Linux kernel hook that knows the destination process of a socket - linux

I'm looking for a specific place to add some kind of hook to the linux kernel that will allow me to do the following:
Detect just the case of a TCP connection going to loopback (127.0.0.1)
Detect if the source process of the packet is the same as the destination process
If this is the case, allow the connection, deny any other loopback connection (one process to another process)
I've been looking at netfilter hooks but they don't seem to fit the need. bpf filters only look at the packet itself.
I was also looking at the LSM hook of socket-connect. I can try to achieve this from the socket-connect hook by looking at what process has the requested port current bound to predict where the connection is going to connect that this sounds quite hackish.

A simple approach could be the use of Linux network namespaces.
Linux Network Namespaces
As the name would imply, network namespaces partition the use of the network—devices, addresses, ports, routes, firewall rules, etc.—into separate boxes, essentially virtualizing the network within a single running kernel instance. Network namespaces entered the kernel in 2.6.24,...
see https://lwn.net/Articles/580893/
Unshare
unshare() allows a process (or thread) to disassociate parts of its execution context that are currently being shared with other processes (or threads).
see http://man7.org/linux/man-pages/man2/unshare.2.html
Testcase
Since a program should be able to comunicate with itself we need a program that communicates with itself via sockes. There is a cool SO answer that shows a simple Java program that transfers the text 'Hello World!' via socket to itself, see here https://stackoverflow.com/a/8119708.
/usr/bin/java SendReceive
works as expected and gives the output 'Hello World!'
With the -n option one can unshare network namespace.
unshare -n -- sh -c '/usr/bin/java SendReceive'
gives a SocketException: Network is unreachable because there is no access to the loopback device.
unshare -n -- sh -c 'ip link set dev lo up; /usr/bin/java SendReceive'
finally transfers 'Hello World!' again via a loopback interface. BTW: this is a private loopback device. You cannot access open ports on the standard loopback device.
See also this cool Unix Stackexchange answer: https://unix.stackexchange.com/a/83348: Block network access of a process?
Screenshot
Here a screenshot of the test case mentioned above executed on Ubuntu 18.10:

Related

Internet socket behavior when communicating within the same host

I am recently writing some tool for testing some network processes that run across different hosts.
I am tempted to the idea that when testing, instead of running the client and server in different hosts, I can run them within one host.
Since the client and server are using TCP to communicate, so I think this should be fine, except one point below:
Is the TCP socket behavior the same when communicating data within the same host as the case of across hosts?
Will the data be physically present to the NIC interface and then routed to the target socket? Or the kernel will bypass the NIC interface under such scenarios? (Let's limit the OS as only Linux here for discussion)
There seems little specification regarding to such case.
==== EDIT ====
I actually notice some difference between intra-host and inter-host communications.
When doing inter-host communications, my program can successfully get hardware timestamp. But with the exact same code to run within the same host, the hardware timestamp disappears. When supported and enabled, hardware timestamp of TCP packet is available, and is returned as the ancillary data of recvmsg along with the received TCP data. Linux kernel timestamp doc has all the related info.
I checked the source code, the only difference is that whether the sender is within the same host of the receiver, no other difference.
So I am wondering whether Linux kernel will bypass the NIC and present the data directly to the receiver when doing intra-host communication, thus cause the issue.
Will the data be physically present to the NIC interface and then routed to the target socket?
No. There is typically no device that provides this capability, nor is there any need for one.
Or the kernel will bypass the NIC interface under such scenarios?
The kernel will not use the NIC unless it needs to send or receive a packet on a network. Typically, NICs can only return local packets if put in a test or loopback mode, which would require them to stop listening to the network.

When using socket locally where the data go through?

Say the system is linux, I use TPC/IP protocol. When I send data to 127.0.0.1:1024 from A process, then B process get all the data.
How does the system handle these local data traffics?
Does the data go through the network interface card from A to B?
Or they are only manipulated in the memory (much faster than network interface card)?
It'll not be processed by your network card as 127.0.0.1 address is not set on any (it's on loopback device) but it'll go through whole ip stack. Benefits are that you can manipulate this traffic with iptables or iproute tools and whatever you made that way will be ready to work between remote hosts.
If you care more about performance and use only local communiaction consider AF_UNIX socket. You can find more in man socket and man unix.
Check man ipc as well.

CentOS server: how to detect network status changes

I have setup a CentOS 6.4 server (minimal install) which is connected to network through an ethernet cable. The problem is that when the network link goes down, the status changes are not automatically detected but if i type "ifconfig" the interface still keeps its IP address (which is assigned by a DHCP server). After some time that the link is down the interface loses the address but when the link comes up again the network connection is not automatically restored like it would happen in a desktop computer. Even the command "dhclient eth0" does not always work to restore things, and I have to restart the whole network service with "/ect/init.d/network restart".
Is there any way to automatically detect network status changes like it happens in desktop installations? I'm thinking about a cron script that every 5 minutes pings a server outside my network and if it doesn't get any response it restarts network service, but this does not sound very efficient... is there another way?
EDIT: I realized I have not explained the situation correctly. My network topology is: server --> switch --> router --> external network (the router is another centos server with DHCPD).
For some reasons (that i'm not getting), when it happens that the router goes down and reboots, the other server becomes unreachable, and I have to manually restart network service on it. So the link does not effectively go down (the switch keeps it up), but the status change is at IP level.
You can check if you have NetworkManager enabled, I usually don't use it in the servers but it can help you in this case because it will monitor automatically the connections (it is quite common in desktop installations).
https://access.redhat.com/site/documentation/en-US/Red_Hat_Enterprise_Linux/6/html/Deployment_Guide/ch-NetworkManager.html
To look for messages indicating an issue with the NIC just check the kernel ring buffer using the dmesg command.
For example when the cable is disconnected from a given interface this is what I get:
igb: eth1 NIC Link is Down
The first word will depend on the name of your network driver.
You could also configure the system to log these messages also to /var/log/messages (by default I am not sure if they appear there). Then it would be just a matter of monitoring the log, look for similar messages and restart the network service.
In any case the NetworkManager, if it is not already enabled, it should be an easier solution.
There is a module called miimon for monitoring the network interface's status. ethtool will give you the link status.
$ethtool eth0
...
Link detected: yes

using netcat for external loop-back test between two ports

I am writing a test script to exercise processor boards for a burn-in cycle during manufacturing. I would like to use netcat to transfer files from one process, out one Ethernet port and back into another Ethernet port to a receiving process. It looks like netcat would be an easy tool to use for this.
The problem is that if I set up the ethernet ports with IP addresses on separate IP sub nets and attempt to transfer data from one to the other, the kernel's protocol stack detects an internal route and although the data transfer completes as expected, it does NOT go out over the wire. The packets are routed internally.
That's great for network optimization but it foils the test I want to do.
Is there easy way to make this work? Is there a trick with iptables that would work? Or maybe things you can do to the route table?
I use network name spaces to do this sort of thing. With each of the adapters in a different namespace the data traffic definitely goes through the wire instead of reflecting in the network stack. The separate namespaces also prevent reverse packet filters and such from getting in the way.
So presume eth0 and eth1, wiht iperf3 as the reflecting agent (ping server or whatever). [DISCLAIMER:text from memory, all typos are typos, YMMV]
ip netns add target
ip link set dev eth1 up netns target
ip netns exec target ip address add dev eth1 xxx.xxx.xxx.xxx/y
ip netns exec target iperf3 --server
So now you've created the namespace "target", moved one of your adapters into that namespace. Set its IP address. And finally run your application in the that target namespace.
You can now run any (compatible) program in the native namespace, and if it references the xxx.xxx.xxx.xxx IP address (which clearly must be reachable with some route) will result in on-wire traffic that, with a proper loop-back path, will find the adapter within the other namespace as if it were a different computer all together.
Once finished, you kill the daemon server and delete the namespace by name and then the namespace members revert and you are back to vanilla.
killall iperf3
ip netns delete target
This also works with "virtual functions" of a single interface, but that example requires teasing out one or more virtual functions --- e.g. SR-IOV type adapters -- and handing out local mac addresses. So I haven't done that enough to have a sample code tidbit ready.
Internal routing is preferred because in the default routing behaviour you have all the internal routes marked as scope link in the local table. Check this out with:
ip rule show
ip route show table local
If your kernel supports multiple routing tables you can simply alter the local table to achieve your goal. You don't need iptables.
Let's say 192.168.1.1 is your target ip address and eth0 is the interface where you want to send your packets out to the wire.
ip route add 192.168.1.1/32 dev eth0 table local

How create a virtual io device in Linux that proxies data to real device?

I have an interesting problem. I am working on an embedded box with multiple instances of Linux running each on an ARM processor. They are connected over internal 1GBps network. I have a serial port device node attached to processor A (Lets say Linux-A running on it). I have a program running on processor B (Lets say on Linux-B) access the serial port device as if it is attached to Linux-B locally.
My program invokes term i/o type api calls on device node to control tty echo, character mode input. What I am wondering is if there is a way to create a virtual serial device that is available on Linux-B somehow talking to real serial device on Linux-A over internal network.
I am thinking something along the lines of:
Linux-B has /dev/ttyvirtual. Anything that gets written to it gets transported over network socket to Linux-A serialserver. The serial server exrcises the api calls on real device lets say /dev/ttys0.
Any data waiting on ttys0 gets transported back to /dev/ttyvirtual.
What are all the things involved to get this done fast?
Thanks
Videoguy
Update:
I found a discussion at
http://fixunix.com/bsd/261068-network-socket-serial-port-question.html with great pointers.
Another useful link is http://blog.philippklaus.de/2011/08/make-rs232-serial-devices-accessible-via-ethernet/
Take a look at openpty(3). This lets you create a pseudo-TTY (like /dev/pts/0, the sort that ssh connections use), which will respond as a normal TTY would, but give you direct programmatic control over the connections.
This way you can host a serial device (eg. /dev/pts/5) that you forward data between a network connection, and then other apps can perform serial operations on it without knowing about the underlying network bridge.
I ended up using socat
Examples can be found here: socat examples
You socat back to back on both the machines. One listens on a tcp port and forwards data to local virtual port or pty. The socat on other box uses real device as input and forwards any data to tcp port.

Resources