GitLab

before deleting it.

Convert most SNMP interaction to use the snmpit*() library, so that
they get support for retrying failures, etc.

Add new library calls for wrapping bulkwalk() - so now, we will
retry those on error as well. Before, we had the bad behavior than
many functions, like listVlans() would just see empty lists instead
of errors.

When making a new Cisco object, we now test network connectivity right
away, by fetching an OID that should exist on all SNMP devices. Before,
we wouldn't find out we couldn't contact the switch until we actually
did something on it.

Also, make VLAN number choosing go a bit faster by converting it to
bulkwalk() (using the new library function) so we can grab all VLAN
numbers at once.

we're going to remove. We used to do all VLANs in one lock for
performace reasons - however, I'm discovering that the lock can get
held for such a long time when many VLANs are being deleted that other
VLAN operations, such as listing VLANs, can fail. And, it's not
actually that much slower to grab a new lock each time.

This seems to be necesary because of what I can only describe as a
bug in CatOS in which not ll ports in a trunk return the proper channel
grou number.

EtherChannel and try another trunk port before giving up altogether.

turns out that IOS now supports the CISCO-STACK-MIB, which is what
we use in CatOS to set them.

2900 series switches acutally use a different MIB for this stuff,
so I added support for determining the 'class' of switch.

operation on our existing stack 3x faster.

The real reason for doing this, though, is robustness - the old
code was terribly fragile, depending on several tables being in order,
etc. This was breaking on some of our new switches.

Also fix a few typos in debugging messages.

so -B now ignores that rather than warning.

snmpit_cisco: hack to get around 2950 switch returning just <port>
rather than <module>.<port> in interface info tables.

Note: snmpit_cisco does not successfully return speed/duplex info for -b,
this may cause problems with whack-on-lan and needs to be fixed!

Non-modular switches with two different types of interfaces give us
trouble under IOS because they appear to be on the same module, but
the duplicate card numbers. ie. Fastethernet0/1 and Gigabitethernet0/1
Hack around this by pretending the gigabit ports on such a switch are
on module 1, not module 0.

This works fine on the 29xx series, but probably won't work well for
anything else. The real solution is to move to iface instead of
card.port, or use a type.card.port triple.

In order to invoke this hack, you must put '-modhack' in the type for
the switch in the nodes table.

trunk.

Added support for Cisco switches to enable and disable trunking for a
port.

Added two new command-line options:
-N <port> <vlans>
	Enable trunking on a port, allowing the given list of VLANs to
	cross the trunk

-U <port>:
	Disable trunking for the port

This functionality is needed for Mike's firewall support.

retries and better error reporting.

file to a tftp server.

Short term, will be used to periodically save off the configurations
of our switches for backup purposes.

Long-term, could be the start of automated configuration for Cisco
devices that reside within the testbed.

information. Rather than pass it all in, which was getting very
cumbersome, and inconsistent between Cisco and Intel switches, the
modules query the database themselves (via a new function in
snmpit_lib.pm).

Also in this commit are two new options for switch stacks - the
ability to specify minimum and maximum VLAN number to use.

dying while holding the VLAN edit buffer lock. Also, avoid uninit.
value warnings in some cases when we don't get a response from the
switch. Finally, add some more debugging information.

More changes on the way.

get an updated copyrights message.

that are being removed.

setting port speed and duplex. Well, they probably do, in some MIB,
but not in the one we use on CatOS, and these features aren't actually
very important right now.

Also, make snmpit itself handle unsupported port operations.

VLAN domain. This means that we have to create and delete VLANs on
all switches, instead of just the 'leader'. This behavior is
controlled through the new single_domain column in the
switch_stack_types table.

Most snmpit_cisco functions now take Cisco-specific VLAN numbers,
rather than testbed VLAN ids.

This is an intermediate step in removing dependance on VTP.

As a nice side-effect, this saves another 25% of the time to
operate on topologies for large experiments.

for large experiments. Not as much of an improvement as I had hoped,
but further improvement will probably take a lot more rearranging.

printed incorrectly.

    Accept [switch.]<module>/<port> format for ports, so that we can
	deal with ports not in the database (mostly for my own
	debugging sanity.)
    A -n option that prevents assign from changing hardware settings
    	(though, unlike TESTMODE, does read some information from
	the switches)
    Private VLAN support, through the -x,-y, and -z switches. There
	are only 5 letters of the alphabet left, so I've given up on
	memnonic switches.
    Worked a bit on making VLAN deletion more efficient, but with
    	little sucess

Private VLANs work like so:
Make a primary private VLAN with:
snmpit -m myvlan-primary -y primary
Attach a community VLAN to it like so:
snmpit -m myvlan-community -y community -x myvlan-primary -z cisco2.1/15
Put some ports into the community VLAN:
snmpit -m myvlan-community pc1:0 pc2:0

set of MIBs, and index some of the tables differently.

This requires some more information to be passed when switch and stack objects
are created.

Also, make the SNMP community to use overridable in the defs file.

Re-worked the snmpitGet functions slightly to make them more
convenient and to work around some ickiness in the SNMP library.

time in between retries.  This should help out when we have dozens
of experiments swapping in and out at once.

(and so, has been renamed setVlansOnTrunk). Part of the work necessary
to get the stack-level setVlanOnTrunks function to accept multiple
VLANs in one call.

take more than one. This can increase efficiency, since, for example,
we now only have to lock the VLAN edit buffer once when tearing down
an experiment, instead of once per VLAN. There are still other
functions (setVlanOnTrunks comes to mind) that could benefit from this
treatment.

These optimization have been made only for Ciscos - minimal changes
necessary to keep Intel support working were made, but they will still
have the same old slow behavior.

return value on VLAN creation. If it reports success, we look for the
VLAN with findVlan(), to make sure it's really there. We retry a
couple times if it didn't work. This will add a small (sub-second)
time penalty per VLAN created.

We ran into one instance, that we've noticed, where the switch
reported that VLAN creation succeeded, but the VLAN did not actually
get created on the switch.

OID from the CISCO-VLAN-MEMBERSHIP MIB. But, this MIB does not seem
to be supported by all Cisco switches. So, we now use vlanPortVlan
from the CISCO-STACK MIB.

So, we no longer need CISCO-VLAN-MEMBERSHIP-MIB at all, which should make
snmpit usable on a wider variety of Ciscos.