Re: kmem_map too small, revisited

On 3/16/2004 5:52 PM, Sven Willenberger wrote:
> On Tue, 2004-03-16 at 18:31, David Malone wrote:
>>On Tue, Mar 16, 2004 at 03:18:37PM -0500, Sven Willenberger wrote:
>>>What exactly is "cred" and how would I go about finding out a) if that
>>>first set of numbers is completely out of line and b) what is causing it
>>>to go so high if it is in fact out of whack.
>>
>>cred is for the credentials of an object (who created it, so the
>>kernel knows what it is permitted to do). A leak of cred objects
>>related to tcp connections was fixed on the 19th of Feb in -current
>>and was merged before the release of 5.2. Is the code you are
>>running recent enough to have this fix?
> 
> I assume you mean 5.2.1? If so, then this bodes quite well actually.
> From the previous reply in which the cred reading was extremely high it
> looks like the build date was just prior to the commit:
> 
> FreeBSD host.dmv.com 5.2.1-RC2 FreeBSD 5.2.1-RC2 #0: Thu Feb 19 11:19:04 EST 2004
> 
> I will update the other machines in this cluster immediately to
> 5.2.1-Release (cvsup'ing with a release tag of RELENG_5_2 should
> suffice, yes?) if they are not already. Since these machines are heavy
> in tcp connections and run under various users (defang, mailnull, etc)
> your explanation makes absolute sense in light of the repaired leak. I
> hope this does the trick :-)

RELENG_5_2 is the correct tag.  The fix was included in 5.2.1-RELEASE.

The original fix on Feb 19:
http://lists.freebsd.org/pipermail/cvs-src/2004-February/018888.html

The MFC to RELENG_5_2 on Feb 23:
http://lists.freebsd.org/pipermail/cvs-src/2004-February/019147.html

Jon