GitLab

from the interfaces table of the parent node. This coincides with
Rob's changes to stash the IP there, and arrange for it to go into
DNS.

allocated to an experiment.

(not to be confused with "tarballs") to download an RPM file.

the types table yesterday, and it is hardwired here. Yuck.

changes; Print a warning when the DB slots are null, and return
standard 255.255.255.0. This is for the transition phase.

Also add version 12 modification to routerconfig; return the "src"
from the virt_routes table, so that the client side can figure out
what interface the route is associated with. There are associated
changes in libsetup to ifconfig,routerconfig,tunnelconfig.

that part out since it is not finished/tested.

inside, but rather I do just enough to get the card booted (using
Abhijeet's minicom/expect scripts, but with changes to support the
configuration coming from tmcd. I also create a file of interface and
routeadd directives, so that the network configures properly, but
thats about it. Getting a more complete client side environment that
includes perl and sshd for the arm will have to wait.

to return the name/type of subnodes. Added a subconfig command that
returns a type-specific configuration that the client will know what
to do with. At the moment, just ixp is known as a subnode. The subnode
support mostly works off the vnode stuff since otherwise it would be
lot of code duplication. Eventually subnode should replace vnode, and
vnode will become a type of subnode (like ixp).

Also changed usage of control_net to control_iface, and made use of
the "role" slot in ifconfig to determine what interfaces to return
info for.

iron out.  TMCD/libsetup now has a plabconfig commands that parallels
the jailconfig command.  The Plab boot process has been added to
libsetup and a -p option has been added to vnodesetup to parallel the
-j option.  Parts of the code that were Jail-specific, but labeled
just as vnode stuff have been renamed.  $vnodedir in vnodesetup has
been removed, since it was redunant with libsetup's CONFDIR, and
CONFDIR is much more intelligent.

syncserver (if any) for the experiment, and whether the node is also
the actual syncserver and should start the daemon.

syslogging to a fraction of its former self. Actually, its mostly been
moved under if (verbose) tests. Instead, just syslog the number of
bytes returned for each request.

Added a signal handler to change the verbosity of a running tmcd.
To turn on verbosity:

	kill -USR1 `cat /var/run/tmcd.pid`

To turn off verbosity:

	kill -USR2 `cat /var/run/tmcd.pid`

You can send the signal to individual children, but that would be
silly and pointless.

experiments created in groups and experiments created in the main
project. Too easy for the queries to get out of sync, and the
advantage is minor anyway.

stand alone (no underlying phys interface) so that both ends of a
virtual link can be on the same node.

Change "startcmd" to return the uid of the swapper, not the creator,
as the UID to run the startup command.

Get creator/swapper uid in the initial query for the node, instead of
as separate queries later. Saves some queries ...

Remove old code.

Add progagents command to return the list of program object names, and
the UID to run those objects as (again, the swapper not the creator).
This allows the program-agent startup code on the node to be more
selective in its elvin subscription, as well as running the programs
as the proper UID.

Changes to jailconfig. If the node has its own IP in the nodes table
(jailip), return that as JAILIP in the configuration for mkjail.

returned for all users. Actually, it was just not resetting the
count back to zero, so all users got their groups, plus a few extra.

dojailconfig: Do not return a jail config unless the node is really
jailed.

doifconfig: Look at the node name, and return ifconfig info specific
to the name instead of always returning all info for the machine. Upon
boot, virtual interfaces will not be setup, but instead will be setup
when the jail is setup. When the jail requests ifconfig info, it gets
only its list of interfaces to setup.

dolinkdelays: Ditto, since linkdelays will be setup after per-jail
interfaces are setup.

to interfaces) so that we return the MAC of the virtual interface not
the physical interface.  I do not much like this arrangement.  We
store veth interfaces in another dynamic table, so must join with both
interfaces and veth_interfaces to see which iface each linkdelay
corresponds to. If there is a veth entry use that, else use the normal
interfaces entry. Maybe we should use the regular interfaces table,
with type veth, entries added/deleted on the fly. I avoided that cause
I view the interfaces table as static and pertaining to physical
interfaces only, but the current arrangement does not scale well
(although if it only ever has to scale to veth_interfaces, so be it).

of device types. INTERFACE=%d is gone, replaced by IFACETYPE=eth or
veth.
Also removed some shark cruft and generally cleaned up the function.

who gets root on widearea nodes, inside and outside of jail. Kinda
brute force; might need to make this more flexible at some point,
perhaps with a node/user mapping table for widearearoot (root outside
the jail), and a widearea_trust slot to the group_membership table
(root inside a jail), but this will do for now since its handled
entirely inside of tmcd.

I was originally using local_root to determine root access inside the
jail, but we need to more finely control who gets root on widearea
nodes. Outside the jail, only tbadmin got jail, and thats definitely
too restrictive!

and /share, all rooted in /netbed. This is convenient! New images have
symlinks pointing /{proj,groups,share} into /netbed.

clients so they can create a proper .forward file. 2) Add some
test code for seeing how feasible it is to do accounts on ops/tips
using tmcd instead of direct ssh. Initial results are not good; too
many accounts; does not scale well. Going to need a smarter protocol
for doing hundreds of accounts, but I will leave this code in place
for now. Note, unlike testbed nodes, control nodes must connect with
ssl.

Bad idea. Use an auto_incrementing idx field instead.

simply compiled into tmcd itself.  We will probably extract them from the
DB a little down the road, possibly setting different values for different
projects, experimenters, etc.

jail. This was attempting to get this from the tunnels table and
the interfaces table, but that was plain irritating. Just get it
from virt_nodes. Hey, its a virtual node, so use a virt table!

INADDRANY: When 1, jail is allowed to bind to INADDR_ANY. When packet
           comes in, the pchlookup checks the prison IPs.

ROUTING:   Jail gets access to its routing table. This presently implies
           that the jail gets its own private routing table via new
	   jail options.

DEVMEM:    Jail gets a real /dev/mem and /dev/kmem instead of a
           symlink to /dev/null. This pretty much bypasses security so
           its not something to do on widearea nodes, but on local
           nodes that fine.

even when the delays table is not setup properly (cause its an
experiment swapped in under the old assign_wrapper).

for lan node delays). Also, return the vnode names for both delays
and linkdelays so that the agent knows what vnode it is operating on
behalf of. This is for the event subscription. The intent is ti get
away from the pipe number stuff.

tweaks.
Add IPALIASES to ifconfig when (vers > = 8). This is the initial
approach for doing emulated links.

but eventually did not use it. Current images still make a call to it
from libsetup but will get an empty string that doesn't affect anything.

be flushed (fingers crossed!).