Date: Mon, 10 Oct 2005 15:55:10 -0400 From: Ben Kelly <bkelly@vadev.org> To: freebsd-stable@freebsd.org Subject: Re: 5.x: how do I get a *swap*-backed /tmp via rc.conf? Message-ID: <200510101555.10416.bkelly@vadev.org> In-Reply-To: <20051010194932.GT47561@bunrab.catwhisker.org> References: <20051010020729.GA56351@bunrab.catwhisker.org> <200510101103.50546.bkelly@vadev.org> <20051010194932.GT47561@bunrab.catwhisker.org>
next in thread | previous in thread | raw e-mail | index | archive | help
On Monday 10 October 2005 3:49 pm, David Wolfskill wrote: > On Mon, Oct 10, 2005 at 11:03:50AM -0400, Ben Kelly wrote: > > On Monday 10 October 2005 10:55 am, Malcolm Kay wrote: > > .... > > > > > These paramaters are used by the startup script /etc/rc.d/tmp > > > which calls mount_md defined in /etc/rc.subr which specifically > > > adds the _M (malloc) option to the mdmfs call. > > > > > > You'll need to invoke your own script (or; not so nice; > > > edit rc.subr). > > > > Is there a reason not to use the ramdisk_* knobs? This seems to work for > > me: > > > > ramdisk_units="10 11" > > > > # tmp > > ramdisk_10_config="-t swap -s 256m" > > ramdisk_10_perms="1777" > > > > # mimedefang spool > > ramdisk_11_config="-t swap -s 192m" > > ramdisk_11_owner="mailnull" > > ramdisk_11_perms="700" > >... > > Well, other than the point that I'm not seeing those knobs, as Yes. Sorry. I realized after I posted that those rc scripts are not hooked into the makefile so they will not get installed by mergemaster. They are, however, in rc.conf(5) and in the src CVS tree. > Lowell Gilbert pointed out (in response to my original message), > the "-M" flag was moved from src/etc/rc.subr to the tmpmfs_flags > and varmfs_flags variables in src/etc/defaults/rc.conf in HEAD (on > 24 Aug), and that change was MFCed to RELENG_6 on 28 Aug. > > I filed a PR, bin/87218 about 3 hours ago, in which I requested > that the change in question also be MFCed to RELENG_5. > > I have, in fact, tested the implementation of the change for RELENG_5, > and it both allows the specification of a swap-backed /tmp (while > preserving the default behavior) and when I put the modified RELENG_5 > box (with the swap-, rather than malloc-backed /tmp) under a superset of > the load that crashed it yesterday, it performed without a problem. > > This would seem to be a Good Thing. And I don't see a downside to the > requested MFC for RELENG_5. I was just trying to offer an alternative. Also, I sent my mail this morning but it was held up at my outgoing mail server, so it probably looked a bit out of place in the conversation. Anyway, sorry for the noise. > > Peace, > david
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200510101555.10416.bkelly>