1 // SPDX-License-Identifier: GPL-2.0
4 #include "btree_key_cache.h"
5 #include "bkey_methods.h"
6 #include "btree_update.h"
9 #include "extent_update.h"
12 #include "subvolume.h"
15 #include <linux/random.h>
17 #include <asm/unaligned.h>
19 const char * const bch2_inode_opts[] = {
20 #define x(name, ...) #name,
26 static const u8 byte_table[8] = { 1, 2, 3, 4, 6, 8, 10, 13 };
27 static const u8 bits_table[8] = {
38 static int inode_decode_field(const u8 *in, const u8 *end,
39 u64 out[2], unsigned *out_bits)
41 __be64 be[2] = { 0, 0 };
42 unsigned bytes, shift;
52 * position of highest set bit indicates number of bytes:
53 * shift = number of bits to remove in high byte:
55 shift = 8 - __fls(*in); /* 1 <= shift <= 8 */
56 bytes = byte_table[shift - 1];
61 p = (u8 *) be + 16 - bytes;
63 *p ^= (1 << 8) >> shift;
65 out[0] = be64_to_cpu(be[0]);
66 out[1] = be64_to_cpu(be[1]);
67 *out_bits = out[0] ? 64 + fls64(out[0]) : fls64(out[1]);
72 void bch2_inode_pack(struct bch_fs *c,
73 struct bkey_inode_buf *packed,
74 const struct bch_inode_unpacked *inode)
76 struct bkey_i_inode_v2 *k = &packed->inode;
77 u8 *out = k->v.fields;
78 u8 *end = (void *) &packed[1];
79 u8 *last_nonzero_field = out;
80 unsigned nr_fields = 0, last_nonzero_fieldnr = 0;
84 bkey_inode_v2_init(&packed->inode.k_i);
85 packed->inode.k.p.offset = inode->bi_inum;
86 packed->inode.v.bi_journal_seq = cpu_to_le64(inode->bi_journal_seq);
87 packed->inode.v.bi_hash_seed = inode->bi_hash_seed;
88 packed->inode.v.bi_flags = cpu_to_le64(inode->bi_flags);
89 packed->inode.v.bi_flags = cpu_to_le64(inode->bi_flags);
90 packed->inode.v.bi_mode = cpu_to_le16(inode->bi_mode);
92 #define x(_name, _bits) \
96 ret = bch2_varint_encode_fast(out, inode->_name); \
102 last_nonzero_field = out; \
103 last_nonzero_fieldnr = nr_fields; \
115 out = last_nonzero_field;
116 nr_fields = last_nonzero_fieldnr;
118 bytes = out - (u8 *) &packed->inode.v;
119 set_bkey_val_bytes(&packed->inode.k, bytes);
120 memset_u64s_tail(&packed->inode.v, 0, bytes);
122 SET_INODEv2_NR_FIELDS(&k->v, nr_fields);
124 if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG)) {
125 struct bch_inode_unpacked unpacked;
127 int ret = bch2_inode_unpack(bkey_i_to_s_c(&packed->inode.k_i),
130 BUG_ON(unpacked.bi_inum != inode->bi_inum);
131 BUG_ON(unpacked.bi_hash_seed != inode->bi_hash_seed);
132 BUG_ON(unpacked.bi_mode != inode->bi_mode);
134 #define x(_name, _bits) if (unpacked._name != inode->_name) \
135 panic("unpacked %llu should be %llu", \
136 (u64) unpacked._name, (u64) inode->_name);
142 static noinline int bch2_inode_unpack_v1(struct bkey_s_c_inode inode,
143 struct bch_inode_unpacked *unpacked)
145 const u8 *in = inode.v->fields;
146 const u8 *end = bkey_val_end(inode);
148 unsigned fieldnr = 0, field_bits;
151 #define x(_name, _bits) \
152 if (fieldnr++ == INODE_NR_FIELDS(inode.v)) { \
153 memset(&unpacked->_name, 0, \
154 sizeof(*unpacked) - \
155 offsetof(struct bch_inode_unpacked, _name)); \
159 ret = inode_decode_field(in, end, field, &field_bits); \
163 if (field_bits > sizeof(unpacked->_name) * 8) \
166 unpacked->_name = field[1]; \
172 /* XXX: signal if there were more fields than expected? */
176 static int bch2_inode_unpack_v2(struct bch_inode_unpacked *unpacked,
177 const u8 *in, const u8 *end,
180 unsigned fieldnr = 0;
184 #define x(_name, _bits) \
185 if (fieldnr < nr_fields) { \
186 ret = bch2_varint_decode_fast(in, end, &v[0]); \
192 ret = bch2_varint_decode_fast(in, end, &v[1]); \
203 unpacked->_name = v[0]; \
204 if (v[1] || v[0] != unpacked->_name) \
211 /* XXX: signal if there were more fields than expected? */
215 int bch2_inode_unpack(struct bkey_s_c k,
216 struct bch_inode_unpacked *unpacked)
219 case KEY_TYPE_inode: {
220 struct bkey_s_c_inode inode = bkey_s_c_to_inode(k);
222 unpacked->bi_inum = inode.k->p.offset;
223 unpacked->bi_journal_seq= 0;
224 unpacked->bi_hash_seed = inode.v->bi_hash_seed;
225 unpacked->bi_flags = le32_to_cpu(inode.v->bi_flags);
226 unpacked->bi_mode = le16_to_cpu(inode.v->bi_mode);
228 if (INODE_NEW_VARINT(inode.v)) {
229 return bch2_inode_unpack_v2(unpacked, inode.v->fields,
231 INODE_NR_FIELDS(inode.v));
233 return bch2_inode_unpack_v1(inode, unpacked);
237 case KEY_TYPE_inode_v2: {
238 struct bkey_s_c_inode_v2 inode = bkey_s_c_to_inode_v2(k);
240 unpacked->bi_inum = inode.k->p.offset;
241 unpacked->bi_journal_seq= le64_to_cpu(inode.v->bi_journal_seq);
242 unpacked->bi_hash_seed = inode.v->bi_hash_seed;
243 unpacked->bi_flags = le64_to_cpu(inode.v->bi_flags);
244 unpacked->bi_mode = le16_to_cpu(inode.v->bi_mode);
246 return bch2_inode_unpack_v2(unpacked, inode.v->fields,
248 INODEv2_NR_FIELDS(inode.v));
255 int bch2_inode_peek(struct btree_trans *trans,
256 struct btree_iter *iter,
257 struct bch_inode_unpacked *inode,
258 subvol_inum inum, unsigned flags)
264 if (0 && trans->c->opts.inodes_use_key_cache)
265 flags |= BTREE_ITER_CACHED;
267 ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot);
271 bch2_trans_iter_init(trans, iter, BTREE_ID_inodes,
272 SPOS(0, inum.inum, snapshot), flags);
273 k = bch2_btree_iter_peek_slot(iter);
278 ret = bkey_is_inode(k.k) ? 0 : -ENOENT;
282 ret = bch2_inode_unpack(k, inode);
288 bch2_trans_iter_exit(trans, iter);
292 int bch2_inode_write(struct btree_trans *trans,
293 struct btree_iter *iter,
294 struct bch_inode_unpacked *inode)
296 struct bkey_inode_buf *inode_p;
298 inode_p = bch2_trans_kmalloc(trans, sizeof(*inode_p));
300 return PTR_ERR(inode_p);
302 bch2_inode_pack(trans->c, inode_p, inode);
303 inode_p->inode.k.p.snapshot = iter->snapshot;
304 return bch2_trans_update(trans, iter, &inode_p->inode.k_i, 0);
307 const char *bch2_inode_invalid(const struct bch_fs *c, struct bkey_s_c k)
309 struct bkey_s_c_inode inode = bkey_s_c_to_inode(k);
310 struct bch_inode_unpacked unpacked;
313 return "nonzero k.p.inode";
315 if (bkey_val_bytes(k.k) < sizeof(struct bch_inode))
316 return "incorrect value size";
318 if (k.k->p.offset < BLOCKDEV_INODE_MAX)
319 return "fs inode in blockdev range";
321 if (INODE_STR_HASH(inode.v) >= BCH_STR_HASH_NR)
322 return "invalid str hash type";
324 if (bch2_inode_unpack(k, &unpacked))
325 return "invalid variable length fields";
327 if (unpacked.bi_data_checksum >= BCH_CSUM_OPT_NR + 1)
328 return "invalid data checksum type";
330 if (unpacked.bi_compression >= BCH_COMPRESSION_OPT_NR + 1)
331 return "invalid data checksum type";
333 if ((unpacked.bi_flags & BCH_INODE_UNLINKED) &&
334 unpacked.bi_nlink != 0)
335 return "flagged as unlinked but bi_nlink != 0";
337 if (unpacked.bi_subvol && !S_ISDIR(unpacked.bi_mode))
338 return "subvolume root but not a directory";
343 const char *bch2_inode_v2_invalid(const struct bch_fs *c, struct bkey_s_c k)
345 struct bkey_s_c_inode_v2 inode = bkey_s_c_to_inode_v2(k);
346 struct bch_inode_unpacked unpacked;
349 return "nonzero k.p.inode";
351 if (bkey_val_bytes(k.k) < sizeof(struct bch_inode))
352 return "incorrect value size";
354 if (k.k->p.offset < BLOCKDEV_INODE_MAX)
355 return "fs inode in blockdev range";
357 if (INODEv2_STR_HASH(inode.v) >= BCH_STR_HASH_NR)
358 return "invalid str hash type";
360 if (bch2_inode_unpack(k, &unpacked))
361 return "invalid variable length fields";
363 if (unpacked.bi_data_checksum >= BCH_CSUM_OPT_NR + 1)
364 return "invalid data checksum type";
366 if (unpacked.bi_compression >= BCH_COMPRESSION_OPT_NR + 1)
367 return "invalid data checksum type";
369 if ((unpacked.bi_flags & BCH_INODE_UNLINKED) &&
370 unpacked.bi_nlink != 0)
371 return "flagged as unlinked but bi_nlink != 0";
373 if (unpacked.bi_subvol && !S_ISDIR(unpacked.bi_mode))
374 return "subvolume root but not a directory";
379 static void __bch2_inode_unpacked_to_text(struct printbuf *out, struct bch_inode_unpacked *inode)
381 pr_buf(out, "mode %o flags %x journal_seq %llu",
382 inode->bi_mode, inode->bi_flags,
383 inode->bi_journal_seq);
385 #define x(_name, _bits) \
386 pr_buf(out, " "#_name " %llu", (u64) inode->_name);
391 void bch2_inode_unpacked_to_text(struct printbuf *out, struct bch_inode_unpacked *inode)
393 pr_buf(out, "inum: %llu ", inode->bi_inum);
394 __bch2_inode_unpacked_to_text(out, inode);
397 void bch2_inode_to_text(struct printbuf *out, struct bch_fs *c,
400 struct bch_inode_unpacked inode;
402 if (bch2_inode_unpack(k, &inode)) {
403 pr_buf(out, "(unpack error)");
407 __bch2_inode_unpacked_to_text(out, &inode);
410 const char *bch2_inode_generation_invalid(const struct bch_fs *c,
414 return "nonzero k.p.inode";
416 if (bkey_val_bytes(k.k) != sizeof(struct bch_inode_generation))
417 return "incorrect value size";
422 void bch2_inode_generation_to_text(struct printbuf *out, struct bch_fs *c,
425 struct bkey_s_c_inode_generation gen = bkey_s_c_to_inode_generation(k);
427 pr_buf(out, "generation: %u", le32_to_cpu(gen.v->bi_generation));
430 void bch2_inode_init_early(struct bch_fs *c,
431 struct bch_inode_unpacked *inode_u)
433 enum bch_str_hash_type str_hash =
434 bch2_str_hash_opt_to_type(c, c->opts.str_hash);
436 memset(inode_u, 0, sizeof(*inode_u));
439 inode_u->bi_flags |= str_hash << INODE_STR_HASH_OFFSET;
440 get_random_bytes(&inode_u->bi_hash_seed,
441 sizeof(inode_u->bi_hash_seed));
444 void bch2_inode_init_late(struct bch_inode_unpacked *inode_u, u64 now,
445 uid_t uid, gid_t gid, umode_t mode, dev_t rdev,
446 struct bch_inode_unpacked *parent)
448 inode_u->bi_mode = mode;
449 inode_u->bi_uid = uid;
450 inode_u->bi_gid = gid;
451 inode_u->bi_dev = rdev;
452 inode_u->bi_atime = now;
453 inode_u->bi_mtime = now;
454 inode_u->bi_ctime = now;
455 inode_u->bi_otime = now;
457 if (parent && parent->bi_mode & S_ISGID) {
458 inode_u->bi_gid = parent->bi_gid;
460 inode_u->bi_mode |= S_ISGID;
464 #define x(_name, ...) inode_u->bi_##_name = parent->bi_##_name;
470 void bch2_inode_init(struct bch_fs *c, struct bch_inode_unpacked *inode_u,
471 uid_t uid, gid_t gid, umode_t mode, dev_t rdev,
472 struct bch_inode_unpacked *parent)
474 bch2_inode_init_early(c, inode_u);
475 bch2_inode_init_late(inode_u, bch2_current_time(c),
476 uid, gid, mode, rdev, parent);
479 static inline u32 bkey_generation(struct bkey_s_c k)
483 case KEY_TYPE_inode_v2:
485 case KEY_TYPE_inode_generation:
486 return le32_to_cpu(bkey_s_c_to_inode_generation(k).v->bi_generation);
493 * This just finds an empty slot:
495 int bch2_inode_create(struct btree_trans *trans,
496 struct btree_iter *iter,
497 struct bch_inode_unpacked *inode_u,
498 u32 snapshot, u64 cpu)
500 struct bch_fs *c = trans->c;
502 u64 min, max, start, pos, *hint;
504 unsigned bits = (c->opts.inodes_32bit ? 31 : 63);
506 if (c->opts.shard_inode_numbers) {
507 bits -= c->inode_shard_bits;
510 max = (cpu << bits) | ~(ULLONG_MAX << bits);
512 min = max_t(u64, min, BLOCKDEV_INODE_MAX);
513 hint = c->unused_inode_hints + cpu;
515 min = BLOCKDEV_INODE_MAX;
516 max = ~(ULLONG_MAX << bits);
517 hint = c->unused_inode_hints;
520 start = READ_ONCE(*hint);
522 if (start >= max || start < min)
526 bch2_trans_iter_init(trans, iter, BTREE_ID_inodes, POS(0, pos),
527 BTREE_ITER_ALL_SNAPSHOTS|
530 while ((k = bch2_btree_iter_peek(iter)).k &&
531 !(ret = bkey_err(k)) &&
532 bkey_cmp(k.k->p, POS(0, max)) < 0) {
533 while (pos < iter->pos.offset) {
534 if (!bch2_btree_key_cache_find(c, BTREE_ID_inodes, POS(0, pos)))
540 if (k.k->p.snapshot == snapshot &&
541 !bkey_is_inode(k.k) &&
542 !bch2_btree_key_cache_find(c, BTREE_ID_inodes, SPOS(0, pos, snapshot))) {
543 bch2_btree_iter_advance(iter);
548 * We don't need to iterate over keys in every snapshot once
549 * we've found just one:
551 pos = iter->pos.offset + 1;
552 bch2_btree_iter_set_pos(iter, POS(0, pos));
555 while (!ret && pos < max) {
556 if (!bch2_btree_key_cache_find(c, BTREE_ID_inodes, POS(0, pos)))
562 if (!ret && start == min)
566 bch2_trans_iter_exit(trans, iter);
570 /* Retry from start */
572 bch2_btree_iter_set_pos(iter, POS(0, pos));
575 bch2_btree_iter_set_pos(iter, SPOS(0, pos, snapshot));
576 k = bch2_btree_iter_peek_slot(iter);
579 bch2_trans_iter_exit(trans, iter);
583 /* We may have raced while the iterator wasn't pointing at pos: */
584 if (bkey_is_inode(k.k) ||
585 bch2_btree_key_cache_find(c, BTREE_ID_inodes, k.k->p))
588 *hint = k.k->p.offset;
589 inode_u->bi_inum = k.k->p.offset;
590 inode_u->bi_generation = bkey_generation(k);
594 static int bch2_inode_delete_keys(struct btree_trans *trans,
595 subvol_inum inum, enum btree_id id)
600 while (!ret || ret == -EINTR) {
601 struct btree_iter iter;
603 struct bkey_i delete;
606 bch2_trans_begin(trans);
608 ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot);
612 bch2_trans_iter_init(trans, &iter, id,
613 SPOS(inum.inum, offset, snapshot),
615 k = bch2_btree_iter_peek(&iter);
617 if (!k.k || iter.pos.inode != inum.inum) {
618 bch2_trans_iter_exit(trans, &iter);
626 bkey_init(&delete.k);
627 delete.k.p = iter.pos;
629 if (btree_node_type_is_extents(iter.btree_id)) {
630 unsigned max_sectors =
631 min_t(u64, U64_MAX - iter.pos.offset,
632 KEY_SIZE_MAX & (~0 << trans->c->block_bits));
634 /* create the biggest key we can */
635 bch2_key_resize(&delete.k, max_sectors);
637 ret = bch2_extent_trim_atomic(trans, &iter, &delete);
642 ret = bch2_trans_update(trans, &iter, &delete, 0) ?:
643 bch2_trans_commit(trans, NULL, NULL,
644 BTREE_INSERT_NOFAIL);
646 offset = iter.pos.offset;
647 bch2_trans_iter_exit(trans, &iter);
653 int bch2_inode_rm(struct bch_fs *c, subvol_inum inum, bool cached)
655 struct btree_trans trans;
656 struct btree_iter iter = { NULL };
657 struct bkey_i_inode_generation delete;
658 struct bch_inode_unpacked inode_u;
660 unsigned iter_flags = BTREE_ITER_INTENT;
664 if (0 && cached && c->opts.inodes_use_key_cache)
665 iter_flags |= BTREE_ITER_CACHED;
667 bch2_trans_init(&trans, c, 0, 1024);
670 * If this was a directory, there shouldn't be any real dirents left -
671 * but there could be whiteouts (from hash collisions) that we should
674 * XXX: the dirent could ideally would delete whiteouts when they're no
677 ret = bch2_inode_delete_keys(&trans, inum, BTREE_ID_extents) ?:
678 bch2_inode_delete_keys(&trans, inum, BTREE_ID_xattrs) ?:
679 bch2_inode_delete_keys(&trans, inum, BTREE_ID_dirents);
683 bch2_trans_begin(&trans);
685 ret = bch2_subvolume_get_snapshot(&trans, inum.subvol, &snapshot);
689 bch2_trans_iter_init(&trans, &iter, BTREE_ID_inodes,
690 SPOS(0, inum.inum, snapshot), iter_flags);
691 k = bch2_btree_iter_peek_slot(&iter);
697 if (!bkey_is_inode(k.k)) {
698 bch2_fs_inconsistent(trans.c,
699 "inode %llu not found when deleting",
705 bch2_inode_unpack(k, &inode_u);
707 /* Subvolume root? */
708 BUG_ON(inode_u.bi_subvol);
710 bkey_inode_generation_init(&delete.k_i);
711 delete.k.p = iter.pos;
712 delete.v.bi_generation = cpu_to_le32(inode_u.bi_generation + 1);
714 ret = bch2_trans_update(&trans, &iter, &delete.k_i, 0) ?:
715 bch2_trans_commit(&trans, NULL, NULL,
716 BTREE_INSERT_NOFAIL);
718 bch2_trans_iter_exit(&trans, &iter);
722 bch2_trans_exit(&trans);
726 int bch2_inode_find_by_inum_trans(struct btree_trans *trans,
728 struct bch_inode_unpacked *inode)
730 struct btree_iter iter;
733 ret = bch2_inode_peek(trans, &iter, inode, inum, 0);
735 bch2_trans_iter_exit(trans, &iter);
739 int bch2_inode_find_by_inum(struct bch_fs *c, subvol_inum inum,
740 struct bch_inode_unpacked *inode)
742 return bch2_trans_do(c, NULL, NULL, 0,
743 bch2_inode_find_by_inum_trans(&trans, inum, inode));