Attachment 712026 Details for Bug 922803 – pnfs: do not reset to mds if wb_offset != wb_pgbase

[patch] pnfs: do not reset to mds if wb_offset != wb_pgbase

0001-pnfs-do-not-reset-to-mds-if-wb_offset-wb_pgbase.patch (text/plain), 2.22 KB, created by Benny Halevy on 2013-03-18 15:01:40 UTC

(hide)

Description:

Filename:

MIME Type:

Creator: Benny Halevy

Created: 2013-03-18 15:01:40 UTC

Size: 2.22 KB

patch

obsolete

>From cfb99ea2af49ad1a8201c15c0b8c0cda0ac35bec Mon Sep 17 00:00:00 2001
>From: Benny Halevy <bhalevy@tonian.com>
>Date: Thu, 14 Mar 2013 07:17:43 +0200
>Subject: [PATCH] pnfs: do not reset to mds if wb_offset != wb_pgbase
>
>We're seeing roughly 20% of the I/Os going to the MDS
>when installing a VM over KVM in "none" caching mode (O_DIRECT).
>Instrumenting the client revealed that this is caused by buffer
>alignment vs. file offset alignment.
>Besides being a performance problem, when the MDS caches data
>this is also manifested as data corruption when data is written
>first via the MDS, then via the DS, eventually the stale data is
>read back from the MDS.
>
>Note that this check exists also for the file layout specific
>pg_init_* functions.
>
>Signed-off-by: Benny Halevy <bhalevy@tonian.com>
>---
> fs/nfs/pnfs.c | 20 ++++++++++++--------
> 1 file changed, 12 insertions(+), 8 deletions(-)
>
>diff --git a/fs/nfs/pnfs.c b/fs/nfs/pnfs.c
>index 793e19f..daf2890 100644
>--- a/fs/nfs/pnfs.c
>+++ b/fs/nfs/pnfs.c
>@@ -1093,10 +1093,12 @@ pnfs_generic_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *r
> {
> 	BUG_ON(pgio->pg_lseg != NULL);
> 
>-	if (req->wb_offset != req->wb_pgbase) {
>-		nfs_pageio_reset_read_mds(pgio);
>-		return;
>-	}
>+	if (req->wb_offset != req->wb_pgbase)
>+		dprintk("%s: inode=%ld: offset=%llu wb_bytes=%u wb_offset=%u wb_pgbase=%u\n",
>+			__func__, pgio->pg_inode->i_ino,
>+			(((unsigned long long)req->wb_index) << PAGE_CACHE_SHIFT) + req->wb_offset,
>+			req->wb_bytes, req->wb_offset, req->wb_pgbase);
>+
> 	pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
> 					   req->wb_context,
> 					   req_offset(req),
>@@ -1115,10 +1117,12 @@ pnfs_generic_pg_init_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *
> {
> 	BUG_ON(pgio->pg_lseg != NULL);
> 
>-	if (req->wb_offset != req->wb_pgbase) {
>-		nfs_pageio_reset_write_mds(pgio);
>-		return;
>-	}
>+	if (req->wb_offset != req->wb_pgbase)
>+		dprintk("%s: inode=%ld: offset=%llu wb_bytes=%u wb_offset=%u wb_pgbase=%u\n",
>+			__func__, pgio->pg_inode->i_ino,
>+			(((unsigned long long)req->wb_index) << PAGE_CACHE_SHIFT) + req->wb_offset,
>+			req->wb_bytes, req->wb_offset, req->wb_pgbase);
>+
> 	pgio->pg_lseg = pnfs_update_layout(pgio->pg_inode,
> 					   req->wb_context,
> 					   req_offset(req),
>-- 
>1.7.11.7
>

Actions: View | Diff

Attachments on bug 922803: 712026