Subject: Re: O_SYNC vs. O_FSYNC
To: Ted Lemon <firstname.lastname@example.org>
From: Chris G Demetriou <Chris_G_Demetriou@LAGAVULIN.PDL.CS.CMU.EDU>
Date: 05/28/1995 02:27:14
> While we're talking about third party applications and header file
> compatibility, does anybody know why SunOS uses O_SYNC as its flag to
> tell the kernel to do synchronous writes, while NetBSD, BSD/OS and
> probably 4.4BSD use O_FSYNC?
> This particular bugaboo is preventing Kerberos 5 from building on
> NetBSD/pmax and NetBSD/i386. I can obviously add hacks to the
> configure tree for Kerberos 5 to conditionalize the use of O_SYNC
> vs. O_FSYNC, but this will be a good deal more work than simply
> changing the header file.
> Is O_FSYNC in the POSIX spec or something?
I don't have a 1003.1 spec, but i DO have a 1003.1B spec (which is
1003.1 with the "Real Time Extensions" from 1003.4 folded in).
It has a bunch of flags that talk about synchronization:
O_SYNC complete sync, including update times, etc.
O_DSYNC sync writes, "random" stuff like update/access
times needn't be updated synchronously.
O_RSYNC sync reads(?). from what i can tell: don't
return a value until you know the state of the
data, apply pending writes before returning
read data. (seems that this is normal for
UFS's... i'm not quite sure i understand
all P.1b implies...)
>From looking at the changebars, and whatnot, it appears that these
were added in .1B. I'm not sure if they're conditionally defined,
based on a compile-time symbolic constant
(e.g. _POSIX_SYNCHRONIZED_IO), but it doesn't look like they are...