shadow@clusterfs.com
2007-Feb-01 06:23 UTC
[Lustre-devel] [Bug 10786] NFS export and use of fsid
Please don''t reply to lustre-devel. Instead, comment in Bugzilla by using the following link: https://bugzilla.lustre.org/show_bug.cgi?id=10786 today i finished investigate this problem... problem is more complex then only lustre problem.>From other posts i see you use RHEL kernels - if i right root of problem innfs-tools and knfsd. nfs-tools from RHEL4 can`t correctly read data from procfs in 2.6 kernels. If MINOR or MAJOR number of device exceed 256 - 2.6 kernel use fsid type 3 to communicate with rpc.mountd, but rpc.mountd from RHEL4 (nfs-utils 1.0.6-70.EL) can`t read data in this format (can be read only type 0,1). After fix this in rpc.mountd, i found other bug in knfsd. knfsd can be read data only with fsid type 1, other types is ignored. after fixing this isses i`m able to correctly export lustre file system without set fsid in /etc/exports. Other distros can be have similar problems. problem don`t exist with rpc.mountd from nfs-utils v1.0.10 and in kernel after 2.6.16 (i inspect SLES10 kernel).
shadow@clusterfs.com
2007-Feb-01 09:13 UTC
[Lustre-devel] [Bug 10786] NFS export and use of fsid
Please don''t reply to lustre-devel. Instead, comment in Bugzilla by using the following link: https://bugzilla.lustre.org/show_bug.cgi?id=10786 Created an attachment (id=9473) Please don''t reply to lustre-devel. Instead, comment in Bugzilla by using the following link: --> (https://bugzilla.lustre.org/attachment.cgi?id=9473&action=view) patch for kernel fs/nfsd/export.c
shadow@clusterfs.com
2007-Feb-01 09:15 UTC
[Lustre-devel] [Bug 10786] NFS export and use of fsid
Please don''t reply to lustre-devel. Instead, comment in Bugzilla by using the following link: https://bugzilla.lustre.org/show_bug.cgi?id=10786 Created an attachment (id=9474) Please don''t reply to lustre-devel. Instead, comment in Bugzilla by using the following link: --> (https://bugzilla.lustre.org/attachment.cgi?id=9474&action=view) patch for nfs-tools obtain from sles10.
shadow@clusterfs.com
2007-Feb-01 11:45 UTC
[Lustre-devel] [Bug 10786] NFS export and use of fsid
Please don''t reply to lustre-devel. Instead, comment in Bugzilla by using the following link: https://bugzilla.lustre.org/show_bug.cgi?id=10786 Created an attachment (id=9475) Please don''t reply to lustre-devel. Instead, comment in Bugzilla by using the following link: --> (https://bugzilla.lustre.org/attachment.cgi?id=9475&action=view) remove necessity set fsid for lustre exports patch set flag FS_REQUIRES_DEV to lustre file system type. it`s safe for 2.6 kernel because none check it`s flag in kernel, except kNFSD (export.c). nfs failover enabled by default for kernels up from 2.6.16 or any 2.4 kernel (witch correctly exchange with mountd), or with set flag --enable-nfs-failover.
shadow@clusterfs.com
2007-Feb-02 02:43 UTC
[Lustre-devel] [Bug 10786] NFS export and use of fsid
Please don''t reply to lustre-devel. Instead, comment in Bugzilla by using the following link: https://bugzilla.lustre.org/show_bug.cgi?id=10786 just for reference. Please don''t reply to lustre-devel. Instead, comment in Bugzilla by using the following link: https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=226746
green@clusterfs.com
2007-Feb-06 20:49 UTC
[Lustre-devel] [Bug 10786] NFS export and use of fsid
Please don''t reply to lustre-devel. Instead, comment in Bugzilla by using the following link: https://bugzilla.lustre.org/show_bug.cgi?id=10786 (From update of attachment 9475)>+# check kernel can support nfs failover >+AC_DEFUN([LC_CORRECT_NFSD], >+[AC_MSG_CHECKING([check nfsd can support failover]) >+LB_LINUX_TRY_COMPILE([ >+ #include <linux/version.h> >+],[ >+ #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0)) && (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,16)) >+ #error "this version nfsd can\`t support export non standart devices" >+ #endifI tend to agree with Andreas that check like this is not enough due to patched kernels. Of course it is hard to find out if there is no kernel source, too. (and if we were to use code like this, no point in configure check, just put that same #if into lustre sources)
shadow@clusterfs.com
2007-Feb-08 00:58 UTC
[Lustre-devel] [Bug 10786] NFS export and use of fsid
Please don''t reply to lustre-devel. Instead, comment in Bugzilla by using the following link: https://bugzilla.lustre.org/show_bug.cgi?id=10786 Created an attachment (id=9541) Please don''t reply to lustre-devel. Instead, comment in Bugzilla by using the following link: --> (https://bugzilla.lustre.org/attachment.cgi?id=9541&action=view) remove necessity set fsid for lustre exports merge fixes from bug 9306. add check CFS kernel patch into autoconf macro.