Update bcachefs sources to 42284b8b2b bcachefs: Fix race between btree updates & journal replay

.bcachefs_revision

@@ -1 +1 @@
-9d28e4a535cac3b88104a4525dd39fc16175f38c
+42284b8b2bb980c80140b640de7cb12bc1e4541c

libbcachefs/bcachefs.h

@@ -561,6 +561,7 @@ struct journal_keys {
 		enum btree_id	btree_id:8;
 		unsigned	level:8;
 		bool		allocated;
+		bool		overwritten;
 		struct bkey_i	*k;
 		u32		journal_seq;
 		u32		journal_offset;

libbcachefs/btree_update_leaf.c

@@ -15,6 +15,7 @@
 #include "journal.h"
 #include "journal_reclaim.h"
 #include "keylist.h"
+#include "recovery.h"
 #include "subvolume.h"
 #include "replicas.h"
 
@@ -624,6 +625,14 @@ fail:
 	return btree_trans_restart(trans);
 }
 
+static noinline void bch2_drop_overwrites_from_journal(struct btree_trans *trans)
+{
+	struct btree_insert_entry *i;
+
+	trans_for_each_update(trans, i)
+		bch2_journal_key_overwritten(trans->c, i->btree_id, i->level, i->k->k.p);
+}
+
 /*
  * Get journal reservation, take write locks, and attempt to do btree update(s):
  */
@@ -701,6 +710,9 @@ static inline int do_bch2_trans_commit(struct btree_trans *trans,
 
 	ret = bch2_trans_commit_write_locked(trans, stopped_at, trace_ip);
 
+	if (!ret && unlikely(!test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags)))
+		bch2_drop_overwrites_from_journal(trans);
+
 	trans_for_each_update(trans, i)
 		if (!same_leaf_as_prev(trans, i))
 			bch2_btree_node_unlock_write_inlined(trans, i->path,

libbcachefs/buckets.c

@@ -531,20 +531,6 @@ void bch2_mark_alloc_bucket(struct bch_fs *c, struct bch_dev *ca,
 	BUG_ON(owned_by_allocator == old.owned_by_allocator);
 }
 
-static inline u8 bkey_alloc_gen(struct bkey_s_c k)
-{
-	switch (k.k->type) {
-	case KEY_TYPE_alloc:
-		return bkey_s_c_to_alloc(k).v->gen;
-	case KEY_TYPE_alloc_v2:
-		return bkey_s_c_to_alloc_v2(k).v->gen;
-	case KEY_TYPE_alloc_v3:
-		return bkey_s_c_to_alloc_v3(k).v->gen;
-	default:
-		return 0;
-	}
-}
-
 static int bch2_mark_alloc(struct btree_trans *trans,
 			   struct bkey_s_c old, struct bkey_s_c new,
 			   unsigned flags)
@@ -589,7 +575,7 @@ static int bch2_mark_alloc(struct btree_trans *trans,
 		return 0;
 
 	percpu_down_read(&c->mark_lock);
-	if (!gc && new_u.gen != bkey_alloc_gen(old))
+	if (!gc && new_u.gen != old_u.gen)
 		*bucket_gen(ca, new.k->p.offset) = new_u.gen;
 
 	g = __bucket(ca, new.k->p.offset, gc);

libbcachefs/journal_io.c

@@ -558,20 +558,20 @@ static void journal_entry_dev_usage_to_text(struct printbuf *out, struct bch_fs
 		container_of(entry, struct jset_entry_dev_usage, entry);
 	unsigned i, nr_types = jset_entry_dev_usage_nr_types(u);
 
-	pr_buf(out, "dev=%u ", le32_to_cpu(u->dev));
+	pr_buf(out, "dev=%u", le32_to_cpu(u->dev));
 
 	for (i = 0; i < nr_types; i++) {
 		if (i < BCH_DATA_NR)
-			pr_buf(out, "%s", bch2_data_types[i]);
+			pr_buf(out, " %s", bch2_data_types[i]);
 		else
-			pr_buf(out, "(unknown data type %u)", i);
+			pr_buf(out, " (unknown data type %u)", i);
 		pr_buf(out, ": buckets=%llu sectors=%llu fragmented=%llu",
 		       le64_to_cpu(u->d[i].buckets),
 		       le64_to_cpu(u->d[i].sectors),
 		       le64_to_cpu(u->d[i].fragmented));
 	}
 
-	pr_buf(out, "buckets_ec: %llu buckets_unavailable: %llu",
+	pr_buf(out, " buckets_ec: %llu buckets_unavailable: %llu",
 	       le64_to_cpu(u->buckets_ec),
 	       le64_to_cpu(u->buckets_unavailable));
 }

libbcachefs/recovery.c

@@ -185,6 +185,19 @@ int bch2_journal_key_delete(struct bch_fs *c, enum btree_id id,
 	return bch2_journal_key_insert(c, id, level, &whiteout);
 }
 
+void bch2_journal_key_overwritten(struct bch_fs *c, enum btree_id btree,
+				  unsigned level, struct bpos pos)
+{
+	struct journal_keys *keys = &c->journal_keys;
+	size_t idx = journal_key_search(keys, btree, level, pos);
+
+	if (idx < keys->nr &&
+	    keys->d[idx].btree_id	== btree &&
+	    keys->d[idx].level		== level &&
+	    !bpos_cmp(keys->d[idx].k->k.p, pos))
+		keys->d[idx].overwritten = true;
+}
+
 static struct bkey_i *bch2_journal_iter_peek(struct journal_iter *iter)
 {
 	struct journal_key *k = iter->idx - iter->keys->nr
@@ -533,6 +546,10 @@ static int __bch2_journal_replay_key(struct btree_trans *trans,
 		BTREE_ITER_NOT_EXTENTS;
 	int ret;
 
+	/* Must be checked with btree locked: */
+	if (k->overwritten)
+		return 0;
+
 	if (!k->level && k->btree_id == BTREE_ID_alloc)
 		iter_flags |= BTREE_ITER_CACHED|BTREE_ITER_CACHED_NOFILL;
 

libbcachefs/recovery.h

@@ -37,6 +37,8 @@ int bch2_journal_key_insert(struct bch_fs *, enum btree_id,
 			    unsigned, struct bkey_i *);
 int bch2_journal_key_delete(struct bch_fs *, enum btree_id,
 			    unsigned, struct bpos);
+void bch2_journal_key_overwritten(struct bch_fs *, enum btree_id,
+				  unsigned, struct bpos);
 
 void bch2_btree_and_journal_iter_advance(struct btree_and_journal_iter *);
 struct bkey_s_c bch2_btree_and_journal_iter_peek(struct btree_and_journal_iter *);