Skip to contentSkip to conversation

Raw CAN socket support in LXC?

DISCUSSION John Ernberg 4 participants 12 messages May 8, 2015 – May 12, 2015 CAN network drivers

lore
Share & tools
lore
#0 lore raw

Hi

We're working on an embedded system where we're planning on wrapping our applications in containers as a resource control safety measurement.
From our understanding Raw CAN sockets are not yet supported in LXC,
and now we're wondering why.

Is there any reasoning we should be aware of as to why support has not been added? Except possibly "We have not seen the need" or "No one has taken the time to".

Our system uses linux kernel 3.10.17.

Thank you in advance for any assistance.

Best regards // John Ernberg

#1 lore raw
Quoted 4 lines · John Ernberg

We're working on an embedded system where we're planning on wrapping our applications in containers as a resource control safety measurement.
From our understanding Raw CAN sockets are not yet supported in LXC,
and now we're wondering why.

I don't know - I've never tried to do so. LXC make heavy use of network namespaces, I think netfilter (NAT, filter) is used to grant access to the containers.

Can you elaborate your use case in more detail. Do you want to limit the sending of CAN frames from the containers to the outside? Do you want to filter on CAN-id or even the contents? What about the receiving side? Do you want to limit the reception to certain CAN-ids, too?

Quoted 3 lines

Is there any reasoning we should be aware of as to why support has not been added? Except possibly "We have not seen the need" or "No one has taken the time to".

I think all of the above.

regards,
Marc

Signature · 4 lines
Pengutronix e.K.                  | Marc Kleine-Budde           |
Industrial Linux Solutions        | Phone: +49-231-2826-924     |
Vertretung West/Dortmund          | Fax:   +49-5121-206917-5555 |
Amtsgericht Hildesheim, HRA 2686  | http://www.pengutronix.de   |
#2 lore raw
Quoted 14 lines · Marc Kleine-Budde
Quoted 4 lines · John Ernberg

We're working on an embedded system where we're planning on wrapping our applications in containers as a resource control safety measurement.
From our understanding Raw CAN sockets are not yet supported in LXC,
and now we're wondering why.

I don't know - I've never tried to do so. LXC make heavy use of network namespaces, I think netfilter (NAT, filter) is used to grant access to the containers.

Can you elaborate your use case in more detail. Do you want to limit the sending of CAN frames from the containers to the outside? Do you want to filter on CAN-id or even the contents? What about the receiving side? Do you want to limit the reception to certain CAN-ids, too?

Quoted 8 lines
Quoted 3 lines

Is there any reasoning we should be aware of as to why support has not been added? Except possibly "We have not seen the need" or "No one has taken the time to".

I think all of the above.

regards,
Marc

Hi Marc

Thank you for your reply.

Our main use-case is to prevent a possible misbehaving application to bring down the entire system through memory leaks, crashes etc. At this time we have not considered any filtering or blocking of directions.

Best regards // John Ernberg

#3 lore raw
Quoted 3 lines · John Ernberg

Our main use-case is to prevent a possible misbehaving application to bring down the entire system through memory leaks, crashes etc. At this time we have not considered any filtering or blocking of directions.

Can you see the your CAN interface inside of the LXC container? Or are the RAW sockets the problem?

Marc

btw: you can use systemd to do memory limiting, too. It makes use of the same feature in the kernel (cgroups).

Signature · 4 lines
Pengutronix e.K.                  | Marc Kleine-Budde           |
Industrial Linux Solutions        | Phone: +49-231-2826-924     |
Vertretung West/Dortmund          | Fax:   +49-5121-206917-5555 |
Amtsgericht Hildesheim, HRA 2686  | http://www.pengutronix.de   |
#4 lore raw
Quoted 4 lines · Marc Kleine-Budde
Quoted 3 lines · John Ernberg

Our main use-case is to prevent a possible misbehaving application to bring down the entire system through memory leaks, crashes etc. At this time we have not considered any filtering or blocking of directions.

Quoted 8 lines

Can you see the your CAN interface inside of the LXC container? Or are the RAW sockets the problem?

Marc

btw: you can use systemd to do memory limiting, too. It makes use of the same feature in the kernel (cgroups).

Hi

With the experimenting I have been able to do on an Ubuntu 14.04 LTS computer, I cannot find any CAN or virtual CAN interfaces inside the container.
When I change the config to specifically forward CAN interfaces, like this:
lxc.network.type = can
lxc.network.link = lxcbr0
lxc.network.flags = up
lxc.network.hwaddr = 00-00-00-00-00-00-00-00-00-00-00-00-00-00-00-00 I get a series of parsing errors from LXC and the container cannot be created.

Creating a privileged container did not change anything.

We will also take the systemd alternative into consideration, but the current state is that we'd like to run with LXC, as they provide us with more options for future changes to our use-cases.

Best regards // John Ernberg

#5 lore raw
Quoted 15 lines · John Ernberg
Quoted 4 lines · Marc Kleine-Budde
Quoted 3 lines · John Ernberg

Our main use-case is to prevent a possible misbehaving application to bring down the entire system through memory leaks, crashes etc. At this time we have not considered any filtering or blocking of directions.

Quoted 8 lines

Can you see the your CAN interface inside of the LXC container? Or are the RAW sockets the problem?

Marc

btw: you can use systemd to do memory limiting, too. It makes use of the same feature in the kernel (cgroups).

Hi

Quoted 8 lines

With the experimenting I have been able to do on an Ubuntu 14.04 LTS computer, I cannot find any CAN or virtual CAN interfaces inside the container.
When I change the config to specifically forward CAN interfaces, like this:
lxc.network.type = can
lxc.network.link = lxcbr0
lxc.network.flags = up
lxc.network.hwaddr = 00-00-00-00-00-00-00-00-00-00-00-00-00-00-00-00

You're probably following an ethernet example here, right? So let's first figure out, that lxc does in the background and see how this translates into the CAN world.

Marc

Signature · 4 lines
Pengutronix e.K.                  | Marc Kleine-Budde           |
Industrial Linux Solutions        | Phone: +49-231-2826-924     |
Vertretung West/Dortmund          | Fax:   +49-5121-206917-5555 |
Amtsgericht Hildesheim, HRA 2686  | http://www.pengutronix.de   |
#6 lore raw
Quoted 24 lines · Marc Kleine-Budde
Quoted 15 lines · John Ernberg
Quoted 4 lines · Marc Kleine-Budde
Quoted 3 lines · John Ernberg

Our main use-case is to prevent a possible misbehaving application to bring down the entire system through memory leaks, crashes etc. At this time we have not considered any filtering or blocking of directions.

Quoted 8 lines

Can you see the your CAN interface inside of the LXC container? Or are the RAW sockets the problem?

Marc

btw: you can use systemd to do memory limiting, too. It makes use of the same feature in the kernel (cgroups).

Hi

Quoted 8 lines

With the experimenting I have been able to do on an Ubuntu 14.04 LTS computer, I cannot find any CAN or virtual CAN interfaces inside the container.
When I change the config to specifically forward CAN interfaces, like this:
lxc.network.type = can
lxc.network.link = lxcbr0
lxc.network.flags = up
lxc.network.hwaddr = 00-00-00-00-00-00-00-00-00-00-00-00-00-00-00-00

Quoted 6 lines

You're probably following an ethernet example here, right? So let's first figure out, that lxc does in the background and see how this translates into the CAN world.

Marc

Hi

I first tried to figure out what kind of settings that would be the most appropriate to tinker with, but I could not find anything more relevant so I copied the ethernet example in the end.
I will dig into the LXC code and see what can be done in terms of getting the CAN network interfaces through.

Best regards // John Ernberg

#7 lore raw
Quoted 5 lines · John Ernberg

I first tried to figure out what kind of settings that would be the most appropriate to tinker with, but I could not find anything more relevant so I copied the ethernet example in the end.
I will dig into the LXC code and see what can be done in terms of getting the CAN network interfaces through.

It probably sets up an etherner bridge and some kind of virtual ethernet device between the container and the host.

Can you start a container with a working ethernet setup and post the output of:

brctl show

Marc

Signature · 4 lines
Pengutronix e.K.                  | Marc Kleine-Budde           |
Industrial Linux Solutions        | Phone: +49-231-2826-924     |
Vertretung West/Dortmund          | Fax:   +49-5121-206917-5555 |
Amtsgericht Hildesheim, HRA 2686  | http://www.pengutronix.de   |
#8 lore raw
Quoted 9 lines · Marc Kleine-Budde
Quoted 5 lines · John Ernberg

I first tried to figure out what kind of settings that would be the most appropriate to tinker with, but I could not find anything more relevant so I copied the ethernet example in the end.
I will dig into the LXC code and see what can be done in terms of getting the CAN network interfaces through.

It probably sets up an etherner bridge and some kind of virtual ethernet device between the container and the host.

Quoted 7 lines

Can you start a container with a working ethernet setup and post the output of:

brctl show

Marc

Here's what my ifconfig and brctl looks like when I am running a container with working ethernet bridging.

$ ifconfig
lo        Link encap:Local Loopback
           inet addr:127.0.0.1  Mask:255.0.0.0
           inet6 addr: ::1/128 Scope:Host
           UP LOOPBACK RUNNING  MTU:65536  Metric:1
           RX packets:62708 errors:0 dropped:0 overruns:0 frame:0
           TX packets:62708 errors:0 dropped:0 overruns:0 carrier:0
           collisions:0 txqueuelen:0
           RX bytes:26174279 (26.1 MB)  TX bytes:26174279 (26.1 MB)

lxcbr0 Link encap:Ethernet HWaddr fe:8e:ec:8f:13:24
inet addr:10.0.1.1 Bcast:10.0.1.255 Mask:255.255.255.0 inet6 addr: fe80::801b:d5ff:fe73:3464/64 Scope:Link UP BROADCAST RUNNING MULTICAST MTU:1500 Metric:1 RX packets:121 errors:0 dropped:0 overruns:0 frame:0 TX packets:153 errors:0 dropped:0 overruns:0 carrier:0 collisions:0 txqueuelen:0
RX bytes:9845 (9.8 KB) TX bytes:20560 (20.5 KB)

vcan0 Link encap:UNSPEC HWaddr
00-00-00-00-00-00-00-00-00-00-00-00-00-00-00-00
UP RUNNING NOARP MTU:16 Metric:1
RX packets:0 errors:0 dropped:0 overruns:0 frame:0 TX packets:0 errors:0 dropped:0 overruns:0 carrier:0 collisions:0 txqueuelen:0
RX bytes:0 (0.0 B) TX bytes:0 (0.0 B)

vethTRGO6W Link encap:Ethernet HWaddr fe:8e:ec:8f:13:24
inet6 addr: fe80::fc8e:ecff:fe8f:1324/64 Scope:Link UP BROADCAST RUNNING MULTICAST MTU:1500 Metric:1 RX packets:39 errors:0 dropped:0 overruns:0 frame:0 TX packets:56 errors:0 dropped:0 overruns:0 carrier:0 collisions:0 txqueuelen:1000
RX bytes:3787 (3.7 KB) TX bytes:8005 (8.0 KB)

$ brctl show
bridge name    bridge id      STP enabled    interfaces
lxcbr0         8000.fe8eec8f1324   no        vethTRGO6W

Best regards // John Ernberg

#9 lore raw
In reply to John Ernberg

Hello John,

Quoted 13 lines · On Friday 08 of May 2015 13:17:52 John Ernberg wrote:

Hi

With the experimenting I have been able to do on an Ubuntu 14.04 LTS computer, I cannot find any CAN or virtual CAN interfaces inside the container.
When I change the config to specifically forward CAN interfaces, like this:
lxc.network.type = can
lxc.network.link = lxcbr0
lxc.network.flags = up
lxc.network.hwaddr = 00-00-00-00-00-00-00-00-00-00-00-00-00-00-00-00 I get a series of parsing errors from LXC and the container cannot be created.

Quoted 5 lines

Creating a privileged container did not change anything.

We will also take the systemd alternative into consideration, but the current state is that we'd like to run with LXC, as they provide us with more options for future changes to our use-cases.

you cannot use Ethernet bridging. This cannot work. But you should be able to forward whole interface to the container

lxc.network.type = phys

I have not tried this for CAN, but it works with Ethernet and I expect that in this case there is no significant difference to CAN. But you lose (at least for Ethernet) access to the interface from the outer system.

If you want to filter frames then you should be able to setup virtual CAN interface in the outer/host system, then setup CANGW to route messages in the interrest and the start container which hides virtual can interface from the outer system but CANGW setup should continue to work after this move of interface to other namespace.

I am out of time with teaching now so I do not expect to find time to test that soon but I hope that hint is in the right direction. If there is some need to do code adaptations somebody of my colleagues could have interrest to look at it if we are sucesfull with gaining one of our partially related grant or we find other funding.

Best wishes,

Pavel

#10 lore raw
Quoted 46 lines · Pavel Pisa

Hello John,

Quoted 13 lines · On Friday 08 of May 2015 13:17:52 John Ernberg wrote:

Hi

With the experimenting I have been able to do on an Ubuntu 14.04 LTS computer, I cannot find any CAN or virtual CAN interfaces inside the container.
When I change the config to specifically forward CAN interfaces, like this:
lxc.network.type = can
lxc.network.link = lxcbr0
lxc.network.flags = up
lxc.network.hwaddr = 00-00-00-00-00-00-00-00-00-00-00-00-00-00-00-00 I get a series of parsing errors from LXC and the container cannot be created.

Quoted 5 lines

Creating a privileged container did not change anything.

We will also take the systemd alternative into consideration, but the current state is that we'd like to run with LXC, as they provide us with more options for future changes to our use-cases.

you cannot use Ethernet bridging. This cannot work. But you should be able to forward whole interface to the container

lxc.network.type = phys

I have not tried this for CAN, but it works with Ethernet and I expect that in this case there is no significant difference to CAN. But you lose (at least for Ethernet) access to the interface from the outer system.

If you want to filter frames then you should be able to setup virtual CAN interface in the outer/host system, then setup CANGW to route messages in the interrest and the start container which hides virtual can interface from the outer system but CANGW setup should continue to work after this move of interface to other namespace.

I am out of time with teaching now so I do not expect to find time to test that soon but I hope that hint is in the right direction. If there is some need to do code adaptations somebody of my colleagues could have interrest to look at it if we are sucesfull with gaining one of our partially related grant or we find other funding.

Quoted 3 lines

Best wishes,

Pavel

Hi Pavel,

I succeeded in forwarding the can interface into the container. When I tried to open a socket, and bind it to the interface to make a test for receiving frames I got an error saying "socket: Address family not supported by protocol".
Looking around in LXC, and then the kernel, I discovered that the AF_CAN implementation does not support net_namespaces yet, and net_namespaces does not implement any handling for CAN, so we cannot open any kind of socket to actually read from the interface.

Is there anything quick and dirty I could try to see if socket traffic can be forwarded into the container on a phys-forwarded CAN interface? Or would a proper net_namespaces implementation for CAN be necessary?

Best regards // John Ernberg

#11 lore raw
Quoted 11 lines · John Ernberg

I succeeded in forwarding the can interface into the container. When I tried to open a socket, and bind it to the interface to make a test for receiving frames I got an error saying "socket: Address family not supported by protocol".
Looking around in LXC, and then the kernel, I discovered that the AF_CAN implementation does not support net_namespaces yet, and net_namespaces does not implement any handling for CAN, so we cannot open any kind of socket to actually read from the interface.

Is there anything quick and dirty I could try to see if socket traffic can be forwarded into the container on a phys-forwarded CAN interface?

You can try to read CAN frames from the CAN netdev via AF_PACKET socket:

https://github.com/linux-can/can-tests/blob/master/tst-packet.c

Or use wireshark which has CAN support too.

Quoted 1 line

Or would a proper net_namespaces implementation for CAN be necessary?

When you use AF_PACKET you bypass the CAN network layer stuff.

Best regards,
Oliver