Re: [PATCH] Drop O_LARGEFILE from F_GETFL for POSIX compliance

[Andrew Morton]

Andi Kleen <ak@xxxxxxx> wrote:
>
> 
> On 64bit architectures open() sets O_LARGEFILE implicitely. This causes the LSB
> testsuite to fail, which checks that F_GETFL only returns the flags set by 
> a previous open.
> 
> According to the POSIX standards gurus the Linux behaviour is not compliant.
> 
> This patch fixes this by just not reporting O_LARGEFILE in F_GETFL.
> 
> This has been in several shipping SuSE releases and the x86-64.org CVS
> treee for a long time, so is unlikely to break anything.
> 
> -Andi
> 
> diff -burpN -X ../KDIFX linux-2.4.26-pre5/fs/fcntl.c linux-merge/fs/fcntl.c
> --- linux-2.4.26-pre5/fs/fcntl.c	2004-01-13 10:29:17.000000000 +0100
> +++ linux-merge/fs/fcntl.c	2003-10-23 15:40:52.000000000 +0200
> @@ -271,7 +271,7 @@ static long do_fcntl(unsigned int fd, un
>  			set_close_on_exec(fd, arg&1);
>  			break;
>  		case F_GETFL:
> -			err = filp->f_flags;
> +			err = filp->f_flags & ~O_LARGEFILE;
>  			break;
>  		case F_SETFL:
>  			lock_kernel();

eh?   If the application on a 64-bit box does

	open("foo", O_LARGEFILE|O_RDWR);

then a subsequent F_GETFL will now return just O_RDWR, will it not?  So
it's still posixly incorrect?

I think open() needs to set O_KERNEL_LARGEFILE, and we mask that off in
F_GETFL, and test for (O_LARGEFILE|O_KERNEL_LARGEFILE) everywhere.

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/