Re: [RFC][PATCH] afs: Work around strnlen() oops with CONFIG_FORTIFIED_SOURCE=y

[Linus Torvalds]

On Mon, Dec 21, 2020 at 8:14 AM David Howells <dhowells@xxxxxxxxxx> wrote:
>
> CONFIG_FORTIFIED_SOURCE=y now causes an oops in strnlen() from afs (see
> attached patch for an explanation).  Is replacing the use with memchr() the
> right approach?  Or should I be calling __real_strnlen() or whatever it's
> called?

Ugh. No.

> AFS has a structured layout in its directory contents (AFS dirs are
> downloaded as files and parsed locally by the client for lookup/readdir).
> The slots in the directory are defined by union afs_xdr_dirent.  This,
> however, only directly allows a name of a length that will fit into that
> union.  To support a longer name, the next 1-8 contiguous entries are
> annexed to the first one and the name flows across these.

I htink the right fix would be to try to create a type that actually
describes that.

IOW, maybe the afs_xdr_dirent union could be written something like

  union afs_xdr_dirent {
          struct {
                  u8              valid;
                  u8              unused[1];
                  __be16          hash_next;
                  __be32          vnode;
                  __be32          unique;
                  u8              name[];
         } u;
          u8                      extended_name[32];
  } __packed;

instead, and have a big comment about how "name[]" is that
"16+overflow+next entries" thing?

I didn't check how you currently use that ->name thing (not a good
identifier to grep for..), so you might want some other model - like
using a separate union case for this "unconstrained name" case.

In fact, maybe that separate union struct is a better model anyway, to
act as even more of documentation about the different cases..

              Linus