Skip to contentSkip to conversation

hfs, hfsplus: remove WARN_ON when new bnode already hashed

PATCH In review Aditya Srivastava 2 participants 2 messages Jun 22 – Jun 23

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 hfs, hfsplus: remove WARN_ON when new bnode already hashed

From: Aditya Prakash Srivastava <[email protected]>

Syzbot reported a warning in hfsplus_bnode_create() when mounting and traversing a fuzzed/corrupted disk image.

The warning is triggered because the node index is found to be already active and hashed in the btree hash table, which indicates on-disk metadata corruption.

Filesystem corruption should be handled gracefully by returning an error and logging a message, rather than triggering kernel-level warnings (which can panic the system under panic_on_warn=1 configurations). The code already prints a critical message and returns -EEXIST (or the node in hfs), so the WARN_ON() is completely redundant and dangerous.

Remove the WARN_ON(1) from both hfs_bnode_create() and hfsplus_bnode_create().

---
 fs/hfs/bnode.c     | 1 -
 fs/hfsplus/bnode.c | 1 -
 2 files changed, 2 deletions(-)
fs/hfs/bnode.c +0 -1
index 13d58c51fc46..335834961741 100644
@@ -517,7 +517,6 @@ struct hfs_bnode *hfs_bnode_create(struct hfs_btree *tree, u32 num)
 	spin_unlock(&tree->hash_lock);
 	if (node) {
 		pr_crit("new node %u already hashed?\n", num);
-		WARN_ON(1);
 		return node;
 	}
 	node = __hfs_bnode_create(tree, num);
fs/hfsplus/bnode.c +0 -1
index f8b5a8ae58ff..bfc2a7f53edd 100644
@@ -631,7 +631,6 @@ struct hfs_bnode *hfs_bnode_create(struct hfs_btree *tree, u32 num)
 	spin_unlock(&tree->hash_lock);
 	if (node) {
 		pr_crit("new node %u already hashed?\n", num);
-		WARN_ON(1);
 		return ERR_PTR(-EEXIST);
 	}
 	node = __hfs_bnode_create(tree, num);
Signature · 1 line
2.47.3
#1 lore raw
Quoted 55 lines · Aditya Srivastava

From: Aditya Prakash Srivastava <[email protected]>

Syzbot reported a warning in hfsplus_bnode_create() when mounting and traversing a fuzzed/corrupted disk image.

The warning is triggered because the node index is found to be already
active and hashed in the btree hash table, which indicates on-disk metadata corruption.

Filesystem corruption should be handled gracefully by returning an error
and logging a message, rather than triggering kernel-level warnings (which can panic the system under panic_on_warn=1 configurations). The code already prints a critical message and returns -EEXIST (or the
node in hfs), so the WARN_ON() is completely redundant and dangerous.

Remove the WARN_ON(1) from both hfs_bnode_create() and hfsplus_bnode_create().

---
 fs/hfs/bnode.c     | 1 -
 fs/hfsplus/bnode.c | 1 -

 2 files changed, 2 deletions(-)
fs/hfs/bnode.c +0 -1
index 13d58c51fc46..335834961741 100644
@@ -517,7 +517,6 @@ struct hfs_bnode *hfs_bnode_create(struct
 hfs_btree *tree, u32 num)
  	spin_unlock(&tree->hash_lock);
  	if (node) {
  		pr_crit("new node %u already hashed?\n", num);
-		WARN_ON(1);
  		return node;
  	}

  node = __hfs_bnode_create(tree, num);

fs/hfsplus/bnode.c +0 -1
index f8b5a8ae58ff..bfc2a7f53edd 100644
@@ -631,7 +631,6 @@ struct hfs_bnode *hfs_bnode_create(struct
 hfs_btree *tree, u32 num)
  	spin_unlock(&tree->hash_lock);
  	if (node) {
  		pr_crit("new node %u already hashed?\n", num);
-		WARN_ON(1);
  		return ERR_PTR(-EEXIST);
  	}
Quoted 1 line

  node = __hfs_bnode_create(tree, num);

If we have corrupted HFS/HFS+ volume and we can detect it during the mount operation, then we should not mount the corrupted volume or mount in READ-ONLY mode. As I remember, we already have HFS+ logic that detects b-tree's map corruption for the case of node 0. We haven't this logic for HFS file system logic yet, if I remember correctly. So, you are welcomed to port this logic for HFS case. Removing WARN_ON() simply hides the potential issues. So, I don't think that we need to remove it.

Thanks,
Slava.