When a 'mount -o remount...' is issued, the mount program reads the current
mount options from /etc/mtab so that it can pass them to the helper program. On
newer distros where /etc/mtab is a symlink to /proc/mounts, this will result in
a call to nfs_show_mount_options(), which has the following snippet:
seq_printf(m, ",proto=%s",
rpc_peeraddr2str(nfss->client, RPC_DISPLAY_NETID));
which in most cases will result in the mount program passing 'proto=tcp' to the
mount.nfs4 program, which mount.nfs4 then passes in the data field of the mount
syscall.
nfs_remount() calls nfs_parse_mount_options(), where the NFS_MOUNT_TCP flag will
be set in the nfs_parsed_mount_data's flag field. But nfs_remount() then calls
nfs_compare_remount_data(), where we will fail the very first check:
static int
nfs_compare_remount_data(struct nfs_server *nfss,
struct nfs_parsed_mount_data *data)
{
if (data->flags != nfss->flags ||
...
return -EINVAL;
...
because NFS_MOUNT_TCP was never set in the nfs_server's flags field in the first
place.
As a result, the remount operation fails:
# mount -o remount,ro /mnt/t
mount.nfs4: an incorrect mount option was specified
The following patch corrects that issue by having nfs4_validate_mount_flags()
set the NFS_MOUNT_TCP flag when appropriate.
Scott Mayhew (1):
nfs: Ensure NFS_MOUNT_TCP is correctly set for v4 mounts
fs/nfs/super.c | 3 +++
1 file changed, 3 insertions(+)
--
1.9.0
The NFS_MOUNT_TCP flag needs to be correctly set in the nfs_server->flags
field, otherwise remount operations are likely to run afoul of the flags
check in nfs_compare_remount_data() on systems where /etc/mtab is
symlinked to /proc/mounts.
Signed-off-by: Scott Mayhew <[email protected]>
---
fs/nfs/super.c | 3 +++
1 file changed, 3 insertions(+)
diff --git a/fs/nfs/super.c b/fs/nfs/super.c
index 2cb5694..0ae4e73 100644
--- a/fs/nfs/super.c
+++ b/fs/nfs/super.c
@@ -2726,6 +2726,9 @@ static void nfs4_validate_mount_flags(struct nfs_parsed_mount_data *args)
{
args->flags &= ~(NFS_MOUNT_NONLM|NFS_MOUNT_NOACL|NFS_MOUNT_VER3|
NFS_MOUNT_LOCAL_FLOCK|NFS_MOUNT_LOCAL_FCNTL);
+
+ if (args->nfs_server.protocol == XPRT_TRANSPORT_TCP)
+ args->flags |= NFS_MOUNT_TCP;
}
/*
--
1.9.0
On Thu, May 29, 2014 at 1:25 PM, Scott Mayhew <[email protected]> wrote:
> The NFS_MOUNT_TCP flag needs to be correctly set in the nfs_server->flags
> field, otherwise remount operations are likely to run afoul of the flags
> check in nfs_compare_remount_data() on systems where /etc/mtab is
> symlinked to /proc/mounts.
>
> Signed-off-by: Scott Mayhew <[email protected]>
> ---
> fs/nfs/super.c | 3 +++
> 1 file changed, 3 insertions(+)
>
> diff --git a/fs/nfs/super.c b/fs/nfs/super.c
> index 2cb5694..0ae4e73 100644
> --- a/fs/nfs/super.c
> +++ b/fs/nfs/super.c
> @@ -2726,6 +2726,9 @@ static void nfs4_validate_mount_flags(struct nfs_parsed_mount_data *args)
> {
> args->flags &= ~(NFS_MOUNT_NONLM|NFS_MOUNT_NOACL|NFS_MOUNT_VER3|
> NFS_MOUNT_LOCAL_FLOCK|NFS_MOUNT_LOCAL_FCNTL);
> +
> + if (args->nfs_server.protocol == XPRT_TRANSPORT_TCP)
> + args->flags |= NFS_MOUNT_TCP;
> }
>
> /*
> --
> 1.9.0
>
Hi Scott,
Why can't we just apply the NFS_MOUNT_CMP_FLAGMASK to
nfs_compare_remount_data? NFS_MOUNT_TCP really is just legacy crud
from the bad old days of binary mount structures.
Cheers
Trond
--
Trond Myklebust
Linux NFS client maintainer, PrimaryData
[email protected]