Assignment 4 - SO2 Transport Protocol¶
- Deadline: Monday, 29 May 2023, 23:00
- This assignment can be made in teams (max 2). Only one of them must submit the assignment, and the names of the student should be listed in a README file.
Implement a simple datagram transport protocol - STP (SO2 Transport Protocol).
Assignment's Objectives¶
- gaining knowledge about the operation of the networking subsystem in the Linux kernel
- obtaining skills to work with the basic structures of the networking subsystem in Linux
- deepening the notions related to communication and networking protocols by implementing a protocol in an existing protocol stack
Statement¶
Implement, in the Linux kernel, a protocol called STP (SO2 Transport Protocol), at network and transport level, that works using datagrams (it is not connection-oriented and does not use flow-control elements).
The STP protocol acts as a Transport layer protocol (port-based multiplexing) but operates at level 3 (Network) of the OSI stack, above the Data Link level.
The STP header is defined by the struct stp_header
structure:
struct stp_header {
__be16 dst;
__be16 src;
__be16 len;
__u8 flags;
__u8 csum;
};
where:
len
is the length of the packet in bytes (including the header);dst
andsrc
are the destination and source ports, respectively;flags
contains various flags, currently unused (marked reserved);csum
is the checksum of the entire package including the header; the checksum is calculated by exclusive OR (XOR) between all bytes.
Sockets using this protocol will use the AF_STP
family.
The protocol must work directly over Ethernet. The ports used are between 1
and 65535
. Port 0
is not used.
The definition of STP-related structures and macros can be found in the assignment support header.
Implementation Details¶
The kernel module will be named af_stp.ko.
You have to define a structure of type net_proto_family, which provides the operation to create STP sockets. Newly created sockets are not associated with any port or interface and cannot receive / send packets. You must initialize the socket ops field with the list of operations specific to the STP family. This field refers to a structure proto_ops which must include the following functions:
release
: releases an STP socketbind
: associates a socket with a port (possibly also an interface) on which packets will be received / sent:- there may be bind sockets only on one port (not on an interface)
- sockets associated with only one port will be able to receive packets sent to that port on all interfaces (analogous to UDP sockets associated with only one port); these sockets cannot send packets because the interface from which they can be sent via the standard sockets API cannot be specified
- two sockets cannot be binded to the same port-interface combination:
- if there is a socket already binded with a port and an interface then a second socket cannot be binded to the same port and the same interface or without a specified interface
- if there is a socket already binded to a port but without a specified interface then a second socket cannot be binded to the same port (with or without a specified interface)
- we recommend using a hash table for bind instead of other data structures (list, array); in the kernel there is a hash table implementation in the hashtable.h header
connect
: associates a socket with a remote port and hardware address (MAC address) to which packets will be sent / received:- this should allow
send
/recv
operations on the socket instead ofsendmsg
/recvmsg
orsendto
/recvfrom
- once connected to a host, sockets will only accept packets from that host
- once connected, the sockets can no longer be disconnected
- this should allow
sendmsg
,recvmsg
: send or receive a datagram on an STP socket:- for the receive part, metainformation about the host that sent the packet can be stored in the cb field in sk_buff
poll
: the default functiondatagram_poll
will have to be used- for the rest of the operations the predefined stubs in the kernel will have to be used (
sock_no_*
)
static const struct proto_ops stp_ops = {
.family = PF_STP,
.owner = THIS_MODULE,
.release = stp_release,
.bind = stp_bind,
.connect = stp_connect,
.socketpair = sock_no_socketpair,
.accept = sock_no_accept,
.getname = sock_no_getname,
.poll = datagram_poll,
.ioctl = sock_no_ioctl,
.listen = sock_no_listen,
.shutdown = sock_no_shutdown,
.setsockopt = sock_no_setsockopt,
.getsockopt = sock_no_getsockopt,
.sendmsg = stp_sendmsg,
.recvmsg = stp_recvmsg,
.mmap = sock_no_mmap,
.sendpage = sock_no_sendpage,
};
Socket operations use a type of address called sockaddr_stp
, a type defined in the assignment support header.
For the bind operation, only the port and the index of the interface on which the socket is bind will be considered.
For the receive operation, only the addr
and port
fields in the structure will be filled in with the MAC address of the host that sent the packet and with the port from which it was sent.
Also, when sending a packet, the destination host will be obtained from the addr
and port
fields of this structure.
You need to register a structure packet_type, using the call dev_add_pack to be able to receive STP packets from the network layer.
The protocol will need to provide an interface through the procfs file system for statistics on sent / received packets.
The file must be named /proc/net/stp_stats
, specified by the STP_PROC_FULL_FILENAME
macro in assignment support header.
The format must be of simple table type with 2
rows: on the first row the header of the table, and on the second row the statistics corresponding to the columns.
The columns of the table must be in order:
RxPkts HdrErr CsumErr NoSock NoBuffs TxPkts
where:
RxPkts
- the number of packets receivedHdrErr
- the number of packets received with header errors (packets too short or with source or destination 0 ports)CsumErr
- the number of packets received with checksum errorsNoSock
- the number of received packets for which no destination socket was foundNoBuffs
- the number of received packets that could not be received because the socket queue was fullTxPkts
- the number of packets sent
To create or delete the entry specified by STP_PROC_FULL_FILENAME
we recommend using the functions proc_create and proc_remove.
Sample Protocol Implementations¶
For examples of protocol implementation, we recommend the implementation of PF_PACKET sockets and the various functions in UDP implementation or IP implementation.
Testing¶
In order to simplify the assignment evaluation process, but also to reduce the mistakes of the submitted assignments, the assignment evaluation will be done automatically with the help of a test script called _checker. The test script assumes that the kernel module is called af_stp.ko.
tcpdump¶
You can use the tcpdump
utility to troubleshoot sent packets.
The tests use the loopback interface; to track sent packets you can use a command line of the form:
tcpdump -i lo -XX
You can use a static version of tcpdump.
To add to the PATH
environment variable in the virtual machine, copy this file to /linux/tools/labs/rootfs/bin
.
Create the directory if it does not exist. Remember to give the tcpdump
file execution permissions:
# Connect to the docker using ./local.sh docker interactive
cd /linux/tools/labs/rootfs/bin
wget http://elf.cs.pub.ro/so2/res/teme/tcpdump
chmod +x tcpdump
QuickStart¶
It is mandatory to start the implementation of the assignment from the code skeleton found in the src directory. There is only one header in the skeleton called stp.h. You will provide the rest of the implementation. You can add as many *.c` sources and additional *.h` headers. You should also provide a Kbuild file that will compile the kernel module called af_stp.ko. Follow the instructions in the README.md file of the assignment's repo.
Tips¶
To increase your chances of getting the highest grade, read and follow the Linux kernel coding style described in the Coding Style document.
Also, use the following static analysis tools to verify the code:
checkpatch.pl
$ linux/scripts/checkpatch.pl --no-tree --terse -f /path/to/your/file.c
sparse
$ sudo apt-get install sparse $ cd linux $ make C=2 /path/to/your/file.c
cppcheck
$ sudo apt-get install cppcheck $ cppcheck /path/to/your/file.c
Penalties¶
Information about assigments penalties can be found on the General Directions page.
In exceptional cases (the assigment passes the tests by not complying with the requirements) and if the assigment does not pass all the tests, the grade will may decrease more than mentioned above.
Submitting the assigment¶
The assignment will be graded automatically using the vmchecker-next infrastructure. The submission will be made on moodle on the course's page to the related assignment. You will find the submission details in the README.md file of the repo.
Resources¶
- Lecture 10 - Networking
- Lab 10 - Networking
- Linux kernel sources
- Understanding Linux Network Internals
- chapters 8-13
- assignment support header
We recommend that you use gitlab to store your homework. Follow the directions in README.
Questions¶
For questions about the topic, you can consult the mailing list archives or you can write a question on the dedicated Teams channel.
Before you ask a question, make sure that:
- you have read the statement of the assigment well
- the question is not already presented on the FAQ page
- the answer cannot be found in the mailing list archives