[krbdev.mit.edu #7804] krb5_cc_retrieve_cred_seq manipulates KRB5_TC_OPENCLOSE in a non-threadsafe way

Andrea Campi via RT rt-comment at krbdev.mit.edu
Mon Dec 16 18:29:26 EST 2013


It's not just efficiency, it can actually cause errors.

I found
http://kerberos.996246.n3.nabble.com/Concurrency-issue-between-krb5-cc-retrieve-cred-and-krb5-cc-store-cred-td28908.html
,
very similar.


On Mon, Dec 16, 2013 at 12:01 PM, Greg Hudson via RT <
rt-comment at krbdev.mit.edu> wrote:

> The OPENCLOSE flag is intended to be a matter of optimization, not
> correctness.  Another thread using the already-open file shouldn't
> generally be a problem; it will just be a little faster.  Are you seeing
> a more severe issue than this?
>
> (I can think of a scenario where there are two overlapping
> krb5_cc_retrieve_cred_seqs running in different threads, and when the
> first one finishes, it turns the OPENCLOSE flag back on, causing the
> other thread to run less efficiently.  But I still don't see that causing
> a user-visible problem, just an efficiency issue.)
>



More information about the krb5-bugs mailing list