I am running CentOS 5.5 with a 14T ext4 volume. We are sharing out a few sub-directories via NFS. Our customer was doing some penetration testing with our web app that writes to one of those NFS shares. We are not sure if they did something to cause the metadata to grow so large or if it is corrupt. Here is the listing:

drwxrwxr-x 1 owner owner 470M Jun 24 18:15 temp.bad

I guess the metadata could actually be that large, however we have been unable to perform any operations on that directory to determine if the directory is just loaded with files or corrupted. We have not run an fsck on the volume because we would need to schedule downtime for out customers to do so. Has anyone come across this before? Do you have any recommendations to examine more detail on this issue? Thanks.

Mike

Volume in question:
/dev/sdb1 ext4 13T 1.7T 11T 14% /data01

Some question:

1. how many files are there in the directory?
2. what command did you use to know the dimension of the directory? Could you post the output of "ls -lsd <dir>"?

Third, a suggestion... you could try running a readonly fsck, by using the "-n" option... this will tell you if there is any problem, but remember, be careful and test this on another filesystem first! I tried on my /boot with ext3 and worked fine...

What were the parameters during creation of the 14 TB file system? (AFAIR there is a default of 5 % of the file system reserved for the journal up to an upper limit of (...darn, memory fails here )).