Re: [PATCH net v3 11/14] afs: Create a server appdata key

Simon Horman [off-list ref] wrote:

[Severity: High]
Does this toksize calculation omit the 4 bytes required for the level field? 

It looks like the sizes added up correspond to starttime (8), lifetime (4),
bytelife (4), and expirationtime (8).

If the level field is omitted, toksize will be 4 bytes too small.

Yeah.

Can this manual advance of the xdr pointer hide the token size mismatch?

Because xdr is advanced based on the miscalculated encsize rather than the
actual number of fields written, it appears this will bypass the adatasize
sanity check on the next line.

As it happens, the encryption padding hides the problem.

When crypto_krb5_encrypt() executes, it seems it will only process the
truncated toksize bytes of plaintext. This would leave the trailing 4 bytes
of the serialized XDR unencrypted and overwritten by the appended Kerberos
checksum. 

The buffer is sized and padded out by crypto_krb5_how_much_buffer() and
kzalloc() preclears it, so crypto_krb5_encrypt() happens to work correctly
most of the time as the data it encrypts is rounded out to a whole encryption
blocks.

David