[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: [Cluster-devel] GFS2: Wait for unlock completion on umount



still not good.

setup:

8 nodes cluster (node1-4 x86, node5-8 x86_64)

mounting one partition -> OK
umounting is OOPS´orama.

http://fabbione.fedorapeople.org/oops.tar.bz2

5 out of 8 nodes do OOPS (node1/2/3 are ok and can continue mounting again).

This happens at the very first iteration.

Kernel is Fedora rawhide (2.6.33-rcX...) + -nwm fixes.

Fabio

On 1/14/2010 10:31 AM, Steven Whitehouse wrote:
> 
> This patch adds a wait on umount between the point at which we
> dispose of all glocks and the point at which we unmount the
> lock protocol. This ensures that we've received all the replies
> to our unlock requests before we stop the locking.
> 
> Signed-off-by: Steven Whitehouse <swhiteho redhat com>
> Reported-by: Fabio M. Di Nitto <fdinitto redhat com>
> 
> diff --git a/fs/gfs2/incore.h b/fs/gfs2/incore.h
> index f93f9b9..b8025e5 100644
> --- a/fs/gfs2/incore.h
> +++ b/fs/gfs2/incore.h
> @@ -543,6 +543,8 @@ struct gfs2_sbd {
>  	struct gfs2_holder sd_live_gh;
>  	struct gfs2_glock *sd_rename_gl;
>  	struct gfs2_glock *sd_trans_gl;
> +	wait_queue_head_t sd_glock_wait;
> +	atomic_t sd_glock_disposal;
>  
>  	/* Inode Stuff */
>  
> diff --git a/fs/gfs2/lock_dlm.c b/fs/gfs2/lock_dlm.c
> index 094839e..484411c 100644
> --- a/fs/gfs2/lock_dlm.c
> +++ b/fs/gfs2/lock_dlm.c
> @@ -21,6 +21,7 @@ static void gdlm_ast(void *arg)
>  {
>  	struct gfs2_glock *gl = arg;
>  	unsigned ret = gl->gl_state;
> +	struct gfs2_sbd *sdp = gl->gl_sbd;
>  
>  	BUG_ON(gl->gl_lksb.sb_flags & DLM_SBF_DEMOTED);
>  
> @@ -33,6 +34,8 @@ static void gdlm_ast(void *arg)
>  			kmem_cache_free(gfs2_glock_aspace_cachep, gl);
>  		else
>  			kmem_cache_free(gfs2_glock_cachep, gl);
> +		if (atomic_dec_and_test(&sdp->sd_glock_disposal))
> +			wake_up(&sdp->sd_glock_wait);
>  		return;
>  	case -DLM_ECANCEL: /* Cancel while getting lock */
>  		ret |= LM_OUT_CANCELED;
> @@ -170,7 +173,8 @@ static unsigned int gdlm_lock(struct gfs2_glock *gl,
>  static void gdlm_put_lock(struct kmem_cache *cachep, void *ptr)
>  {
>  	struct gfs2_glock *gl = ptr;
> -	struct lm_lockstruct *ls = &gl->gl_sbd->sd_lockstruct;
> +	struct gfs2_sbd *sdp = gl->gl_sbd;
> +	struct lm_lockstruct *ls = &sdp->sd_lockstruct;
>  	int error;
>  
>  	if (gl->gl_lksb.sb_lkid == 0) {
> @@ -186,6 +190,7 @@ static void gdlm_put_lock(struct kmem_cache *cachep, void *ptr)
>  		       (unsigned long long)gl->gl_name.ln_number, error);
>  		return;
>  	}
> +	atomic_inc(&sdp->sd_glock_disposal);
>  }
>  
>  static void gdlm_cancel(struct gfs2_glock *gl)
> diff --git a/fs/gfs2/ops_fstype.c b/fs/gfs2/ops_fstype.c
> index 968a99f..9baa566 100644
> --- a/fs/gfs2/ops_fstype.c
> +++ b/fs/gfs2/ops_fstype.c
> @@ -81,6 +81,8 @@ static struct gfs2_sbd *init_sbd(struct super_block *sb)
>  
>  	gfs2_tune_init(&sdp->sd_tune);
>  
> +	init_waitqueue_head(&sdp->sd_glock_wait);
> +	atomic_set(&sdp->sd_glock_disposal, 0);
>  	spin_lock_init(&sdp->sd_statfs_spin);
>  
>  	spin_lock_init(&sdp->sd_rindex_spin);
> diff --git a/fs/gfs2/super.c b/fs/gfs2/super.c
> index c008b08..e2bf19f 100644
> --- a/fs/gfs2/super.c
> +++ b/fs/gfs2/super.c
> @@ -21,6 +21,7 @@
>  #include <linux/gfs2_ondisk.h>
>  #include <linux/crc32.h>
>  #include <linux/time.h>
> +#include <linux/wait.h>
>  
>  #include "gfs2.h"
>  #include "incore.h"
> @@ -860,6 +861,8 @@ restart:
>  	/*  Take apart glock structures and buffer lists  */
>  	invalidate_inodes(sdp->sd_vfs);
>  	gfs2_gl_hash_clear(sdp);
> +	/* Wait for dlm to reply to all our unlock requests */
> +	wait_event(sdp->sd_glock_wait, atomic_read(&sdp->sd_glock_disposal) == 0);
>  	/*  Unmount the locking protocol  */
>  	gfs2_lm_unmount(sdp);
>  
> 


[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]