Hello NFS developers,
reposting the two patches inline as requested by Steve.
I'm currently systemd-ifying our nfs-utils Ubuntu package. For testing I put
the NFS server and client (i. e. localhost:/foo/bar mounts) on the same
machine. With that I get long hangs during shutdown on stopping the NFS .mount
units, as when that happens the NFS server is already shut down.
This is certainly a corner case as you'd usually not NFS-mount a share from
localhost; but fixing it is relatively simple with the first patch, which makes
sure that if NFS server and client are installed, the server starts before the
client, and the client stops before the server.
For a client without installed server this is harmless as Before= does not
imply a dependency. Likewise, for an NFS server which does not mount shares by
itself, it's also a no-op as remote-fs.target is empty. This would only
slightly reorder the boot sequence for machines which both are an NFS server
and have some remote NFS mounts, but I don't see an issue with that.
The second patch make NFS start earlier in the boot (i. e. before
basic.target), so that you can do things like put /var/ on NFS, or have rcS
SysV init scripts which depend on $remote_fs work. I tested this on both a
server and a client. This is certainly a bit more intrusive, but could be
worthwhile; what do you think?
Thanks for considering,
Martin
This makes mounting NFS shares from localhost work reliably, as you need to
start the server before attempting (client) mounts, and conversely on shutdown
need to unmount all shares before stopping the server to avoid hangs.
---
systemd/nfs-server.service | 3 +++
1 file changed, 3 insertions(+)
diff --git a/systemd/nfs-server.service b/systemd/nfs-server.service
index 8010aad..b35e193 100644
--- a/systemd/nfs-server.service
+++ b/systemd/nfs-server.service
@@ -13,6 +13,9 @@ Before= rpc-statd-notify.service
Wants=auth-rpcgss-module.service
After=rpc-gssd.service gssproxy.service rpc-svcgssd.service
+# start/stop server before/after client
+Before=remote-fs-pre.target
+
Wants=nfs-config.service
After=nfs-config.service
--
2.1.4
Stop depending on basic.target in the daemons which still do; i. e. add
DefaultDependencies=no. This makes it possible to run NFS during early boot,
and helps if you e. g. have /var on NFS. We don't require much else than
local-fs.
---
systemd/auth-rpcgss-module.service | 1 +
systemd/nfs-config.service | 2 ++
systemd/nfs-idmapd.service | 3 ++-
systemd/nfs-mountd.service | 3 ++-
systemd/nfs-server.service | 2 ++
systemd/rpc-statd-notify.service | 3 ++-
systemd/rpc-svcgssd.service | 3 ++-
7 files changed, 13 insertions(+), 4 deletions(-)
diff --git a/systemd/auth-rpcgss-module.service b/systemd/auth-rpcgss-module.service
index 0355e13..5241f7b 100644
--- a/systemd/auth-rpcgss-module.service
+++ b/systemd/auth-rpcgss-module.service
@@ -6,6 +6,7 @@
# unit will fail. But that's OK.)
[Unit]
Description=Kernel Module supporting RPCSEC_GSS
+DefaultDependencies=no
Before=gssproxy.service rpc-svcgssd.service rpc-gssd.service
Wants=gssproxy.service rpc-svcgssd.service rpc-gssd.service
ConditionPathExists=/etc/krb5.keytab
diff --git a/systemd/nfs-config.service b/systemd/nfs-config.service
index 64010e6..7f65305 100644
--- a/systemd/nfs-config.service
+++ b/systemd/nfs-config.service
@@ -1,5 +1,7 @@
[Unit]
Description=Preprocess NFS configuration
+After=local-fs.target
+DefaultDependencies=no
[Service]
Type=oneshot
diff --git a/systemd/nfs-idmapd.service b/systemd/nfs-idmapd.service
index e84f8c8..df3dd9d 100644
--- a/systemd/nfs-idmapd.service
+++ b/systemd/nfs-idmapd.service
@@ -1,7 +1,8 @@
[Unit]
Description=NFSv4 ID-name mapping service
+DefaultDependencies=no
Requires=var-lib-nfs-rpc_pipefs.mount
-After=var-lib-nfs-rpc_pipefs.mount
+After=var-lib-nfs-rpc_pipefs.mount local-fs.target
BindsTo=nfs-server.service
diff --git a/systemd/nfs-mountd.service b/systemd/nfs-mountd.service
index d908afe..8a39f3e 100644
--- a/systemd/nfs-mountd.service
+++ b/systemd/nfs-mountd.service
@@ -1,8 +1,9 @@
[Unit]
Description=NFS Mount Daemon
+DefaultDependencies=no
Requires=proc-fs-nfsd.mount
After=proc-fs-nfsd.mount
-After=network.target
+After=network.target local-fs.target
BindsTo=nfs-server.service
Wants=nfs-config.service
diff --git a/systemd/nfs-server.service b/systemd/nfs-server.service
index b35e193..db801cb 100644
--- a/systemd/nfs-server.service
+++ b/systemd/nfs-server.service
@@ -1,10 +1,12 @@
[Unit]
Description=NFS server and services
+DefaultDependencies=no
Requires= network.target proc-fs-nfsd.mount rpcbind.target
Requires= nfs-mountd.service
Wants=rpc-statd.service nfs-idmapd.service
Wants=rpc-statd-notify.service
+After= local-fs.target
After= network.target proc-fs-nfsd.mount rpcbind.target nfs-mountd.service
After= nfs-idmapd.service rpc-statd.service
Before= rpc-statd-notify.service
diff --git a/systemd/rpc-statd-notify.service b/systemd/rpc-statd-notify.service
index a655445..a7e2f8e 100644
--- a/systemd/rpc-statd-notify.service
+++ b/systemd/rpc-statd-notify.service
@@ -1,7 +1,8 @@
[Unit]
Description=Notify NFS peers of a restart
+DefaultDependencies=no
Requires=network-online.target
-After=network.target nss-lookup.target
+After=local-fs.target network.target nss-lookup.target
# if we run an nfs server, it needs to be running before we
# tell clients that it has restarted.
diff --git a/systemd/rpc-svcgssd.service b/systemd/rpc-svcgssd.service
index f7424b0..41177b6 100644
--- a/systemd/rpc-svcgssd.service
+++ b/systemd/rpc-svcgssd.service
@@ -1,7 +1,8 @@
[Unit]
Description=RPC security service for NFS server
+DefaultDependencies=no
Requires=var-lib-nfs-rpc_pipefs.mount
-After=var-lib-nfs-rpc_pipefs.mount
+After=var-lib-nfs-rpc_pipefs.mount local-fs.target
PartOf=nfs-server.service
PartOf=nfs-utils.service
--
2.1.4
On Tue, Mar 03, 2015 at 06:28:09PM +0100, Martin Pitt wrote:
> This makes mounting NFS shares from localhost work reliably, as you need to
> start the server before attempting (client) mounts, and conversely on shutdown
> need to unmount all shares before stopping the server to avoid hangs.
> ---
> systemd/nfs-server.service | 3 +++
> 1 file changed, 3 insertions(+)
>
> diff --git a/systemd/nfs-server.service b/systemd/nfs-server.service
> index 8010aad..b35e193 100644
> --- a/systemd/nfs-server.service
> +++ b/systemd/nfs-server.service
> @@ -13,6 +13,9 @@ Before= rpc-statd-notify.service
> Wants=auth-rpcgss-module.service
> After=rpc-gssd.service gssproxy.service rpc-svcgssd.service
>
> +# start/stop server before/after client
> +Before=remote-fs-pre.target
> +
> Wants=nfs-config.service
> After=nfs-config.service
Looks reasonable.
Zbyszek
On Tue, Mar 03, 2015 at 06:28:10PM +0100, Martin Pitt wrote:
> Stop depending on basic.target in the daemons which still do; i. e. add
> DefaultDependencies=no. This makes it possible to run NFS during early boot,
> and helps if you e. g. have /var on NFS. We don't require much else than
> local-fs.
Are you sure that all of those nfs daemons do not require
sockets.target and other stuff provided by basic.target to be ready?
Zbyszek
> ---
> systemd/auth-rpcgss-module.service | 1 +
> systemd/nfs-config.service | 2 ++
> systemd/nfs-idmapd.service | 3 ++-
> systemd/nfs-mountd.service | 3 ++-
> systemd/nfs-server.service | 2 ++
> systemd/rpc-statd-notify.service | 3 ++-
> systemd/rpc-svcgssd.service | 3 ++-
> 7 files changed, 13 insertions(+), 4 deletions(-)
>
> diff --git a/systemd/auth-rpcgss-module.service b/systemd/auth-rpcgss-module.service
> index 0355e13..5241f7b 100644
> --- a/systemd/auth-rpcgss-module.service
> +++ b/systemd/auth-rpcgss-module.service
> @@ -6,6 +6,7 @@
> # unit will fail. But that's OK.)
> [Unit]
> Description=Kernel Module supporting RPCSEC_GSS
> +DefaultDependencies=no
> Before=gssproxy.service rpc-svcgssd.service rpc-gssd.service
> Wants=gssproxy.service rpc-svcgssd.service rpc-gssd.service
> ConditionPathExists=/etc/krb5.keytab
> diff --git a/systemd/nfs-config.service b/systemd/nfs-config.service
> index 64010e6..7f65305 100644
> --- a/systemd/nfs-config.service
> +++ b/systemd/nfs-config.service
> @@ -1,5 +1,7 @@
> [Unit]
> Description=Preprocess NFS configuration
> +After=local-fs.target
> +DefaultDependencies=no
>
> [Service]
> Type=oneshot
> diff --git a/systemd/nfs-idmapd.service b/systemd/nfs-idmapd.service
> index e84f8c8..df3dd9d 100644
> --- a/systemd/nfs-idmapd.service
> +++ b/systemd/nfs-idmapd.service
> @@ -1,7 +1,8 @@
> [Unit]
> Description=NFSv4 ID-name mapping service
> +DefaultDependencies=no
> Requires=var-lib-nfs-rpc_pipefs.mount
> -After=var-lib-nfs-rpc_pipefs.mount
> +After=var-lib-nfs-rpc_pipefs.mount local-fs.target
>
> BindsTo=nfs-server.service
>
> diff --git a/systemd/nfs-mountd.service b/systemd/nfs-mountd.service
> index d908afe..8a39f3e 100644
> --- a/systemd/nfs-mountd.service
> +++ b/systemd/nfs-mountd.service
> @@ -1,8 +1,9 @@
> [Unit]
> Description=NFS Mount Daemon
> +DefaultDependencies=no
> Requires=proc-fs-nfsd.mount
> After=proc-fs-nfsd.mount
> -After=network.target
> +After=network.target local-fs.target
> BindsTo=nfs-server.service
>
> Wants=nfs-config.service
> diff --git a/systemd/nfs-server.service b/systemd/nfs-server.service
> index b35e193..db801cb 100644
> --- a/systemd/nfs-server.service
> +++ b/systemd/nfs-server.service
> @@ -1,10 +1,12 @@
> [Unit]
> Description=NFS server and services
> +DefaultDependencies=no
> Requires= network.target proc-fs-nfsd.mount rpcbind.target
> Requires= nfs-mountd.service
> Wants=rpc-statd.service nfs-idmapd.service
> Wants=rpc-statd-notify.service
>
> +After= local-fs.target
> After= network.target proc-fs-nfsd.mount rpcbind.target nfs-mountd.service
> After= nfs-idmapd.service rpc-statd.service
> Before= rpc-statd-notify.service
> diff --git a/systemd/rpc-statd-notify.service b/systemd/rpc-statd-notify.service
> index a655445..a7e2f8e 100644
> --- a/systemd/rpc-statd-notify.service
> +++ b/systemd/rpc-statd-notify.service
> @@ -1,7 +1,8 @@
> [Unit]
> Description=Notify NFS peers of a restart
> +DefaultDependencies=no
> Requires=network-online.target
> -After=network.target nss-lookup.target
> +After=local-fs.target network.target nss-lookup.target
>
> # if we run an nfs server, it needs to be running before we
> # tell clients that it has restarted.
> diff --git a/systemd/rpc-svcgssd.service b/systemd/rpc-svcgssd.service
> index f7424b0..41177b6 100644
> --- a/systemd/rpc-svcgssd.service
> +++ b/systemd/rpc-svcgssd.service
> @@ -1,7 +1,8 @@
> [Unit]
> Description=RPC security service for NFS server
> +DefaultDependencies=no
> Requires=var-lib-nfs-rpc_pipefs.mount
> -After=var-lib-nfs-rpc_pipefs.mount
> +After=var-lib-nfs-rpc_pipefs.mount local-fs.target
> PartOf=nfs-server.service
> PartOf=nfs-utils.service
>
> --
> 2.1.4
>
> _______________________________________________
> systemd-devel mailing list
> [email protected]
> http://lists.freedesktop.org/mailman/listinfo/systemd-devel
Hello all,
Zbigniew Jędrzejewski-Szmek [2015-03-03 19:08 +0100]:
> Are you sure that all of those nfs daemons do not require
> sockets.target and other stuff provided by basic.target to be ready?
The corresponding upstart jobs trigger on virtual file systems (/sys,
etc.) and rpcbind, and we've used them for years. Also, NFS does not
yet use socket activation, or talks to other services on sockets
(except for rpcbind), so we don't need sockets.target either. The
other dependencies (some network.target, some nss-lookup.target, etc.)
are already specified explicitly. So I'm quite sure.
That said, there's of course always a nonzero chance that this breaks
in a case which I haven't tested. In particular, I didn't test
kerberos/gssd, I'd appreciate if someone who has a real-world setup
with that could give this a spin.
Thanks,
Martin
--
Martin Pitt | http://www.piware.de
Ubuntu Developer (http://www.ubuntu.com) | Debian Developer (http://www.debian.org)
On 03/03/2015 03:12 PM, Martin Pitt wrote:
> Hello all,
>
> Zbigniew Jędrzejewski-Szmek [2015-03-03 19:08 +0100]:
>> Are you sure that all of those nfs daemons do not require
>> sockets.target and other stuff provided by basic.target to be ready?
>
> The corresponding upstart jobs trigger on virtual file systems (/sys,
> etc.) and rpcbind, and we've used them for years. Also, NFS does not
> yet use socket activation, or talks to other services on sockets
> (except for rpcbind), so we don't need sockets.target either. The
> other dependencies (some network.target, some nss-lookup.target, etc.)
> are already specified explicitly. So I'm quite sure.
You are correct. rpcbind is the only service we have that uses socket
activation, which I don't think works very well...
Just last week I notice if you reboot a vm and the first command
you type is rpcinfo -p. That command will time out trying to talk
with rpcbind. After the timeout everything works fine...
I thought I opened a bz but I can't seem to find it.
steved.
>
> That said, there's of course always a nonzero chance that this breaks
> in a case which I haven't tested. In particular, I didn't test
> kerberos/gssd, I'd appreciate if someone who has a real-world setup
> with that could give this a spin.
>
> Thanks,
>
> Martin
>
On 03/03/2015 12:28 PM, Martin Pitt wrote:
> Hello NFS developers,
>
> reposting the two patches inline as requested by Steve.
>
> I'm currently systemd-ifying our nfs-utils Ubuntu package. For testing I put
> the NFS server and client (i. e. localhost:/foo/bar mounts) on the same
> machine. With that I get long hangs during shutdown on stopping the NFS .mount
> units, as when that happens the NFS server is already shut down.
>
> This is certainly a corner case as you'd usually not NFS-mount a share from
> localhost; but fixing it is relatively simple with the first patch, which makes
> sure that if NFS server and client are installed, the server starts before the
> client, and the client stops before the server.
>
> For a client without installed server this is harmless as Before= does not
> imply a dependency. Likewise, for an NFS server which does not mount shares by
> itself, it's also a no-op as remote-fs.target is empty. This would only
> slightly reorder the boot sequence for machines which both are an NFS server
> and have some remote NFS mounts, but I don't see an issue with that.
>
> The second patch make NFS start earlier in the boot (i. e. before
> basic.target), so that you can do things like put /var/ on NFS, or have rcS
> SysV init scripts which depend on $remote_fs work. I tested this on both a
> server and a client. This is certainly a bit more intrusive, but could be
> worthwhile; what do you think?
>
> Thanks for considering,
Committed....
steved.
>
> Martin
>
> _______________________________________________
> systemd-devel mailing list
> [email protected]
> http://lists.freedesktop.org/mailman/listinfo/systemd-devel
>