Re: (ITS#5827) Syncprov persistent operations info leaking

[hyc@symas.com]

Pierangelo Masarati wrote:
> hyc@symas.com wrote:
>> ando@sys-net.it wrote:
>>> Full_Name: Pierangelo Masarati
>>> Version:
>>> OS:
>>> URL: ftp://ftp.openldap.org/incoming/
>>> Submission from: (NULL) (81.72.89.40)
>>> Submitted by: ando
>>>
>>>
>>> After heavily loading a MMR pool, valgrind finds the following:
>>>
>>> ==29650== 2,437 (240 direct, 2,197 indirect) bytes in 3 blocks are definitely
>>> lo
>>> st in loss record 13 of 15
>> Most likely because we increment s_inuse at line 1250, at the beginning of a
>> non-Delete write op, and don't decrement it at the end of the op. I think this
>> should be simple to fix, will take a look.
>
> Seems to work.  Thanks, p.

I don't think the current fix is complete. Due to the Abandon check at 1190, 
it's possible (though extremely unlikely) for an Abandon to arrive between the 
start and end of a mod op, and cause the final decrement to be skipped still. 
Probably need to do some restructuring to catch this, I'll look at this some more.
-- 
   -- Howard Chu
   CTO, Symas Corp.           http://www.symas.com
   Director, Highland Sun     http://highlandsun.com/hyc/
   Chief Architect, OpenLDAP  http://www.openldap.org/project/