On Wed, 2002-03-06 at 04:00, Buchan Milne wrote:

> |>>Brad Felmey wrote:
> |>>
> |>>>-Uvh'ing to this build is causing smb mountpoints to hang permanently.
> |>>>I've verified this on seven separate machines so far. If a share is
> |>>>mounted via smbmount, and anything happens to that mount (network drop
> |>>>or whatever) during access. That mountpoint hangs _forever_. It's
> |>>>impossible to kill it with 'kill -9'. It's as bad as NFS. It used to
> |>>>time out and give up after a while, and you could certainly
> |>>>
> |>kill it, but
> |>
> |>>>this is no longer the case.
> 
> What happens after network connectivity returns?

It's dead, and doesn't return.

> |>>Are you sure it is samba? Have you tried 2.2.2 on the same box? Of
> |>>course smb mounts do not olny involve samba, but the kernel also.
> |>>
> |>Yes, 2.2.1, 2.2.2, 2.2.3a, on several machines.
> 
> Did you get the same or different behaviour?

As I state below, 2.2.1 at least works just fine, as does 2.2.2,
although I haven't tested it as extensively as I have 2.2.1 and 2.2.3a.

> |>Syslog shows a kernel failure:

<snip bunch of kernel puke>
 
> Well, this looks like a kernel smbfs issue ... not a samba issue.

Then why does it not show up with 2.2.1?

Presuming it is a kernel issue, should I then bring this to Juan's
attention?

> OK, last night I did the following test on each of Mandrake 8.2beta4
> /kernel-2.4.18-2mdk/samba-2.2.3a-6mdk, Mandrake
> 8.1/kernel-2.4.17-5mdk/samba-2.2.1a-15mdk, Mandrake
> 8.0/kernel-2.4.3-20mdk/samba-2.0.7-25mdk :
> 
> $smbmount //screamer/bgmilne screamer
> #screamer was running Mandrake 8.2beta3 / samba-2.2.3a-6mdk (I think)
> $ ls screamer
> #returns a directory listing
> #remove ethernet cable
> $ls screamer
> #does not return
> #plug ethernet cable back in
> $ ls screamer # on another VC
> # both the new and the previously non-returned ls screamer now return a
> # directory listing
> 
> It is not possible to unmount the mount while the network is down, but
> as soon as network connectivity returns, and another process accesses
> the share, it works fine (except it cannot be umounted at all after
> disconnected operation). I can't reproduce your problem.

I'm sorry to hear you can't. I can. It's making my life a living hell.
I'm getting sick of getting paged at 02:00 because some process failed
due to a blown mount.

> This really isn't what smbfs and samba are for. The samba team
> recommends using NFS between samba boxes rather than smbfs cross-mounts.

Thank you, but the idea is to have live connectivity. The un-live part
comes from a rotten power grid. We have about 40% of the machines in
this building on UPS. Some of the machines I have to connect to are not
on UPS.

The whole reason I chose Samba is because of its previous behavior as a
rather forgiving platform when connectivity drops. Unlike NFS, I could
usually recover from a hung Samba session. I personally view NFS as the
bastard spawn of Satan. I hate it with a passion. I like the idea of a
UNIX-to-UNIX shared filesystem, but it's dog-slow, unstable, and when it
crashes, it WILL NOT recover. When and if the Linux NFS people can make
it work like something I can rely upon, I'll give it another shot.

> Well, as I say, 2.2.3a is working fine for me, and this looks more like
> a kernel issue than anything else. If I can reproduce it, I can try and
> track it down, but I can't reproduce this ...

Okay, that's your prerogative. I doubt the wisdom of putting out Mdk 8.2
with a crashing and unrecoverable Samba/smbfs/whateverItIs, but it's not
my call, I guess.
-- 
Brad Felmey


Reply via email to