NFSD(8) FreeBSD System Manager's Manual NFSD(8)
NAME
nfsd - remote NFS server
SYNOPSIS
nfsd [-arduteN] [-n num_servers] [-h bindip] [-p pnfs_setup]
[-m mirror_level] [-P pidfile] [-V virtual_hostname]
[--maxthreads max_threads] [--minthreads min_threads]
DESCRIPTION
The nfsd utility runs on a server machine to service NFSv4 and/or NFSv3
requests from client machines. At least one nfsd must be running for a
machine to operate as a server.
Unless otherwise specified, eight servers per CPU for UDP transport are
started.
The nfsd server may be configured to serve NFSv3, NFSv4 or both of them.
When nfs_server_enable="YES" is set in rc.conf(5), nfsd is configured for
NFSv3 only, by default. To configure the nfsd to serve NFSv4, you must
also:
o Add a "V4:" line to exports(5). (This is needed as well as either
the ZFS sharenfs property or an exports line for each exported file
system.)
o Either choose to do uids in "strings", which is probably the easiest
unless you are doing kerberos(7) mounts or run nfsuserd(8) (choose
one, not both of these). Setting the sysctls MIBs
vfs.nfs.enable_uidtostring and vfs.nfsd.enable_stringtouid to one and
not running the nfsuserd chooses uids in "strings".
o Set nfsv4_server_enable="YES" in rc.conf(5) or set
nfsv4_server_only="YES" in rc.conf(5) if you do not want NFSv3 to be
served.
When nfsd is run in an appropriately configured vnet jail, the server is
restricted to TCP transport and no pNFS service. Therefore, the -t
option must be specified and none of the -u, -p and -m options can be
specified when run in a vnet jail. See jail(8) for more information.
The following options are available:
-r Register the NFS service with rpcbind(8) without creating any
servers. This option can be used along with the -u or -t options
to re-register NFS if the rpcbind server is restarted.
-d Unregister the NFS service with rpcbind(8) without creating any
servers.
-P pidfile
Specify alternative location of a file where main process PID
will be stored. The default location is /var/run/nfsd.pid.
-V virtual_hostname
Specifies a hostname to be used as a principal name, instead of
the default hostname.
-n threads
This option is deprecated and is limited to a maximum of 256
threads. The options --maxthreads and --minthreads should now be
used. The threads argument for --minthreads and --maxthreads may
be set to the same value to avoid dynamic changes to the number
of threads.
--maxthreads threads
Specifies the maximum servers that will be kept around to service
requests.
--minthreads threads
Specifies the minimum servers that will be kept around to service
requests.
-h bindip
Specifies which IP address or hostname to bind to on the local
host. This option is recommended when a host has multiple
interfaces. Multiple -h options may be specified.
-a Specifies that nfsd should bind to the wildcard IP address. This
is the default if no -h options are given. It may also be
specified in addition to any -h options given. Note that NFS/UDP
does not operate properly when bound to the wildcard IP address
whether you use -a or do not use -h.
-p pnfs_setup
Enables pNFS support in the server and specifies the information
that the daemon needs to start it. This option can only be used
on one server and specifies that this server will be the MetaData
Server (MDS) for the pNFS service. This can only be done if
there is at least one FreeBSD system configured as a Data Server
(DS) for it to use.
The pnfs_setup string is a set of fields separated by ','
characters: Each of these fields specifies one DS. It consists
of a server hostname, followed by a ':' and the directory path
where the DS's data storage file system is mounted on this MDS
server. This can optionally be followed by a '#' and the
mds_path, which is the directory path for an exported file system
on this MDS. If this is specified, it means that this DS is to
be used to store data files for this mds_path file system only.
If this optional component does not exist, the DS will be used to
store data files for all exported MDS file systems. The DS
storage file systems must be mounted on this system before the
nfsd is started with this option specified.
For example:
nfsv4-data0:/data0,nfsv4-data1:/data1
would specify two DS servers called nfsv4-data0 and nfsv4-data1
that comprise the data storage component of the pNFS service.
These two DSs would be used to store data files for all exported
file systems on this MDS. The directories "/data0" and "/data1"
are where the data storage servers exported storage directories
are mounted on this system (which will act as the MDS).
Whereas, for the example:
nfsv4-data0:/data0#/export1,nfsv4-data1:/data1#/export2
would specify two DSs as above, however nfsv4-data0 will be used
to store data files for "/export1" and nfsv4-data1 will be used
to store data files for "/export2".
When using IPv6 addresses for DSs be wary of using link local
addresses. The IPv6 address for the DS is sent to the client and
there is no scope zone in it. As such, a link local address may
not work for a pNFS client to DS TCP connection. When parsed,
nfsd will only use a link local address if it is the only address
returned by getaddrinfo(3) for the DS hostname.
-m mirror_level
This option is only meaningful when used with the -p option. It
specifies the "mirror_level", which defines how many of the DSs
will have a copy of a file's data storage file. The default of
one implies no mirroring of data storage files on the DSs. The
"mirror_level" would normally be set to 2 to enable mirroring,
but can be as high as NFSDEV_MAXMIRRORS. There must be at least
"mirror_level" DSs for each exported file system on the MDS, as
specified in the -p option. This implies that, for the above
example using "#/export1" and "#/export2", mirroring cannot be
done. There would need to be two DS entries for each of
"#/export1" and "#/export2" in order to support a "mirror_level"
of two.
If mirroring is enabled, the server must use the Flexible File
layout. If mirroring is not enabled, the server will use the
File layout by default, but this default can be changed to the
Flexible File layout if the sysctl(8) vfs.nfsd.default_flexfile
is set non-zero.
-t Serve TCP NFS clients.
-u Serve UDP NFS clients.
-e Ignored; included for backward compatibility.
-N Cause nfsd to execute in the foreground instead of in daemon
mode.
For example, "nfsd -u -t --minthreads 6 --maxthreads 6" serves UDP and
TCP transports using six kernel threads (servers).
For a system dedicated to servicing NFS RPCs, the number of threads
(servers) should be sufficient to handle the peak client RPC load. For
systems that perform other services, the number of threads (servers) may
need to be limited, so that resources are available for these other
services.
The nfsd utility listens for service requests at the port indicated in
the NFS server specification; see Network File System Protocol
Specification, RFC1094, NFS: Network File System Version 3 Protocol
Specification, RFC1813, Network File System (NFS) Version 4 Protocol,
RFC7530, Network File System (NFS) Version 4 Minor Version 1 Protocol,
RFC5661, Network File System (NFS) Version 4 Minor Version 2 Protocol,
RFC7862, File System Extended Attributes in NFSv4, RFC8276 and Parallel
NFS (pNFS) Flexible File Layout, RFC8435.
If nfsd detects that NFS is not loaded in the running kernel, it will
attempt to load a loadable kernel module containing NFS support using
kldload(2). If this fails, or no NFS KLD is available, nfsd will exit
with an error.
If nfsd is to be run on a host with multiple interfaces or interface
aliases, use of the -h option is recommended. If you do not use the
option NFS may not respond to UDP packets from the same IP address they
were sent to. Use of this option is also recommended when securing NFS
exports on a firewalling machine such that the NFS sockets can only be
accessed by the inside interface. The ipfw utility would then be used to
block NFS-related packets that come in on the outside interface.
If the server has stopped servicing clients and has generated a console
message like "nfsd server cache flooded...", the value for
vfs.nfsd.tcphighwater needs to be increased. This should allow the
server to again handle requests without a reboot. Also, you may want to
consider decreasing the value for vfs.nfsd.tcpcachetimeo to several
minutes (in seconds) instead of 12 hours when this occurs.
Unfortunately making vfs.nfsd.tcphighwater too large can result in the
mbuf limit being reached, as indicated by a console message like
"kern.ipc.nmbufs limit reached". If you cannot find values of the above
sysctl values that work, you can disable the DRC cache for TCP by setting
vfs.nfsd.cachetcp to 0.
The nfsd utility has to be terminated with SIGUSR1 and cannot be killed
with SIGTERM or SIGQUIT. The nfsd utility needs to ignore these signals
in order to stay alive as long as possible during a shutdown, otherwise
loopback mounts will not be able to unmount. If you have to kill nfsd
just do a "kill -USR1 <PID of master nfsd>"
EXIT STATUS
The nfsd utility exits 0 on success, and >0 if an error occurs.
SEE ALSO
nfsstat(1), kldload(2), nfssvc(2), nfsv4(4), pnfs(4), pnfsserver(4),
exports(5), rc.conf(5), stablerestart(5), kerberos(7), gssd(8), ipfw(8),
jail(8), mountd(8), nfsiod(8), nfsrevoke(8), nfsuserd(8), rpcbind(8)
HISTORY
The nfsd utility first appeared in 4.4BSD.
BUGS
If nfsd is started when gssd(8) is not running, it will service AUTH_SYS
requests only. To fix the problem you must kill nfsd and then restart
it, after the gssd(8) is running.
For a Flexible File Layout pNFS server, if there are Linux clients doing
NFSv4.1 or NFSv4.2 mounts, those clients might need the sysctl(8)
vfs.nfsd.flexlinuxhack to be set to one on the MDS as a workaround.
Linux 5.n kernels appear to have been patched such that this sysctl(8)
does not need to be set.
For NFSv4.2, a Copy operation can take a long time to complete. If there
is a concurrent ExchangeID or DelegReturn operation which requires the
exclusive lock on all NFSv4 state, this can result in a "stall" of the
nfsd server. If your storage is on ZFS without block cloning enabled,
setting the sysctl(8) vfs.zfs.dmu_offset_next_sync to 0 can often avoid
this problem. It is also possible to set the sysctl(8)
vfs.nfsd.maxcopyrange to 10-100 megabytes to try and reduce Copy
operation times. As a last resort, setting sysctl(8)
vfs.nfsd.maxcopyrange to 0 disables the Copy operation.
FreeBSD 15.1-STABLE-HBSD December 23, 2025 NFSD(8)