi_flags,
372 inode->bi_journal_seq);
374 #define x(_name, _bits) \
375 pr_buf(out, " "#_name " %llu", (u64) inode->_name);
380 void bch2_inode_unpacked_to_text(struct printbuf *out, struct bch_inode_unpacked *inode)
382 pr_buf(out, "inum: %llu ", inode->bi_inum);
383 __bch2_inode_unpacked_to_text(out, inode);
386 void bch2_inode_to_text(struct printbuf *out, struct bch_fs *c,
389 struct bch_inode_unpacked inode;
391 if (bch2_inode_unpack(k, &inode)) {
392 pr_buf(out, "(unpack error)");
396 __bch2_inode_unpacked_to_text(out, &inode);
399 const char *bch2_inode_generation_invalid(const struct bch_fs *c,
403 return "nonzero k.p.inode";
405 if (bkey_val_bytes(k.k) != sizeof(struct bch_inode_generation))
406 return "incorrect value size";
411 void bch2_inode_generation_to_text(struct printbuf *out, struct bch_fs *c,
414 struct bkey_s_c_inode_generation gen = bkey_s_c_to_inode_generation(k);
416 pr_buf(out, "generation: %u", le32_to_cpu(gen.v->bi_generation));
419 void bch2_inode_init_early(struct bch_fs *c,
420 struct bch_inode_unpacked *inode_u)
422 enum bch_str_hash_type str_hash =
423 bch2_str_hash_opt_to_type(c, c->opts.str_hash);
425 memset(inode_u, 0, sizeof(*inode_u));
428 inode_u->bi_flags |= str_hash << INODE_STR_HASH_OFFSET;
429 get_random_bytes(&inode_u->bi_hash_seed,
430 sizeof(inode_u->bi_hash_seed));
433 void bch2_inode_init_late(struct bch_inode_unpacked *inode_u, u64 now,
434 uid_t uid, gid_t gid, umode_t mode, dev_t rdev,
435 struct bch_inode_unpacked *parent)
437 inode_u->bi_mode = mode;
438 inode_u->bi_uid = uid;
439 inode_u->bi_gid = gid;
440 inode_u->bi_dev = rdev;
441 inode_u->bi_atime = now;
442 inode_u->bi_mtime = now;
443 inode_u->bi_ctime = now;
444 inode_u->bi_otime = now;
446 if (parent && parent->bi_mode & S_ISGID) {
447 inode_u->bi_gid = parent->bi_gid;
449 inode_u->bi_mode |= S_ISGID;
453 #define x(_name, ...) inode_u->bi_##_name = parent->bi_##_name;
459 void bch2_inode_init(struct bch_fs *c, struct bch_inode_unpacked *inode_u,
460 uid_t uid, gid_t gid, umode_t mode, dev_t rdev,
461 struct bch_inode_unpacked *parent)
463 bch2_inode_init_early(c, inode_u);
464 bch2_inode_init_late(inode_u, bch2_current_time(c),
465 uid, gid, mode, rdev, parent);
468 static inline u32 bkey_generation(struct bkey_s_c k)
472 case KEY_TYPE_inode_v2:
474 case KEY_TYPE_inode_generation:
475 return le32_to_cpu(bkey_s_c_to_inode_generation(k).v->bi_generation);
482 * This just finds an empty slot:
484 int bch2_inode_create(struct btree_trans *trans,
485 struct btree_iter *iter,
486 struct bch_inode_unpacked *inode_u,
487 u32 snapshot, u64 cpu)
489 struct bch_fs *c = trans->c;
491 u64 min, max, start, pos, *hint;
493 unsigned bits = (c->opts.inodes_32bit ? 31 : 63);
495 if (c->opts.shard_inode_numbers) {
496 bits -= c->inode_shard_bits;
499 max = (cpu << bits) | ~(ULLONG_MAX << bits);
501 min = max_t(u64, min, BLOCKDEV_INODE_MAX);
502 hint = c->unused_inode_hints + cpu;
504 min = BLOCKDEV_INODE_MAX;
505 max = ~(ULLONG_MAX << bits);
506 hint = c->unused_inode_hints;
509 start = READ_ONCE(*hint);
511 if (start >= max || start < min)
515 bch2_trans_iter_init(trans, iter, BTREE_ID_inodes, POS(0, pos),
516 BTREE_ITER_ALL_SNAPSHOTS|
519 while ((k = bch2_btree_iter_peek(iter)).k &&
520 !(ret = bkey_err(k)) &&
521 bkey_cmp(k.k->p, POS(0, max)) < 0) {
522 while (pos < iter->pos.offset) {
523 if (!bch2_btree_key_cache_find(c, BTREE_ID_inodes, POS(0, pos)))
529 if (k.k->p.snapshot == snapshot &&
530 !bkey_is_inode(k.k) &&
531 !bch2_btree_key_cache_find(c, BTREE_ID_inodes, SPOS(0, pos, snapshot))) {
532 bch2_btree_iter_advance(iter);
537 * We don't need to iterate over keys in every snapshot once
538 * we've found just one:
540 pos = iter->pos.offset + 1;
541 bch2_btree_iter_set_pos(iter, POS(0, pos));
544 while (!ret && pos < max) {
545 if (!bch2_btree_key_cache_find(c, BTREE_ID_inodes, POS(0, pos)))
551 if (!ret && start == min)
555 bch2_trans_iter_exit(trans, iter);
559 /* Retry from start */
561 bch2_btree_iter_set_pos(iter, POS(0, pos));
564 bch2_btree_iter_set_pos(iter, SPOS(0, pos, snapshot));
565 k = bch2_btree_iter_peek_slot(iter);
568 bch2_trans_iter_exit(trans, iter);
572 /* We may have raced while the iterator wasn't pointing at pos: */
573 if (bkey_is_inode(k.k) ||
574 bch2_btree_key_cache_find(c, BTREE_ID_inodes, k.k->p))
577 *hint = k.k->p.offset;
578 inode_u->bi_inum = k.k->p.offset;
579 inode_u->bi_generation = bkey_generation(k);
583 static int bch2_inode_delete_keys(struct btree_trans *trans,
584 subvol_inum inum, enum btree_id id)
586 struct btree_iter iter;
588 struct bkey_i delete;
593 * We're never going to be deleting extents, no need to use an extent
596 bch2_trans_iter_init(trans, &iter, id, POS(inum.inum, 0),
597 BTREE_ITER_NOT_EXTENTS|
601 bch2_trans_begin(trans);
603 ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot);
607 bch2_btree_iter_set_snapshot(&iter, snapshot);
609 k = bch2_btree_iter_peek(&iter);
614 if (!k.k || iter.pos.inode != inum.inum)
617 bkey_init(&delete.k);
618 delete.k.p = iter.pos;
620 ret = bch2_trans_update(trans, &iter, &delete, 0) ?:
621 bch2_trans_commit(trans, NULL, NULL,
622 BTREE_INSERT_NOFAIL);
624 if (ret && ret != -EINTR)
628 bch2_trans_iter_exit(trans, &iter);
632 int bch2_inode_rm(struct bch_fs *c, subvol_inum inum)
634 struct btree_trans trans;
635 struct btree_iter iter = { NULL };
636 struct bkey_i_inode_generation delete;
637 struct bch_inode_unpacked inode_u;
642 bch2_trans_init(&trans, c, 0, 1024);
645 * If this was a directory, there shouldn't be any real dirents left -
646 * but there could be whiteouts (from hash collisions) that we should
649 * XXX: the dirent could ideally would delete whiteouts when they're no
652 ret = bch2_inode_delete_keys(&trans, inum, BTREE_ID_extents) ?:
653 bch2_inode_delete_keys(&trans, inum, BTREE_ID_xattrs) ?:
654 bch2_inode_delete_keys(&trans, inum, BTREE_ID_dirents);
658 bch2_trans_begin(&trans);
660 ret = bch2_subvolume_get_snapshot(&trans, inum.subvol, &snapshot);
664 bch2_trans_iter_init(&trans, &iter, BTREE_ID_inodes,
665 SPOS(0, inum.inum, snapshot),
666 BTREE_ITER_INTENT|BTREE_ITER_CACHED);
667 k = bch2_btree_iter_peek_slot(&iter);
673 if (!bkey_is_inode(k.k)) {
674 bch2_fs_inconsistent(trans.c,
675 "inode %llu not found when deleting",
681 bch2_inode_unpack(k, &inode_u);
683 /* Subvolume root? */
684 BUG_ON(inode_u.bi_subvol);
686 bkey_inode_generation_init(&delete.k_i);
687 delete.k.p = iter.pos;
688 delete.v.bi_generation = cpu_to_le32(inode_u.bi_generation + 1);
690 ret = bch2_trans_update(&trans, &iter, &delete.k_i, 0) ?:
691 bch2_trans_commit(&trans, NULL, NULL,
692 BTREE_INSERT_NOFAIL);
694 bch2_trans_iter_exit(&trans, &iter);
698 bch2_trans_exit(&trans);
702 int bch2_inode_find_by_inum_trans(struct btree_trans *trans,
704 struct bch_inode_unpacked *inode)
706 struct btree_iter iter;
709 ret = bch2_inode_peek(trans, &iter, inode, inum, 0);
711 bch2_trans_iter_exit(trans, &iter);
715 int bch2_inode_find_by_inum(struct bch_fs *c, subvol_inum inum,
716 struct bch_inode_unpacked *inode)
718 return bch2_trans_do(c, NULL, NULL, 0,
719 bch2_inode_find_by_inum_trans(&trans, inum, inode));