[Cluster-devel] [GFS2 PATCH] GFS2: Request demote when a "try" flock fails
Steven Whitehouse
swhiteho at redhat.com
Thu Aug 21 15:51:23 UTC 2014
Hi,
Now in the -nmw git tree. Thanks,
Steve.
On 20/08/14 17:44, Bob Peterson wrote:
> Hi,
>
> This patch changes the flock code so that it uses the TRY_1CB flag
> instead of the TRY flag on the first attempt. That forces any holding
> nodes to issue a dlm callback, which requests a demote of the glock.
> Then, if the "try" failed, it sleeps a small amount of time for the
> demote to occur. Then it tries again, for an increasing amount of time.
> Subsequent attempts to gain the "try" lock don't use "_1CB" so that
> only one callback is issued.
>
> Regards,
>
> Bob Peterson
> Red Hat File Systems
>
> Signed-off-by: Bob Peterson <rpeterso at redhat.com>
> ---
> diff --git a/fs/gfs2/file.c b/fs/gfs2/file.c
> index 26b3f95..7f4ed3d 100644
> --- a/fs/gfs2/file.c
> +++ b/fs/gfs2/file.c
> @@ -26,6 +26,7 @@
> #include <linux/dlm.h>
> #include <linux/dlm_plock.h>
> #include <linux/aio.h>
> +#include <linux/delay.h>
>
> #include "gfs2.h"
> #include "incore.h"
> @@ -979,9 +980,10 @@ static int do_flock(struct file *file, int cmd, struct file_lock *fl)
> unsigned int state;
> int flags;
> int error = 0;
> + int sleeptime;
>
> state = (fl->fl_type == F_WRLCK) ? LM_ST_EXCLUSIVE : LM_ST_SHARED;
> - flags = (IS_SETLKW(cmd) ? 0 : LM_FLAG_TRY) | GL_EXACT;
> + flags = (IS_SETLKW(cmd) ? 0 : LM_FLAG_TRY_1CB) | GL_EXACT;
>
> mutex_lock(&fp->f_fl_mutex);
>
> @@ -1001,7 +1003,14 @@ static int do_flock(struct file *file, int cmd, struct file_lock *fl)
> gfs2_holder_init(gl, state, flags, fl_gh);
> gfs2_glock_put(gl);
> }
> - error = gfs2_glock_nq(fl_gh);
> + for (sleeptime = 1; sleeptime <= 4; sleeptime <<= 1) {
> + error = gfs2_glock_nq(fl_gh);
> + if (error != GLR_TRYFAILED)
> + break;
> + fl_gh->gh_flags = LM_FLAG_TRY | GL_EXACT;
> + fl_gh->gh_error = 0;
> + msleep(sleeptime);
> + }
> if (error) {
> gfs2_holder_uninit(fl_gh);
> if (error == GLR_TRYFAILED)
> @@ -1024,7 +1033,7 @@ static void do_unflock(struct file *file, struct file_lock *fl)
> mutex_lock(&fp->f_fl_mutex);
> flock_lock_file_wait(file, fl);
> if (fl_gh->gh_gl) {
> - gfs2_glock_dq_wait(fl_gh);
> + gfs2_glock_dq(fl_gh);
> gfs2_holder_uninit(fl_gh);
> }
> mutex_unlock(&fp->f_fl_mutex);
>
More information about the Cluster-devel
mailing list