[PATCH] io_uring: fix filename leak in __io_openat_prep()

[PATCH] io_uring: fix filename leak in __io_openat_prep()

[Prithvi Tambewagh]

__io_openat_prep() allocates a struct filename using getname(), but
it isn't freed in case the present file is installed in the fixed file
table and simultaneously, it has the flag O_CLOEXEC set in the
open->how.flags field.

This is an erroneous condition, since for a file installed in the fixed
file table, it won't be installed in the normal file table, due to which
the file cannot support close on exec. Earlier, the code just returned
-EINVAL error code for this condition, however, the memory allocated for
that struct filename wasn't freed, resulting in a memory leak.

Hence, the case of file being installed in the fixed file table as well
as having O_CLOEXEC flag in open->how.flags set, is adressed by using
putname() to release the memory allocated to the struct filename, then
setting the field open->filename to NULL, and after that, returning
-EINVAL.

Reported-by: syzbot+00e61c43eb5e4740438f@syzkaller.appspotmail.com
Closes: https://syzkaller.appspot.com/bug?extid=00e61c43eb5e4740438f
Tested-by: syzbot+00e61c43eb5e4740438f@syzkaller.appspotmail.com
Cc: stable@vger.kernel.org
Signed-off-by: Prithvi Tambewagh <activprithvi@gmail.com>
---
 io_uring/openclose.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/io_uring/openclose.c b/io_uring/openclose.c
index bfeb91b31bba..fc190a3d8112 100644
--- a/io_uring/openclose.c
+++ b/io_uring/openclose.c
@@ -75,8 +75,11 @@ static int __io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe
 	}
 
 	open->file_slot = READ_ONCE(sqe->file_index);
-	if (open->file_slot && (open->how.flags & O_CLOEXEC))
+	if (open->file_slot && (open->how.flags & O_CLOEXEC)) {
+		putname(open->filename);
+		open->filename = NULL;
 		return -EINVAL;
+	}
 
 	open->nofile = rlimit(RLIMIT_NOFILE);
 	req->flags |= REQ_F_NEED_CLEANUP;

base-commit: b927546677c876e26eba308550207c2ddf812a43
-- 
2.34.1

Re: [PATCH] io_uring: fix filename leak in __io_openat_prep()

[Jens Axboe]

On 12/24/25 9:42 AM, Prithvi Tambewagh wrote:
> __io_openat_prep() allocates a struct filename using getname(), but
> it isn't freed in case the present file is installed in the fixed file
> table and simultaneously, it has the flag O_CLOEXEC set in the
> open->how.flags field.
> 
> This is an erroneous condition, since for a file installed in the fixed
> file table, it won't be installed in the normal file table, due to which
> the file cannot support close on exec. Earlier, the code just returned
> -EINVAL error code for this condition, however, the memory allocated for
> that struct filename wasn't freed, resulting in a memory leak.
> 
> Hence, the case of file being installed in the fixed file table as well
> as having O_CLOEXEC flag in open->how.flags set, is adressed by using
> putname() to release the memory allocated to the struct filename, then
> setting the field open->filename to NULL, and after that, returning
> -EINVAL.
> 
> Reported-by: syzbot+00e61c43eb5e4740438f@syzkaller.appspotmail.com
> Closes: https://syzkaller.appspot.com/bug?extid=00e61c43eb5e4740438f
> Tested-by: syzbot+00e61c43eb5e4740438f@syzkaller.appspotmail.com
> Cc: stable@vger.kernel.org
> Signed-off-by: Prithvi Tambewagh <activprithvi@gmail.com>
> ---
>  io_uring/openclose.c | 5 ++++-
>  1 file changed, 4 insertions(+), 1 deletion(-)
> 
> diff --git a/io_uring/openclose.c b/io_uring/openclose.c
> index bfeb91b31bba..fc190a3d8112 100644
> --- a/io_uring/openclose.c
> +++ b/io_uring/openclose.c
> @@ -75,8 +75,11 @@ static int __io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe
>  	}
>  
>  	open->file_slot = READ_ONCE(sqe->file_index);
> -	if (open->file_slot && (open->how.flags & O_CLOEXEC))
> +	if (open->file_slot && (open->how.flags & O_CLOEXEC)) {
> +		putname(open->filename);
> +		open->filename = NULL;
>  		return -EINVAL;
> +	}
>  
>  	open->nofile = rlimit(RLIMIT_NOFILE);
>  	req->flags |= REQ_F_NEED_CLEANUP;

You can probably fix it similarly by just having REQ_F_NEED_CLEANUP set
earlier in the process, then everything that needs undoing will get
undone as part of ending the request.

-- 
Jens Axboe