NFS: Fix byte accounting for generic NFS reads

Currently, the NFS I/O counters count the number of bytes requested
by applications, rather than the number of bytes actually read by the
system calls.

The number of bytes requested for reads is actually not that useful,
because the value is usually a buffer size for reads.  That is, that
requested number is usually a maximum, and frequently doesn't reflect
the actual number of bytes read.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>

diff --git a/fs/nfs/file.c b/fs/nfs/file.c
index 57cf94f..7f4910c 100644 (file)
--- a/fs/nfs/file.c
+++ b/fs/nfs/file.c
@@ -262,9 +262,11 @@ nfs_file_read(struct kiocb *iocb, const struct iovec *iov,
                (unsigned long) count, (unsigned long) pos);
 
        result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
-       nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, count);
-       if (!result)
+       if (!result) {
                result = generic_file_aio_read(iocb, iov, nr_segs, pos);
+               if (result > 0)
+                       nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result);
+       }
        return result;
 }