I'm using aufs to run a system with aufs on /. The union is made of
several ro squashfs branches, with a rw tmpfs on top.
I've run into a problem where log rotation (repeated delete/recreate of
a single path) causes unbounded disk usage for the xino file. I'm using
aufs cvs 20070709 on mostly-vanilla kernel.org linux 2.6.22.13 i386.
Here is a simple way to reproduce the behavior:
# cd /tmp
# mkdir a b x u
# mount -t tmpfs tmpfs a
# mount -t tmpfs tmpfs x
# mount -t tmpfs -o ro tmpfs b
# mount -t aufs -o br:/tmp/a=rw:/tmp/b=ro,xino=/tmp/x/xino aufs u
# df /tmp/x; cat /sys/fs/aufs/18876c00/xino
Filesystem 1K-blocks Used Available Use% Mounted on
tmpfs 192188 60 192128 1% /tmp/x
8x4096 4096
0: 1, 88x4096 5988068
1: 1, 24x4096 5954360
# while true; do : > /tmp/u/x; rm /tmp/u/x; done
...
^C
# df /tmp/x; cat /sys/fs/aufs/18876c00/xino
Filesystem 1K-blocks Used Available Use% Mounted on
tmpfs 192188 244 191944 1% /tmp/x
8x4096 4096
0: 1, 456x4096 6174460
1: 1, 24x4096 5954360
The longer I wait, the more space is used for the xino file on /tmp/x.
Is this normal? Is there any limit to the amount of space used? If not,
this is a disk space leak. I found this problem because I ran into
systems that were using 10 times as much space for xino as for the
actual files in the rw branch.
Does it even make sense to use xino when the writable branch is tmpfs?
thanks,
Jason
-------------------------------------------------------------------------
This SF.net email is sponsored by: Microsoft
Defy all challenges. Microsoft(R) Visual Studio 2005.
http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/
