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"
12 #include "extent_update.h"
16 #include "subvolume.h"
19 #include <linux/random.h>
21 #include <asm/unaligned.h>
23 const char * const bch2_inode_opts[] = {
24 #define x(name, ...) #name,
30 static const u8 byte_table[8] = { 1, 2, 3, 4, 6, 8, 10, 13 };
32 static int inode_decode_field(const u8 *in, const u8 *end,
33 u64 out[2], unsigned *out_bits)
35 __be64 be[2] = { 0, 0 };
36 unsigned bytes, shift;
46 * position of highest set bit indicates number of bytes:
47 * shift = number of bits to remove in high byte:
49 shift = 8 - __fls(*in); /* 1 <= shift <= 8 */
50 bytes = byte_table[shift - 1];
55 p = (u8 *) be + 16 - bytes;
57 *p ^= (1 << 8) >> shift;
59 out[0] = be64_to_cpu(be[0]);
60 out[1] = be64_to_cpu(be[1]);
61 *out_bits = out[0] ? 64 + fls64(out[0]) : fls64(out[1]);
66 static inline void bch2_inode_pack_inlined(struct bkey_inode_buf *packed,
67 const struct bch_inode_unpacked *inode)
69 struct bkey_i_inode_v3 *k = &packed->inode;
70 u8 *out = k->v.fields;
71 u8 *end = (void *) &packed[1];
72 u8 *last_nonzero_field = out;
73 unsigned nr_fields = 0, last_nonzero_fieldnr = 0;
77 bkey_inode_v3_init(&packed->inode.k_i);
78 packed->inode.k.p.offset = inode->bi_inum;
79 packed->inode.v.bi_journal_seq = cpu_to_le64(inode->bi_journal_seq);
80 packed->inode.v.bi_hash_seed = inode->bi_hash_seed;
81 packed->inode.v.bi_flags = cpu_to_le64(inode->bi_flags);
82 packed->inode.v.bi_sectors = cpu_to_le64(inode->bi_sectors);
83 packed->inode.v.bi_size = cpu_to_le64(inode->bi_size);
84 packed->inode.v.bi_version = cpu_to_le64(inode->bi_version);
85 SET_INODEv3_MODE(&packed->inode.v, inode->bi_mode);
86 SET_INODEv3_FIELDS_START(&packed->inode.v, INODEv3_FIELDS_START_CUR);
89 #define x(_name, _bits) \
93 ret = bch2_varint_encode_fast(out, inode->_name); \
99 last_nonzero_field = out; \
100 last_nonzero_fieldnr = nr_fields; \
108 BCH_INODE_FIELDS_v3()
112 out = last_nonzero_field;
113 nr_fields = last_nonzero_fieldnr;
115 bytes = out - (u8 *) &packed->inode.v;
116 set_bkey_val_bytes(&packed->inode.k, bytes);
117 memset_u64s_tail(&packed->inode.v, 0, bytes);
119 SET_INODEv3_NR_FIELDS(&k->v, nr_fields);
121 if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG)) {
122 struct bch_inode_unpacked unpacked;
124 ret = bch2_inode_unpack(bkey_i_to_s_c(&packed->inode.k_i), &unpacked);
126 BUG_ON(unpacked.bi_inum != inode->bi_inum);
127 BUG_ON(unpacked.bi_hash_seed != inode->bi_hash_seed);
128 BUG_ON(unpacked.bi_sectors != inode->bi_sectors);
129 BUG_ON(unpacked.bi_size != inode->bi_size);
130 BUG_ON(unpacked.bi_version != inode->bi_version);
131 BUG_ON(unpacked.bi_mode != inode->bi_mode);
133 #define x(_name, _bits) if (unpacked._name != inode->_name) \
134 panic("unpacked %llu should be %llu", \
135 (u64) unpacked._name, (u64) inode->_name);
136 BCH_INODE_FIELDS_v3()
141 void bch2_inode_pack(struct bkey_inode_buf *packed,
142 const struct bch_inode_unpacked *inode)
144 bch2_inode_pack_inlined(packed, inode);
147 static noinline int bch2_inode_unpack_v1(struct bkey_s_c_inode inode,
148 struct bch_inode_unpacked *unpacked)
150 const u8 *in = inode.v->fields;
151 const u8 *end = bkey_val_end(inode);
153 unsigned fieldnr = 0, field_bits;
156 #define x(_name, _bits) \
157 if (fieldnr++ == INODE_NR_FIELDS(inode.v)) { \
158 unsigned offset = offsetof(struct bch_inode_unpacked, _name);\
159 memset((void *) unpacked + offset, 0, \
160 sizeof(*unpacked) - offset); \
164 ret = inode_decode_field(in, end, field, &field_bits); \
168 if (field_bits > sizeof(unpacked->_name) * 8) \
171 unpacked->_name = field[1]; \
174 BCH_INODE_FIELDS_v2()
177 /* XXX: signal if there were more fields than expected? */
181 static int bch2_inode_unpack_v2(struct bch_inode_unpacked *unpacked,
182 const u8 *in, const u8 *end,
185 unsigned fieldnr = 0;
189 #define x(_name, _bits) \
190 if (fieldnr < nr_fields) { \
191 ret = bch2_varint_decode_fast(in, end, &v[0]); \
197 ret = bch2_varint_decode_fast(in, end, &v[1]); \
208 unpacked->_name = v[0]; \
209 if (v[1] || v[0] != unpacked->_name) \
213 BCH_INODE_FIELDS_v2()
216 /* XXX: signal if there were more fields than expected? */
220 static int bch2_inode_unpack_v3(struct bkey_s_c k,
221 struct bch_inode_unpacked *unpacked)
223 struct bkey_s_c_inode_v3 inode = bkey_s_c_to_inode_v3(k);
224 const u8 *in = inode.v->fields;
225 const u8 *end = bkey_val_end(inode);
226 unsigned nr_fields = INODEv3_NR_FIELDS(inode.v);
227 unsigned fieldnr = 0;
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_sectors = le64_to_cpu(inode.v->bi_sectors);
236 unpacked->bi_size = le64_to_cpu(inode.v->bi_size);
237 unpacked->bi_version = le64_to_cpu(inode.v->bi_version);
238 unpacked->bi_mode = INODEv3_MODE(inode.v);
240 #define x(_name, _bits) \
241 if (fieldnr < nr_fields) { \
242 ret = bch2_varint_decode_fast(in, end, &v[0]); \
248 ret = bch2_varint_decode_fast(in, end, &v[1]); \
259 unpacked->_name = v[0]; \
260 if (v[1] || v[0] != unpacked->_name) \
264 BCH_INODE_FIELDS_v3()
267 /* XXX: signal if there were more fields than expected? */
271 static noinline int bch2_inode_unpack_slowpath(struct bkey_s_c k,
272 struct bch_inode_unpacked *unpacked)
274 memset(unpacked, 0, sizeof(*unpacked));
277 case KEY_TYPE_inode: {
278 struct bkey_s_c_inode inode = bkey_s_c_to_inode(k);
280 unpacked->bi_inum = inode.k->p.offset;
281 unpacked->bi_journal_seq= 0;
282 unpacked->bi_hash_seed = inode.v->bi_hash_seed;
283 unpacked->bi_flags = le32_to_cpu(inode.v->bi_flags);
284 unpacked->bi_mode = le16_to_cpu(inode.v->bi_mode);
286 if (INODE_NEW_VARINT(inode.v)) {
287 return bch2_inode_unpack_v2(unpacked, inode.v->fields,
289 INODE_NR_FIELDS(inode.v));
291 return bch2_inode_unpack_v1(inode, unpacked);
295 case KEY_TYPE_inode_v2: {
296 struct bkey_s_c_inode_v2 inode = bkey_s_c_to_inode_v2(k);
298 unpacked->bi_inum = inode.k->p.offset;
299 unpacked->bi_journal_seq= le64_to_cpu(inode.v->bi_journal_seq);
300 unpacked->bi_hash_seed = inode.v->bi_hash_seed;
301 unpacked->bi_flags = le64_to_cpu(inode.v->bi_flags);
302 unpacked->bi_mode = le16_to_cpu(inode.v->bi_mode);
304 return bch2_inode_unpack_v2(unpacked, inode.v->fields,
306 INODEv2_NR_FIELDS(inode.v));
313 int bch2_inode_unpack(struct bkey_s_c k,
314 struct bch_inode_unpacked *unpacked)
316 if (likely(k.k->type == KEY_TYPE_inode_v3))
317 return bch2_inode_unpack_v3(k, unpacked);
318 return bch2_inode_unpack_slowpath(k, unpacked);
321 static int bch2_inode_peek_nowarn(struct btree_trans *trans,
322 struct btree_iter *iter,
323 struct bch_inode_unpacked *inode,
324 subvol_inum inum, unsigned flags)
330 ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot);
334 k = bch2_bkey_get_iter(trans, iter, BTREE_ID_inodes,
335 SPOS(0, inum.inum, snapshot),
336 flags|BTREE_ITER_CACHED);
341 ret = bkey_is_inode(k.k) ? 0 : -BCH_ERR_ENOENT_inode;
345 ret = bch2_inode_unpack(k, inode);
351 bch2_trans_iter_exit(trans, iter);
355 int bch2_inode_peek(struct btree_trans *trans,
356 struct btree_iter *iter,
357 struct bch_inode_unpacked *inode,
358 subvol_inum inum, unsigned flags)
360 int ret = bch2_inode_peek_nowarn(trans, iter, inode, inum, flags);
361 bch_err_msg(trans->c, ret, "looking up inum %u:%llu:", inum.subvol, inum.inum);
365 int bch2_inode_write(struct btree_trans *trans,
366 struct btree_iter *iter,
367 struct bch_inode_unpacked *inode)
369 struct bkey_inode_buf *inode_p;
371 inode_p = bch2_trans_kmalloc(trans, sizeof(*inode_p));
373 return PTR_ERR(inode_p);
375 bch2_inode_pack_inlined(inode_p, inode);
376 inode_p->inode.k.p.snapshot = iter->snapshot;
377 return bch2_trans_update(trans, iter, &inode_p->inode.k_i, 0);
380 struct bkey_i *bch2_inode_to_v3(struct btree_trans *trans, struct bkey_i *k)
382 struct bch_inode_unpacked u;
383 struct bkey_inode_buf *inode_p;
386 if (!bkey_is_inode(&k->k))
387 return ERR_PTR(-ENOENT);
389 inode_p = bch2_trans_kmalloc(trans, sizeof(*inode_p));
391 return ERR_CAST(inode_p);
393 ret = bch2_inode_unpack(bkey_i_to_s_c(k), &u);
397 bch2_inode_pack(inode_p, &u);
398 return &inode_p->inode.k_i;
401 static int __bch2_inode_invalid(struct bkey_s_c k, struct printbuf *err)
403 struct bch_inode_unpacked unpacked;
406 prt_printf(err, "nonzero k.p.inode");
407 return -BCH_ERR_invalid_bkey;
410 if (k.k->p.offset < BLOCKDEV_INODE_MAX) {
411 prt_printf(err, "fs inode in blockdev range");
412 return -BCH_ERR_invalid_bkey;
415 if (bch2_inode_unpack(k, &unpacked)) {
416 prt_printf(err, "invalid variable length fields");
417 return -BCH_ERR_invalid_bkey;
420 if (unpacked.bi_data_checksum >= BCH_CSUM_OPT_NR + 1) {
421 prt_printf(err, "invalid data checksum type (%u >= %u",
422 unpacked.bi_data_checksum, BCH_CSUM_OPT_NR + 1);
423 return -BCH_ERR_invalid_bkey;
426 if (unpacked.bi_compression &&
427 !bch2_compression_opt_valid(unpacked.bi_compression - 1)) {
428 prt_printf(err, "invalid compression opt %u",
429 unpacked.bi_compression - 1);
430 return -BCH_ERR_invalid_bkey;
433 if ((unpacked.bi_flags & BCH_INODE_UNLINKED) &&
434 unpacked.bi_nlink != 0) {
435 prt_printf(err, "flagged as unlinked but bi_nlink != 0");
436 return -BCH_ERR_invalid_bkey;
439 if (unpacked.bi_subvol && !S_ISDIR(unpacked.bi_mode)) {
440 prt_printf(err, "subvolume root but not a directory");
441 return -BCH_ERR_invalid_bkey;
447 int bch2_inode_invalid(const struct bch_fs *c, struct bkey_s_c k,
448 enum bkey_invalid_flags flags,
449 struct printbuf *err)
451 struct bkey_s_c_inode inode = bkey_s_c_to_inode(k);
453 if (INODE_STR_HASH(inode.v) >= BCH_STR_HASH_NR) {
454 prt_printf(err, "invalid str hash type (%llu >= %u)",
455 INODE_STR_HASH(inode.v), BCH_STR_HASH_NR);
456 return -BCH_ERR_invalid_bkey;
459 return __bch2_inode_invalid(k, err);
462 int bch2_inode_v2_invalid(const struct bch_fs *c, struct bkey_s_c k,
463 enum bkey_invalid_flags flags,
464 struct printbuf *err)
466 struct bkey_s_c_inode_v2 inode = bkey_s_c_to_inode_v2(k);
468 if (INODEv2_STR_HASH(inode.v) >= BCH_STR_HASH_NR) {
469 prt_printf(err, "invalid str hash type (%llu >= %u)",
470 INODEv2_STR_HASH(inode.v), BCH_STR_HASH_NR);
471 return -BCH_ERR_invalid_bkey;
474 return __bch2_inode_invalid(k, err);
477 int bch2_inode_v3_invalid(const struct bch_fs *c, struct bkey_s_c k,
478 enum bkey_invalid_flags flags,
479 struct printbuf *err)
481 struct bkey_s_c_inode_v3 inode = bkey_s_c_to_inode_v3(k);
483 if (INODEv3_FIELDS_START(inode.v) < INODEv3_FIELDS_START_INITIAL ||
484 INODEv3_FIELDS_START(inode.v) > bkey_val_u64s(inode.k)) {
485 prt_printf(err, "invalid fields_start (got %llu, min %u max %zu)",
486 INODEv3_FIELDS_START(inode.v),
487 INODEv3_FIELDS_START_INITIAL,
488 bkey_val_u64s(inode.k));
489 return -BCH_ERR_invalid_bkey;
492 if (INODEv3_STR_HASH(inode.v) >= BCH_STR_HASH_NR) {
493 prt_printf(err, "invalid str hash type (%llu >= %u)",
494 INODEv3_STR_HASH(inode.v), BCH_STR_HASH_NR);
495 return -BCH_ERR_invalid_bkey;
498 return __bch2_inode_invalid(k, err);
501 static void __bch2_inode_unpacked_to_text(struct printbuf *out,
502 struct bch_inode_unpacked *inode)
504 prt_printf(out, "mode %o flags %x journal_seq %llu bi_size %llu bi_sectors %llu bi_version %llu",
505 inode->bi_mode, inode->bi_flags,
506 inode->bi_journal_seq,
511 #define x(_name, _bits) \
512 prt_printf(out, " "#_name " %llu", (u64) inode->_name);
513 BCH_INODE_FIELDS_v3()
517 void bch2_inode_unpacked_to_text(struct printbuf *out, struct bch_inode_unpacked *inode)
519 prt_printf(out, "inum: %llu ", inode->bi_inum);
520 __bch2_inode_unpacked_to_text(out, inode);
523 void bch2_inode_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c k)
525 struct bch_inode_unpacked inode;
527 if (bch2_inode_unpack(k, &inode)) {
528 prt_printf(out, "(unpack error)");
532 __bch2_inode_unpacked_to_text(out, &inode);
535 static inline u64 bkey_inode_flags(struct bkey_s_c k)
539 return le32_to_cpu(bkey_s_c_to_inode(k).v->bi_flags);
540 case KEY_TYPE_inode_v2:
541 return le64_to_cpu(bkey_s_c_to_inode_v2(k).v->bi_flags);
542 case KEY_TYPE_inode_v3:
543 return le64_to_cpu(bkey_s_c_to_inode_v3(k).v->bi_flags);
549 static inline bool bkey_is_deleted_inode(struct bkey_s_c k)
551 return bkey_inode_flags(k) & BCH_INODE_UNLINKED;
554 int bch2_trans_mark_inode(struct btree_trans *trans,
555 enum btree_id btree_id, unsigned level,
560 int nr = bkey_is_inode(&new->k) - bkey_is_inode(old.k);
561 bool old_deleted = bkey_is_deleted_inode(old);
562 bool new_deleted = bkey_is_deleted_inode(bkey_i_to_s_c(new));
565 int ret = bch2_replicas_deltas_realloc(trans, 0);
566 struct replicas_delta_list *d = trans->fs_usage_deltas;
574 if (old_deleted != new_deleted) {
575 int ret = bch2_btree_bit_mod(trans, BTREE_ID_deleted_inodes, new->k.p, new_deleted);
583 int bch2_mark_inode(struct btree_trans *trans,
584 enum btree_id btree_id, unsigned level,
585 struct bkey_s_c old, struct bkey_s_c new,
588 struct bch_fs *c = trans->c;
589 struct bch_fs_usage *fs_usage;
590 u64 journal_seq = trans->journal_res.seq;
592 if (flags & BTREE_TRIGGER_INSERT) {
593 struct bch_inode_v3 *v = (struct bch_inode_v3 *) new.v;
595 BUG_ON(!journal_seq);
596 BUG_ON(new.k->type != KEY_TYPE_inode_v3);
598 v->bi_journal_seq = cpu_to_le64(journal_seq);
601 if (flags & BTREE_TRIGGER_GC) {
602 percpu_down_read(&c->mark_lock);
605 fs_usage = fs_usage_ptr(c, journal_seq, flags & BTREE_TRIGGER_GC);
606 fs_usage->nr_inodes += bkey_is_inode(new.k);
607 fs_usage->nr_inodes -= bkey_is_inode(old.k);
610 percpu_up_read(&c->mark_lock);
615 int bch2_inode_generation_invalid(const struct bch_fs *c, struct bkey_s_c k,
616 enum bkey_invalid_flags flags,
617 struct printbuf *err)
620 prt_printf(err, "nonzero k.p.inode");
621 return -BCH_ERR_invalid_bkey;
627 void bch2_inode_generation_to_text(struct printbuf *out, struct bch_fs *c,
630 struct bkey_s_c_inode_generation gen = bkey_s_c_to_inode_generation(k);
632 prt_printf(out, "generation: %u", le32_to_cpu(gen.v->bi_generation));
635 void bch2_inode_init_early(struct bch_fs *c,
636 struct bch_inode_unpacked *inode_u)
638 enum bch_str_hash_type str_hash =
639 bch2_str_hash_opt_to_type(c, c->opts.str_hash);
641 memset(inode_u, 0, sizeof(*inode_u));
644 inode_u->bi_flags |= str_hash << INODE_STR_HASH_OFFSET;
645 get_random_bytes(&inode_u->bi_hash_seed,
646 sizeof(inode_u->bi_hash_seed));
649 void bch2_inode_init_late(struct bch_inode_unpacked *inode_u, u64 now,
650 uid_t uid, gid_t gid, umode_t mode, dev_t rdev,
651 struct bch_inode_unpacked *parent)
653 inode_u->bi_mode = mode;
654 inode_u->bi_uid = uid;
655 inode_u->bi_gid = gid;
656 inode_u->bi_dev = rdev;
657 inode_u->bi_atime = now;
658 inode_u->bi_mtime = now;
659 inode_u->bi_ctime = now;
660 inode_u->bi_otime = now;
662 if (parent && parent->bi_mode & S_ISGID) {
663 inode_u->bi_gid = parent->bi_gid;
665 inode_u->bi_mode |= S_ISGID;
669 #define x(_name, ...) inode_u->bi_##_name = parent->bi_##_name;
675 void bch2_inode_init(struct bch_fs *c, struct bch_inode_unpacked *inode_u,
676 uid_t uid, gid_t gid, umode_t mode, dev_t rdev,
677 struct bch_inode_unpacked *parent)
679 bch2_inode_init_early(c, inode_u);
680 bch2_inode_init_late(inode_u, bch2_current_time(c),
681 uid, gid, mode, rdev, parent);
684 static inline u32 bkey_generation(struct bkey_s_c k)
688 case KEY_TYPE_inode_v2:
690 case KEY_TYPE_inode_generation:
691 return le32_to_cpu(bkey_s_c_to_inode_generation(k).v->bi_generation);
698 * This just finds an empty slot:
700 int bch2_inode_create(struct btree_trans *trans,
701 struct btree_iter *iter,
702 struct bch_inode_unpacked *inode_u,
703 u32 snapshot, u64 cpu)
705 struct bch_fs *c = trans->c;
707 u64 min, max, start, pos, *hint;
709 unsigned bits = (c->opts.inodes_32bit ? 31 : 63);
711 if (c->opts.shard_inode_numbers) {
712 bits -= c->inode_shard_bits;
715 max = (cpu << bits) | ~(ULLONG_MAX << bits);
717 min = max_t(u64, min, BLOCKDEV_INODE_MAX);
718 hint = c->unused_inode_hints + cpu;
720 min = BLOCKDEV_INODE_MAX;
721 max = ~(ULLONG_MAX << bits);
722 hint = c->unused_inode_hints;
725 start = READ_ONCE(*hint);
727 if (start >= max || start < min)
731 bch2_trans_iter_init(trans, iter, BTREE_ID_inodes, POS(0, pos),
732 BTREE_ITER_ALL_SNAPSHOTS|
735 while ((k = bch2_btree_iter_peek(iter)).k &&
736 !(ret = bkey_err(k)) &&
737 bkey_lt(k.k->p, POS(0, max))) {
738 if (pos < iter->pos.offset)
742 * We don't need to iterate over keys in every snapshot once
743 * we've found just one:
745 pos = iter->pos.offset + 1;
746 bch2_btree_iter_set_pos(iter, POS(0, pos));
749 if (!ret && pos < max)
752 if (!ret && start == min)
753 ret = -BCH_ERR_ENOSPC_inode_create;
756 bch2_trans_iter_exit(trans, iter);
760 /* Retry from start */
762 bch2_btree_iter_set_pos(iter, POS(0, pos));
765 bch2_btree_iter_set_pos(iter, SPOS(0, pos, snapshot));
766 k = bch2_btree_iter_peek_slot(iter);
769 bch2_trans_iter_exit(trans, iter);
773 *hint = k.k->p.offset;
774 inode_u->bi_inum = k.k->p.offset;
775 inode_u->bi_generation = bkey_generation(k);
779 static int bch2_inode_delete_keys(struct btree_trans *trans,
780 subvol_inum inum, enum btree_id id)
782 struct btree_iter iter;
784 struct bkey_i delete;
785 struct bpos end = POS(inum.inum, U64_MAX);
790 * We're never going to be deleting partial extents, no need to use an
793 bch2_trans_iter_init(trans, &iter, id, POS(inum.inum, 0),
797 bch2_trans_begin(trans);
799 ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot);
803 bch2_btree_iter_set_snapshot(&iter, snapshot);
805 k = bch2_btree_iter_peek_upto(&iter, end);
813 bkey_init(&delete.k);
814 delete.k.p = iter.pos;
816 if (iter.flags & BTREE_ITER_IS_EXTENTS)
817 bch2_key_resize(&delete.k,
818 bpos_min(end, k.k->p).offset -
821 ret = bch2_trans_update(trans, &iter, &delete, 0) ?:
822 bch2_trans_commit(trans, NULL, NULL,
823 BTREE_INSERT_NOFAIL);
825 if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
829 bch2_trans_iter_exit(trans, &iter);
833 int bch2_inode_rm(struct bch_fs *c, subvol_inum inum)
835 struct btree_trans *trans = bch2_trans_get(c);
836 struct btree_iter iter = { NULL };
837 struct bkey_i_inode_generation delete;
838 struct bch_inode_unpacked inode_u;
844 * If this was a directory, there shouldn't be any real dirents left -
845 * but there could be whiteouts (from hash collisions) that we should
848 * XXX: the dirent could ideally would delete whiteouts when they're no
851 ret = bch2_inode_delete_keys(trans, inum, BTREE_ID_extents) ?:
852 bch2_inode_delete_keys(trans, inum, BTREE_ID_xattrs) ?:
853 bch2_inode_delete_keys(trans, inum, BTREE_ID_dirents);
857 bch2_trans_begin(trans);
859 ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot);
863 k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes,
864 SPOS(0, inum.inum, snapshot),
865 BTREE_ITER_INTENT|BTREE_ITER_CACHED);
870 if (!bkey_is_inode(k.k)) {
871 bch2_fs_inconsistent(c,
872 "inode %llu:%u not found when deleting",
873 inum.inum, snapshot);
878 bch2_inode_unpack(k, &inode_u);
880 bkey_inode_generation_init(&delete.k_i);
881 delete.k.p = iter.pos;
882 delete.v.bi_generation = cpu_to_le32(inode_u.bi_generation + 1);
884 ret = bch2_trans_update(trans, &iter, &delete.k_i, 0) ?:
885 bch2_trans_commit(trans, NULL, NULL,
886 BTREE_INSERT_NOFAIL);
888 bch2_trans_iter_exit(trans, &iter);
889 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
892 bch2_trans_put(trans);
896 int bch2_inode_find_by_inum_nowarn_trans(struct btree_trans *trans,
898 struct bch_inode_unpacked *inode)
900 struct btree_iter iter;
903 ret = bch2_inode_peek_nowarn(trans, &iter, inode, inum, 0);
905 bch2_trans_iter_exit(trans, &iter);
909 int bch2_inode_find_by_inum_trans(struct btree_trans *trans,
911 struct bch_inode_unpacked *inode)
913 struct btree_iter iter;
916 ret = bch2_inode_peek(trans, &iter, inode, inum, 0);
918 bch2_trans_iter_exit(trans, &iter);
922 int bch2_inode_find_by_inum(struct bch_fs *c, subvol_inum inum,
923 struct bch_inode_unpacked *inode)
925 return bch2_trans_do(c, NULL, NULL, 0,
926 bch2_inode_find_by_inum_trans(trans, inum, inode));
929 int bch2_inode_nlink_inc(struct bch_inode_unpacked *bi)
931 if (bi->bi_flags & BCH_INODE_UNLINKED)
932 bi->bi_flags &= ~BCH_INODE_UNLINKED;
934 if (bi->bi_nlink == U32_MAX)
943 void bch2_inode_nlink_dec(struct btree_trans *trans, struct bch_inode_unpacked *bi)
945 if (bi->bi_nlink && (bi->bi_flags & BCH_INODE_UNLINKED)) {
946 bch2_trans_inconsistent(trans, "inode %llu unlinked but link count nonzero",
951 if (bi->bi_flags & BCH_INODE_UNLINKED) {
952 bch2_trans_inconsistent(trans, "inode %llu link count underflow", bi->bi_inum);
959 bi->bi_flags |= BCH_INODE_UNLINKED;
962 struct bch_opts bch2_inode_opts_to_opts(struct bch_inode_unpacked *inode)
964 struct bch_opts ret = { 0 };
965 #define x(_name, _bits) \
966 if (inode->bi_##_name) \
967 opt_set(ret, _name, inode->bi_##_name - 1);
973 void bch2_inode_opts_get(struct bch_io_opts *opts, struct bch_fs *c,
974 struct bch_inode_unpacked *inode)
976 #define x(_name, _bits) opts->_name = inode_opt_get(c, inode, _name);
981 opts->compression = opts->background_compression = opts->data_checksum = opts->erasure_code = 0;
984 int bch2_inum_opts_get(struct btree_trans *trans, subvol_inum inum, struct bch_io_opts *opts)
986 struct bch_inode_unpacked inode;
987 int ret = lockrestart_do(trans, bch2_inode_find_by_inum_trans(trans, inum, &inode));
992 bch2_inode_opts_get(opts, trans->c, &inode);
996 int bch2_inode_rm_snapshot(struct btree_trans *trans, u64 inum, u32 snapshot)
998 struct bch_fs *c = trans->c;
999 struct btree_iter iter = { NULL };
1000 struct bkey_i_inode_generation delete;
1001 struct bch_inode_unpacked inode_u;
1006 ret = bch2_btree_delete_range_trans(trans, BTREE_ID_extents,
1007 SPOS(inum, 0, snapshot),
1008 SPOS(inum, U64_MAX, snapshot),
1010 bch2_btree_delete_range_trans(trans, BTREE_ID_dirents,
1011 SPOS(inum, 0, snapshot),
1012 SPOS(inum, U64_MAX, snapshot),
1014 bch2_btree_delete_range_trans(trans, BTREE_ID_xattrs,
1015 SPOS(inum, 0, snapshot),
1016 SPOS(inum, U64_MAX, snapshot),
1018 } while (ret == -BCH_ERR_transaction_restart_nested);
1022 bch2_trans_begin(trans);
1024 k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes,
1025 SPOS(0, inum, snapshot), BTREE_ITER_INTENT);
1030 if (!bkey_is_inode(k.k)) {
1031 bch2_fs_inconsistent(c,
1032 "inode %llu:%u not found when deleting",
1038 bch2_inode_unpack(k, &inode_u);
1040 /* Subvolume root? */
1041 if (inode_u.bi_subvol)
1042 bch_warn(c, "deleting inode %llu marked as unlinked, but also a subvolume root!?", inode_u.bi_inum);
1044 bkey_inode_generation_init(&delete.k_i);
1045 delete.k.p = iter.pos;
1046 delete.v.bi_generation = cpu_to_le32(inode_u.bi_generation + 1);
1048 ret = bch2_trans_update(trans, &iter, &delete.k_i, 0) ?:
1049 bch2_trans_commit(trans, NULL, NULL,
1050 BTREE_INSERT_NOFAIL);
1052 bch2_trans_iter_exit(trans, &iter);
1053 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
1056 return ret ?: -BCH_ERR_transaction_restart_nested;
1059 static int may_delete_deleted_inode(struct btree_trans *trans, struct bpos pos)
1061 struct bch_fs *c = trans->c;
1062 struct btree_iter iter;
1064 struct bch_inode_unpacked inode;
1067 if (bch2_snapshot_is_internal_node(c, pos.snapshot))
1070 if (!fsck_err_on(c->sb.clean, c,
1071 "filesystem marked as clean but have deleted inode %llu:%u",
1072 pos.offset, pos.snapshot))
1075 k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes, pos, BTREE_ITER_CACHED);
1080 ret = bkey_is_inode(k.k) ? 0 : -BCH_ERR_ENOENT_inode;
1081 if (fsck_err_on(!bkey_is_inode(k.k), c,
1082 "nonexistent inode %llu:%u in deleted_inodes btree",
1083 pos.offset, pos.snapshot))
1086 ret = bch2_inode_unpack(k, &inode);
1090 if (fsck_err_on(S_ISDIR(inode.bi_mode), c,
1091 "directory %llu:%u in deleted_inodes btree",
1092 pos.offset, pos.snapshot))
1095 if (fsck_err_on(!(inode.bi_flags & BCH_INODE_UNLINKED), c,
1096 "non-deleted inode %llu:%u in deleted_inodes btree",
1097 pos.offset, pos.snapshot))
1105 return bch2_btree_bit_mod(trans, BTREE_ID_deleted_inodes, pos, false);
1108 int bch2_delete_dead_inodes(struct bch_fs *c)
1110 struct btree_trans *trans = bch2_trans_get(c);
1111 struct btree_iter iter;
1115 ret = bch2_btree_write_buffer_flush_sync(trans);
1120 * Weird transaction restart handling here because on successful delete,
1121 * bch2_inode_rm_snapshot() will return a nested transaction restart,
1122 * but we can't retry because the btree write buffer won't have been
1123 * flushed and we'd spin:
1125 for_each_btree_key(trans, iter, BTREE_ID_deleted_inodes, POS_MIN,
1126 BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
1127 ret = lockrestart_do(trans, may_delete_deleted_inode(trans, k.k->p));
1132 if (!test_bit(BCH_FS_RW, &c->flags)) {
1133 bch2_trans_unlock(trans);
1137 ret = bch2_inode_rm_snapshot(trans, k.k->p.offset, k.k->p.snapshot);
1138 if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
1142 bch2_trans_iter_exit(trans, &iter);
1144 bch2_trans_put(trans);