Re: [cobalt-users] [Raq2/Qube2] Sleeping nfsoid loaded in Kernels?

[flash22@xxxxxxx]

On Fri, 1 Dec 2000, James Hoaggs wrote:

> [bigger snip]
> 
> IMHO i'm not sure. I'm just wondering if the NFS client nfsiods can be
> awoken to some remote NFS server via some port by UDP without TCP & portmapper.
>  BTW, email is not perfect though your responses are always 3 days late,
> also when I hit reply it only goes back to you, not the list. Is your
> mail server on the fritz? 
> 
> Also, can you please explain what this script in /usr/sbin/exportfs 
> is doing regarding the nfs and rpc services? :
> #!/bin/sh
> 
> usage () {
> 	echo "usage: exportfs [-a]"

> Though it looks like the command actually stops the service, as though
> it was already running, like when done for SSH.

Because nfsd will only mount local filesystems listed in /etc/exports, and
you probably (hopefully) have none listed in there so there's nothing
that's mountable...

Normally that script just restarts nfs services so the file will be
reread, tho interestingly, it's not a very safe script, it doesn't check
for current active mounts, so removing a mount point from exports and
restarting will probably make the rpc mount daemon very unhappy -/

All the nfsiod's do is keep a list of resources that need to be in kernel
space, linux nfs runs in user space and the nfs daemon attaches to the
kernel and passes info back and forth with the nfsiod's to impliment the
whole nfs system, if you don't have nfsd, you have no way to access data,
nfsiod can't access the disk by itself...

(This is also how linux manages to share nfs mounts across physical mount
points, something many unices can't do because they map nfs internally
directly to the filesystem drivers and want everything to map to the same
physical device...)

btw, the biggest risk in using nfs isn't the security, it's really not all
that much worse than say, telnet (yea yea, telnet is about as secure as
putting scotch tape on the door) but rather due to admins not
understanding what they are doinf when exporting filesystems, and thus
allwing people to create things on the filesystem they have no business
being able to create in the first place (like device nodes or suid files)
---
Yeah, i know it takes forever for things i post to show up on this list,
it seems to be a (known) quirk in mailman, it doesn't seem to like it
when the delivery address is different than the original subscribe
address, even tho it allows you to subscribe that way...kinda sad,
majordomo can handle that just fine;0

My mail server is 2 hops from the backbone, so i would hope it's not an
issue;0


g