[PATCH] cifs: potential buffer overflow in handling symlinks

[Steve French]

Wouldn't it be easier and clearer to do the compare vs the maximum len ie

if (link_len > CIFS_MF_SYMLINK_LINK_MAXLEN)

instead of

if (link_len > buf_len - CIFS_MF_SYMLINK_LINK_OFFSET)

since buf_len is  CIFS_MF_SYMLINK_FILE_SIZE and looking at link.c line
26 and 27 this means we can use CIFS_MF_SYMLINK_LINK_OFFSET for the
comparison:

#define CIFS_MF_SYMLINK_LINK_MAXLEN (1024)
#define CIFS_MF_SYMLINK_FILE_SIZE \
        (CIFS_MF_SYMLINK_LINK_OFFSET + CIFS_MF_SYMLINK_LINK_MAXLEN)

On Tue, Apr 12, 2022 at 1:01 AM Harshit Mogalapalli
<harshit.m.mogalapalli at oracle.com> wrote:
>
> Smatch printed a warning:
>         arch/x86/crypto/poly1305_glue.c:198 poly1305_update_arch() error:
>         __memcpy() 'dctx->buf' too small (16 vs u32max)
>
> It's caused because Smatch marks 'link_len' as untrusted since it comes
> from sscanf(). Add a check to ensure that 'link_len' is not larger than
> the size of the 'link_str' buffer.
>
> Fixes: c69c1b6eaea1 ("cifs: implement CIFSParseMFSymlink()")
> Signed-off-by: Harshit Mogalapalli <harshit.m.mogalapalli at oracle.com>
> ---
>  fs/cifs/link.c | 3 +++
>  1 file changed, 3 insertions(+)
>
> diff --git a/fs/cifs/link.c b/fs/cifs/link.c
> index 852e54e..ebfedae 100644
> --- a/fs/cifs/link.c
> +++ b/fs/cifs/link.c
> @@ -85,6 +85,9 @@
>         if (rc != 1)
>                 return -EINVAL;
>
> +       if (link_len > buf_len - CIFS_MF_SYMLINK_LINK_OFFSET)
> +               return -EINVAL;
> +
>         rc = symlink_hash(link_len, link_str, md5_hash);
>         if (rc) {
>                 cifs_dbg(FYI, "%s: MD5 hash failure: %d\n", __func__, rc);
> --
> 1.8.3.1
>


-- 
Thanks,

Steve