Re: Crash while trying to destroy memory cache with Heimdal 1.4.1rc2

Hello,

Think about krb5_cc_destroy as a krb5_cc_close() combined with delete all data.

> Hi,
> 
> I need some help in understanding the Heimdal KRB5 Memory cache usage.  
> In the file 'lib/krb5/test_cc.c ', I see the following line in documentation. 
> 
>  93 /*
>  94  * Check that a closed cc still keeps it data and that it's no longer
>  95  * there when it's destroyed.
>  96  */
> 

This checks that if a credential cache doesn't go away on krb5_cc_close() and that is really gone when calling krb5_cc_destroy().


> This implies, that a KRB5 cache needs to be destroyed after it is close, or else it will retain the data.
> But if you look at the code, it suggests that ref counting is done and the data is freed when you do a close.
> The ref counting can be seen in the file 'lib/krb5/mcache.c ',  in following lines: 

Reference counting is done on the name, so that multiple holders of the same cache wont destroy the underlaying handle. The credential data can be destroyed though.

> So I chose to still do a krb5_cc_destroy to make sure I'm not leaking memory in some other scenarios. 
> But, when my class tries to do a krb5_cc_destroy() in the destructor, it hits an krb5_abortx() in the same file on following line numbers:
> 
> 200     if (m->refcnt == 0)
> 201       krb5_abortx(context, "mcc_destroy: refcnt already 0");
> 202 

krb5_cc_destroy() is an implicit krb5_cc_close().


> I need some help in understanding the use of KRB5 cache of type MEMORY.  Following are the specific questions I need help with:
> 1] Is Krb5 Mem cache data ref-counted or needs to be explicitly destroyed?

Its refcounted and the data is destroyed.

> 2] While destroying the object, should I use krb5_cc_close() or krb5_cc_destroy() or both?

krb5_cc_destroy(), destroy lead to the marking the object as dead, and when the last reference is dropped, its gone.
It can be resurected by krb5_cc_initalize().

Love