[Ocfs2-devel] 2.6 I/O path totally messed up

[Christoph Hellwig]

The second argumentto generic_file_write_nolock is an iovec, not an
actual userbuffer.  The current code doesn't even have the slightest
chance to work.  Maybe it's time to decouple the 2.6 read/write code
from the 2.4 code and implement proper vectored operations everywhere?

(and kill tge O_DIRECT vs O_APPEN hack, and the broken fallback code,
 and..)