Hi,
On Fri 11-09-09 21:16:00, Christoph Hellwig wrote:
> While Linux provided an O_SYNC flag basically since day 1, it took until
> Linux 2.4.0-test12pre2 to actually get it implemented for filesystems,
> since that day we had generic_osync_around with only minor changes and the
> great "For now, when the user asks for O_SYNC, we'll actually give O_DSYNC"
> comment. This patch intends to actually give us real O_SYNC semantics
> in addition to the O_DSYNC semantics. After Jan's O_SYNC patches which
> are required before this patch it's actually surprisingly simple, we
> just need to figure out when to set the datasync flag to vfs_fsync_range
> and when not.
>
> This patch renames the existing O_SYNC flag to O_DSYNC while keeping
> it's numerical value to keep binary compatibility, and adds a new real
> O_SYNC flag. To guarantee backwards compatiblity it is defined as
> expanding to both the O_DSYNC and the new additional binary flag
> (__O_SYNC) to make sure we are backwards-compatible when compiled against
> the new headers.
>
> This also means that all places that don't care about the differences
> can just check O_DSYNC and get the right behaviour for O_SYNC, too - only
> places that actuall care need to check __O_SYNC in addition. Drivers
> and network filesystems have been updated in a fail safe way to always
> do the full sync magic if O_DSYNC is set. The few places setting O_SYNC
> for lower layers are kept that way for now to stay failsafe.
>
> Note that parisc really fucked up their headers as they already define
> a O_DSYNC that has always been a no-op. We try to repair it by using it
> for the new O_DSYNC and redefinining O_SYNC to send both the traditional
> O_SYNC numerical value _and_ the O_DSYNC one.
I've sent Linus a pull request without this patch (I have some comments
to it). When this patch is ready, you can merge it yourself or I can do
it if you like.
> Index: linux-2.6/fs/afs/write.c
> ===================================================================
> --- linux-2.6.orig/fs/afs/write.c 2009-09-10 21:02:06.710003950 -0300
> +++ linux-2.6/fs/afs/write.c 2009-09-11 16:11:50.439008144 -0300
> @@ -692,8 +692,9 @@ ssize_t afs_file_write(struct kiocb *ioc
> }
>
> /* return error values for O_SYNC and IS_SYNC() */
> - if (IS_SYNC(&vnode->vfs_inode) || iocb->ki_filp->f_flags & O_SYNC) {
> - ret = afs_fsync(iocb->ki_filp, dentry, 1);
> + if (IS_SYNC(&vnode->vfs_inode) || iocb->ki_filp->f_flags & O_DSYNC) {
> + ret = afs_fsync(iocb->ki_filp, dentry,
> + (iocb->ki_filp->f_flags & __O_SYNC) ? 0 : 1);
> if (ret < 0)
> result = ret;
> }
This code can go away because generic_file_aio_write() already calls
fsync()...
> Index: linux-2.6/arch/mips/include/asm/fcntl.h
> ===================================================================
> --- linux-2.6.orig/arch/mips/include/asm/fcntl.h 2009-09-10 21:02:06.443262027 -0300
> +++ linux-2.6/arch/mips/include/asm/fcntl.h 2009-09-11 16:11:50.495015560 -0300
> @@ -10,7 +10,7 @@
>
>
> #define O_APPEND 0x0008
> -#define O_SYNC 0x0010
> +#define O_DSYNC 000010 /* used to be O_SYNC, see below */
The value used to be in hex, not in octal. Moreover I don't see O_SYNC
defined in the header now...
> Index: linux-2.6/arch/mips/kernel/kspd.c
> ===================================================================
> --- linux-2.6.orig/arch/mips/kernel/kspd.c 2009-09-10 21:02:06.465005782 -0300
> +++ linux-2.6/arch/mips/kernel/kspd.c 2009-09-11 16:11:50.499009085 -0300
> @@ -82,6 +82,7 @@ static int sp_stopping = 0;
> #define MTSP_O_SHLOCK 0x0010
> #define MTSP_O_EXLOCK 0x0020
> #define MTSP_O_ASYNC 0x0040
> +/* XXX: check which of these is actually O_SYNC vs O_DSYNC */
> #define MTSP_O_FSYNC O_SYNC
> #define MTSP_O_NOFOLLOW 0x0100
> #define MTSP_O_SYNC 0x0080
Since noone uses MTSP_O_FSYNC and it's not exported, I guess it's your
choice ;). Looking at the code, it looks slightly incomplete - probably
open_flags_table should contain all the MTSP_O_... flags but I don't really
know.
> Index: linux-2.6/arch/parisc/include/asm/fcntl.h
> ===================================================================
> --- linux-2.6.orig/arch/parisc/include/asm/fcntl.h 2009-09-10 21:02:06.618023193 -0300
> +++ linux-2.6/arch/parisc/include/asm/fcntl.h 2009-09-11 16:11:50.512006342 -0300
> @@ -1,14 +1,13 @@
> #ifndef _PARISC_FCNTL_H
> #define _PARISC_FCNTL_H
>
> -/* open/fcntl - O_SYNC is only implemented on blocks devices and on files
> - located on an ext2 file system */
> #define O_APPEND 000000010
> #define O_BLKSEEK 000000100 /* HPUX only */
> #define O_CREAT 000000400 /* not fcntl */
> #define O_EXCL 000002000 /* not fcntl */
> #define O_LARGEFILE 000004000
> -#define O_SYNC 000100000
> +#define __O_SYNC 000100000
> +#define O_SYNC (__O_SYNC|O_DSYNC)
> #define O_NONBLOCK 000200004 /* HPUX has separate NDELAY & NONBLOCK */
> #define O_NOCTTY 000400000 /* not fcntl */
> #define O_DSYNC 001000000 /* HPUX only */
So for parisc, programs compiled against old headers will fail open
O_SYNC because of the check in open() you've added will bail out with
EINVAL. I don't like it but I'm not sure we can do better...
> Index: linux-2.6/fs/sync.c
> ===================================================================
> --- linux-2.6.orig/fs/sync.c 2009-09-11 16:11:49.725278522 -0300
> +++ linux-2.6/fs/sync.c 2009-09-11 16:11:50.516015792 -0300
> @@ -287,10 +287,11 @@ SYSCALL_DEFINE1(fdatasync, unsigned int,
> */
> int generic_write_sync(struct file *file, loff_t pos, loff_t count)
> {
> - if (!(file->f_flags & O_SYNC) && !IS_SYNC(file->f_mapping->host))
> + if (!(file->f_flags & O_DSYNC) && !IS_SYNC(file->f_mapping->host))
> return 0;
> return vfs_fsync_range(file, file->f_path.dentry, pos,
> - pos + count - 1, 1);
> + pos + count - 1,
> + (file->f_flags & __O_SYNC) ? 1 : 0);
The logic is inverted here, isn't it?
Honza
--
Jan Kara <jack@xxxxxxx>
SUSE Labs, CR
--
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
