Date: Sat, 10 Nov 2012 20:36:34 +0000 From: Chris Rees <crees@FreeBSD.org> To: Kevin Day <toasty@dragondata.com> Cc: "freebsd-fs@freebsd.org" <freebsd-fs@freebsd.org>, Bryan Drewery <bryan@shatow.net> Subject: Re: ZFS can't delete files when over quota Message-ID: <CADLo83-mtBnwwB7qXmJ4yNp9LtVtoaqTFvV0MtR5GMrJc_XzZQ@mail.gmail.com> In-Reply-To: <67C1C89F-4A91-4595-8EA7-19AF3EC4656F@dragondata.com> References: <509E79C7.10300@shatow.net> <CADLo83-mZ0v=zXJnHF01%2BtOvCtBWtrOTZpcd-V%2BWTF7qXNMJUg@mail.gmail.com> <CAFMmRNx2Xqo%2B3FZ6cT%2Ben-u26SVAecN%2B9ULfoV==fXRRTG7fDg@mail.gmail.com> <CADLo83_wEvUtYDgzoCoE4oNF7j3AMWe-AX6OvpxM12xd5AzJ7w@mail.gmail.com> <509EA942.9060801@shatow.net> <67C1C89F-4A91-4595-8EA7-19AF3EC4656F@dragondata.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On 10 November 2012 20:29, Kevin Day <toasty@dragondata.com> wrote: > > On Nov 10, 2012, at 1:21 PM, Bryan Drewery <bryan@shatow.net> wrote: > >> On 11/10/2012 12:55 PM, Chris Rees wrote: >>> Bryan, >>> >>> Please try the patch at [1]; if it works I'll document it. >>> >>> Chris >>> >>> [1] http://www.bayofrum.net/~crees/patches/bdrewery.diff >>> >> >> Hmm, I'm not a fan of -T. I think it should just work out-of-the-box her= e. >> >> Something like: >> >> http://people.freebsd.org/~bdrewery/rm-quota.txt >> >> Bryan > > > This also may cause unintended or weird behavior with regard to open/runn= ing binaries or processes that want to keep a file open. > > If you're trying to rm a binary that's currently in use, this is normally= a supported feature. Old copies of the binary keep running and the disk sp= ace isn't actually freed until after everything holding the old version exi= ts. With the patch, truncate will fail with ETXTBSY. Having rm fail with ET= XTBSY is probably wrong. > > Programs are also allowed to keep reading/writing to a file that another = process has unlinked. Unlinking a file is not supposed to destroy the conte= nts of it until after everyone is done with it. Throwing a truncate in ther= e changes that behavior. > > I realize this is only happening under already broken circumstances, but = changing unlink to occasionally doing unlink-truncate-unlink is going to ma= ke some hard to debug situations occur. > > Also, you need to be testing errno =3D=3D EDQUOT, not rval. These are the reasons I added the -T option, which I realise would have been more correct as -t; it's undesirable to have as default behaviour. In my patch, errno is tested. Chris
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?CADLo83-mtBnwwB7qXmJ4yNp9LtVtoaqTFvV0MtR5GMrJc_XzZQ>