When hfs_bnode_create() finds that a node is already hashed (which should
not happen in normal operation), it currently returns the existing node
without incrementing its reference count. This causes a reference count
inconsistency that leads to a kernel panic when the node is later freed
in hfs_bnode_put():

    kernel BUG at fs/hfsplus/bnode.c:676!
    BUG_ON(!atomic_read(&node->refcnt))

This scenario can occur when hfs_bmap_alloc() attempts to allocate a node
that is already in use (e.g., when node 0's bitmap bit is incorrectly
unset), or due to filesystem corruption.

Returning an existing node from a create path is not normal operation.

Fix this by returning ERR_PTR(-EEXIST) instead of the node when it's
already hashed. This properly signals the error condition to callers,
which already check for IS_ERR() return values.

Reported-by: syzbot+1c8ff72d0cd8a50dfeaa@syzkaller.appspotmail.com
Link: https://syzkaller.appspot.com/bug?extid=1c8ff72d0cd8a50dfeaa
Link: https://lore.kernel.org/all/784415834694f39902088fa8946850fc1779a318.camel@ibm.com/
Fixes: 634725a92938 ("[PATCH] hfs: cleanup HFS+ prints")
Signed-off-by: Shardul Bankar <shardul.b@mpiricsoftware.com>
---
v3 changes:
  - This is posted standalone as discussed in the v2 thread.
v2 changes:
  - Implement Slava's suggestion: return ERR_PTR(-EEXIST) for already-hashed nodes.
  - Keep the node-0 allocation guard as a minimal, targeted hardening measure.

 fs/hfsplus/bnode.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/fs/hfsplus/bnode.c b/fs/hfsplus/bnode.c
index 191661af9677..250a226336ea 100644
--- a/fs/hfsplus/bnode.c
+++ b/fs/hfsplus/bnode.c
@@ -629,7 +629,7 @@ struct hfs_bnode *hfs_bnode_create(struct hfs_btree *tree, u32 num)
 	if (node) {
 		pr_crit("new node %u already hashed?\n", num);
 		WARN_ON(1);
-		return node;
+		return ERR_PTR(-EEXIST);
 	}
 	node = __hfs_bnode_create(tree, num);
 	if (!node)
-- 
2.34.1

On Tue, 2025-12-30 at 02:19 +0530, Shardul Bankar wrote:
> When hfs_bnode_create() finds that a node is already hashed (which should
> not happen in normal operation), it currently returns the existing node
> without incrementing its reference count. This causes a reference count
> inconsistency that leads to a kernel panic when the node is later freed
> in hfs_bnode_put():
> 
>     kernel BUG at fs/hfsplus/bnode.c:676!
>     BUG_ON(!atomic_read(&node->refcnt))
> 
> This scenario can occur when hfs_bmap_alloc() attempts to allocate a node
> that is already in use (e.g., when node 0's bitmap bit is incorrectly
> unset), or due to filesystem corruption.
> 
> Returning an existing node from a create path is not normal operation.
> 
> Fix this by returning ERR_PTR(-EEXIST) instead of the node when it's
> already hashed. This properly signals the error condition to callers,
> which already check for IS_ERR() return values.
> 
> Reported-by: syzbot+1c8ff72d0cd8a50dfeaa@syzkaller.appspotmail.com
> Link: https://syzkaller.appspot.com/bug?extid=1c8ff72d0cd8a50dfeaa  
> Link: https://lore.kernel.org/all/784415834694f39902088fa8946850fc1779a318.camel@ibm.com/  
> Fixes: 634725a92938 ("[PATCH] hfs: cleanup HFS+ prints")
> Signed-off-by: Shardul Bankar <shardul.b@mpiricsoftware.com>
> ---
> v3 changes:
>   - This is posted standalone as discussed in the v2 thread.
> v2 changes:
>   - Implement Slava's suggestion: return ERR_PTR(-EEXIST) for already-hashed nodes.
>   - Keep the node-0 allocation guard as a minimal, targeted hardening measure.
> 
>  fs/hfsplus/bnode.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/fs/hfsplus/bnode.c b/fs/hfsplus/bnode.c
> index 191661af9677..250a226336ea 100644
> --- a/fs/hfsplus/bnode.c
> +++ b/fs/hfsplus/bnode.c
> @@ -629,7 +629,7 @@ struct hfs_bnode *hfs_bnode_create(struct hfs_btree *tree, u32 num)
>  	if (node) {
>  		pr_crit("new node %u already hashed?\n", num);
>  		WARN_ON(1);
> -		return node;
> +		return ERR_PTR(-EEXIST);
>  	}
>  	node = __hfs_bnode_create(tree, num);
>  	if (!node)

Looks good. Thank you for the fix.

Reviewed-by: Viacheslav Dubeyko <slava@dubeyko.com>

Thanks,
Slava.