i just wanted to confirm this issue and provide additional information.
i can trigger the bug quite reliably when trying to "mv" directories to
or within an xfs-partition. the last directory was only 40MB containing
8000 files (a git-repo). the "mv" process stays in "D" state, is
unkillable and strace'ing the process makes strace going into "D" state.
fwiw, i've strace'd the strace:
the box is running 2.6.15-rc2 (i386) with CONFIG_PREEMPT_NONE=y and
CONFIG_4KSTACKS=y. more details here:
no raid-controller involved.
BOFH excuse #84:
Someone is standing on the ethernet cable, causing a kink in the cable