Trond Myklebust wrote: > On Sat, 2008-01-05 at 23:35 +0200, Timo Sirainen wrote: >> On 5.1.2008, at 17.28, Trond Myklebust wrote: >> >>> On Sat, 2008-01-05 at 05:07 +0100, Michael Kerrisk wrote: >>>> I think you missed my question on O_APPEND though? >>> Oh, sorry... >>> >>> O_APPEND is supported for normal writes, but not for O_DIRECT writes. >>> The reason for this is that the NFS protocol does not have the >>> equivalent of an atomic 'append()' RPC request. >> I'd rather say that O_APPEND is "emulated", not "supported". For >> example if two NFS clients try to append to the same log file, >> there's no way to do that with O_APPEND without the possibility of >> file corruption. > > I strongly disagree. By 'normal' writes we mean _cached_ writes. There > is no support within that model for having two clients writing to the > same file.
Hi Trond, I suspect you and Timo are coming at this from different angles, and I think Timo's angle is closer to what I was thinking of when I asked my question. What I'm wondering about is this: two naive programs that use O_APPEND, unaware that the file they are writing to is on NFS. Will O_APPEND work in the expected way. It sounds to me as though the answer is (still, even with NFSv4) no, right? Cheers, Michael -- Michael Kerrisk Maintainer of the Linux man-pages project http://www.kernel.org/doc/man-pages/ Want to report a man-pages bug? Look here: http://www.kernel.org/doc/man-pages/reporting_bugs.html -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]