Add support for Linux kernel anyIP feature

[pumper]

Hi

I have a kimailio box that receives messages from one interface and send it from another interface and I use dispatcher module successfully.

Now I assigned an I range to Linux loopback interface through Linux kernel anyip feature. And I can ping the range.

ip -4 route add local 192.168.0.0/24 dev lo

this linux kernel feature is described in https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=ab79ad14a2d51e95f0ac3cef7cd116a57089ba82 and https://blog.widodh.nl/2016/04/anyip-bind-a-whole-subnet-to-your-linux-machine/
another linux kernel feature that is required is Non-Local Binds in the Kernel which van be found in https://oswalt.dev/2022/02/non-local-address-binds-in-linux/ and after enabling that we must be able to send data with any ip address with kamailio this feature is described in https://oswalt.dev/2022/02/non-local-address-binds-in-linux/

I want to send messages from source of the ip in this range but kamailio cant do this in the right way.

I added this two directives to kamailio.cfg:

mhomed=yes

ip_free_bind = 1

and when I want to use an ip from the list I use

$fs=”udp:192.168.0.120:5060”;

With no luck

If I use

force_send_socket(udp:192.168.0.120:5060);

kamailio returns an error

bad force_send_socket argument: 192.168.0.120:5060 (kamailio doesn't listen on it)

and if I add listen directive for one ip like:

listen=udp:192.168.0.120:5060

everything works fine but I want to be able to use any ip from list not just one ip.

Thanks.

[henningw]

This is probably not supported yet

[pumper]

Thanks for quick response I know this linux feature is not supported in kamailio and because of that I put this feature request here.
Thanks.

[joelsdc]

Is there any limitation that stops you from using multiple listen= directives?

In fact, given enough resources, is there even a limit on the number of listen= directives that Kamailio can handle?

[jaybinks]

I'm really curious about the usecase you see here, and how it differs from what is achievable with net.ipv4.ip_nonlocal_bind and or a bgp daemon on your host . I know they are not the same, but I can't (right now ) think of a reason I'd want to bind to a whole subnet. Maybe something TLS related, but in that case you have to do all the TLS config, so is avoiding the listen line really a big advantage? I'm honestly curious though

…

On Sat, 18 Mar 2023, 12:02 pm Joel Serrano, ***@***.***> wrote: Is there any limitation that stops you from using multiple listen= directives? In fact, given enough resources, is there even a limit on the number of listen= directives that Kamailio can handle? — Reply to this email directly, view it on GitHub <#3394 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AALQR65OKEQDCDYCST6SWUTW4UJSFANCNFSM6AAAAAAV5HE5GI> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[sergey-safarov]

I think here is case like enterprise PBX in the cloud. Many PBX should have own IP. Example to create outbound call to another PBX required whitelist PBX IP address. Before somebody asked, what is prevents you to bind on 65536 IP addresses (/16 network). Could answer?

…

On Sat, Mar 18, 2023, 10:15 AM Jay Binks ***@***.***> wrote: I'm really curious about the usecase you see here, and how it differs from what is achievable with net.ipv4.ip_nonlocal_bind and or a bgp daemon on your host . I know they are not the same, but I can't (right now ) think of a reason I'd want to bind to a whole subnet. Maybe something TLS related, but in that case you have to do all the TLS config, so is avoiding the listen line really a big advantage? I'm honestly curious though On Sat, 18 Mar 2023, 12:02 pm Joel Serrano, ***@***.***> wrote: > Is there any limitation that stops you from using multiple listen= > directives? > > In fact, given enough resources, is there even a limit on the number of > listen= directives that Kamailio can handle? > > — > Reply to this email directly, view it on GitHub > < #3394 (comment)>, > or unsubscribe > < https://github.com/notifications/unsubscribe-auth/AALQR65OKEQDCDYCST6SWUTW4UJSFANCNFSM6AAAAAAV5HE5GI > > . > You are receiving this because you are subscribed to this thread.Message > ID: ***@***.***> > — Reply to this email directly, view it on GitHub <#3394 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AATRRQKUBTBPBQ5S6PGA673W4VVJ7ANCNFSM6AAAAAAV5HE5GI> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[joelsdc]

AFAIR, depending on the number of children or socket_workers combined with the amount listen= and the PKG_SIZE setting, RAM would be the limitation.

In an example where you have say 16mb of PKG_SIZE combined with children=1 you would need (16mb * 1 children * 65535 IPs) 1TB of RAM for a /16.

[henningw]

Sure, having a complete network on individual workers will waste a lot of memory. Just to add a bit more context to this request, this is the feature in question link.

This blog post provides some good summary of the option and also explain the related IP_FREEBIND option: link

[github-actions]

This issue is stale because it has been open 6 weeks with no activity. Remove stale label or comment or this will be closed in 2 weeks.

[sergey-safarov]

@pumper could you help me understand how to bind to 127.0.0.0/8.
At https://oswalt.dev/2022/02/non-local-address-binds-in-linux/ described use for bind to any IP

import socket
s = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
s.bind(("0.0.0.0", 0))

But not for network addr with mask.
Hope you can suggest.

Probable it can be done using SO_BINDTODEVICE socket option.