1 // SPDX-License-Identifier: GPL-2.0
4 #include "btree_key_cache.h"
5 #include "btree_write_buffer.h"
6 #include "bkey_methods.h"
7 #include "btree_update.h"
11 #include "extent_update.h"
14 #include "subvolume.h"
17 #include <linux/random.h>
19 #include <asm/unaligned.h>
21 const char * const bch2_inode_opts[] = {
22 #define x(name, ...) #name,
28 static const u8 byte_table[8] = { 1, 2, 3, 4, 6, 8, 10, 13 };
30 static int inode_decode_field(const u8 *in, const u8 *end,
31 u64 out[2], unsigned *out_bits)
33 __be64 be[2] = { 0, 0 };
34 unsigned bytes, shift;
44 * position of highest set bit indicates number of bytes:
45 * shift = number of bits to remove in high byte:
47 shift = 8 - __fls(*in); /* 1 <= shift <= 8 */
48 bytes = byte_table[shift - 1];
53 p = (u8 *) be + 16 - bytes;
55 *p ^= (1 << 8) >> shift;
57 out[0] = be64_to_cpu(be[0]);
58 out[1] = be64_to_cpu(be[1]);
59 *out_bits = out[0] ? 64 + fls64(out[0]) : fls64(out[1]);
64 static inline void bch2_inode_pack_inlined(struct bkey_inode_buf *packed,
65 const struct bch_inode_unpacked *inode)
67 struct bkey_i_inode_v3 *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_v3_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_sectors = cpu_to_le64(inode->bi_sectors);
81 packed->inode.v.bi_size = cpu_to_le64(inode->bi_size);
82 packed->inode.v.bi_version = cpu_to_le64(inode->bi_version);
83 SET_INODEv3_MODE(&packed->inode.v, inode->bi_mode);
84 SET_INODEv3_FIELDS_START(&packed->inode.v, INODEv3_FIELDS_START_CUR);
87 #define x(_name, _bits) \
91 ret = bch2_varint_encode_fast(out, inode->_name); \
97 last_nonzero_field = out; \
98 last_nonzero_fieldnr = nr_fields; \
106 BCH_INODE_FIELDS_v3()
110 out = last_nonzero_field;
111 nr_fields = last_nonzero_fieldnr;
113 bytes = out - (u8 *) &packed->inode.v;
114 set_bkey_val_bytes(&packed->inode.k, bytes);
115 memset_u64s_tail(&packed->inode.v, 0, bytes);
117 SET_INODEv3_NR_FIELDS(&k->v, nr_fields);
119 if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG)) {
120 struct bch_inode_unpacked unpacked;
122 int ret = bch2_inode_unpack(bkey_i_to_s_c(&packed->inode.k_i),
125 BUG_ON(unpacked.bi_inum != inode->bi_inum);
126 BUG_ON(unpacked.bi_hash_seed != inode->bi_hash_seed);
127 BUG_ON(unpacked.bi_sectors != inode->bi_sectors);
128 BUG_ON(unpacked.bi_size != inode->bi_size);
129 BUG_ON(unpacked.bi_version != inode->bi_version);
130 BUG_ON(unpacked.bi_mode != inode->bi_mode);
132 #define x(_name, _bits) if (unpacked._name != inode->_name) \
133 panic("unpacked %llu should be %llu", \
134 (u64) unpacked._name, (u64) inode->_name);
135 BCH_INODE_FIELDS_v3()
140 void bch2_inode_pack(struct bkey_inode_buf *packed,
141 const struct bch_inode_unpacked *inode)
143 bch2_inode_pack_inlined(packed, inode);
146 static noinline int bch2_inode_unpack_v1(struct bkey_s_c_inode inode,
147 struct bch_inode_unpacked *unpacked)
149 const u8 *in = inode.v->fields;
150 const u8 *end = bkey_val_end(inode);
152 unsigned fieldnr = 0, field_bits;
155 #define x(_name, _bits) \
156 if (fieldnr++ == INODE_NR_FIELDS(inode.v)) { \
157 unsigned offset = offsetof(struct bch_inode_unpacked, _name);\
158 memset((void *) unpacked + offset, 0, \
159 sizeof(*unpacked) - offset); \
163 ret = inode_decode_field(in, end, field, &field_bits); \
167 if (field_bits > sizeof(unpacked->_name) * 8) \
170 unpacked->_name = field[1]; \
173 BCH_INODE_FIELDS_v2()
176 /* XXX: signal if there were more fields than expected? */
180 static int bch2_inode_unpack_v2(struct bch_inode_unpacked *unpacked,
181 const u8 *in, const u8 *end,
184 unsigned fieldnr = 0;
188 #define x(_name, _bits) \
189 if (fieldnr < nr_fields) { \
190 ret = bch2_varint_decode_fast(in, end, &v[0]); \
196 ret = bch2_varint_decode_fast(in, end, &v[1]); \
207 unpacked->_name = v[0]; \
208 if (v[1] || v[0] != unpacked->_name) \
212 BCH_INODE_FIELDS_v2()
215 /* XXX: signal if there were more fields than expected? */
219 static int bch2_inode_unpack_v3(struct bkey_s_c k,
220 struct bch_inode_unpacked *unpacked)
222 struct bkey_s_c_inode_v3 inode = bkey_s_c_to_inode_v3(k);
223 const u8 *in = inode.v->fields;
224 const u8 *end = bkey_val_end(inode);
225 unsigned nr_fields = INODEv3_NR_FIELDS(inode.v);
226 unsigned fieldnr = 0;
230 unpacked->bi_inum = inode.k->p.offset;
231 unpacked->bi_journal_seq= le64_to_cpu(inode.v->bi_journal_seq);
232 unpacked->bi_hash_seed = inode.v->bi_hash_seed;
233 unpacked->bi_flags = le64_to_cpu(inode.v->bi_flags);
234 unpacked->bi_sectors = le64_to_cpu(inode.v->bi_sectors);
235 unpacked->bi_size = le64_to_cpu(inode.v->bi_size);
236 unpacked->bi_version = le64_to_cpu(inode.v->bi_version);
237 unpacked->bi_mode = INODEv3_MODE(inode.v);
239 #define x(_name, _bits) \
240 if (fieldnr < nr_fields) { \
241 ret = bch2_varint_decode_fast(in, end, &v[0]); \
247 ret = bch2_varint_decode_fast(in, end, &v[1]); \
258 unpacked->_name = v[0]; \
259 if (v[1] || v[0] != unpacked->_name) \
263 BCH_INODE_FIELDS_v3()
266 /* XXX: signal if there were more fields than expected? */
270 static noinline int bch2_inode_unpack_slowpath(struct bkey_s_c k,
271 struct bch_inode_unpacked *unpacked)
273 memset(unpacked, 0, sizeof(*unpacked));
276 case KEY_TYPE_inode: {
277 struct bkey_s_c_inode inode = bkey_s_c_to_inode(k);
279 unpacked->bi_inum = inode.k->p.offset;
280 unpacked->bi_journal_seq= 0;
281 unpacked->bi_hash_seed = inode.v->bi_hash_seed;
282 unpacked->bi_flags = le32_to_cpu(inode.v->bi_flags);
283 unpacked->bi_mode = le16_to_cpu(inode.v->bi_mode);
285 if (INODE_NEW_VARINT(inode.v)) {
286 return bch2_inode_unpack_v2(unpacked, inode.v->fields,
288 INODE_NR_FIELDS(inode.v));
290 return bch2_inode_unpack_v1(inode, unpacked);
294 case KEY_TYPE_inode_v2: {
295 struct bkey_s_c_inode_v2 inode = bkey_s_c_to_inode_v2(k);
297 unpacked->bi_inum = inode.k->p.offset;
298 unpacked->bi_journal_seq= le64_to_cpu(inode.v->bi_journal_seq);
299 unpacked->bi_hash_seed = inode.v->bi_hash_seed;
300 unpacked->bi_flags = le64_to_cpu(inode.v->bi_flags);
301 unpacked->bi_mode = le16_to_cpu(inode.v->bi_mode);
303 return bch2_inode_unpack_v2(unpacked, inode.v->fields,
305 INODEv2_NR_FIELDS(inode.v));
312 int bch2_inode_unpack(struct bkey_s_c k,
313 struct bch_inode_unpacked *unpacked)
315 if (likely(k.k->type == KEY_TYPE_inode_v3))
316 return bch2_inode_unpack_v3(k, unpacked);
317 return bch2_inode_unpack_slowpath(k, unpacked);
320 int bch2_inode_peek(struct btree_trans *trans,
321 struct btree_iter *iter,
322 struct bch_inode_unpacked *inode,
323 subvol_inum inum, unsigned flags)
329 ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot);
333 k = bch2_bkey_get_iter(trans, iter, BTREE_ID_inodes,
334 SPOS(0, inum.inum, snapshot),
335 flags|BTREE_ITER_CACHED);
340 ret = bkey_is_inode(k.k) ? 0 : -BCH_ERR_ENOENT_inode;
344 ret = bch2_inode_unpack(k, inode);
350 bch2_trans_iter_exit(trans, iter);
351 if (!bch2_err_matches(ret, BCH_ERR_transaction_restart))
352 bch_err_msg(trans->c, ret, "looking up inum %u:%llu:", inum.subvol, inum.inum);
356 int bch2_inode_write(struct btree_trans *trans,
357 struct btree_iter *iter,
358 struct bch_inode_unpacked *inode)
360 struct bkey_inode_buf *inode_p;
362 inode_p = bch2_trans_kmalloc(trans, sizeof(*inode_p));
364 return PTR_ERR(inode_p);
366 bch2_inode_pack_inlined(inode_p, inode);
367 inode_p->inode.k.p.snapshot = iter->snapshot;
368 return bch2_trans_update(trans, iter, &inode_p->inode.k_i, 0);
371 struct bkey_i *bch2_inode_to_v3(struct btree_trans *trans, struct bkey_i *k)
373 struct bch_inode_unpacked u;
374 struct bkey_inode_buf *inode_p;
377 if (!bkey_is_inode(&k->k))
378 return ERR_PTR(-ENOENT);
380 inode_p = bch2_trans_kmalloc(trans, sizeof(*inode_p));
382 return ERR_CAST(inode_p);
384 ret = bch2_inode_unpack(bkey_i_to_s_c(k), &u);
388 bch2_inode_pack(inode_p, &u);
389 return &inode_p->inode.k_i;
392 static int __bch2_inode_invalid(struct bkey_s_c k, struct printbuf *err)
394 struct bch_inode_unpacked unpacked;
397 prt_printf(err, "nonzero k.p.inode");
398 return -BCH_ERR_invalid_bkey;
401 if (k.k->p.offset < BLOCKDEV_INODE_MAX) {
402 prt_printf(err, "fs inode in blockdev range");
403 return -BCH_ERR_invalid_bkey;
406 if (bch2_inode_unpack(k, &unpacked)) {
407 prt_printf(err, "invalid variable length fields");
408 return -BCH_ERR_invalid_bkey;
411 if (unpacked.bi_data_checksum >= BCH_CSUM_OPT_NR + 1) {
412 prt_printf(err, "invalid data checksum type (%u >= %u",
413 unpacked.bi_data_checksum, BCH_CSUM_OPT_NR + 1);
414 return -BCH_ERR_invalid_bkey;
417 if (unpacked.bi_compression >= BCH_COMPRESSION_OPT_NR + 1) {
418 prt_printf(err, "invalid data checksum type (%u >= %u)",
419 unpacked.bi_compression, BCH_COMPRESSION_OPT_NR + 1);
420 return -BCH_ERR_invalid_bkey;
423 if ((unpacked.bi_flags & BCH_INODE_UNLINKED) &&
424 unpacked.bi_nlink != 0) {
425 prt_printf(err, "flagged as unlinked but bi_nlink != 0");
426 return -BCH_ERR_invalid_bkey;
429 if (unpacked.bi_subvol && !S_ISDIR(unpacked.bi_mode)) {
430 prt_printf(err, "subvolume root but not a directory");
431 return -BCH_ERR_invalid_bkey;
437 int bch2_inode_invalid(const struct bch_fs *c, struct bkey_s_c k,
438 enum bkey_invalid_flags flags,
439 struct printbuf *err)
441 struct bkey_s_c_inode inode = bkey_s_c_to_inode(k);
443 if (INODE_STR_HASH(inode.v) >= BCH_STR_HASH_NR) {
444 prt_printf(err, "invalid str hash type (%llu >= %u)",
445 INODE_STR_HASH(inode.v), BCH_STR_HASH_NR);
446 return -BCH_ERR_invalid_bkey;
449 return __bch2_inode_invalid(k, err);
452 int bch2_inode_v2_invalid(const struct bch_fs *c, struct bkey_s_c k,
453 enum bkey_invalid_flags flags,
454 struct printbuf *err)
456 struct bkey_s_c_inode_v2 inode = bkey_s_c_to_inode_v2(k);
458 if (INODEv2_STR_HASH(inode.v) >= BCH_STR_HASH_NR) {
459 prt_printf(err, "invalid str hash type (%llu >= %u)",
460 INODEv2_STR_HASH(inode.v), BCH_STR_HASH_NR);
461 return -BCH_ERR_invalid_bkey;
464 return __bch2_inode_invalid(k, err);
467 int bch2_inode_v3_invalid(const struct bch_fs *c, struct bkey_s_c k,
468 enum bkey_invalid_flags flags,
469 struct printbuf *err)
471 struct bkey_s_c_inode_v3 inode = bkey_s_c_to_inode_v3(k);
473 if (INODEv3_FIELDS_START(inode.v) < INODEv3_FIELDS_START_INITIAL ||
474 INODEv3_FIELDS_START(inode.v) > bkey_val_u64s(inode.k)) {
475 prt_printf(err, "invalid fields_start (got %llu, min %u max %zu)",
476 INODEv3_FIELDS_START(inode.v),
477 INODEv3_FIELDS_START_INITIAL,
478 bkey_val_u64s(inode.k));
479 return -BCH_ERR_invalid_bkey;
482 if (INODEv3_STR_HASH(inode.v) >= BCH_STR_HASH_NR) {
483 prt_printf(err, "invalid str hash type (%llu >= %u)",
484 INODEv3_STR_HASH(inode.v), BCH_STR_HASH_NR);
485 return -BCH_ERR_invalid_bkey;
488 return __bch2_inode_invalid(k, err);
491 static void __bch2_inode_unpacked_to_text(struct printbuf *out,
492 struct bch_inode_unpacked *inode)
494 prt_printf(out, "mode %o flags %x journal_seq %llu bi_size %llu bi_sectors %llu bi_version %llu",
495 inode->bi_mode, inode->bi_flags,
496 inode->bi_journal_seq,
501 #define x(_name, _bits) \
502 prt_printf(out, " "#_name " %llu", (u64) inode->_name);
503 BCH_INODE_FIELDS_v3()
507 void bch2_inode_unpacked_to_text(struct printbuf *out, struct bch_inode_unpacked *inode)
509 prt_printf(out, "inum: %llu ", inode->bi_inum);
510 __bch2_inode_unpacked_to_text(out, inode);
513 void bch2_inode_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c k)
515 struct bch_inode_unpacked inode;
517 if (bch2_inode_unpack(k, &inode)) {
518 prt_printf(out, "(unpack error)");
522 __bch2_inode_unpacked_to_text(out, &inode);
525 static inline u64 bkey_inode_flags(struct bkey_s_c k)
529 return le32_to_cpu(bkey_s_c_to_inode(k).v->bi_flags);
530 case KEY_TYPE_inode_v2:
531 return le64_to_cpu(bkey_s_c_to_inode_v2(k).v->bi_flags);
532 case KEY_TYPE_inode_v3:
533 return le64_to_cpu(bkey_s_c_to_inode_v3(k).v->bi_flags);
539 static inline bool bkey_is_deleted_inode(struct bkey_s_c k)
541 return bkey_inode_flags(k) & BCH_INODE_UNLINKED;
544 int bch2_trans_mark_inode(struct btree_trans *trans,
545 enum btree_id btree_id, unsigned level,
550 int nr = bkey_is_inode(&new->k) - bkey_is_inode(old.k);
551 bool old_deleted = bkey_is_deleted_inode(old);
552 bool new_deleted = bkey_is_deleted_inode(bkey_i_to_s_c(new));
555 int ret = bch2_replicas_deltas_realloc(trans, 0);
556 struct replicas_delta_list *d = trans->fs_usage_deltas;
564 if (old_deleted != new_deleted) {
565 int ret = bch2_btree_bit_mod(trans, BTREE_ID_deleted_inodes, new->k.p, new_deleted);
573 int bch2_mark_inode(struct btree_trans *trans,
574 enum btree_id btree_id, unsigned level,
575 struct bkey_s_c old, struct bkey_s_c new,
578 struct bch_fs *c = trans->c;
579 struct bch_fs_usage *fs_usage;
580 u64 journal_seq = trans->journal_res.seq;
582 if (flags & BTREE_TRIGGER_INSERT) {
583 struct bch_inode_v3 *v = (struct bch_inode_v3 *) new.v;
585 BUG_ON(!journal_seq);
586 BUG_ON(new.k->type != KEY_TYPE_inode_v3);
588 v->bi_journal_seq = cpu_to_le64(journal_seq);
591 if (flags & BTREE_TRIGGER_GC) {
592 percpu_down_read(&c->mark_lock);
595 fs_usage = fs_usage_ptr(c, journal_seq, flags & BTREE_TRIGGER_GC);
596 fs_usage->nr_inodes += bkey_is_inode(new.k);
597 fs_usage->nr_inodes -= bkey_is_inode(old.k);
600 percpu_up_read(&c->mark_lock);
605 int bch2_inode_generation_invalid(const struct bch_fs *c, struct bkey_s_c k,
606 enum bkey_invalid_flags flags,
607 struct printbuf *err)
610 prt_printf(err, "nonzero k.p.inode");
611 return -BCH_ERR_invalid_bkey;
617 void bch2_inode_generation_to_text(struct printbuf *out, struct bch_fs *c,
620 struct bkey_s_c_inode_generation gen = bkey_s_c_to_inode_generation(k);
622 prt_printf(out, "generation: %u", le32_to_cpu(gen.v->bi_generation));
625 void bch2_inode_init_early(struct bch_fs *c,
626 struct bch_inode_unpacked *inode_u)
628 enum bch_str_hash_type str_hash =
629 bch2_str_hash_opt_to_type(c, c->opts.str_hash);
631 memset(inode_u, 0, sizeof(*inode_u));
634 inode_u->bi_flags |= str_hash << INODE_STR_HASH_OFFSET;
635 get_random_bytes(&inode_u->bi_hash_seed,
636 sizeof(inode_u->bi_hash_seed));
639 void bch2_inode_init_late(struct bch_inode_unpacked *inode_u, u64 now,
640 uid_t uid, gid_t gid, umode_t mode, dev_t rdev,
641 struct bch_inode_unpacked *parent)
643 inode_u->bi_mode = mode;
644 inode_u->bi_uid = uid;
645 inode_u->bi_gid = gid;
646 inode_u->bi_dev = rdev;
647 inode_u->bi_atime = now;
648 inode_u->bi_mtime = now;
649 inode_u->bi_ctime = now;
650 inode_u->bi_otime = now;
652 if (parent && parent->bi_mode & S_ISGID) {
653 inode_u->bi_gid = parent->bi_gid;
655 inode_u->bi_mode |= S_ISGID;
659 #define x(_name, ...) inode_u->bi_##_name = parent->bi_##_name;
665 void bch2_inode_init(struct bch_fs *c, struct bch_inode_unpacked *inode_u,
666 uid_t uid, gid_t gid, umode_t mode, dev_t rdev,
667 struct bch_inode_unpacked *parent)
669 bch2_inode_init_early(c, inode_u);
670 bch2_inode_init_late(inode_u, bch2_current_time(c),
671 uid, gid, mode, rdev, parent);
674 static inline u32 bkey_generation(struct bkey_s_c k)
678 case KEY_TYPE_inode_v2:
680 case KEY_TYPE_inode_generation:
681 return le32_to_cpu(bkey_s_c_to_inode_generation(k).v->bi_generation);
688 * This just finds an empty slot:
690 int bch2_inode_create(struct btree_trans *trans,
691 struct btree_iter *iter,
692 struct bch_inode_unpacked *inode_u,
693 u32 snapshot, u64 cpu)
695 struct bch_fs *c = trans->c;
697 u64 min, max, start, pos, *hint;
699 unsigned bits = (c->opts.inodes_32bit ? 31 : 63);
701 if (c->opts.shard_inode_numbers) {
702 bits -= c->inode_shard_bits;
705 max = (cpu << bits) | ~(ULLONG_MAX << bits);
707 min = max_t(u64, min, BLOCKDEV_INODE_MAX);
708 hint = c->unused_inode_hints + cpu;
710 min = BLOCKDEV_INODE_MAX;
711 max = ~(ULLONG_MAX << bits);
712 hint = c->unused_inode_hints;
715 start = READ_ONCE(*hint);
717 if (start >= max || start < min)
721 bch2_trans_iter_init(trans, iter, BTREE_ID_inodes, POS(0, pos),
722 BTREE_ITER_ALL_SNAPSHOTS|
725 while ((k = bch2_btree_iter_peek(iter)).k &&
726 !(ret = bkey_err(k)) &&
727 bkey_lt(k.k->p, POS(0, max))) {
728 if (pos < iter->pos.offset)
732 * We don't need to iterate over keys in every snapshot once
733 * we've found just one:
735 pos = iter->pos.offset + 1;
736 bch2_btree_iter_set_pos(iter, POS(0, pos));
739 if (!ret && pos < max)
742 if (!ret && start == min)
743 ret = -BCH_ERR_ENOSPC_inode_create;
746 bch2_trans_iter_exit(trans, iter);
750 /* Retry from start */
752 bch2_btree_iter_set_pos(iter, POS(0, pos));
755 bch2_btree_iter_set_pos(iter, SPOS(0, pos, snapshot));
756 k = bch2_btree_iter_peek_slot(iter);
759 bch2_trans_iter_exit(trans, iter);
763 *hint = k.k->p.offset;
764 inode_u->bi_inum = k.k->p.offset;
765 inode_u->bi_generation = bkey_generation(k);
769 static int bch2_inode_delete_keys(struct btree_trans *trans,
770 subvol_inum inum, enum btree_id id)
772 struct btree_iter iter;
774 struct bkey_i delete;
779 * We're never going to be deleting partial extents, no need to use an
782 bch2_trans_iter_init(trans, &iter, id, POS(inum.inum, 0),
783 BTREE_ITER_INTENT|BTREE_ITER_NOT_EXTENTS);
786 bch2_trans_begin(trans);
788 ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot);
792 bch2_btree_iter_set_snapshot(&iter, snapshot);
794 k = bch2_btree_iter_peek_upto(&iter, POS(inum.inum, U64_MAX));
802 bkey_init(&delete.k);
803 delete.k.p = iter.pos;
805 ret = bch2_trans_update(trans, &iter, &delete, 0) ?:
806 bch2_trans_commit(trans, NULL, NULL,
807 BTREE_INSERT_NOFAIL);
809 if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
813 bch2_trans_iter_exit(trans, &iter);
817 int bch2_inode_rm(struct bch_fs *c, subvol_inum inum)
819 struct btree_trans trans;
820 struct btree_iter iter = { NULL };
821 struct bkey_i_inode_generation delete;
822 struct bch_inode_unpacked inode_u;
827 bch2_trans_init(&trans, c, 0, 1024);
830 * If this was a directory, there shouldn't be any real dirents left -
831 * but there could be whiteouts (from hash collisions) that we should
834 * XXX: the dirent could ideally would delete whiteouts when they're no
837 ret = bch2_inode_delete_keys(&trans, inum, BTREE_ID_extents) ?:
838 bch2_inode_delete_keys(&trans, inum, BTREE_ID_xattrs) ?:
839 bch2_inode_delete_keys(&trans, inum, BTREE_ID_dirents);
843 bch2_trans_begin(&trans);
845 ret = bch2_subvolume_get_snapshot(&trans, inum.subvol, &snapshot);
849 k = bch2_bkey_get_iter(&trans, &iter, BTREE_ID_inodes,
850 SPOS(0, inum.inum, snapshot),
851 BTREE_ITER_INTENT|BTREE_ITER_CACHED);
856 if (!bkey_is_inode(k.k)) {
857 bch2_fs_inconsistent(trans.c,
858 "inode %llu:%u not found when deleting",
859 inum.inum, snapshot);
864 bch2_inode_unpack(k, &inode_u);
866 bkey_inode_generation_init(&delete.k_i);
867 delete.k.p = iter.pos;
868 delete.v.bi_generation = cpu_to_le32(inode_u.bi_generation + 1);
870 ret = bch2_trans_update(&trans, &iter, &delete.k_i, 0) ?:
871 bch2_trans_commit(&trans, NULL, NULL,
872 BTREE_INSERT_NOFAIL);
874 bch2_trans_iter_exit(&trans, &iter);
875 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
878 bch2_trans_exit(&trans);
882 int bch2_inode_find_by_inum_trans(struct btree_trans *trans,
884 struct bch_inode_unpacked *inode)
886 struct btree_iter iter;
889 ret = bch2_inode_peek(trans, &iter, inode, inum, 0);
891 bch2_trans_iter_exit(trans, &iter);
895 int bch2_inode_find_by_inum(struct bch_fs *c, subvol_inum inum,
896 struct bch_inode_unpacked *inode)
898 return bch2_trans_do(c, NULL, NULL, 0,
899 bch2_inode_find_by_inum_trans(&trans, inum, inode));
902 int bch2_inode_nlink_inc(struct bch_inode_unpacked *bi)
904 if (bi->bi_flags & BCH_INODE_UNLINKED)
905 bi->bi_flags &= ~BCH_INODE_UNLINKED;
907 if (bi->bi_nlink == U32_MAX)
916 void bch2_inode_nlink_dec(struct btree_trans *trans, struct bch_inode_unpacked *bi)
918 if (bi->bi_nlink && (bi->bi_flags & BCH_INODE_UNLINKED)) {
919 bch2_trans_inconsistent(trans, "inode %llu unlinked but link count nonzero",
924 if (bi->bi_flags & BCH_INODE_UNLINKED) {
925 bch2_trans_inconsistent(trans, "inode %llu link count underflow", bi->bi_inum);
932 bi->bi_flags |= BCH_INODE_UNLINKED;
935 struct bch_opts bch2_inode_opts_to_opts(struct bch_inode_unpacked *inode)
937 struct bch_opts ret = { 0 };
938 #define x(_name, _bits) \
939 if (inode->bi_##_name) \
940 opt_set(ret, _name, inode->bi_##_name - 1);
946 void bch2_inode_opts_get(struct bch_io_opts *opts, struct bch_fs *c,
947 struct bch_inode_unpacked *inode)
949 #define x(_name, _bits) opts->_name = inode_opt_get(c, inode, _name);
954 opts->compression = opts->background_compression = opts->data_checksum = opts->erasure_code = 0;
957 int bch2_inode_rm_snapshot(struct btree_trans *trans, u64 inum, u32 snapshot)
959 struct bch_fs *c = trans->c;
960 struct btree_iter iter = { NULL };
961 struct bkey_i_inode_generation delete;
962 struct bch_inode_unpacked inode_u;
967 ret = bch2_btree_delete_range_trans(trans, BTREE_ID_extents,
968 SPOS(inum, 0, snapshot),
969 SPOS(inum, U64_MAX, snapshot),
971 bch2_btree_delete_range_trans(trans, BTREE_ID_dirents,
972 SPOS(inum, 0, snapshot),
973 SPOS(inum, U64_MAX, snapshot),
975 bch2_btree_delete_range_trans(trans, BTREE_ID_xattrs,
976 SPOS(inum, 0, snapshot),
977 SPOS(inum, U64_MAX, snapshot),
979 } while (ret == -BCH_ERR_transaction_restart_nested);
983 bch2_trans_begin(trans);
985 k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes,
986 SPOS(0, inum, snapshot), BTREE_ITER_INTENT);
991 if (!bkey_is_inode(k.k)) {
992 bch2_fs_inconsistent(c,
993 "inode %llu:%u not found when deleting",
999 bch2_inode_unpack(k, &inode_u);
1001 /* Subvolume root? */
1002 if (inode_u.bi_subvol)
1003 bch_warn(c, "deleting inode %llu marked as unlinked, but also a subvolume root!?", inode_u.bi_inum);
1005 bkey_inode_generation_init(&delete.k_i);
1006 delete.k.p = iter.pos;
1007 delete.v.bi_generation = cpu_to_le32(inode_u.bi_generation + 1);
1009 ret = bch2_trans_update(trans, &iter, &delete.k_i, 0) ?:
1010 bch2_trans_commit(trans, NULL, NULL,
1011 BTREE_INSERT_NOFAIL);
1013 bch2_trans_iter_exit(trans, &iter);
1014 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
1017 return ret ?: -BCH_ERR_transaction_restart_nested;
1020 static int may_delete_deleted_inode(struct btree_trans *trans, struct bpos pos)
1022 struct bch_fs *c = trans->c;
1023 struct btree_iter iter;
1025 struct bch_inode_unpacked inode;
1028 if (bch2_snapshot_is_internal_node(c, pos.snapshot))
1031 if (!fsck_err_on(c->sb.clean, c,
1032 "filesystem marked as clean but have deleted inode %llu:%u",
1033 pos.offset, pos.snapshot))
1036 k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes, pos, BTREE_ITER_CACHED);
1041 ret = bkey_is_inode(k.k) ? 0 : -BCH_ERR_ENOENT_inode;
1042 if (fsck_err_on(!bkey_is_inode(k.k), c,
1043 "nonexistent inode %llu:%u in deleted_inodes btree",
1044 pos.offset, pos.snapshot))
1047 ret = bch2_inode_unpack(k, &inode);
1051 if (fsck_err_on(!(inode.bi_flags & BCH_INODE_UNLINKED), c,
1052 "non-deleted inode %llu:%u in deleted_inodes btree",
1053 pos.offset, pos.snapshot))
1061 return bch2_btree_bit_mod(trans, BTREE_ID_deleted_inodes, pos, false);
1064 int bch2_delete_dead_inodes(struct bch_fs *c)
1066 struct btree_trans trans;
1067 struct btree_iter iter;
1071 bch2_trans_init(&trans, c, 0, 0);
1073 ret = bch2_btree_write_buffer_flush_sync(&trans);
1078 * Weird transaction restart handling here because on successful delete,
1079 * bch2_inode_rm_snapshot() will return a nested transaction restart,
1080 * but we can't retry because the btree write buffer won't have been
1081 * flushed and we'd spin:
1083 for_each_btree_key(&trans, iter, BTREE_ID_deleted_inodes, POS_MIN,
1084 BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
1085 ret = lockrestart_do(&trans, may_delete_deleted_inode(&trans, k.k->p));
1090 if (!test_bit(BCH_FS_RW, &c->flags)) {
1091 bch2_trans_unlock(&trans);
1095 ret = bch2_inode_rm_snapshot(&trans, k.k->p.offset, k.k->p.snapshot);
1096 if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
1100 bch2_trans_iter_exit(&trans, &iter);
1102 bch2_trans_exit(&trans);