2015-09-04 21:01:53

by Trond Myklebust

Subject: [PATCH v2 1/2] NFS: Optimise away the close-to-open getattr if there is no cached data

If there is no cached data, then there is no need to track the file
change attribute on close.

Signed-off-by: Trond Myklebust <[email protected]>
---
fs/nfs/inode.c | 13 ++++++++++---
1 file changed, 10 insertions(+), 3 deletions(-)

diff --git a/fs/nfs/inode.c b/fs/nfs/inode.c
index 99a68bd9c178..6307d8de103d 100644
--- a/fs/nfs/inode.c
+++ b/fs/nfs/inode.c
@@ -761,11 +761,13 @@ EXPORT_SYMBOL_GPL(nfs_put_lock_context);
* @ctx: pointer to context
* @is_sync: is this a synchronous close
*
- * always ensure that the attributes are up to date if we're mounted
- * with close-to-open semantics
+ * Ensure that the attributes are up to date if we're mounted
+ * with close-to-open semantics and we have cached data that will
+ * need to be revalidated on open.
*/
void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
{
+ struct nfs_inode *nfsi;
struct inode *inode;
struct nfs_server *server;

@@ -774,7 +776,12 @@ void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
if (!is_sync)
return;
inode = d_inode(ctx->dentry);
- if (!list_empty(&NFS_I(inode)->open_files))
+ nfsi = NFS_I(inode);
+ if (inode->i_mapping->nrpages == 0)
+ return;
+ if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
+ return;
+ if (!list_empty(&nfsi->open_files))
return;
server = NFS_SERVER(inode);
if (server->flags & NFS_MOUNT_NOCTO)
--
2.4.3

2015-09-04 21:01:54

by Trond Myklebust

Subject: [PATCH v2 2/2] Revert "NFS: Make close(2) asynchronous when closing NFS O_DIRECT files"

This reverts commit f895c53f8ace3c3e49ebf9def90e63fc6d46d2bf.

This commit causes a NFSv4 regression in that close()+unlink() can end
up failing. The reason is that we no longer have a guarantee that the
CLOSE has completed on the server, meaning that the subsequent call to
REMOVE may fail with NFS4ERR_FILE_OPEN if the server implements Windows
unlink() semantics.

Reported-by: <Olga Kornievskaia <[email protected]>
Cc: Chuck Lever <[email protected]>
Signed-off-by: Trond Myklebust <[email protected]>
---
fs/nfs/inode.c | 7 ++++++-
1 file changed, 6 insertions(+), 1 deletion(-)

diff --git a/fs/nfs/inode.c b/fs/nfs/inode.c
index 6307d8de103d..326d9e10d833 100644
--- a/fs/nfs/inode.c
+++ b/fs/nfs/inode.c
@@ -853,6 +853,11 @@ void put_nfs_open_context(struct nfs_open_context *ctx)
}
EXPORT_SYMBOL_GPL(put_nfs_open_context);

+static void put_nfs_open_context_sync(struct nfs_open_context *ctx)
+{
+ __put_nfs_open_context(ctx, 1);
+}
+
/*
* Ensure that mmap has a recent RPC credential for use when writing out
* shared pages
@@ -908,7 +913,7 @@ void nfs_file_clear_open_context(struct file *filp)
spin_lock(&inode->i_lock);
list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
spin_unlock(&inode->i_lock);
- __put_nfs_open_context(ctx, filp->f_flags & O_DIRECT ? 0 : 1);
+ put_nfs_open_context_sync(ctx);
}
}

--
2.4.3

2015-09-04 21:14:55

by Chuck Lever III

Subject: Re: [PATCH v2 2/2] Revert "NFS: Make close(2) asynchronous when closing NFS O_DIRECT files"

On Sep 4, 2015, at 5:01 PM, Trond Myklebust <[email protected]> wrote:

> This reverts commit f895c53f8ace3c3e49ebf9def90e63fc6d46d2bf.
>
> This commit causes a NFSv4 regression in that close()+unlink() can end
> up failing. The reason is that we no longer have a guarantee that the
> CLOSE has completed on the server, meaning that the subsequent call to
> REMOVE may fail with NFS4ERR_FILE_OPEN if the server implements Windows
> unlink() semantics.
>
> Reported-by: <Olga Kornievskaia <[email protected]>
> Cc: Chuck Lever <[email protected]>

I'm fine with this as long as the current NFSv3 behavior stays
the same. I haven't tested, but I believe 1/2 will guarantee
that NFSv3 close(2) behavior does not change.

Reviewed-by: Chuck Lever <[email protected]>


> Signed-off-by: Trond Myklebust <[email protected]>
> ---
> fs/nfs/inode.c | 7 ++++++-
> 1 file changed, 6 insertions(+), 1 deletion(-)
>
> diff --git a/fs/nfs/inode.c b/fs/nfs/inode.c
> index 6307d8de103d..326d9e10d833 100644
> --- a/fs/nfs/inode.c
> +++ b/fs/nfs/inode.c
> @@ -853,6 +853,11 @@ void put_nfs_open_context(struct nfs_open_context *ctx)
> }
> EXPORT_SYMBOL_GPL(put_nfs_open_context);
>
> +static void put_nfs_open_context_sync(struct nfs_open_context *ctx)
> +{
> + __put_nfs_open_context(ctx, 1);
> +}
> +
> /*
> * Ensure that mmap has a recent RPC credential for use when writing out
> * shared pages
> @@ -908,7 +913,7 @@ void nfs_file_clear_open_context(struct file *filp)
> spin_lock(&inode->i_lock);
> list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
> spin_unlock(&inode->i_lock);
> - __put_nfs_open_context(ctx, filp->f_flags & O_DIRECT ? 0 : 1);
> + put_nfs_open_context_sync(ctx);
> }
> }
>
> --
> 2.4.3
>

--
Chuck Lever