Btrfs devel - Mar 2010 - [PATCH] Btrfs: force delalloc flushing when things get desperate

When testing with max_extents=4k, we enospc out really really early.  The reason
for this is we really overwhelm the system with our worst case calculation.
When we try to flush delalloc, we don''t want everybody to wait around
forever,
so we wake up the waiters when we''ve done some of the work in hopes
that its
enough work to get everything they need done.  The problem with this is we
don''t
wait long enough sometimes.  So if we''ve already done a flush_delalloc
and
didn''t find what we need, do it again and this time wait for the
flushing to be
completely finished before returning.  This makes my ENOSPC test actually
finish, instead of finishing in about 20 seconds.  Thanks,

Signed-off-by: Josef Bacik <josef@redhat.com>
---
 fs/btrfs/extent-tree.c |   25 +++++++++++++++++--------
 1 files changed, 17 insertions(+), 8 deletions(-)

diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index 0085dcb..aeef481 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -2873,7 +2873,7 @@ static noinline void flush_delalloc_async(struct
btrfs_work *work)
 	kfree(async);
 }
 
-static void wait_on_flush(struct btrfs_root *root, struct btrfs_space_info
*info)
+static void wait_on_flush(struct btrfs_root *root, struct btrfs_space_info
*info, int soft)
 {
 	DEFINE_WAIT(wait);
 	u64 num_bytes;
@@ -2895,6 +2895,12 @@ static void wait_on_flush(struct btrfs_root *root, struct
btrfs_space_info *info
 			break;
 		}
 
+		if (!soft) {
+			spin_unlock(&info->lock);
+			schedule();
+			continue;
+		}
+
 		free = 0;
 		for_each_possible_cpu(i) {
 			struct btrfs_reserved_space_pool *pool;
@@ -2924,7 +2930,7 @@ static void wait_on_flush(struct btrfs_root *root, struct
btrfs_space_info *info
 }
 
 static void flush_delalloc(struct btrfs_root *root,
-				 struct btrfs_space_info *info)
+			   struct btrfs_space_info *info, int soft)
 {
 	struct async_flush *async;
 	bool wait = false;
@@ -2939,7 +2945,7 @@ static void flush_delalloc(struct btrfs_root *root,
 	spin_unlock(&info->lock);
 
 	if (wait) {
-		wait_on_flush(root, info);
+		wait_on_flush(root, info, soft);
 		return;
 	}
 
@@ -2953,7 +2959,7 @@ static void flush_delalloc(struct btrfs_root *root,
 
 	btrfs_queue_worker(&root->fs_info->enospc_workers,
 			   &async->work);
-	wait_on_flush(root, info);
+	wait_on_flush(root, info, soft);
 	return;
 
 flush:
@@ -3146,14 +3152,17 @@ again:
 
 	if (!delalloc_flushed) {
 		delalloc_flushed = true;
-		flush_delalloc(root, meta_sinfo);
+		flush_delalloc(root, meta_sinfo, 1);
 		goto again;
 	}
 
 	if (!chunk_allocated) {
+		int ret;
+
 		chunk_allocated = true;
-		btrfs_wait_ordered_extents(root, 0, 0);
-		maybe_allocate_chunk(root, meta_sinfo);
+		ret = maybe_allocate_chunk(root, meta_sinfo);
+		if (!ret)
+			flush_delalloc(root, meta_sinfo, 0);
 		goto again;
 	}
 
@@ -3338,7 +3347,7 @@ again:
 
 	if (!delalloc_flushed) {
 		delalloc_flushed = true;
-		flush_delalloc(root, meta_sinfo);
+		flush_delalloc(root, meta_sinfo, 0);
 		goto again;
 	}
 
-- 
1.6.6

--
To unsubscribe from this list: send the line "unsubscribe linux-btrfs"
in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

On Fri, Mar 12, 2010 at 04:23:09PM -0500, Josef Bacik
wrote:
> When testing with max_extents=4k, we enospc out really really early.  The
reason
> for this is we really overwhelm the system with our worst case calculation.
> When we try to flush delalloc, we don''t want everybody to wait
around forever,
> so we wake up the waiters when we''ve done some of the work in
hopes that its
> enough work to get everything they need done.  The problem with this is we
don''t
> wait long enough sometimes.  So if we''ve already done a
flush_delalloc and
> didn''t find what we need, do it again and this time wait for the
flushing to be
> completely finished before returning.  This makes my ENOSPC test actually
> finish, instead of finishing in about 20 seconds.  Thanks,
Thanks Josef, was this one against the per-cpu work?  It doesn''t apply
cleanly but is simple enough that I can just bang it in there ;)

-chris
> 
> Signed-off-by: Josef Bacik <josef@redhat.com>
> ---
>  fs/btrfs/extent-tree.c |   25 +++++++++++++++++--------
>  1 files changed, 17 insertions(+), 8 deletions(-)
> 
> diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
> index 0085dcb..aeef481 100644
> --- a/fs/btrfs/extent-tree.c
> +++ b/fs/btrfs/extent-tree.c
> @@ -2873,7 +2873,7 @@ static noinline void flush_delalloc_async(struct
btrfs_work *work)
>  	kfree(async);
>  }
>  
> -static void wait_on_flush(struct btrfs_root *root, struct btrfs_space_info
*info)
> +static void wait_on_flush(struct btrfs_root *root, struct btrfs_space_info
*info, int soft)
>  {
>  	DEFINE_WAIT(wait);
>  	u64 num_bytes;
> @@ -2895,6 +2895,12 @@ static void wait_on_flush(struct btrfs_root *root,
struct btrfs_space_info *info
>  			break;
>  		}
>  
> +		if (!soft) {
> +			spin_unlock(&info->lock);
> +			schedule();
> +			continue;
> +		}
> +
>  		free = 0;
>  		for_each_possible_cpu(i) {
>  			struct btrfs_reserved_space_pool *pool;
> @@ -2924,7 +2930,7 @@ static void wait_on_flush(struct btrfs_root *root,
struct btrfs_space_info *info
>  }
>  
>  static void flush_delalloc(struct btrfs_root *root,
> -				 struct btrfs_space_info *info)
> +			   struct btrfs_space_info *info, int soft)
>  {
>  	struct async_flush *async;
>  	bool wait = false;
> @@ -2939,7 +2945,7 @@ static void flush_delalloc(struct btrfs_root *root,
>  	spin_unlock(&info->lock);
>  
>  	if (wait) {
> -		wait_on_flush(root, info);
> +		wait_on_flush(root, info, soft);
>  		return;
>  	}
>  
> @@ -2953,7 +2959,7 @@ static void flush_delalloc(struct btrfs_root *root,
>  
>  	btrfs_queue_worker(&root->fs_info->enospc_workers,
>  			   &async->work);
> -	wait_on_flush(root, info);
> +	wait_on_flush(root, info, soft);
>  	return;
>  
>  flush:
> @@ -3146,14 +3152,17 @@ again:
>  
>  	if (!delalloc_flushed) {
>  		delalloc_flushed = true;
> -		flush_delalloc(root, meta_sinfo);
> +		flush_delalloc(root, meta_sinfo, 1);
>  		goto again;
>  	}
>  
>  	if (!chunk_allocated) {
> +		int ret;
> +
>  		chunk_allocated = true;
> -		btrfs_wait_ordered_extents(root, 0, 0);
> -		maybe_allocate_chunk(root, meta_sinfo);
> +		ret = maybe_allocate_chunk(root, meta_sinfo);
> +		if (!ret)
> +			flush_delalloc(root, meta_sinfo, 0);
>  		goto again;
>  	}
>  
> @@ -3338,7 +3347,7 @@ again:
>  
>  	if (!delalloc_flushed) {
>  		delalloc_flushed = true;
> -		flush_delalloc(root, meta_sinfo);
> +		flush_delalloc(root, meta_sinfo, 0);
>  		goto again;
>  	}
>  
> -- 
> 1.6.6
> 
> --
> To unsubscribe from this list: send the line "unsubscribe
linux-btrfs" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
--
To unsubscribe from this list: send the line "unsubscribe linux-btrfs"
in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

On Tue, Mar 30, 2010 at 09:25:54PM -0400, Chris Mason
wrote:
> On Fri, Mar 12, 2010 at 04:23:09PM -0500, Josef Bacik wrote:
> > When testing with max_extents=4k, we enospc out really really early. 
The reason
> > for this is we really overwhelm the system with our worst case
calculation.
> > When we try to flush delalloc, we don''t want everybody to
wait around forever,
> > so we wake up the waiters when we''ve done some of the work in
hopes that its
> > enough work to get everything they need done.  The problem with this
is we don''t
> > wait long enough sometimes.  So if we''ve already done a
flush_delalloc and
> > didn''t find what we need, do it again and this time wait for
the flushing to be
> > completely finished before returning.  This makes my ENOSPC test
actually
> > finish, instead of finishing in about 20 seconds.  Thanks,
> 
> Thanks Josef, was this one against the per-cpu work?  It doesn''t
apply
> cleanly but is simple enough that I can just bang it in there ;)
> 
Yeah I think it was, I can redo it if you like.  It''s not as big of a
deal
without max_extent, but it could still be usefull for some corner cases.
Thanks,

Josef
--
To unsubscribe from this list: send the line "unsubscribe linux-btrfs"
in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html