RFC - NFS startup order

3.1. SERVER STARTUP

A/ mount -t nfsd /proc/fs/nfsd
This filesystem needs to be mount before most daemons, particularly exportfs, mountd, svcgssd, idmapd.
It could be mounted once, or the script that starts each daemon could test if it is mounted and mount it if not.

B/ svcgssd ; idmapd
These supply services to nfsd and so should be started before rpc.nfsd. Where they come between mounting the nfsd filesystem and starting the nfsd server is not important. idmapd is only need for NFSv4 support.

3.2. CLIENT STARTUP

A/ sm-notify
This should be run shortly after boot and before any NFS filesystems are mounted with remote-locking support - filesystems can be mounted with "-o nolock" before sm-notify. This is appropriate for '/', '/usr', and '/var'.

B/ gssd ; idmapd
(I need some guidance here).

Quoted 3 lines · Neil Brown

3.1. SERVER STARTUP

Quoted 5 lines

A/ mount -t nfsd /proc/fs/nfsd
This filesystem needs to be mount before most daemons, particularly exportfs, mountd, svcgssd, idmapd.
It could be mounted once, or the script that starts each daemon could test if it is mounted and mount it if not.

mount -t rpc_pipefs /var/lib/nfs/rpc_pipefs
This filesystem must be mounted before svcgssd and idmapd daemons.

Quoted 11 lines · On Tuesday April 3, [email protected] wrote:

Quoted 8 lines · Neil Brown

3.1. SERVER STARTUP

A/ mount -t nfsd /proc/fs/nfsd
This filesystem needs to be mount before most daemons, particularly exportfs, mountd, svcgssd, idmapd.
It could be mounted once, or the script that starts each daemon could test if it is mounted and mount it if not.

Quoted 3 lines

mount -t rpc_pipefs /var/lib/nfs/rpc_pipefs
This filesystem must be mounted before svcgssd and idmapd daemons.

Is rpc_pipefs really needed for svcgssd? or just for 'gssd' and 'idmapd'.

I gather that a SIGHUP to idmapd will cause it to open the serve channels after they have been made available. It is also sufficient to SIGHUP idmapd after mounting rpc_pipefs if rpc_pipefs is mounted later for some reason?

3. DAEMON STARTUP ORDER

This nfs-utils packages does not provide any scripts for starting various daemons as mast distributions replace them with their own, so

any scripts we package would not get much testing. Instead, we explain the dependencies involved in startup so that scripts can be written to work correctly.

3.0 PREREQUISITES

Name service (host name lookup) should be working before any NFS services are started.

"portmap" must be running before any NFS services (server or client) are started.

Normally network interfaces should be configured first as well, though this isn't critical for the NFS server (providing name service is handled locally).

3.1. SERVER STARTUP

A/ mount -t nfsd /proc/fs/nfsd
This filesystem needs to be mount before most daemons, particularly exportfs, mountd, svcgssd, idmapd.
It could be mounted once, or the script that starts each daemon could test if it is mounted and mount it if not.

B/ svcgssd ; idmapd
These supply services to nfsd and so should be started before rpc.nfsd. Where they come between mounting the nfsd filesystem and starting the nfsd server is not important. idmapd is only need for NFSv4 support.

C/ exportfs -av ; rpc.mountd
If it important that exportfs be run before mountd so that

mountd is working from current information (in /var/lib/nfs/etab).
It is also important that both of these are run before rpc.nfsd.
If not, any NFS requests that arrive before mountd is started will get replied to with a 'Stale NFS File handle' error.

D/ rpc.statd --no-notify
It is best if statd is started before nfsd though this isn't critical. Certainly it should be at most a few seconds after nfsd.
When nfsd starts it will start lockd. If lockd then receives a lock request it will communicate with statd. If statd is not running lockd will retry, but it won't wait forever for a reply.
Note that if statd is started before nfsd, the --no-notify option must be used. If notify requests are sent out before nfsd start, clients may try to reclaim locks and, on finding that lockd isn't running, they will give up and never reclaim the lock.
rpc.statd is only needed for NFSv2 and NFSv3 support.

E/ rpc.nfsd
Starting nfsd will automatically start lockd. The nfs server will now be fully active and respond to any requests from clients.

F/ sm-notify
This will notify any client which might have locks from before a reboot to try to reclaim their locks. This should start immediately after rpc.nfsd is started so that clients have a chance to reclaim locks within the 90 second grace period. sm-notify is only needed for NFSv2 and NFSv3 support.

3.2. CLIENT STARTUP

A/ sm-notify
This should be run shortly after boot and before any NFS filesystems are mounted with remote-locking support - filesystems can be mounted with "-o nolock" before sm-notify. This is appropriate for '/', '/usr', and '/var'.

B/ gssd ; idmapd
(I need some guidance here).

C/ statd should be run before any NFSv2 or NFSv3 filesystem is
mounted with remote locking (i.e. without -o nolock). 'mount' will try to use "/usr/sbin/start-statd" to start statd if it is not already running, so there is no need to explicitly start statd in boot-time scripts.

3.3. SERVER/CLIENT INTERACTIONS

A/ sm-notify
Both the server and the client need sm-notify to be run. It should be run after the NFS server is started, but before and NFS filesystems are mounted with remote locking.

B/ rpc.statd
Both the server and the client need rpc.statd to be running. Each should try to start when they need it.

C/ idmapd
(I need some guidance here).

idmapd should be started before mounting any NFSv4 filesystems. gssd is only needed when filesystems are mounted securely (with Kerberos or SPKM3). It must be started before mounting any NFS filesystems securely.

Both the server and client need idmapd to be running. If idmapd is started (for the client) before starting nfsd (or is it mountd?), then idmapd should be sent a HUP signal after starting nfsd (mountd?) to signal that the server channels should be opened.