🚀 go-pugleaf

Quoth GEustace@massey.ac.nz (Glen Eustace) (in <1039@massey.ac.nz>):
#[A series of questions about sharing failures]

When you enable locking and sharing services by mounting a file system
with /MS (or the equivalent in NFSCONF), PC-NFS calls the portmapper on
the server to obtain the UDP port for the Network Lock Manager (NLM),
squirrels this away in the mount table and sets an internal "MS" flag
on the drive. The effect of MS is twofold.  First (and most obviously)
DOS Lock and Unlock (Int21H 5C00H and 5C01H) calls are intercepted and
translated into the equivalent RPCs to the NLM. Secondly, all file
opens and closes become subject to file sharing verification, as
described in the DOS Technical Reference for the Open (Int21H 3DH)
call. When a file is opened, an NLM_SHARE call is made to the NLM to
check that it is OK to open the file in the requested mode for the
requested use. And obviously, when the file is closed an NLM_UNSHARE
call is made to tell the NLM that the file is no longer in use.  Thus
even if you don't think "any locking is going on", every file open and
close is causing at least one RPC call to the NLM on the server.

So far, so good. What happens when things break down? Unlike the
NFS server, the NLM is a regular user level process, and on a busy
server it is possible for the process to get swapped out and not
respond for an extended period. Obviously this is more likely on a
smaller or slower server. 

There is another potential problem. Suppose the server is rebooted,
or even that the NLM is killed and restarted. Any locks/shares that it
was holding on behalf of its clients are lost. In Unix there is a
mechanism to re-establish locks, based on the status monitor. (For
more details, see the Usenix paper by Jo-Mei Chang from ~1986.) Unfortunately
this is too heavyweight a scheme to shoehorn into a PC. And even
if  there were no open files, and thus no outstanding locks or shares
to be recovered, the UDP port in the mount table is going to be
incorrect after the NLM is restarted for any reason.

In PC-NFS 2.0, 3.0 and 3.0.1 any server failure - real or perceived -
is treated as fatal to the current application if it touches the
affected drive. This is prudent; it is also the way in which most other
PC networks handle things. However we also mark the drive as FAILED
(which is where the "?" in the "NET USE" display comes from) and
disallow any further access to the drive until it has been remounted.
Any attempted access will provoke the critical error message you
describe. This is actually overkill, and in the next release of PC-NFS
we'll fix this and call the portmapper to reacquire the NLM port. This
means that if you have a drive mounted /MS and someone reboots the
server, the next time you touch a file on the drive PC-NFS will try the
old port, time out (which takes a few seconds) and then get the new
port and proceed.

Geoff

-- Geoff Arnold, PC-NFS architect, Sun Microsystems. (geoff@East.Sun.COM)   --
   *** "Now is no time to speculate or hypothecate, but rather a time ***
   *** for action, or at least not a time to rule it out, though not  ***
   *** necessarily a time to rule it in, either." - George Bush       ***