I'm just curious --- what is your use case for wanting to use Direct
I/O in mke2fs? In general, using buffered I/O is much faster, and the
historically, Direct I/O option was primarily used as a workaround for
kernels (long ago) that were buggy with respect to write throttling,
and userspace programs which issued a large number of number of
buffered writes, especially in a constrained memory cgroup, could
potentially get OOM-killed.
What the kernel *should* do is to throttled the buffered writes, by
cause the buffered write(2) to block until there is enough free memory
in the cgroup so the user space program doesn't get OOM-killed.
I don't believe modern kernels has had write throttling bugs in quite
a while, however.
One of the reasons I didn't bother backporting the bug to Debian
Stable was because I had assumed the -D option isn't commonly used.
That's why I'm curious why you were using the -D option and ran across
this bug.
Thanks,
- Ted
