[PATCH] bcachefs: Relocate rcu_read_lock()/rcu_read_unlock() to snapshot_t()

Ahmed Ehab posted 1 patch 2 months, 1 week ago
fs/bcachefs/snapshot.h | 26 +++++++-------------------
1 file changed, 7 insertions(+), 19 deletions(-)
[PATCH] bcachefs: Relocate rcu_read_lock()/rcu_read_unlock() to snapshot_t()
Posted by Ahmed Ehab 2 months, 1 week ago
Syzbot reports a problem that a warning is triggered due to suspicious
use of rcu_dereference_check(). That is triggered by a call of
bch2_snapshot_tree_oldest_subvol().

The cause of the warning is that the rcu_read_lock() is called in wrapper
methods instead of calling it directly before calling rcu_dereference()
in snapshot_t().For example in this case, snapshot_t() is called
directly from bch2_snapshot_tree_oldest_subvol() without holding the
read lock. This also results in duplicating the rcu_read_lock()
and rcu_read_unlock() calls, which may lead to future errors in the case
of forgetting to hold the read locks as in this case.

To fix this, move rcu_read_lock() and rcu_read_unlock() to snapshot_t().
This will make sure that rcu_dereference_check() is never called without
holding the read lock.

Reported-by: <syzbot+f7c41a878676b72c16a6@syzkaller.appspotmail.com>
Signed-off-by: Ahmed Ehab <bottaawesome633@gmail.com>
---
 fs/bcachefs/snapshot.h | 26 +++++++-------------------
 1 file changed, 7 insertions(+), 19 deletions(-)

diff --git a/fs/bcachefs/snapshot.h b/fs/bcachefs/snapshot.h
index eb5ef64221d6..04f18fac56fe 100644
--- a/fs/bcachefs/snapshot.h
+++ b/fs/bcachefs/snapshot.h
@@ -42,15 +42,19 @@ static inline struct snapshot_t *__snapshot_t(struct snapshot_table *t, u32 id)
 
 static inline const struct snapshot_t *snapshot_t(struct bch_fs *c, u32 id)
 {
-	return __snapshot_t(rcu_dereference(c->snapshots), id);
+	struct snapshot_table *temp;
+
+	rcu_read_lock();
+	temp = rcu_dereference(c->snapshots);
+	rcu_read_unlock();
+
+	return __snapshot_t(temp, id);
 }
 
 static inline u32 bch2_snapshot_tree(struct bch_fs *c, u32 id)
 {
-	rcu_read_lock();
 	const struct snapshot_t *s = snapshot_t(c, id);
 	id = s ? s->tree : 0;
-	rcu_read_unlock();
 
 	return id;
 }
@@ -63,9 +67,7 @@ static inline u32 __bch2_snapshot_parent_early(struct bch_fs *c, u32 id)
 
 static inline u32 bch2_snapshot_parent_early(struct bch_fs *c, u32 id)
 {
-	rcu_read_lock();
 	id = __bch2_snapshot_parent_early(c, id);
-	rcu_read_unlock();
 
 	return id;
 }
@@ -89,19 +91,15 @@ static inline u32 __bch2_snapshot_parent(struct bch_fs *c, u32 id)
 
 static inline u32 bch2_snapshot_parent(struct bch_fs *c, u32 id)
 {
-	rcu_read_lock();
 	id = __bch2_snapshot_parent(c, id);
-	rcu_read_unlock();
 
 	return id;
 }
 
 static inline u32 bch2_snapshot_nth_parent(struct bch_fs *c, u32 id, u32 n)
 {
-	rcu_read_lock();
 	while (n--)
 		id = __bch2_snapshot_parent(c, id);
-	rcu_read_unlock();
 
 	return id;
 }
@@ -112,10 +110,8 @@ static inline u32 bch2_snapshot_root(struct bch_fs *c, u32 id)
 {
 	u32 parent;
 
-	rcu_read_lock();
 	while ((parent = __bch2_snapshot_parent(c, id)))
 		id = parent;
-	rcu_read_unlock();
 
 	return id;
 }
@@ -128,19 +124,15 @@ static inline u32 __bch2_snapshot_equiv(struct bch_fs *c, u32 id)
 
 static inline u32 bch2_snapshot_equiv(struct bch_fs *c, u32 id)
 {
-	rcu_read_lock();
 	id = __bch2_snapshot_equiv(c, id);
-	rcu_read_unlock();
 
 	return id;
 }
 
 static inline int bch2_snapshot_is_internal_node(struct bch_fs *c, u32 id)
 {
-	rcu_read_lock();
 	const struct snapshot_t *s = snapshot_t(c, id);
 	int ret = s ? s->children[0] : -BCH_ERR_invalid_snapshot_node;
-	rcu_read_unlock();
 
 	return ret;
 }
@@ -157,9 +149,7 @@ static inline u32 bch2_snapshot_depth(struct bch_fs *c, u32 parent)
 {
 	u32 depth;
 
-	rcu_read_lock();
 	depth = parent ? snapshot_t(c, parent)->depth + 1 : 0;
-	rcu_read_unlock();
 
 	return depth;
 }
@@ -175,10 +165,8 @@ static inline bool bch2_snapshot_is_ancestor(struct bch_fs *c, u32 id, u32 ances
 
 static inline bool bch2_snapshot_has_children(struct bch_fs *c, u32 id)
 {
-	rcu_read_lock();
 	const struct snapshot_t *t = snapshot_t(c, id);
 	bool ret = t && (t->children[0]|t->children[1]) != 0;
-	rcu_read_unlock();
 
 	return ret;
 }
-- 
2.46.0
Re: [PATCH] bcachefs: Relocate rcu_read_lock()/rcu_read_unlock() to snapshot_t()
Posted by Kent Overstreet 2 months, 1 week ago
On Thu, Sep 19, 2024 at 10:36:03AM GMT, Ahmed Ehab wrote:
> Syzbot reports a problem that a warning is triggered due to suspicious
> use of rcu_dereference_check(). That is triggered by a call of
> bch2_snapshot_tree_oldest_subvol().
> 
> The cause of the warning is that the rcu_read_lock() is called in wrapper
> methods instead of calling it directly before calling rcu_dereference()
> in snapshot_t().For example in this case, snapshot_t() is called
> directly from bch2_snapshot_tree_oldest_subvol() without holding the
> read lock. This also results in duplicating the rcu_read_lock()
> and rcu_read_unlock() calls, which may lead to future errors in the case
> of forgetting to hold the read locks as in this case.
> 
> To fix this, move rcu_read_lock() and rcu_read_unlock() to snapshot_t().
> This will make sure that rcu_dereference_check() is never called without
> holding the read lock.
> 
> Reported-by: <syzbot+f7c41a878676b72c16a6@syzkaller.appspotmail.com>
> Signed-off-by: Ahmed Ehab <bottaawesome633@gmail.com>
> ---
>  fs/bcachefs/snapshot.h | 26 +++++++-------------------
>  1 file changed, 7 insertions(+), 19 deletions(-)
> 
> diff --git a/fs/bcachefs/snapshot.h b/fs/bcachefs/snapshot.h
> index eb5ef64221d6..04f18fac56fe 100644
> --- a/fs/bcachefs/snapshot.h
> +++ b/fs/bcachefs/snapshot.h
> @@ -42,15 +42,19 @@ static inline struct snapshot_t *__snapshot_t(struct snapshot_table *t, u32 id)
>  
>  static inline const struct snapshot_t *snapshot_t(struct bch_fs *c, u32 id)
>  {
> -	return __snapshot_t(rcu_dereference(c->snapshots), id);
> +	struct snapshot_table *temp;
> +
> +	rcu_read_lock();
> +	temp = rcu_dereference(c->snapshots);
> +	rcu_read_unlock();
> +
> +	return __snapshot_t(temp, id);

This is very wrong - as in, you need to study up on how RCU works.

We need to be holding rcu_read_lock() while we're accessing the object
we got to from the rcu pointer, so rcu_read_lock() always needs to be
taken by the caller.