Date: Tue, 11 Aug 2009 22:56:47 +0200 (CEST) From: Peter Much <pmc@citylink.dinoex.sub.org> To: freebsd-fs@freebsd.org, kmacy@freebsd.org Subject: Re: zfs/panic: short after rollback Message-ID: <200908112056.n7BKulp2029129@gate.oper.dinoex.org> References: <4A325E9F.2080802@icyb.net.ua> <3c1674c90906121354s6d6ae7ben5082708b1586e94f@mail.gmail.com>
next in thread | previous in thread | raw e-mail | index | archive | help
<kmacy@freebsd.org> aka Kip Macy schrieb mit Datum Fri, 12 Jun 2009 13:54:40 -0700 in m2n.fbsd.stable: |show sleepchain |show thread 100263 | |On Fri, Jun 12, 2009 at 6:56 AM, Andriy Gapon<avg@icyb.net.ua> wrote: |> |> I did zfs rollback xxx@yyy |> And then did ls on a directory in the rolled-back fs. |> panic: sleeping thread This is quite likely the same problem as I experience. And it is maybe also the same problem as in kern/137037 and kern/129148. It seems to show up in some different flavours, while the bottomline is this: do a rollback, and soon after (usually at the next filesystem-related action) the kernel has gone fishing. I experienced it first when doing a rollback of a mounted filesystem. It crashed right after the first try, and it did so reproducible. (Well, more or less reproducible - another day under similar circumstances it did not crash.) Then I started thinking, and came to the conclusion that a rollback of a mounted filesystem (with possibly open files) could easily bring a lot of things into an undefined state, and should not be something one wants to do normally. So maybe it is not supposed to work at all. Anyway, when trying this, I do either get the "sleeping thread" message (as above), or a panic from _sx_xlock() (as shown in my addendum to kern/137037, and in the addendum to kern/129148). So I started to do rollbacks on unmounted filesystems (quite an excessive amount of them), and while this seemed to work at first, later on the system failures reappeared. These system failures took various shapes - I experienced immediate resets without dump, and system hangs. When deliberately trying to reproduce that (after installing a kernel with debugging info and watching the console), I also captured a panic coming from _sx_xlock() - so it seems to be the same problem as without unmounting, only that it takes a couple of rollbacks (a dozen or more) to hit. Over all, there was never any data loss or persistent damage. So, I consider rollback still functional and safe to use, but I consider a system no longer production stable after doing a rollback. rgds, PMc
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200908112056.n7BKulp2029129>