With recent CVE fixes, ad_enty() may now return NULL. This
commit adds basic error handling for these cases and asserting
where such a return is totally unexpected. In case of
ad_getid() and ad_forcegetid(), return CNID_INVALID rather
than 0 to clarify for future people investigating this that
a 0 here is an indication of error.

In case of new_ad_header(), the valid_data_len of the
adouble data may still be zero. This causes subsequent
ad_entry() calls within new_ad_header() to fail. As such,
overwrite valid_data-Len with AD_DATASZ2 or AD_DATASZ_EA
depending on how adouble data is stored on disk.

Another side-effect of the fix is that ad_entry() for
ADEID_DID on ea-backed adouble data will return NULL. In
this case, add explicit check for the backend before
attempting to get the DID.

Signed-off-by: Andrew Walker <[email protected]>

Netatalk appears to not attempt to read adouble
metadata for symlinks (no lgetxtattr call) and so
for the sake of consistency, we will also avoid
setting them in the ad.

In some edge cases the comment in finder info xattr may be
zero-length. This means that ad_entry() will return NULL
with latest CVE fixes. This means we need alternate validation
for comment.

Since the ad_entry handling changed significantly in 3.12->3.13
update, AFP_ASSERT() when metadata is flagged as invalid rather
than attempting to remove it. This will give us an opportunity
to fix any other issues introduced by the original changeset
without affecting user metadata.