latest stable13.0-ALPHA3 can not start varnish anymore.

Kyle Evans kevans at freebsd.org
Tue Feb 2 16:44:39 UTC 2021


On Tue, Feb 2, 2021 at 10:42 AM Kristof Provost <kp at freebsd.org> wrote:
>
> On 2 Feb 2021, at 14:05, Johan Hendriks wrote:
> > On 01/02/2021 22:48, Johan Hendriks wrote:
> >> I just updated my FreeBSD 13.0-APLPHA3 to the latest revision and now
> >> i can not start varnish anymore.
> >> This is on two machines.
> >>
> >> if i start varnish it errors out as the startup script does a config
> >> file check.
> >> If i try to do a test of the config file i get the following error.
> >>
> >> root at jhost002:~ # varnishd -C -f /usr/local/etc/varnish/default.vcl
> >> Error: Cannot create working directory '/tmp/varnishd_C_dwbl7mn/': Is
> >> a directory
> >> Error: Cannot create working directory (/tmp/varnishd_C_dwbl7mn/): No
> >> error: 0
> >> (-? gives usage)
> >> root at jhost002:~ #
> >>
> >> This is on:
> >> FreeBSD jhost002.mydomain.nl 13.0-ALPHA3 FreeBSD 13.0-ALPHA3 #35
> >> stable/13-c256281-gc415d0df47fc: Mon Feb  1 17:04:49 CET 2021
> >> root at srv-01.home.local:/usr/obj/usr/src/amd64.amd64/sys/KRNL amd64
> >>
> >> I did not update the package or installed any other software besides
> >> the buildworld.
> >>
> >> regards,
> >> Johan
> >>
> >>
> > I have tried some bisecting as far as my understanding of git goes. I
> > do not know which one is the latest, but on these revisions varnish
> > works.
> >
> > FreeBSD jhost002 13.0-ALPHA3 FreeBSD 13.0-ALPHA3 #8
> > c256261-g9375a93b6c22: Tue Feb  2 13:33:05 CET 2021
> > root at jhost002:/usr/obj/usr/src/amd64.amd64/sys/GENERIC  amd64
> >
> > uname -a
> > FreeBSD jhost002 13.0-ALPHA3 FreeBSD 13.0-ALPHA3 #7
> > c256260-g247f652e622d: Tue Feb  2 13:07:37 CET 2021
> > root at jhost002:/usr/obj/usr/src/amd64.amd64/sys/GENERIC  amd64
> >
> Can you try setting `sysctl vfs.cache_fast_lookup=0` ?
>
> (As suggested by Mateusz elsewhere.)
>

We confirmed that reverting 006ec2ed fixes it, I'm still trying to get
an idea of what's happening. My initial analysis suggests that
fplookup hits the trailing slash and aborts, leaving us at the
degenerate state back up in the slow-path.

I have a small reproducer that I'm working with:
https://people.freebsd.org/~kevans/namereg.c


More information about the freebsd-stable mailing list