Date: Mon, 29 May 2006 12:50:52 -0500 (CDT) From: Greg Rivers <gcr+freebsd-stable@tharned.org> To: Ulrich Spoerlein <uspoerlein@gmail.com> Cc: Paul Allen <nospam@ugcs.caltech.edu>, freebsd-stable@freebsd.org Subject: Re: kmem leak in tmpmfs? Message-ID: <20060529121831.F43434@nc8000.tharned.org> In-Reply-To: <20060529144610.GD1117@roadrunner.q.local> References: <wpy7wq6qlh.fsf@heho.labo> <44764417.4030004@alumni.rice.edu> <20060526004343.GE28128@groat.ugcs.caltech.edu> <200605252220.44739.gcr%2Bfreebsd-stable@tharned.org> <20060529144610.GD1117@roadrunner.q.local>
next in thread | previous in thread | raw e-mail | index | archive | help
On Mon, 29 May 2006, Ulrich Spoerlein wrote: > Greg Rivers wrote: >> On Thursday 25 May 2006 19:43, Paul Allen wrote: >>>> From Jonathan Noack <noackjr@alumni.rice.edu>, Thu, May 25, 2006 at >>>> 07:56:07PM -0400: I am currently running with the following in >>>> /etc/rc.conf and haven't experienced any problems: >>>> tmpmfs_flags="-S -o async" >>> >>> Is there a way to accomplish this with an fstab entry? >>> >>> md /tmp mfs rw,async,-s1024m,-S >>> >>> no,I don't think so. But surely it would be better to just fixup the >>> standard mount -t path to not call mount_mfs... >>> >> >> Actually there is a way. I too have not been satisfied with the tmpmfs >> features in rc, so for some time now I've simply created a hard link: >> >> cd /sbin && ln -f mdmfs mount_md >> >> and then used an entry like this in /etc/fstab: >> >> md /tmp md rw,async,-Sp1777,-s768M 0 0 >> >> This works great for me. A simple patch to mtree could make this hard link >> part of the base system. Nothing else is needed. > > You should use the 'mfs' file system. This works out of the box: > > md /tmp mfs rw,-s256m,-S,-Otime,async,noatime 0 0 > > Ulrich Spoerlein > Using 'mfs' as you suggest doesn't work; you'll only get a usage message when you try to mount: usage: mount_mfs -C [-lNU] [-a maxcontig] [-b block-size] [-c cylinders] [-d rotdelay] [-e maxbpg] [-F file] [-f frag-size] [-i bytes] [-m percent-free] [-n rotational-positions] [-O optimization] [-o mount-options] [-s size] [-v version] md-device mount-point This is because not all mdmfs options are available in "mount_mfs compatibility" mode as described at the bottom of mdmfs(8). Conspicuously absent are -p, -S, and -w. With 'mfs', you can't turn off softupdates (ie. can't have async), and you can't set arbitrary modes and ownerships on the file system. Creating the "mount_md" hard link to mdmfs and using type 'md' in fstab allows you to use all the features of mdmfs. -- Greg
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20060529121831.F43434>