Copyright (c) 2009 Rick Macklem, University of Guelph All rights reserved. Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met: 1. Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer. 2. Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the docu...
NAMENFSv4 - NFS Version 4 Protocol
DESCRIPTIONThe NFS client and server provides support for the NFSv4 specification; see "Network File System (NFS) Version 4 Protocol RFC 3530" . The protocol is somewhat similar to NFS Version 3, but differs in significant ways. It uses a single compound RPC that concatenates operations to-gether. Each of these operations are similar to the RPCs of NFS Version 3. The operations in the compound are performed in order, until one of them fails (returns an error) and then the RPC terminates at that point.
It has integrated locking support, which implies that the server is no longer stateless. As such, the sysctl Cm net.inet.tcp.syncookies server remains in recovery mode for a grace period (always greater than the lease duration the server uses) after a reboot. During this grace period, clients may recover state but not perform other open/lock state changing operations. To provide for correct recovery semantics, a small file described by stablerestart(5) is used by the server during the recovery phase. If this file is missing or empty, there is a backup copy maintained by nfsd(8) that will be used. If either file is missing, they will be created by the nfsd(8). If both the file and the backup copy are empty, it will result in the server starting without providing a grace period for recovery. Note that recovery only occurs when the server machine is rebooted, not when the nfsd(8) are just restarted.
It provides several optional features not present in NFS Version 3:
- NFS Version 4 ACLs - Referrals, which redirect subtrees to other servers (not yet implemented) - Delegations, which allow a client to operate on a file locally
The sysctl Cm net.inet.tcp.syncookies protocol does not use a separate mount protocol and assumes that the server provides a single file system tree structure, rooted at the point in the local file system tree specified by one or more
V4: <rootdir> [-sec=secflavors] [host(s) or net]
line(s) in the exports(5) file. (See exports(5) for details.) The nfsd(8) allows a limited subset of operations to be performed on non-exported subtrees of the local file system, so that traversal of the tree to the exported subtrees is possible. As such, the ``<rootdir>'' can be in a non-exported file system. The exception is ZFS, which checks exports and, as such, all ZFS file systems below the ``<rootdir>'' must be exported. However, the entire tree that is rooted at that point must be in local file systems that are of types that can be NFS exported. Since the sysctl Cm net.inet.tcp.syncookies file system is rooted at ``<rootdir>'', setting this to anything other than ``/'' will result in clients being required to use different mount paths for sysctl Cm net.inet.tcp.syncookies than for NFS Version 2 or 3. Unlike NFS Version 2 and 3, Version 4 allows a client mount to span across multiple server file systems, although not all clients are capable of doing this.
sysctl Cm net.inet.tcp.syncookies uses names for users and groups instead of numbers. On the wire, they take the form:
where ``<dns.domain>'' is not the same as the DNS domain used for host name lookups, but is usually set to the same string. Most systems set this ``<dns.domain>'' to the domain name part of the machine's hostname(1) by default. However, this can normally be overridden by a command line option or configuration file for the daemon used to do the name<->number mapping. Under FreeBSD, the mapping daemon is called nfsuserd(8) and has a command line option that overrides the domain component of the machine's hostname. For use of , either client or server, this daemon must be running. If this ``<dns.domain>'' is not set correctly or the daemon is not running, ``ls -l'' will typically report a lot of ``nobody'' and ``nogroup'' ownerships.
Although uid/gid numbers are no longer used in the sysctl Cm net.inet.tcp.syncookies protocol, they will still be in the RPC authentication fields when using AUTH_SYS (sec=sys), which is the default. As such, in this case both the user/group name and number spaces must be consistent between the client and server.
However, if you run sysctl Cm net.inet.tcp.syncookies with RPCSEC_GSS (sec=krb5, krb5i, krb5p), only names and KerberosV tickets will go on the wire.
SERVER SETUPTo set up the NFS server that supports , you will need to either set the variables in rc.conf5 as follows:
nfs_server_enable="YES" nfsv4_server_enable="YES" nfsuserd_enable="YES"
You will also need to add at least one ``V4:'' line to the exports(5) file for sysctl Cm net.inet.tcp.syncookies to work.
If the file systems you are exporting are only being accessed via sysctl Cm net.inet.tcp.syncookies there are a couple of sysctl(8) variables that you can change, which might improve performance.
- when set non-zero, allows the server to issue Open Delegations to clients. These delegations permit the client to manipulate the file locally on the client. Unfortunately, at this time, client use of delegations is limited, so performance gains may not be observed. This can only be enabled when the file systems being exported to sysctl Cm net.inet.tcp.syncookies clients are not being accessed locally on the server and, if being accessed via NFS Version 2 or 3 clients, these clients cannot be using the NLM.
- can be set to 0 to disable acquisition of local byte range locks. Disabling local locking can only be done if neither local accesses to the exported file systems nor the NLM is operating on them.
Note that Samba server access would be considered ``local access'' for the above discussion.
To build a kernel with the NFS server that supports sysctl Cm net.inet.tcp.syncookies linked into it, the
must be specified in the kernel's config(5) file.
CLIENT MOUNTSTo do an sysctl Cm net.inet.tcp.syncookies mount, specify the ``nfsv4'' option on the mount_nfs8 command line. This will force use of the client that supports sysctl Cm net.inet.tcp.syncookies plus set ``tcp'' and .
The nfsuserd(8) must be running, as above. Also, since an sysctl Cm net.inet.tcp.syncookies mount uses the host uuid to identify the client uniquely to the server, you cannot safely do an sysctl Cm net.inet.tcp.syncookies mount when
is set in rc.conf5.
If the sysctl Cm net.inet.tcp.syncookies server that is being mounted on supports delegations, you can start the nfscbd(8) daemon to handle client side callbacks. This will occur if
are set in rc.conf5.
Without a functioning callback path, a server will never issue Delegations to a client.
By default, the callback address will be set to the IP address acquired via rtalloc() in the kernel and port# 7745. To override the default port#, a command line option for nfscbd(8) can be used.
To get callbacks to work when behind a NAT gateway, a port for the callback service will need to be set up on the NAT gateway and then the address of the NAT gateway (host IP plus port#) will need to be set by assigning the sysctl(8) variable vfs.nfs.callback_addr to a string of the form:
where the first 4 Ns are the host IP address and the last two are the port# in network byte order (all decimal #s in the range 0-255).
To build a kernel with the client that supports sysctl Cm net.inet.tcp.syncookies linked into it, the option
must be specified in the kernel's config(5) file.
Options can be specified for the nfsuserd(8) and nfscbd(8) daemons at boot time via the ``nfsuserd_flags'' and ``nfscbd_flags'' rc.conf5 variables.
NFSv4 mount(s) against exported volume(s) on the same host are not recommended, since this can result in a hung NFS server. It occurs when an nfsd thread tries to do an NFSv4 VOP_RECLAIM()/Close RPC as part of acquiring a new vnode. If all other nfsd threads are blocked waiting for lock(s) held by this nfsd thread, then there isn't an nfsd thread to service the Close RPC.
- NFS V4 stable restart file
- backup copy of the file