Yes, this is not totally surprising, XFS has an inode revalidate operation,
the regular interfaces such as the stat call will call this, NFS appears
not to use this at all, and assumes that the linux inode always contains
all the correct information.
See what this patch does for you.
> I think I have found a bug somewhere between XFS and knfsd. If anyone has
> seen this or if I am missing something, I would appreciate the help.
> On a dual PIII machine, I have XFS installed on a 130G IDE raid array. I
> want to set the machine up as a NFS server. Of all the available
> journalling filesystems, XFS seems to be the most complete with my set of
> criteria. (Backup, works with knfsd, journals, relatively stable.)
> The problem occured when I tried to use XFS over NFS to compile something.
> Time stamps on files are not getting updated. If I run touch explicitly,
> then the time is updated, but if I do something like
> cp /dev/null filename
> it keeps the old date, despite the fact that the file was truncated to
> zero. Scripts that recreate files if they are out of date do not work
> correctly because rewritten files do not get the current date. The date
> is not even updated to the current date. Also, if I do the following:
> cp my_old_file my_other_old_file
> The time stamp of my_other_old_file is also not updated. These problems
> do not occur if I do the same commands on the local machine. The problems
> also do not occur on ext2 filesystems that are exported from the same
> There is one other odd behavior. It seems that in the underlying XFS file
> system the time stamps are being updated. It is just not getting back to
> the NFS server :(. If I do one of the above commands, log into the NFS
> server and look there, the date is correct. After doing that if I look at
> the NFS exported version it is correct. So, I could continuously run
> find processes to make sure that the files have been recenly re-read, but
> I don't think that is a very good solution.
> The clients I have used have been 2.4.2,2.4.3-XFS,2.2.* and even HPUX.
> They all seem to exhibit the same behavior. An ext2 exported file system
> does not have the errors, so I am led to believe that there is some
> internal linux interaction between XFS and the kernel nfs server.
> Any ideas?
> Justin Leonard Tripp jtripp at i(e)^3 org
> Configurable Computing Laboratory Research Assistant CB 461 x8-7206
> Electrical and Computer Engineering Department Brigham Young University