bcachefs: Debug asserts for ca->ref

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

fs/bcachefs/bcachefs.h

@@ -537,7 +537,13 @@ struct io_count {
 
 struct bch_dev {
 	struct kobject		kobj;
+#ifdef CONFIG_BCACHEFS_DEBUG
+	atomic_long_t		ref;
+	bool			dying;
+	unsigned long		last_put;
+#else
 	struct percpu_ref	ref;
+#endif
 	struct completion	ref_completion;
 	struct percpu_ref	io_ref;
 	struct completion	io_ref_completion;

fs/bcachefs/sb-members.h

@@ -107,12 +107,25 @@ static inline struct bch_dev *__bch2_next_dev(struct bch_fs *c, struct bch_dev *
 
 static inline void bch2_dev_get(struct bch_dev *ca)
 {
+#ifdef CONFIG_BCACHEFS_DEBUG
+	BUG_ON(atomic_long_inc_return(&ca->ref) <= 1L);
+#else
 	percpu_ref_get(&ca->ref);
+#endif
 }
 
 static inline void __bch2_dev_put(struct bch_dev *ca)
 {
+#ifdef CONFIG_BCACHEFS_DEBUG
+	long r = atomic_long_dec_return(&ca->ref);
+	if (r < (long) !ca->dying)
+		panic("bch_dev->ref underflow, last put: %pS\n", (void *) ca->last_put);
+	ca->last_put = _THIS_IP_;
+	if (!r)
+		complete(&ca->ref_completion);
+#else
 	percpu_ref_put(&ca->ref);
+#endif
 }
 
 static inline void bch2_dev_put(struct bch_dev *ca)

fs/bcachefs/super.c

@@ -656,6 +656,7 @@ void bch2_fs_free(struct bch_fs *c)
 		struct bch_dev *ca = rcu_dereference_protected(c->devs[i], true);
 
 		if (ca) {
+			EBUG_ON(atomic_long_read(&ca->ref) != 1);
 			bch2_free_super(&ca->disk_sb);
 			bch2_dev_free(ca);
 		}
@@ -1200,7 +1201,9 @@ static void bch2_dev_free(struct bch_dev *ca)
 	bch2_time_stats_quantiles_exit(&ca->io_latency[READ]);
 
 	percpu_ref_exit(&ca->io_ref);
+#ifndef CONFIG_BCACHEFS_DEBUG
 	percpu_ref_exit(&ca->ref);
+#endif
 	kobject_put(&ca->kobj);
 }
 
@@ -1227,12 +1230,14 @@ static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
 	bch2_dev_journal_exit(ca);
 }
 
+#ifndef CONFIG_BCACHEFS_DEBUG
 static void bch2_dev_ref_complete(struct percpu_ref *ref)
 {
 	struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
 
 	complete(&ca->ref_completion);
 }
+#endif
 
 static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
 {
@@ -1301,9 +1306,14 @@ static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
 	ca->nr_btree_reserve = DIV_ROUND_UP(BTREE_NODE_RESERVE,
 			     ca->mi.bucket_size / btree_sectors(c));
 
-	if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
-			    0, GFP_KERNEL) ||
-	    percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
+#ifndef CONFIG_BCACHEFS_DEBUG
+	if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete, 0, GFP_KERNEL))
+		goto err;
+#else
+	atomic_long_set(&ca->ref, 1);
+#endif
+
+	if (percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
 			    PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
 	    !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) ||
 	    bch2_dev_buckets_alloc(c, ca) ||
@@ -1665,7 +1675,12 @@ int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
 	rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
 	mutex_unlock(&c->sb_lock);
 
+#ifndef CONFIG_BCACHEFS_DEBUG
 	percpu_ref_kill(&ca->ref);
+#else
+	ca->dying = true;
+	bch2_dev_put(ca);
+#endif
 	wait_for_completion(&ca->ref_completion);
 
 	bch2_dev_free(ca);