T. Horsnell wrote:
T. Horsnell wrote:
This is just the job. I dont even have to use udev to get
persistent device names if I use LABEL= in fstab. One question
though:
In the absence of an fsid= option, how does Linux compute a
filesystem handle? Unless I know this, or unless I provide
an fsid= option for *all* my exported filesystems, I risk
choosing an id which may collide with the Linux-generated one.
Thanks,
Terry
Trond (nfs maintainer) indicated to someone
else on the kernel list:
stat --printf "%D\n" /filesystem
Roger, thanks for this pointer. My 'stat' (FC6 updated as of today)
seems to have some oddities. To get filesystem-id as opposed to
device-id I have to:
stat --printf "%i\n" -f /filesystem
and when I do this all my filesystem-id's are zero.
stat --printf "%D\n" /filesystem
gets me a device-id which seems to be 256*majorno+minorno
That is the correct numbering I believe. fsid is a generated
id by nfs, it does not actually exist on the filesystem, doing
the %D shows what nfs would generate.
Maybe I should try and contact Trond.
Will print out the fsid for that filesystem on the
nfs server, the stat command does seem to have some
different arguements rather than --printf so if it yells
read the man page and change the arguement.
In situations where the major/minor could change, I add
fsid to all filesystems.
Also note that if you export the same filesystem twice
then it has the same fsid and only one set of permissions
ie both mounts are either rw or ro (probably whichever
is last), and if you use separate fsid's for each, the
permissions for each can now be different.
Do you mean that an /etc/exports which contains:
/fs1 client1(rw,fsid=1) client2(ro,fsid=1)
will be readonly by client1 and client2, whereas
/fs1 client1(rw,fsid=1) client2(ro,fsid=2)
will be rw by client1 and ro by client2 ?
You probably need to separate that on two lines,
maybe, I have not played with it enough to know.
And it may need to be different directories.
I would probably join the NFS list if you are heavily
using NFS, and ask/search their archives.
I wonder what happens with:
/fs1 client1(rw,fsid=1) client2(ro)
or
/fs1 client1(rw) client2(ro,fsid=1)
maybe I'll do the experiment, but I wish there was a spec for
this. Undocumented properties have a habit of changing at the
drop of a kernel version.
Yeah. The fsid stuff is used heavily by the
high-availability stuff so no one will probably touch it
too much.
Roger
