df and "Value too large for defined data type"
Subsystems (2)
Using a (vanilla) 2.6.3 NFS client and using df on a on a particular NFS mounted file system gives:
df: `/mnt/tmp': Value too large for defined data type
/mnt/tmp is mounted from a server running 2.4.21 that has a 3.5TB RAID split into 4 partitions of less than 1TB each. The file system on the server is XFS.
/proc/mounts on the client shows:
lead:/disk1 /mnt/tmp nfs rw,v3,rsize=8192,wsize=8192,hard,udp,lock,addr=lead 0 0
When I use NFSv2, df (or actually statfs()) works as expected - as do mounts to other NFS servers (it appears to be just this one server).
Also, clients running 2.4.X, have no problems with this server.
Any ideas?
James Pearson
-------------------------------------------------------
SF.Net is sponsored by: Speed Start Your Linux Apps Now. Build and deploy apps & Web services for Linux with a free DVD software kit from IBM. Click Now!
http://ads.osdn.com/?ad_id=1356&alloc_id=3438&op=click
_______________________________________________
NFS maillist - [email protected]
https://lists.sourceforge.net/lists/listinfo/nfs
I think I've found part of the reason for this - and it's not an NFS problem - the 'Value too large for defined data type' comes from the 'return -EOVERFLOW' from vfs_statfs_native() in fs/open.c - the value of f_files (total number of nodes in file system) returned by FSSTAT from the server is 2^64 - 1 (18446744073709551615) !!
It looks like 2.4.X doesn't check for this overflow - and just takes the
lower 32 bits.
The underlying remote file system is XFS - which in this case appears to be saying that it has 2^64 - 1 inodes - which doesn't seem correct...
James Pearson
Quoted 24 lines · James Pearson
Using a (vanilla) 2.6.3 NFS client and using df on a on a particular NFS mounted file system gives:
df: `/mnt/tmp': Value too large for defined data type
/mnt/tmpis mounted from a server running 2.4.21 that has a 3.5TB RAID split into 4 partitions of less than 1TB each. The file system on the server is XFS.
/proc/mountson the client shows:lead:/disk1
/mnt/tmpnfs rw,v3,rsize=8192,wsize=8192,hard,udp,lock,addr=lead 0 0When I use NFSv2, df (or actually
statfs()) works as expected - as do mounts to other NFS servers (it appears to be just this one server).Also, clients running 2.4.X, have no problems with this server.
Any ideas?
James Pearson
Quoted 11 lines
-------------------------------------------------------
SF.Net is sponsored by: Speed Start Your Linux Apps Now. Build and deploy apps & Web services for Linux with a free DVD software kit from IBM. Click Now!
http://ads.osdn.com/?ad_id=1356&alloc_id=3438&op=click
_______________________________________________
NFS maillist - [email protected]
https://lists.sourceforge.net/lists/listinfo/nfs
-------------------------------------------------------
SF.Net is sponsored by: Speed Start Your Linux Apps Now. Build and deploy apps & Web services for Linux with a free DVD software kit from IBM. Click Now!
http://ads.osdn.com/?ad_id=1356&alloc_id=3438&op=click
_______________________________________________
NFS maillist - [email protected]
https://lists.sourceforge.net/lists/listinfo/nfs
Quoted 2 lines · James Pearson
The underlying remote file system is XFS - which in this case appears to be saying that it has 2^64 - 1 inodes - which doesn't seem correct...
XFS supports 64 bit inodes, so this can be correct if it's either
1. on Linux, and very recent, and mounted with the right options, or
2. on IRIX
Greg.
Signature · 12 lines
Greg Banks, R&D Software Engineer, SGI Australian Software Group. I don't speak for SGI. ------------------------------------------------------- SF.Net is sponsored by: Speed Start Your Linux Apps Now. Build and deploy apps & Web services for Linux with a free DVD software kit from IBM. Click Now! http://ads.osdn.com/?ad_id=1356&alloc_id=3438&op=click _______________________________________________ NFS maillist - [email protected] https://lists.sourceforge.net/lists/listinfo/nfs
I'll probally send an email to the XFS list about this - the problem is that we've done nothing different with the file system set up on this server compared with the many others we have - i.e. the file systems were set up with:
mkfs -t xfs /dev/sdb1
etc.
It's just that the file systems in question seem to be saying that the number of inodes on the file system is 2^64 - whereas all the other XFS file systems we have (of a similar size) have nowhere near 2^32 inodes
...
However, the one big difference is that the RAID array in question is 3.5TB - which is partioned on the RAID into 4 (i.e. Linux sees 4 devices on different LUNs) - therefore the overall size of the RAID array is a lot bigger than anything else we've used before (we don't have anything else over 2TB) - and the version of xfsprogs used was very old (1.3.17-0) ...
I'm going to remake one of the partitions with a newer version of mkfs.xfs to see if that makes a difference ...
James Pearson
Quoted 7 lines · Greg Banks
Quoted 2 lines · James Pearson
The underlying remote file system is XFS - which in this case appears to be saying that it has 2^64 - 1 inodes - which doesn't seem correct...
XFS supports 64 bit inodes, so this can be correct if it's either
Quoted 8 lines
1. on Linux, and very recent, and mounted with the right options, or
2. on IRIX
Greg.
--
Greg Banks, R&D Software Engineer, SGI Australian Software Group. I don't speak for SGI.
-------------------------------------------------------
SF.Net is sponsored by: Speed Start Your Linux Apps Now. Build and deploy apps & Web services for Linux with a free DVD software kit from IBM. Click Now!
http://ads.osdn.com/?ad_id=1356&alloc_id=3438&op=click
_______________________________________________
NFS maillist - [email protected]
https://lists.sourceforge.net/lists/listinfo/nfs