Skip to contentSkip to conversation

fuse: check fi is not NULL before calling fuse_passthrough_release()

PATCH In review Tristan Madani 2 participants 2 messages Jun 19 – Jun 22

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

PATCH 1/1 fuse: check fi is not NULL before calling fuse_passthrough_release()

From: Tristan Madani <[email protected]>

fuse_create_open() calls fuse_sync_release() with a NULL fuse_inode when fuse_iget() fails. This propagates to fuse_prepare_release(), which passes the NULL fi to fuse_inode_backing() via the fuse_passthrough_release() call, resulting in a NULL pointer dereference.

The existing comment in fuse_prepare_release() documents that the inode can be NULL on the error path of fuse_create_open(), and the fi->lock access below is already guarded with if (likely(fi)), but the passthrough release path added by commit 4a90451bbc7f ("fuse:
implement open in passthrough mode") was not given the same protection.

Add the missing NULL check for fi before calling
fuse_passthrough_release().

Found by syzkaller.

Fixes 4a90451bbc7f ("fuse: implement open in passthrough mode")
---
 fs/fuse/file.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
fs/fuse/file.c +1 -1
index c59452d60b8d..9b368eab159c 100644
@@ -313,7 +313,7 @@ static void fuse_prepare_release(struct fuse_inode *fi, struct fuse_file *ff,
 	struct fuse_conn *fc = ff->fm->fc;
 	struct fuse_release_args *ra = &ff->args->release_args;
 
-	if (fuse_file_passthrough(ff))
+	if (fi && fuse_file_passthrough(ff))
 		fuse_passthrough_release(ff, fuse_inode_backing(fi));
 
 	/* Inode is NULL on error path of fuse_create_open() */
Signature · 1 line
2.47.3
#1 lore raw
Quoted 9 lines · Tristan Madani

From: Tristan Madani <[email protected]>

fuse_create_open() calls fuse_sync_release() with a NULL fuse_inode when fuse_iget() fails. This propagates to fuse_prepare_release(), which passes the NULL fi to fuse_inode_backing() via the fuse_passthrough_release() call, resulting in a NULL pointer dereference.

Quoted 6 lines

The existing comment in fuse_prepare_release() documents that the inode can be NULL on the error path of fuse_create_open(), and the fi->lock access below is already guarded with if (likely(fi)), but the passthrough release path added by commit 4a90451bbc7f ("fuse:
implement open in passthrough mode") was not given the same protection.

But passthrough open happens after this error patch, so this sounds like an invented problem

Quoted 5 lines

Add the missing NULL check for fi before calling
fuse_passthrough_release().

Found by syzkaller.

Where is the report? where is the repro?

Quoted 13 lines
Fixes 4a90451bbc7f ("fuse: implement open in passthrough mode")
---
 fs/fuse/file.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
fs/fuse/file.c +0 -0
index c59452d60b8d..9b368eab159c 100644

fs/fuse/file.c static void fuse_prepare_release(struct fuse_inode *fi, struct fuse_file *ff, · line 318

Quoted 7 lines
fs/fuse/file.c +1 -1
@@ -313,7 +313,7 @@ static void fuse_prepare_release(struct fuse_inode *fi, struct fuse_file *ff,
        struct fuse_conn *fc = ff->fm->fc;
        struct fuse_release_args *ra = &ff->args->release_args;
 
-       if (fuse_file_passthrough(ff))
+       if (fi && fuse_file_passthrough(ff))
                fuse_passthrough_release(ff, fuse_inode_backing(fi));
 

This should not be possible and if it were possible, this code would be leaking ff->passthrough.

I don't know how you got to this issue and why, but if you are sending me code fixes from an hallucinating AI agent without really understanding and really reproducing the problem please respect my time and don't!

Thanks,
Amir.