freebsd-dev/usr.sbin/nfsd/pnfsserver.4

.\" Copyright (c) 2018 Rick Macklem
.\"
.\" 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
.\"    documentation and/or other materials provided with the distribution.
.\"
.\" THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
.\" ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
.\" IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
.\" ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
.\" FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
.\" DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
.\" OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
.\" HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
.\" LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
.\" OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
.\" SUCH DAMAGE.
.\"
.\" $FreeBSD$
.\"
.Dd August 8, 2018
.Dt PNFSSERVER 4
.Os
.Sh NAME
.Nm pNFSserver
.Nd NFS Version 4.1 Parallel NFS Protocol Server
.Sh DESCRIPTION
A set of FreeBSD servers may be configured to provide a
.Xr pnfs 4
service.
One FreeBSD system needs to be configured as a MetaData Server (MDS) and
at least one additional FreeBSD system needs to be configured as one or
more Data Servers (DS)s.
.Pp
These FreeBSD systems are configured to be NFSv4.1 servers, see
.Xr nfsd 8
and
.Xr exports 5
if you are not familiar with configuring a NFSv4.1 server.
.Sh DS server configuration
The DS(s) need to be configured as NFSv4.1 server(s), with a top level exported
directory used for storage of data files.
This directory must be owned by
.Dq root
and would normally have a mode of
.Dq 700 .
Within this directory there needs to be additional directories named
ds0,...,dsN (where N is 19 by default) also owned by
.Dq root
with mode
.Dq 700 .
These are the directories where the data files are stored.
The following command can be run by root when in the top level exported
directory to create these subdirectories.
.Bd -literal -offset indent
jot -w ds 20 0 | xargs mkdir -m 700
.Ed
.sp
Note that
.Dq 20
is the default and can be set to a larger value on the MDS as shown below.
.sp
The top level exported directory used for storage of data files must be
exported to the MDS with the
.Dq maproot=root sec=sys
export options so that the MDS can create entries in these subdirectories.
It must also be exported to all pNFS aware clients, but these clients do
not require the
.Dq maproot=root
export option and this directory should be exported to them with the same
options as used by the MDS to export file system(s) to the clients.
.Pp
It is possible to have multiple DSs on the same FreeBSD system, but each
of these DSs must have a separate top level exported directory used for storage
of data files and each
of these DSs must be mountable via a separate IP address.
Alias addresses can be set on the DS server system for a network
interface via
.Xr ifconfig 8
to create these different IP addresses.
Multiple DSs on the same server may be useful when data for different file systems
on the MDS are being stored on different file system volumes on the FreeBSD
DS system.
.Sh MDS server configuration
The MDS must be a separate FreeBSD system from the FreeBSD DS system(s) and
NFS clients.
It is configured as a NFSv4.1 server with file system(s) exported to
clients.
However, the
.Dq -p
command line argument for
.Xr nfsd
is used to indicate that it is running as the MDS for a pNFS server.
.Pp
The DS(s) must all be mounted on the MDS using the following mount options:
.Bd -literal -offset indent
nfsv4,minorversion=1,soft,retrans=2
.Ed
.sp
so that they can be defined as DSs in the
.Dq -p
option.
Normally these mounts would be entered in the
.Xr fstab 5
on the MDS.
For example, if there are four DSs named nfsv4-data[0-3], the
.Xr fstab 5
lines might look like:
.Bd -literal -offset
nfsv4-data0:/ /data0 nfs rw,nfsv4,minorversion=1,soft,retrans=2 0 0
nfsv4-data1:/ /data1 nfs rw,nfsv4,minorversion=1,soft,retrans=2 0 0
nfsv4-data2:/ /data2 nfs rw,nfsv4,minorversion=1,soft,retrans=2 0 0
nfsv4-data3:/ /data3 nfs rw,nfsv4,minorversion=1,soft,retrans=2 0 0
.Ed
.sp
The
.Xr nfsd 8
command line option
.Dq -p
indicates that the NFS server is a pNFS MDS and specifies what
DSs are to be used.
.br
For the above
.Xr fstab 5
example, the
.Xr nfsd 8
nfs_server_flags line in your
.Xr rc.conf 5
might look like:
.Bd -literal -offset
nfs_server_flags="-u -t -n 128 -p nfsv4-data0:/data0,nfsv4-data1:/data1,nfsv4-data2:/data2,nfsv4-data3:/data3"
.Ed
.sp
This example specifies that the data files should be distributed over the
four DSs and File layouts will be issued to pNFS enabled clients.
If issuing Flexible File layouts is desired for this case, setting the sysctl
.Dq vfs.nfsd.default_flexfile
non-zero in your
.Xr sysctl.conf 5
file will make the
.Nm
do that.
.br
Alternately, this variant of
.Dq nfs_server_flags
will specify that two way mirroring is to be done, via the
.Dq -m
command line option.
.Bd -literal -offset
nfs_server_flags="-u -t -n 128 -p nfsv4-data0:/data0,nfsv4-data1:/data1,nfsv4-data2:/data2,nfsv4-data3:/data3 -m 2"
.Ed
.sp
With two way mirroring, the data file for each exported file on the MDS
will be stored on two of the DSs.
When mirroring is enabled, the server will always issue Flexible File layouts.
.Pp
It is also possible to specify which DSs are to be used to store data files for
specific exported file systems on the MDS.
For example, if the MDS has exported two file systems
.Dq /export1
and
.Dq /export2
to clients, the following variant of
.Dq nfs_server_flags
will specify that data files for
.Dq /export1
will be stored on nfsv4-data0 and nfsv4-data1, whereas the data files for
.Dq /export2
will be store on nfsv4-data2 and nfsv4-data3.
.Bd -literal -offset
nfs_server_flags="-u -t -n 128 -p nfsv4-data0:/data0#/export1,nfsv4-data1:/data1#/export1,nfsv4-data2:/data2#/export2,nfsv4-data3:/data3#/export2"
.Ed
.sp
This can be used by system administrators to control where data files are
stored and might be useful for control of storage use.
For this case, it may be convenient to co-locate more than one of the DSs
on the same FreeBSD server, using separate file systems on the DS system
for storage of the respective DS's data files.
If mirroring is desired for this case, the
.Dq -m
option also needs to be specified.
There must be enough DSs assigned to each exported file system on the MDS
to support the level of mirroring.
The above example would be fine for two way mirroring, but four way mirroring
would not work, since there are only two DSs assigned to each exported file
system on the MDS.
.Pp
The number of subdirectories in each DS is defined by the
.Dq vfs.nfs.dsdirsize
sysctl on the MDS.
This value can be increased from the default of 20, but only when the
.Xr nfsd 8
is not running and after the additional ds20,... subdirectories have been
created on all the DSs.
For a service that will store a large number of files this sysctl should be
set much larger, to avoid the number of entries in a subdirectory from
getting too large.
.Sh Client mounts
Once operational, NFSv4.1 FreeBSD client mounts done with the
.Dq pnfs
option should do I/O directly on the DSs.
The clients mounting the MDS must be running the
.Xr nfscbd
daemon for pNFS to work.
Set
.Bd -literal -offset indent
nfscbd_enable="YES"
.Ed
.sp
in the
.Xr rc.conf 5
on these clients.
Non-pNFS aware clients or NFSv3 mounts will do all I/O RPCs on the MDS,
which acts as a proxy for the appropriate DS(s).
.Sh Backing up a pNFS service
Since the data is separated from the metadata, the simple way to back up
a pNFS service is to do so from an NFS client that has the service mounted
on it.
If you back up the MDS exported file system(s) on the MDS, you must do it
in such a way that the
.Dq system
namespace extended attributes get backed up.
.Sh Handling of failed mirrored DSs
When a mirrored DS fails, it can be disabled one of three ways:
.sp
1 - The MDS detects a problem when trying to do proxy
operations on the DS.
This can take a couple of minutes
after the DS failure or network partitioning occurs.
.sp
2 - A pNFS client can report an I/O error that occurred for a DS to the MDS in
the arguments for a LayoutReturn operation.
.sp
3 - The system administrator can perform the pnfsdskill(8) command on the MDS
to disable it. If the system administrator does a pnfsdskill(8) and it fails
with ENXIO (Device not configured) that normally means the DS was already
disabled via #1 or #2. Since doing this is harmless, once a system
administrator knows that there is a problem with a mirrored DS, doing the
command is recommended.
.sp
Once a system administrator knows that a mirrored DS has malfunctioned
or has been network partitioned, they should do the following as root/su
on the MDS:
.Bd -literal -offset indent
# pnfsdskill <mounted-on-path-of-DS>
# umount -N <mounted-on-path-of-DS>
.Ed
.sp
Note that the <mounted-on-path-of-DS> must be the exact mounted-on path
string used when the DS was mounted on the MDS.
.Pp
Once the mirrored DS has been disabled, the pNFS service should continue to
function, but file updates will only happen on the DS(s) 
that have not been disabled. Assuming two way mirroring, that implies
the one DS of the pair stored in the
.Dq pnfsd.dsfile
extended attribute for the file on the MDS, for files stored on the disabled DS.
.Pp
The next step is to clear the IP address in the
.Dq pnfsd.dsfile
extended attribute on all files on the MDS for the failed DS.
This is done so that, when the disabled DS is repaired and brought back online,
the data files on this DS will not be used, since they may be out of date.
The command that clears the IP address is
.Xr pnfsdsfile 8
with the
.Dq -r
option.
.Bd -literal -offset
For example:
# pnfsdsfile -r nfsv4-data3 yyy.c
yyy.c:	nfsv4-data2.home.rick	ds0/207508569ff983350c000000ec7c0200e4c57b2e0000000000000000	0.0.0.0	ds0/207508569ff983350c000000ec7c0200e4c57b2e0000000000000000
.Ed
.sp
replaces nfsv4-data3 with an IPv4 address of 0.0.0.0, so that nfsv4-data3
will not get used.
.Pp
Normally this will be called within a
.Xr find 1
command for all regular
files in the exported directory tree and must be done on the MDS.
When used with
.Xr find 1 ,
you will probably also want the
.Dq -q
option so that it won't spit out the results for every file.
If the disabled/repaired DS is nfsv4-data3, the commands done on the MDS
would be:
.Bd -literal -offset
# cd <top-level-exported-dir>
# find . -type f -exec pnfsdsfile -q -r nfsv4-data3 {} \;
.Ed
.sp
There is a problem with the above command if the file found by
.Xr find 1
is renamed or unlinked before the
.Xr pnfsdsfile 8
command is done on it.
This should normally generate an error message.
A simple unlink is harmless
but a link/unlink or rename might result in the file not having been processed
under its new name.
To check that all files have their IP addresses set to 0.0.0.0 these
commands can be used (assuming the
.Xr sh 1
shell):
.Bd -literal -offset
# cd <top-level-exported-dir>
# find . -type f -exec pnfsdsfile {} \; | sed "/nfsv4-data3/!d"
.Ed
.sp
Any line(s) printed require the
.Xr pnfsdsfile 8
with
.Dq -r
to be done again.
Once this is done, the replaced/repaired DS can be brought back online.
It should have empty ds0,...,dsN directories under the top level exported
directory for storage of data files just like it did when first set up.
Mount it on the MDS exactly as you did before disabling it.
For the nfsv4-data3 example, the command would be:
.Bd -literal -offset
# mount -t nfs -o nfsv4,minorversion=1,soft,retrans=2 nfsv4-data3:/ /data3
.Ed
.sp
Then restart the nfsd to re-enable the DS.
.Bd -literal -offset
# /etc/rc.d/nfsd restart
.Ed
.sp
Now, new files can be stored on nfsv4-data3,
but files with the IP address zeroed out on the MDS will not yet use the
repaired DS (nfsv4-data3).
The next step is to go through the exported file tree on the MDS and,
for each of the
files with an IPv4 address of 0.0.0.0 in its extended attribute, copy the file
data to the repaired DS and re-enable use of this mirror for it.
This command for copying the file data for one MDS file is
.Xr pnfsdscopymr 8
and it will also normally be used in a
.Xr find 1 .
For the example case, the commands on the MDS would be:
.Bd -literal -offset
# cd <top-level-exported-dir>
# find . -type f -exec pnfsdscopymr -r /data3 {} \;
.Ed
.sp
When this completes, the recovery should be complete or at least nearly so.
As noted above, if a link/unlink or rename occurs on a file name while the
above
.Xr find 1
is in progress, it may not get copied.
To check for any file(s) not yet copied, the commands are:
.Bd -literal -offset
# cd <top-level-exported-dir>
# find . -type f -exec pnfsdsfile {} \; | sed "/0\.0\.0\.0/!d"
.Ed
.sp
If this command prints out any file name(s), these files must
have the
.Xr pnfsdscopymr 8
command done on them to complete the recovery.
.Bd -literal -offset
# pnfsdscopymr -r /data3 <file-path-reported>
.Ed
.sp
If this commmand fails with the error
.br
.Dq pnfsdscopymr: Copymr failed for file <path>: Device not configured
.br
repeatedly, this may be caused by a Read/Write layout that has not
been returned.
The only way to get rid of such a layout is to restart the
.Xr nfsd 8 .
.sp
All of these commands are designed to be
done while the pNFS service is running and can be re-run safely.
.Pp
For a more detailed discussion of the setup and management of a pNFS service
see:
.Bd -literal -offset indent
http://people.freebsd.org/~rmacklem/pnfs-planb-setup.txt
.Ed
.sp
.Sh SEE ALSO
.Xr nfsv4 4 ,
.Xr pnfs 4 ,
.Xr exports 5 ,
.Xr fstab 5 ,
.Xr rc.conf 5 ,
.Xr sysctl.conf 5 ,
.Xr nfscbd 8 ,
.Xr nfsd 8 ,
.Xr nfsuserd 8 ,
.Xr pnfsdscopymr 8 ,
.Xr pnfsdsfile 8 ,
.Xr pnfsdskill 8
.Sh HISTORY
The
.Nm
command first appeared in
.Fx 12.0 .
.Sh BUGS
Since the MDS cannot be mirrored, it is a single point of failure just
as a non
.Tn pNFS
server is.
For non-mirrored configurations, all FreeBSD systems used in the service
are single points of failure.