Greetings Neil,
> On Mon, Dec 26 2016, daggs wrote:
>
> > Greetings,
> >
> >> On Mon, Dec 26 2016, daggs wrote:
> >>
> >> >> Can you strace mountd while you attempt a mount?
> >> >> e.g.
> >> >> strace -o /tmp/trace -s 1000 -p 241
> >> >>
> >> >> and send the /tmp/trace.
> >> >> Also, after the attempt fails, run
> >> >> rpcdebug -m rpc -s cache
> >> >> grep . /proc/net/rpc/*/c*
> >> >> cat /proc/fs/nfsd/exports
> >> >>
> >> >> and report the output.
> >> >>
> >> > here:
> >> >
> >> > # cat /tmp/trace
> >> > pselect6(1024, [3 4 5 7 8 9 10 11 12], NULL, NULL, NULL, NULL) = 1 (in [3])
> >> > read(3, "nfsd 10.0.0.1\n", 32768) = 14
> >> > openat(AT_FDCWD, "/run/nfs/etab", O_RDONLY) = 14
> >> > fstat(14, {st_mode=S_IFREG|0644, st_size=435, ...}) = 0
> >> > close(14) = 0
> >> > write(3, "nfsd 10.0.0.1 2079 10.0.0.0/24 \n", 32) = 32
> >>
> >> This is weird.
> >> Here mountd is telling nfsd that when a request comes from IP address
> >> 10.0.0.1, it should look for export entries associated with the client
> >> name "10.0.0.0/24", which is good.
> >> However the expiry time for that information is "2079", which is back in
> >> January 1970.
> >> When mountd writes that number, it computes it as
> >> time(0) + DEFAULT_TTL
> >> where DEFAULT_TTL is (30 * 60)
> >> Which suggests time(0) is "279".
> >>
> >> What is the current time on this system?
> >>
> >> If it really was very early on Jan 1st 1970, it should work, however...
> >>
> >>
> >> > pselect6(1024, [3 4 5 7 8 9 10 11 12], NULL, NULL, NULL, NULL <detached ...>
> >> > # rpcdebug -m rpc -s cache
> >> > rpc cache
> >> > # grep . /proc/net/rpc/*/c*
> >> > /proc/net/rpc/auth.unix.gid/content:#uid cnt: gids...
> >> > /proc/net/rpc/auth.unix.ip/channel:nfsd 10.0.0.1
> >> > /proc/net/rpc/auth.unix.ip/channel:nfsd 10.0.0.1
> >> > /proc/net/rpc/auth.unix.ip/content:#class IP domain
> >> > /proc/net/rpc/auth.unix.ip/content:# expiry=2079 refcnt=1 flags=1
> >> > /proc/net/rpc/auth.unix.ip/content:# nfsd 10.0.0.1 10.0.0.0/24
> >>
> >> ...the fact that this line is commented out indicates that the entry in
> >> the cache is already expired. So the time must be after 2079.
> >>
> >> Maybe the time is getting set from the network at an awkward time that
> >> races with NFS service some how.
> >> Can you find a way to run "exportfs -f" after the time has been set
> >> correctly?
> >>
> >> NeilBrown
> >>
> >>
> >
> > wait, I think I've seen this somewhere, does this feature needs rtc? this board doesn't have rtc component.
> > for example, I cannot use openssh as ssh server because it needs rtc. I have to use dropbear.
> > if so, this looks like it will affect nfsv3 mounts, am I right?
>
> No, you shouldn't need an RTC.
> You need the synchronize the clock with ntp or similar, else time stamps
> on files will look wrong.
> Though I think we fixed issues with wall-clock-time jumping in 2.6.37...
>
> If you could try using "exportfs -f", and explain what does happen with
> time - do you use ntp ?? - we might be able to make progress.
I'll build ntp into the image and try. does this affects nfsv3 too?
what should I do with the "exportfs -f"? jsut run it and retry?
Thanks,
Dagg.
--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
