On Mon, 2 Feb 2009 11:20:08 -0700
Jonathan Corbet <corbet@xxxxxxx> wrote:
> Turn struct file->f_flags into an unsigned long and use bitops to change
> its contents. This allows the safe manipulation of these flags without the
> need for locks. BKL use is removed where possible, but manipulations of
> the FASYNC bit still require it (that will be fixed in a later patch).
>
> ...
>
> diff --git a/arch/alpha/include/asm/fcntl.h b/arch/alpha/include/asm/fcntl.h
> index 25da001..de7172f 100644
> --- a/arch/alpha/include/asm/fcntl.h
> +++ b/arch/alpha/include/asm/fcntl.h
> @@ -9,13 +9,18 @@
> #define O_NOCTTY 010000 /* not fcntl */
>
> #define O_NONBLOCK 00004
> +#define NR_O_NONBLOCK 2
> #define O_APPEND 00010
> +#define NR_O_APPEND 3
> #define O_SYNC 040000
> +#define NR_O_SYNC 14
> #define O_DIRECTORY 0100000 /* must be a directory */
> #define O_NOFOLLOW 0200000 /* don't follow links */
> #define O_LARGEFILE 0400000 /* will be set by the kernel on every open */
> #define O_DIRECT 02000000 /* direct disk access - should check with OSF/1 */
> +#define NR_O_DIRECT 19
> #define O_NOATIME 04000000
> +#define NR_O_NOATIME 20
> #define O_CLOEXEC 010000000 /* set close_on_exec */
>
> #define F_GETLK 7
> diff --git a/arch/arm/include/asm/fcntl.h b/arch/arm/include/asm/fcntl.h
> index a80b660..e74b8d1 100644
> --- a/arch/arm/include/asm/fcntl.h
> +++ b/arch/arm/include/asm/fcntl.h
> @@ -4,6 +4,7 @@
> #define O_DIRECTORY 040000 /* must be a directory */
> #define O_NOFOLLOW 0100000 /* don't follow links */
> #define O_DIRECT 0200000 /* direct disk access hint - currently ignored */
> +#define NR_O_DIRECT 16
> #define O_LARGEFILE 0400000
>
> #include <asm-generic/fcntl.h>
hm, it's not exactly a vision of splendour, is it.
I suppose we could/should do
#define O_NONBLOCK (1 << NR_O_NONBLOCK)
but that doesn't prevent people from accidentally doing open-coded &/|
in the future. Perhaps renaming f_flags to f__flags_atomic (stupid
name?) would help.
> +static void tweak_flags_bit(int nr, int on, unsigned long *flags)
> +{
> + if (on)
> + set_bit(nr, flags);
> + else
> + clear_bit(nr, flags);
> +}
I see this construct fairly frequently. I think. Someone(tm) should
check ;) If correct I guess we should have some generic version, put it
in include/linux/bitops.h.
>
> static int setfl(int fd, struct file * filp, unsigned long arg)
> {
> @@ -187,9 +195,17 @@ static int setfl(int fd, struct file * filp, unsigned long arg)
> if (error < 0)
> goto out;
> }
> + change_bit(NR_FASYNC, &filp->f_flags);
> }
>
> - filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
> + /*
> + * Now that we use bitops we have to tweak each bit individually.
> + */
> + tweak_flags_bit(NR_O_APPEND, arg & O_APPEND, &filp->f_flags);
> + tweak_flags_bit(NR_O_NONBLOCK, arg & O_NONBLOCK, &filp->f_flags);
> + tweak_flags_bit(NR_O_NDELAY, arg & O_NDELAY, &filp->f_flags);
> + tweak_flags_bit(NR_O_DIRECT, arg & O_DIRECT, &filp->f_flags);
> + tweak_flags_bit(NR_O_NOATIME, arg & O_NOATIME, &filp->f_flags);
Could pass the file* to tweak_flags_bit() - that would generate less
code. Unless the compiler chooses to inline tweak_flags_bit().
> out:
> unlock_kernel();
> return error;
> diff --git a/fs/ioctl.c b/fs/ioctl.c
> index 240ec63..fc0db36 100644
> --- a/fs/ioctl.c
> +++ b/fs/ioctl.c
> @@ -392,22 +392,24 @@ static int file_ioctl(struct file *filp, unsigned int cmd,
>
> static int ioctl_fionbio(struct file *filp, int __user *argp)
> {
> - unsigned int flag;
> int on, error;
>
> error = get_user(on, argp);
> if (error)
> return error;
> - flag = O_NONBLOCK;
> #ifdef __sparc__
> /* SunOS compatibility item. */
> - if (O_NONBLOCK != O_NDELAY)
> - flag |= O_NDELAY;
> + if (O_NONBLOCK != O_NDELAY) {
> + if (on)
> + set_bit(NR_O_NDELAY, &filp->f_flags);
> + else
> + clear_bit(NR_O_NDELAY, &filp->f_flags);
hey, I recognise that!
> + }
> #endif
> if (on)
> - filp->f_flags |= flag;
> + set_bit(NR_O_NONBLOCK, &filp->f_flags);
> else
> - filp->f_flags &= ~flag;
> + clear_bit(NR_O_NONBLOCK, &filp->f_flags);
and that.
as for the whole patchset: gee, that global lock you had is looking
attractive ;)
--
To unsubscribe from this list: send the line "unsubscribe linux-api" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
