From: Russell Coker Reply-To: russell@coker.com.au To: Daniel J Walsh Subject: Re: A filename to label translation daemon Date: Fri, 10 Aug 2012 12:28:11 +1000 Cc: Colin Walters , Eric Paris , selinux@tycho.nsa.gov, lennart@poettering.net, sds@tycho.nsa.gov References: <1344454290.25533.12.camel@localhost> <201208100037.21877.russell@coker.com.au> <5023EE1C.5060205@redhat.com> In-Reply-To: <5023EE1C.5060205@redhat.com> MIME-Version: 1.0 Content-Type: Text/Plain; charset="utf-8" Message-Id: <201208101228.11637.russell@coker.com.au> Sender: owner-selinux@tycho.nsa.gov List-Id: selinux@tycho.nsa.gov On Fri, 10 Aug 2012, Daniel J Walsh wrote: > On 08/09/2012 10:37 AM, Russell Coker wrote: > > On Thu, 9 Aug 2012, Colin Walters wrote: > >> Seems to make sense...though someone could also probably get fairly far > >> by writing a regular expression optimizer. It might not even be that > >> hard to write a multi-regexp matching engine which took a set of regexps > >> at once and constructed a single matching DFA for them. > > > > Is this really going to help? My slowest system is a P3-866 which takes > > less than 30ms of user time for "restorecon /bin/bash" and takes a total > > of 136ms of wall time if the cache is cold. On a 1.8GHz 64bit system > > it's only 8ms of user time. > > > > What benefit are we expecting to get here? > > kerberos library currently does a matchpathcon on /tmp/BLAH files and sets > the label correctly. With this change in the library we are seeing huge > performance hits of apache services caused by loading the regex. What is kerberos doing under /tmp and why is it being done repeatedly by different processes? > Running make install has caused a huge hit if you are running thousands of > install commands which caused the remove of labeling from the install > command. http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=638304 I believe that is a design bug in the SE Linux code in install, I've filed the above Debian bug report about it. I think that correct design of install wouldn't have a "make install" performed as part of a dpkg or rpm build do any SE Linux checks. That would be faster than any other option. > Systemd has been is executing the load load many many times and is showing > up to 1 second slow down on startup. If the startup is 10 seconds, it is > kind of hard to justify 10% slowdown on boot. Wow, who's got a 10 second boot? Is systemd loading the file contexts 100 times? If not then why is it taking a second? > I believe we just add support for this service and have the labeling fall > back to the default if the labeling socket does not exists, and then > distributions can decide whether or not they want to use it. I'm not opposed to making such changes and a 10% performance improvement is definitely worth getting. But first I think we should consider other approaches to some of these issues such as changing the way install works. -- My Main Blog http://etbe.coker.com.au/ My Documents Blog http://doc.coker.com.au/ -- This message was distributed to subscribers of the selinux mailing list. If you no longer wish to subscribe, send mail to majordomo@tycho.nsa.gov with the words "unsubscribe selinux" without quotes as the message.