Hello Nathaniel,
I have a separated remark about updating the mods.
modifications of the entry occurs in two phases:
* call BE_PREOP plugins then apply the mods on the entry
* call BE_TXN_PREOP plugin then apply *only* extra mods on the entry
The plugin (BE_TXN_PREOP) translates a MOD/replace from MOD/DEL
(current value) + MOD/ADD(new value).
The translation is done like this:
MOD[0]: MOD/REPL ipatokenHOTPcounter new_value
MOD[1]: MOD/REPL modifiersname
MOD[2]: MOD/REPL modifytimestamp
MOD[3]: MOD/REPL entryusn
-->
MOD[0]: MOD/DEL ipatokenHOTPcounter current_value
MOD[1]: MOD/ADD ipatokenHOTPcounter new_value
MOD[2]: MOD/REPL modifiersname
MOD[3]: MOD/REPL modifytimestamp
MOD[4]: MOD/REPL entryusn
That means, we have one added mod (MOD[4]=entryusn) after
BE_TXN_PREOP that will be (re)applied on the second phase.
So the MOD/DEL+MOD/ADD will not be processed. MOD/REPL will be
processed before MOD/DEL+MOD/ADD are generated.
If MOD/DEL+MOD/ADD is needed, then we may need to replace
REPL->DEL/ADD in BE_PREOP callback.
The control done in plugin BE_TXN_PREOP is working fine and
guaranties the counter does not decrease.
thanks
thierry
On 09/19/2014 10:15 PM, Nathaniel McCallum wrote:
This new version fixes a small style issue pointed out to me by richm
(thanks!).
On Fri, 2014-09-19 at 13:39 -0400, Nathaniel McCallum wrote:
The attached version of the patch should solve all of these issues. It
should also be more performant and use less memory.
Nathaniel
On Wed, 2014-09-17 at 15:33 +0200, thierry bordaz wrote:
On 09/15/2014 09:05 PM, Nathaniel McCallum wrote:
This plugin ensures that all counter/watermark operations are atomic
and never decrement. Also, deletion is not permitted.
https://fedorahosted.org/freeipa/ticket/4494
_______________________________________________
Freeipa-devel mailing list
Freeipa-devel@redhat.com
https://www.redhat.com/mailman/listinfo/freeipa-devel
Hello Nathaniel,
More thoughts.. I think being "betxnpreoperation" you are safe
with parallel client ops and the check is atomic. I have few
comments:
* Shouldn't be implemented
SLAPI_PLUGIN_CLOSE_FN/SLAPI_PLUGIN_START_FN callback,
even if they are empty.
* In load_counters, in case we have a target entry that
has not 'objectclass' 'ipatokenHOTP|ipatokenTOTP' and
the mod operation:
dn: <entry>
changetype: modify
replace: ipatokenHOTPcounter
ipatokenHOTPcounter: 200
-
add: objectclass
objectclass: ipatokenHOTP
I wonder if the operation will not fail although IMHO
it should succeeds.
Shouldn't let the schema checking reject the operation
if the attribute is not granted by the entry
objectclass
* in load_counters, I am under the impression it may
return ipatokenHOTPcounter or ipatokenTOTPwatermark
(ipatokenHOTPcounter is missing).
But then how the caller knows that the returned value
is a counter or a watermark ?
* in ldapmod_is_attrs you may prefer PL_strcasecmp to
strcasecmp
About replicated updates, if updates of counters/watermark are
done on several servers. Then a replicated operation may want
to set counters/watermark with a smaller value that the
existing one. In that case, it will return unwilling to
perform. That could break replication.
If the update comes from replication and the value is going
backward, we could make the operation a nop operation (setting
the attribute to its current value).
thanks
theirry
_______________________________________________
Freeipa-devel mailing list
Freeipa-devel@redhat.com
https://www.redhat.com/mailman/listinfo/freeipa-devel
_______________________________________________
Freeipa-devel mailing list
Freeipa-devel@redhat.com
https://www.redhat.com/mailman/listinfo/freeipa-devel