Date: Mon, 21 Jul 1997 01:14:04 -0700 From: Dmitry Kohmanyuk =?KOI8-R?B?5M3J1NLJyiDrz8jNwc7Ayw==?= <dk@farm.org> To: Terry Lambert <terry@lambert.org> Cc: dk+@ua.net, toor@dyson.iquest.net, freebsd-current@FreeBSD.ORG, sia@nest.org Subject: Re: -current :) Message-ID: <19970721011404.23066@dog.farm.org> In-Reply-To: <199707202356.QAA10117@phaeton.artisoft.com>; from "Terry Lambert" on Sun, Jul 20, 1997 at 04:56:46PM -0700 References: <199707192216.PAA21949@dog.farm.org> <199707202356.QAA10117@phaeton.artisoft.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On Sun, Jul 20, 1997 at 04:56:46PM -0700, Terry Lambert wrote: > > > > Probably MMAP related as ACT_STYLE is MMAP in my config.data file. > > > > > > > > John, any idea ? > > > > > > > I have checked the code over and over and over again. I sure wish we > > > had a repeatable example :-(. > > > > this _exactly_ (zero-filled page in active, then new newsgroups) happened > > to me on our news server running 2.2.1-RELEASE. So, it should be old code. > > I have to say that I believe that this is not an NFS specific problem. > > As I said, I have been able to repeat it with an app *not* using mmap() > at all, and *not* over NFS. The app was "vi". I forgot to mention: my news server is not running over NFS (should be obvious, yes? Not really - I know at least 1 article claiming performance improvement on SunOS using NFS-mounted news spool from NetApp server. Oh well.) So, we have different problem reports: zero-filled page written with mmap() without NFS (append??), zero-filled pages appended over NFS (mail client, etc.) yours case (zero-filled pages without NFS or mmap). Do you have code which reproduces this bug using shell scripts?? (please post it if so). It can be several problems. > I *truly* believe that the page containing a frag is not being marked > dirty, and therefore is not allocated or written. > > This is in the append case. > > Obviously, if the file size were updated, and the frag was not, > a zero-fill would occur on read, just like with sparse blocks. > > The operation which should have "dirtied" the block is a write to > a frag. > > One thing to note is that I am running 8k FS blocks; as a result, my > frag size is 1024 bytes. This is double the block size, and it may > be related to the problem. > > All the problems have occurred when the entire file is a frag, and > is extended and remains in the same frag... ie: my file was 40 or > so characters in length and was only grown by 6 characters (I changed > the configuration file name for slapd). And then you got 46 zero bytes, right? > Before Nate jumps in and starts screaming "TerryBSD! TerryBSD!" at > the top of his electronic lungs, as he is wont to do, the problem > repeated on stock kernels. The problem did *not* repeat on a > vanilla 2.2 stock kernel, however. As I have said, my INN mmap() active file on UFS happened under 2.2.1-R.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?19970721011404.23066>