2024-04-03 21:54:11

by Kees Cook

Subject: [PATCH] fs: Set file_handle::handle_bytes before referencing file_handle::f_handle

With adding __counted_by(handle_bytes) to struct file_handle, we need
to explicitly set it in the one place it wasn't yet happening prior to
accessing the flex array "f_handle".

Fixes: 1b43c4629756 ("fs: Annotate struct file_handle with __counted_by() and use struct_size()")
Signed-off-by: Kees Cook <[email protected]>
---
Cc: Christian Brauner <[email protected]>
Cc: "Gustavo A. R. Silva" <[email protected]>
Cc: Alexander Viro <[email protected]>
Cc: Jan Kara <[email protected]>
Cc: Chuck Lever <[email protected]>
Cc: Jeff Layton <[email protected]>
Cc: Amir Goldstein <[email protected]>
Cc: [email protected]
Cc: [email protected]
Cc: [email protected]
---
fs/fhandle.c | 1 +
1 file changed, 1 insertion(+)

diff --git a/fs/fhandle.c b/fs/fhandle.c
index 53ed54711cd2..08ec2340dd22 100644
--- a/fs/fhandle.c
+++ b/fs/fhandle.c
@@ -40,6 +40,7 @@ static long do_sys_name_to_handle(const struct path *path,
GFP_KERNEL);
if (!handle)
return -ENOMEM;
+ handle->handle_bytes = f_handle.handle_bytes;

/* convert handle size to multiple of sizeof(u32) */
handle_dwords = f_handle.handle_bytes >> 2;
--
2.34.1

2024-04-03 22:11:54

by Gustavo A. R. Silva

Subject: Re: [PATCH] fs: Set file_handle::handle_bytes before referencing file_handle::f_handle

On 03/04/24 15:54, Kees Cook wrote:
> With adding __counted_by(handle_bytes) to struct file_handle, we need
> to explicitly set it in the one place it wasn't yet happening prior to
> accessing the flex array "f_handle".

Yes, which (access to `f_handle`) happens here:

48 retval = exportfs_encode_fh(path->dentry,
49 (struct fid *)handle->f_handle,
50 &handle_dwords, fh_flags);

>
> Fixes: 1b43c4629756 ("fs: Annotate struct file_handle with __counted_by() and use struct_size()")
> Signed-off-by: Kees Cook <[email protected]>

Reviewed-by: Gustavo A. R. Silva <[email protected]>

Thanks for catching this!
--
Gustavo

> ---
> Cc: Christian Brauner <[email protected]>
> Cc: "Gustavo A. R. Silva" <[email protected]>
> Cc: Alexander Viro <[email protected]>
> Cc: Jan Kara <[email protected]>
> Cc: Chuck Lever <[email protected]>
> Cc: Jeff Layton <[email protected]>
> Cc: Amir Goldstein <[email protected]>
> Cc: [email protected]
> Cc: [email protected]
> Cc: [email protected]
> ---
> fs/fhandle.c | 1 +
> 1 file changed, 1 insertion(+)
>
> diff --git a/fs/fhandle.c b/fs/fhandle.c
> index 53ed54711cd2..08ec2340dd22 100644
> --- a/fs/fhandle.c
> +++ b/fs/fhandle.c
> @@ -40,6 +40,7 @@ static long do_sys_name_to_handle(const struct path *path,
> GFP_KERNEL);
> if (!handle)
> return -ENOMEM;
> + handle->handle_bytes = f_handle.handle_bytes;
>
> /* convert handle size to multiple of sizeof(u32) */
> handle_dwords = f_handle.handle_bytes >> 2;

2024-04-04 09:19:15

by Jan Kara

Subject: Re: [PATCH] fs: Set file_handle::handle_bytes before referencing file_handle::f_handle

On Wed 03-04-24 14:54:03, Kees Cook wrote:
> With adding __counted_by(handle_bytes) to struct file_handle, we need
> to explicitly set it in the one place it wasn't yet happening prior to
> accessing the flex array "f_handle".
>
> Fixes: 1b43c4629756 ("fs: Annotate struct file_handle with __counted_by() and use struct_size()")
> Signed-off-by: Kees Cook <[email protected]>

OK, so this isn't really a functional bug AFAIU but the compiler will
wrongly complain we are accessing handle->f_handle beyond claimed array
size (because handle->handle_bytes == 0 at that point). Am I right? If
that's the case, please add a short comment explaining this (because it
looks odd we set handle->handle_bytes and then reset it a few lines later).
With the comment feel free to add:

Reviewed-by: Jan Kara <[email protected]>

Honza

> ---
> Cc: Christian Brauner <[email protected]>
> Cc: "Gustavo A. R. Silva" <[email protected]>
> Cc: Alexander Viro <[email protected]>
> Cc: Jan Kara <[email protected]>
> Cc: Chuck Lever <[email protected]>
> Cc: Jeff Layton <[email protected]>
> Cc: Amir Goldstein <[email protected]>
> Cc: [email protected]
> Cc: [email protected]
> Cc: [email protected]
> ---
> fs/fhandle.c | 1 +
> 1 file changed, 1 insertion(+)
>
> diff --git a/fs/fhandle.c b/fs/fhandle.c
> index 53ed54711cd2..08ec2340dd22 100644
> --- a/fs/fhandle.c
> +++ b/fs/fhandle.c
> @@ -40,6 +40,7 @@ static long do_sys_name_to_handle(const struct path *path,
> GFP_KERNEL);
> if (!handle)
> return -ENOMEM;
> + handle->handle_bytes = f_handle.handle_bytes;
>
> /* convert handle size to multiple of sizeof(u32) */
> handle_dwords = f_handle.handle_bytes >> 2;
> --
> 2.34.1
>
--
Jan Kara <[email protected]>
SUSE Labs, CR

2024-04-05 11:00:40

by Christian Brauner

Subject: Re: [PATCH] fs: Set file_handle::handle_bytes before referencing file_handle::f_handle

On Thu, Apr 04, 2024 at 11:19:00AM +0200, Jan Kara wrote:
> On Wed 03-04-24 14:54:03, Kees Cook wrote:
> > With adding __counted_by(handle_bytes) to struct file_handle, we need
> > to explicitly set it in the one place it wasn't yet happening prior to
> > accessing the flex array "f_handle".
> >
> > Fixes: 1b43c4629756 ("fs: Annotate struct file_handle with __counted_by() and use struct_size()")
> > Signed-off-by: Kees Cook <[email protected]>
>
> OK, so this isn't really a functional bug AFAIU but the compiler will
> wrongly complain we are accessing handle->f_handle beyond claimed array
> size (because handle->handle_bytes == 0 at that point). Am I right? If

And really, this also needs to please be mentioned in the commit message
because from reading the commit message I'm not even sure what this
patch is trying to fix.