Re: [PATCH] NFS - Fix for Infinite loop during syncing

[Steve Dickson <SteveD@redhat.com>]

[-- Attachment #1: Type: text/plain, Size: 361 bytes --]

Steve Dickson wrote:

> Now that I think about it... probably makes sense to set NFS_INO_STALE in
> nfs_commit_done() as well... From my debugging, it just seems the 
> earlier we
> break the inode... the better....

I liked this idea some much that here is the patch for it... I really do 
think
the sooner we know an inode is broken... the better...

steved.

[-- Attachment #2: linux-2.4.28-nfs-syncloop.patch --]
[-- Type: text/plain, Size: 807 bytes --]

--- linux-2.4.28/fs/nfs/write.c.orig	2004-04-14 09:05:40.000000000 -0400
+++ linux-2.4.28/fs/nfs/write.c	2004-12-13 19:04:23.000000000 -0500
@@ -1073,6 +1073,9 @@ nfs_writeback_done(struct rpc_task *task
 			SetPageError(page);
 			if (req->wb_file)
 				req->wb_file->f_error = task->tk_status;
+			if (task->tk_status == -ESTALE)
+				NFS_FLAGS(inode) |= NFS_INO_STALE;
+
 			nfs_inode_remove_request(req);
 			dprintk(", error = %d\n", task->tk_status);
 			goto next;
@@ -1223,6 +1226,9 @@ nfs_commit_done(struct rpc_task *task)
 		if (task->tk_status < 0) {
 			if (req->wb_file)
 				req->wb_file->f_error = task->tk_status;
+			if (task->tk_status == -ESTALE)
+				NFS_FLAGS(inode) |= NFS_INO_STALE;
+
 			nfs_inode_remove_request(req);
 			dprintk(", error = %d\n", task->tk_status);
 			goto next;