On Mon, 2007-10-29 at 14:16 +0000, Massimo Mongardini wrote: > Ian Kent wrote: > > On Fri, 2007-10-26 at 12:59 +0100, Massimo Mongardini wrote: > > > > > Hi everyone, > > > I am experiencing an issue in the following circumstances: > > > > > > os: rhel4.5 > > > kernel client: 2.6.9-55.0.9.ELsmp > > > kernel server: 2.6.9-55.0.2.ELsmp > > > autofs version 4.1.3-199.3 > > > > > > server exports folder /nfs/folder with parameters: > > > async,rw,fsid=1,no_root_squash > > > > > > client automounts /nfs/folder with: > > > > > > auto.master: > > > /nfs /etc/auto.nfs > > > > > > auto.nfs: > > > folder -fstype=nfs,udp,rsize=32768,wsize=32768,noatime > > > fs-cluster:/nfs/folder > > > > > > default values for automount are --timeout 60 > > > > > > If I run a script that checks with a series of increasing sleep > > > intervals < 120 if an exported file is executable, > > > I notice that if there is a coincidence between an unmount request and a > > > file request the file test fails > > > > > > > This sounds like the mount/expire race patch and it's friends that > > aren't present in that kernel. > > > > There have been a few issues discussed around these patches (or the lack > > of them) on this list lately. > > > > > > > Thu Oct 25 17:00:28 GMT 2007 Error at DELAY: 38 > > > Oct 25 17:00:28 server mountd[8014]: authenticated unmount request from > > > client:989 for /nfs/folder (/nfs/folder) > > > Fri Oct 26 04:48:41 GMT 2007 Error at DELAY: 25 > > > Oct 26 04:48:41 server mountd[8014]: authenticated unmount request from > > > client:846 for /nfs/folder (/nfs/folder) > > > > > > the test script looks like this: > > > > > > EXECUTABLE="/nfs/folder/somepath/exec" > > > DELAY="1" > > > STOPCOUNT=120 > > > OUT=$1 > > > while true > > > do > > > while [ "$DELAY" -lt "$STOPCOUNT" ] > > > do > > > test -x $EXECUTABLE > > > if [ "$?" -ne "0" ] > > > then > > > echo "$(date) Error at DELAY: $DELAY" >> $OUT > > > else > > > echo "$(date) Executable found at DELAY: $DELAY" >> $OUT > > > fi > > > sleep $DELAY > > > (( DELAY++ )) > > > done > > > done > > > > > > any help is much appreciated > > > > > > cheers, > > > Massimo > > > > > > > > > > > Ian, > thanks for your answer. > I saw that bug but I thought was just related to autofs4. Since the > client is running a maybe-patched 2.6.9-55.0.9.ELsmp version I thought > would be wnough. Am I missing something or some useful post? > > $ rpm -q --changelog kernel-smp-2.6.9-55.0.9.EL | grep -e autofs | > grep race > -revert: autofs4 fix for race between mount and expire [248126]
This is the bit you should notice. The needed patches just aren't present. > -autofs4: fix race between mount and expire (Ian Kent) [248126] > -autofs4: fix race between mount/expire (Jeff Moyer) [175778] > > The server has just: > $ rpm -q --changelog kernel-smp-2.6.9-55.0.2.EL | grep -e autofs | > grep race > -autofs4: fix race between mount/expire (Jeff Moyer) [175778] And this version would be OK if it had the needed corrections. Ian _______________________________________________ autofs mailing list [email protected] http://linux.kernel.org/mailman/listinfo/autofs
