gregtampa / coreemu Goto Github PK

if you launch the CORE GUI as a normal user, remote execution mode is
turned on and a warning is printed, but this behavior is confusing for new
users who then try to build and run a topology... connecting attempts to
the default execution servers will result in timeout delays.

The FreeBSD 8.1-RELEASE kernel warns of "freed uma keg was not empty" and 
memory being lost. After several emulation start/stops, the kernel may crash.

Unknown if 8.1-STABLE or -CURRENT branches solve this issue.

1. create at least two nodes and change their names
2. start the emulation
3. click on two-node tool and try to select the two nodes and an error box
pops up:
"Error: can't read "hostname": no such variable"
(where "hostname" is the new name chosen for the selected node 1)

backtrace:

can't read "hostname": no such variable
    while executing
"set $node"
    (procedure "ifcByPeer" line 4)
    invoked from within
"ifcByPeer $src $node2"
    (procedure "setTwoNodeCommand" line 22)
    invoked from within
"setTwoNodeCommand"
    (procedure "selectTwoNode" line 26)
    invoked from within
"selectTwoNode $curnode"
    (procedure "button1" line 150)
    invoked from within
"button1 .c 751 209 none"
    (command bound to event)

What steps will reproduce the problem?
1. goto ~/.core/configs or where ever you have the *.imn files
2. run core -b sample1.imn (or anyother *.imn file in that directory)
3. get couldn't open error

batch execute sample1.imn
couldn't open "sample1.imn": no such file or directory
    while executing
"open $argv r"
    invoked from within
"if {$execMode == "interactive"} {
    # GUI-related files
    source "$LIBDIR/widget.tcl"
    source "$LIBDIR/tooltips.tcl"
    source "$LIBDIR/initgu..."
    (file "/usr/local/lib/core/core.tcl" line 252)

You get a similar error when using core -s

What is the expected output? What do you see instead?
I expect that the file should be found and opened.  The user expects that 
relative paths will work and there is no indication that the user must use an 
absolute path.  This causes confusion because the user may assume there is a 
problem with other parts of the system (like permission or connectivity if 
mounted over NFS) before discovering the source of the problem

What version of the product are you using? On what operating system?
4.1 built from source release on linux sabyon
4.1 from deb on site in linux ubuntu 10.04 VM

What steps will reproduce the problem?

run the CPU widget on a FreeBSD 8.0 machine

What is the expected output? What do you see instead?

error box pops up; the output of vimage -l no longer displays CPU usage

What steps will reproduce the problem?
1. if user does not install ebtables on a system but installs core, core
will build and run, but api calls to instantiate links will fail in a
cryptic (corepyd.log) manner

What version of the product are you using? On what operating system?
netns

Please provide any additional information below.

Probably would suggest to check when starting core that ebtables is working
on the platform.  A configure check may help too.

Installing CORE 4.0 for Linux from source fails in subtle ways if packages are 
missing.

If brctl is not installed, one will get a runtime error:
EnvironmentError executable not found: no/brctl
due to the configure program not finding a path to brctl and substituting the 
string "no"

If libtk-img is not installed, a Tk error is raised about a "Window" and if one 
goes back to the console, one learns that 
*** error: Img package not found
CORE now requires the Tcl/Tk Img package, for supporting other image
types besides GIFs (e.g. JPG, PNG, etc.). Please install it with:
  yum install tkimg   (RedHat/Fedora)
  sudo apt-get install libtk-img   (Debian/Ubuntu)

Suggestion:  fail to configure successfully if ebtables, bridge-utils, and 
libtk-img are not found

CORE writes files to the /etc/core directory. This should be moved to a /var/* 
directory and/or a user configuration dir.

The files modified are:
- plugins.conf
- servers.conf
- widgets.conf
- any imn file in the configs/ subdir

The feature has been requested to allow comments within the imn config file, 
e.g. within a node declaration block.

the 2.6.35-23-generic kernel under Ubuntu 10.10 on Dell M4500 exhibits vnoded 
lockups when shutting down emulation

cause is currently unknown (related to IPv6 flush bug?)

workaround is to install latest 2.6.37 rc kernel from ppa, where problem is 
fixed

What steps will reproduce the problem?
1. Opening many different nodes 30 with custom-post-config-commands in openvz
2. Programs access and write to files.
3. Typing "vzctl start <containerid>" results in
vzquota : (error) Quota on syscall for id <containerid>: Device or resource
busy
...
...
...
vzquota : (error) Currently used file(s):

big list of files which are used

What is the expected output? What do you see instead?
All the nodes should start up.

What version of the product are you using? On what operating system?
Openvz

Please provide any additional information below.

Having all the nodes start before running custom-post-config-commands
should make this work correctly.

ethereal interface capture is broken on OpenVZ, possibly BSD 8.0 also

this should be renamed to wireshark

when saving/loading imn config files, relative instead of absolute
pathnames should be used.

sample1.imn is not working because /usr/local/etc/core/sample1-bg.gif is
hardcoded in the file, may be /etc/core/sample1-bg.gif

filenames should be relative to the location of the config file, so
/etc/core/configs/sample1.imn could refer to
/etc/core/configs/sample1-bg.gif with simply "wallpaper {sample1-bg.gif}"

user cannot set custom-post-config or custom-pre-config commands using the GUI
they must be edited in the imn topology file

the command-line utilities provided by CORE should have man pages or some form 
of documentation

1. applying link effects during runtime to N1---N2---N3, qdisc show only 
changes the N1N2 link

2. assign delay of 100ms during edit, each link gets 50ms
   assign 100ms after starting experiment and each link gets 100ms

3. removing link effects (0 each field) has no effect

What steps will reproduce the problem?
1. start a CORE scenario
2. open the services dialog and choose customize
3. the node/link messaging used to build custom configs will kill all existing 
nodes and processes

What is the expected output? What do you see instead?

Need to add a runtime check and suppress sending of node/link info, which 
cored.py will already have during runtime.

The Tcl/Tk GUI launches the global Experiment Script immediately, but according 
to the help text in the Experiment Script dialog, this script should be 
launched after the emulation has been started.

The GUI needs to take into account the emulation/session states during startup.

GUI feature request for adding the ability to hide items while the scenario is 
running, e.g.:
.c itemconfigure "node && n3" -state hidden

When I try to create a mobility scenario for Core OpenVZ (installed from
rpm) it defaults to looking for configs in /usr/local/lib/core.

If I create this scenario instead in my home directory /home/core and
navigate to it via the GUI, and select it, the scenario is not loaded, 

Instead, this warning prints out on the console:

***Warning: could not open script file '/usr/local/lib/core/10node.scen'

So, it seems that the GUI looks for this script from the relative path
/usr/local/lib/core regardless of what is selected from the GUI.

on a quagga router, change the IPv4 address and a new network statement is
added to the router ospf block of the configuration, however the old
network statement is not deleted

this could cause routing problems, depending on the addressing scheme used
in the network

What steps will reproduce the problem?
1. Start an emulation with at least one node not positioned at 0,0
2. Change the icon of a node with the coreapisend code
3. The node will pop back to 0,0 with a new icon

What is the expected output? What do you see instead?
Just change the icon no motion should be effected.

What version of the product are you using? On what operating system?
openvz from svn source with 2.6.27-briullov.1 kernel

Please provide any additional information below.
Incorrect behaviour also occurs if while moving a node with the mouse, core
receives a coreapisend call for that node.  It pops the node back to the
initial starting position when the node was initially clicked, but not the
link information.  Having an node "selected" should block on all
coreapisend calls which have to do with position. (ie the icon should not
be placed unless the current position is known from the mouse not the "old"
position)

For Linux version, the manual (section 2.5.5) says to look at config/s
directory; however, this is not distributed or not easily found from the
GUI when running core.

during execution mode you can change the link effects

under BSD and OpenVZ these are applied correctly, but under netns instead
of getting delay/2 on two interfaces you get delay on one interface

Error window on my GUI after it was running overnight (Core 3.4 OpenVz):

Error: divide by zero

divide by zero
divide by zero
    while executing
"expr { 100 * $usage_time / $total_time }"
    (procedure "getCPUUsage" line 25)
    invoked from within
"getCPUUsage"
    (procedure "monitor_loop" line 11)
    invoked from within
"monitor_loop "
    ("after" script)

batch mode does not work on Linux

the option should be removed in favor of Python scripting or disconnected GUI 
sessions; otherwise the 'core --batch file.imn' option could be made to work by 
starting a new session and disconnecting from it

What steps will reproduce the problem?
1. Create a wlan and 2 routers
2. Link the routers to the wlan and assure they are in range of each other
3. Run emulation
4. Attempt to ping one router from the other

What is the expected output? What do you see instead?
No traffic is sent between the nodes.  Ping prints "connect: Network is 
unreachable".

What version of the product are you using? On what operating system?
$ core -v
CORE version 4.1pre1 (20100806)

$ uname -a
Linux aaron-work-laptop 2.6.32-21-generic #32-Ubuntu SMP Fri Apr 16 08:10:02 
UTC 2010 i686 GNU/Linux

$ cat /etc/lsb-release
DISTRIB_ID=Ubuntu
DISTRIB_RELEASE=10.04
DISTRIB_CODENAME=lucid
DISTRIB_DESCRIPTION="Ubuntu 10.04 LTS"

Tried re-installed CORE and quagga.  "/etc/init.d/core status" shows both 
cored.py and span running.  "netstat -nr" shows a totally blank routing table 
on all routers.  Routing works fine with anything other than wlans (e.g. 
switches, hubs, static links).

CORE 4.0 on Ubuntu Linux.
Connect an RJ45 to a hub or switch and connect that to a router. Press start. 
You'll end up with the RJ45 device part of its own bridge, no connectivity.

Reported by Dan Martin

Ubuntu 10.04, connect RJ45 tool to a hub/switch and assign link attributes. No 
effects are applied to the link. When RJ45 is connected with a router it works 
fine.

When reconnecting to a running session, node types and icons are not preserved.

Nodes appear only as routers. The node type and icon should be stored in the 
cored.py objects. Note that custom icon info is not currently sent via API from 
GUI -> cored.py, only node type and model.

the Python daemon applies the MAC address provided by the GUI, so this would 
involve adding GUI support for configuring the MAC address instead of having it 
auto-generated

What steps will reproduce the problem?
1. Start core and cored
2. Add at least one node (network-layer or link-layer) to canvas.
3. Start experiment
4. Core segfaults immediatlly

Output:
Connecting to "cored.py" (127.0.0.1:4038)...connected.
Segmentation fault


What is the expected output? What do you see instead?
Core should not segfault

What version of the product are you using? On what operating system?

I'm using 4.1 20101222, built from source code
linux 2.6.33-sabayon on x86_64
kernel build #1 SMP
TCL version 8.5.7 (/usr/lib64/tcl8.5)
Tk version 8.5.7 (/usr/lib64/tk8.5)
X11 server X11R0 The X.Org Foundation 10903000

Sabayon is a gentoo variant

Please provide any additional information below.
I've tried different versions including a svn build from early December, and 
4.0 the issue was the same.   Also the /etc/init.d/core start script doesn't 
work and I have to run sudo /usr/bin/python /usr/local/sbin/cored.py -d to 
start the core daemon.  If the daemon is not started then core gives an error 
about it not being started but doesn't segfault.

By running scripts from the command line (core -s ~/.core/configs/sample1.imn) 
I also get the same segfault I see the window load briefly before it segfaults. 

If I run using batch mode core -b ~/.core/configs/sample1.imn I do manage to 
start the experiment and the bridges and stuff are created.  I can then start 
the core gui and connect the the existing session.  At this point core appears 
to work as intended, except if I stop the experiment and then try to start it 
again in which case I get a segfault.  In this case stopping the experiment 
dose not destroy the bridges and network interfaces setup for the experiment, 
and I can still connect to the experiment though another instance of the GUI.

1. add a square or oval annotation object
2. try to place a new router node on top of it
3. the only way to place the new node is by clicking somewhere on the empty
background and moving dragging the node back onto the annotation

this event is no longer a canvas background click event because the
annotation is in front of the background

What steps will reproduce the problem?
1. Compile CORE v. 4.0 or nightly (older versions not tested).
2. Set up any scenario.
3. Start emulation

What is the expected output? What do you see instead?
 The emulation does not start, an 'Application Error' window appears and generates the output in the attached .log file.

What version of the product are you using? On what operating system?

 $ core -v
 CORE version 4.0 (20100806)

 $ uname -a
 Linux ln22di 2.6.32-24-generic #39-Ubuntu SMP Wed Jul 28 06:07:29 UTC 2010 i686 GNU/Linux

 $ cat /etc/lsb-release 
 DISTRIB_ID=Ubuntu
 DISTRIB_RELEASE=10.04
 DISTRIB_CODENAME=lucid
 DISTRIB_DESCRIPTION="Ubuntu 10.04.1 LTS"

I’m using core 4.2pre1. I just noticed that RJ45 nodes don’t work when 
assigned to a WLAN node. When starting a network consisting of a router and a 
RJ45 node, both assigned to a WLAN node, the following error message appears in 
the log:

Thread-2: exception while handling msg:
CoreLinkMessage <msgtype = CORE_API_LINK_MSG, flags = 0x1 <CORE_API_ADD_FLAG>>
  CORE_TLV_LINK_TYPE: 0
  CORE_TLV_LINK_N1NUMBER: 1
  CORE_TLV_LINK_N2NUMBER: 3
  CORE_TLV_LINK_DELAY: 0
  CORE_TLV_LINK_BW: 0
  CORE_TLV_LINK_PER: 0
  CORE_TLV_LINK_DUP: 0
Traceback (most recent call last):
  File "/usr/local/sbin/cored.py", line 301, in handlemsg
    reply = msghandler(msg)
  File "/usr/local/sbin/cored.py", line 556, in handlelinkmsg
    hwaddr = mac1, ifindex = ifindex1)
  File "/usr/local/lib/python2.6/dist-packages/core/netns/vnode.py", line 431, in newnetif
    self.attachnet(ifindex, net)
  File "/usr/local/lib/python2.6/dist-packages/core/netns/vnode.py", line 460, in attachnet
    self._netif[ifindex].attachnet(net)
  File "/usr/local/lib/python2.6/dist-packages/core/netns/vnode.py", line 53, in attachnet
    net.attach(self)
  File "/usr/local/lib/python2.6/dist-packages/core/netns/nodes.py", line 313, in attach
    "RJ45 networks support at most 1 network interface"
ValueError: RJ45 networks support at most 1 network interface

using CORE 4.1, create a scenario having an RJ45 interface assigned to a 
physical Ethernet and press start

the underlying physical interface is not brought up

with CORE 4.0 the interface is normally brought up

with tcpdump logging turned on, you can get 0 byte dump files in
/tmp/core_logs, indicating that tcpdump was killed without writing its
buffered data to the log file

if you kill tcpdump manually, the dump file is populated

under OpenVZ, all of the containers think they have the same hostname

the individual hostnames n1, n2, n3, etc are correctly set in the
/etc/vz/10*.conf files, but if you open shells and type 'hostname' they all
report the same name

the vzctl utility says "setting hostname n3", and works correctly when
running vzctl stop/start from the command line, but maybe can't handle the
concurrent operation of coreexecd starting several nodes simultaneously

furthermore, the "HOSTNAME=" line in /etc/sysconfig/network is simply set
to the last node that was created; this file should be moved to per-node state

when configuring a tunnel, the TCP or UDP selection dropdown does not
properly reflect the current setting, and remains the same when you open
the dialog again

What steps will reproduce the problem?
1. Run a core emulation on Ubuntu 10.10. 
2. Stop the emulation.
3. Attempt to restart the emulation

What is the expected output? What do you see instead?
Core should work.  The red indicators do not go away.

What version of the product are you using? On what operating system?
Core namespaces from svn. Ubuntu 10.10

Please provide any additional information below.
After stopping the emulation there is an UNKILLABLE process 

/usr/sbin/vnoded -v -c /tmp/pycore.57088/n1 -l /tmp/pycore.57088/n1.log -C 
/tmp/pycore.57088/n1.conf

Furthermore one of the sys log files has...
unregister_netdevice: waiting for lo to become free. Usage count = 3
being sent to it every 10 seconds without end.

Here is a link which seems some other people have had the same issue using some 
of the same underlying code bits that I assume core namespaces is using. 
http://kerneltrap.org/mailarchive/linux-kernel/2010/8/5/4602782
It doesn't seem like its a core bug but a kernel or other bug but in the mean 
time core should have a work around.

Justin Dean

Linux netem queuing discipline offers packet loss percentage and not bit
error rate. This should be reflected in the GUI and explained in the manual
in a section about porting imn files between BSD and Linux.

Would be nice if the GUI queried the CORE services about the underlying
capabilities of the emulation server. This way we could support various
effects available in netem, asymmetrical link effects, etc.

What steps will reproduce the problem?

Enable SSHD service on a node.

What is the expected output? What do you see instead?

Instead of starting the sshd daemon, the following exception is thrown.

Thread-2: exception while handling msg:
CoreEventMessage <msgtype = CORE_API_EVENT_MSG, flags = 0x0 <>>
  CORE_TLV_EVENT_TYPE: 3
Traceback (most recent call last):
  File "/usr/local/sbin/cored.py", line 301, in handlemsg
    reply = msghandler(msg)
  File "/usr/local/sbin/cored.py", line 821, in handleeventmsg
    self.bootnodes()
  File "/usr/local/sbin/cored.py", line 173, in bootnodes
    n.boot()
  File "/usr/local/lib/python2.6/site-packages/core/bsd/vnode.py", line 405, in boot
    self.session.services.bootnodeservices(self)
  File "/usr/local/lib/python2.6/site-packages/core/service.py", line 180, in bootnodeservices
    self.bootnodeservice(node, s, services)
  File "/usr/local/lib/python2.6/site-packages/core/service.py", line 198, in bootnodeservice
    cfg = s.generateconfig(node,  filename, services)
  File "/usr/local/lib/python2.6/site-packages/core/services/utility.py", line 164, in generateconfig
    sshcfgdir = cls._dirs[0]
IndexError: tuple index out of range

What version of the product are you using? On what operating system?

CORE-4.1 and CORE-4.2beta on FreeBSD-8.2-PRERELEASE

Please provide any additional information below.

The attached patch resolves the issue.

the tunnel node no longer works in the FreeBSD 8.x version as of CORE 4.1

when the GUI is scrolled right/left or down/up and the mouse floats over a
node with an observer widget activated, the gray box pops up in the wrong
location and may also move

each node should have one script of commands that will run when it starts
up; this single script should be viewable/editable; this would eliminate
the confusing custom-config, custom-post-config, and custom-pre-config

instead of various tcl files for different router models, there should be a
protocols directory where tcl files are automatically sucked in. This will
be used to build a list of checkboxes where you can enable the protocols
that you want, such as sshd, dhcpd, ospfd, nrlolsrd, etc.

CORE automatically builds Quagga configs for wired and wireless networks.
When you place a hybrid router that has both wired and wireless interfaces,
redistribution is not properly setup between the two. OSPFv3 instance IDs
become mismatched.

What steps will reproduce the problem?
1. Create a wired link between two nodes
2. Run the experiment

At lower level I believe what is happening is a link add message is sent but 
neither iface 1 number or iface 2 number is specified.

What is the expected output? What do you see instead?

You see an error like this on the server with verbose mode turned on
CoreLinkMessage <msgtype = CORE_API_LINK_MSG, flags = 0x1 <CORE_API_ADD_FLAG>>
  CORE_TLV_LINK_TYPE: 1
  CORE_TLV_LINK_N1NUMBER: 1
  CORE_TLV_LINK_N2NUMBER: 2
  CORE_TLV_LINK_DELAY: 0
  CORE_TLV_LINK_BW: 0
  CORE_TLV_LINK_PER: 0
  CORE_TLV_LINK_DUP: 0
Traceback (most recent call last):
  File "/usr/local/sbin/cored.py", line 301, in handlemsg
    reply = msghandler(msg)
  File "/usr/local/sbin/cored.py", line 557, in handlelinkmsg
    net.linkconfig(node1.netif(ifindex1, net), bw = bw,
  File "/usr/local/lib64/python2.6/site-packages/core/netns/vnode.py", line 400, in netif
    return self._netif[ifindex]
KeyError: None

There should not be any error thrown here.  Looking at the code the previous 
line call the function node1.newnetif which return the iface number of the 
interface created, this can be used to prevent the key error.

Also, I believe from reading the code, but have not tested that this prevents 
the link from receiving the bandwidth delay, loss etc configuration.


What version of the product are you using? On what operating system?
core 4.1 on Linux Sabyon (Gentoo)


Please provide any additional information below.
I have attached a patch to fix this issue.

What steps will reproduce the problem?
1. configure a node with the DHCP server service
2. customize the DHCP service (e.g. change the lease time); save imn file
3. start and runs OK, but when restarting CORE, loading from imn file, the 
/etc/dhcp3/dhcpd.conf file appears truncated

What is the expected output? What do you see instead?

saving/loading is mangling the customized config file

starting CORE and placing a node on a fresh install of CORE on CentOS
produces this error; should have a more descriptive error message

With CORE 4.1 something causes the OSPF hello messages to become synchronized 
over time, e.g. running emulation for 1hr. In conjunction with a link-layer 
such as ns-3 wifi that drops packets due to channel contention, this causes the 
OSPF adjacency to be lost.