bcachefs: btree node scan: fall back to comparing by journal seq

highly damaged filesystems, or filesystems that have been damaged and
repair and damaged again, may have sequence numbers we can't fully trust
- which in itself is something we need to debug.

Add a journal_seq fallback so that repair doesn't get stuck.

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
This commit is contained in:
Kent Overstreet 2024-07-12 14:16:01 -04:00
parent 375476c414
commit 6ec8623f7c
3 changed files with 41 additions and 15 deletions

View File

@ -1006,6 +1006,7 @@ int bch2_btree_node_read_done(struct bch_fs *c, struct bch_dev *ca,
BTREE_PTR_RANGE_UPDATED(&bkey_i_to_btree_ptr_v2(&b->key)->v);
unsigned u64s;
unsigned ptr_written = btree_ptr_sectors_written(bkey_i_to_s_c(&b->key));
u64 max_journal_seq = 0;
struct printbuf buf = PRINTBUF;
int ret = 0, retry_read = 0, write = READ;
u64 start_time = local_clock();
@ -1181,6 +1182,8 @@ int bch2_btree_node_read_done(struct bch_fs *c, struct bch_dev *ca,
sort_iter_add(iter,
vstruct_idx(i, 0),
vstruct_last(i));
max_journal_seq = max(max_journal_seq, le64_to_cpu(i->journal_seq));
}
if (ptr_written) {
@ -1217,6 +1220,7 @@ int bch2_btree_node_read_done(struct bch_fs *c, struct bch_dev *ca,
swap(sorted, b->data);
set_btree_bset(b, b->set, &b->data->keys);
b->nsets = 1;
b->data->keys.journal_seq = cpu_to_le64(max_journal_seq);
BUG_ON(b->nr.live_u64s != u64s);

View File

@ -22,7 +22,9 @@ struct find_btree_nodes_worker {
static void found_btree_node_to_text(struct printbuf *out, struct bch_fs *c, const struct found_btree_node *n)
{
prt_printf(out, "%s l=%u seq=%u cookie=%llx ", bch2_btree_id_str(n->btree_id), n->level, n->seq, n->cookie);
prt_printf(out, "%s l=%u seq=%u journal_seq=%llu cookie=%llx ",
bch2_btree_id_str(n->btree_id), n->level, n->seq,
n->journal_seq, n->cookie);
bch2_bpos_to_text(out, n->min_key);
prt_str(out, "-");
bch2_bpos_to_text(out, n->max_key);
@ -63,19 +65,37 @@ static void found_btree_node_to_key(struct bkey_i *k, const struct found_btree_n
memcpy(bp->v.start, f->ptrs, sizeof(struct bch_extent_ptr) * f->nr_ptrs);
}
static inline u64 bkey_journal_seq(struct bkey_s_c k)
{
switch (k.k->type) {
case KEY_TYPE_inode_v3:
return le64_to_cpu(bkey_s_c_to_inode_v3(k).v->bi_journal_seq);
default:
return 0;
}
}
static bool found_btree_node_is_readable(struct btree_trans *trans,
struct found_btree_node *f)
{
struct { __BKEY_PADDED(k, BKEY_BTREE_PTR_VAL_U64s_MAX); } k;
struct { __BKEY_PADDED(k, BKEY_BTREE_PTR_VAL_U64s_MAX); } tmp;
found_btree_node_to_key(&k.k, f);
found_btree_node_to_key(&tmp.k, f);
struct btree *b = bch2_btree_node_get_noiter(trans, &k.k, f->btree_id, f->level, false);
struct btree *b = bch2_btree_node_get_noiter(trans, &tmp.k, f->btree_id, f->level, false);
bool ret = !IS_ERR_OR_NULL(b);
if (!ret)
return ret;
f->sectors_written = b->written;
f->journal_seq = le64_to_cpu(b->data->keys.journal_seq);
struct bkey_s_c k;
struct bkey unpacked;
struct btree_node_iter iter;
for_each_btree_node_key_unpack(b, k, &iter, &unpacked)
f->journal_seq = max(f->journal_seq, bkey_journal_seq(k));
six_unlock_read(&b->c.lock);
/*
@ -84,7 +104,7 @@ static bool found_btree_node_is_readable(struct btree_trans *trans,
* this node
*/
if (b != btree_node_root(trans->c, b))
bch2_btree_node_evict(trans, &k.k);
bch2_btree_node_evict(trans, &tmp.k);
return ret;
}
@ -105,7 +125,8 @@ static int found_btree_node_cmp_cookie(const void *_l, const void *_r)
static int found_btree_node_cmp_time(const struct found_btree_node *l,
const struct found_btree_node *r)
{
return cmp_int(l->seq, r->seq);
return cmp_int(l->seq, r->seq) ?:
cmp_int(l->journal_seq, r->journal_seq);
}
static int found_btree_node_cmp_pos(const void *_l, const void *_r)
@ -309,15 +330,15 @@ again:
} else if (n->level) {
n->overwritten = true;
} else {
struct printbuf buf = PRINTBUF;
prt_str(&buf, "overlapping btree nodes with same seq! halting\n ");
found_btree_node_to_text(&buf, c, start);
prt_str(&buf, "\n ");
found_btree_node_to_text(&buf, c, n);
bch_err(c, "%s", buf.buf);
printbuf_exit(&buf);
return -BCH_ERR_fsck_repair_unimplemented;
if (bpos_cmp(start->max_key, n->max_key) >= 0)
n->overwritten = true;
else {
n->range_updated = true;
n->min_key = bpos_successor(start->max_key);
n->range_updated = true;
bubble_up(n, end);
goto again;
}
}
}

View File

@ -11,6 +11,7 @@ struct found_btree_node {
u8 level;
unsigned sectors_written;
u32 seq;
u64 journal_seq;
u64 cookie;
struct bpos min_key;