Writing a new device driver

[Matthias Brugger]

Hi Malice,

On 06/25/2012 09:42 PM, Malice wrote:
> Hi
>
> I'm planning to write a new device driver on linux kernel . The device i
> intend to write will be a device that'll pipe the data coming into it to
> another computer that'll be also having the same device . So in short
> the device'll be readable/writable and should communicate through
> network . I know it sounds like a socket but the idea would be to
> implement a interface in shell so that if you type
>
> |     root at host$echo "Hi">>mydevice
> |
>
> will transfer the word hi to the system on the other side of network
> (and may be configure it to store on file). This is a leisure time
> project i intend to do and it sounds much like sockets and other already
> existing implementations . Well the questions regarding the design for
> device implementations are : 1)Should i use a block or a character
> device . The advantage i see for a block device is that if i am a
> initiating a complete file transfer through the device(which i in tend
> to support later) , then it would be fast .
>
> 2)How do i write the data on the network . I'm familiar with sockets and
> stuff , but using sockets from with in the kernel has performance
> issues(I read it somewhere, after all sockets was desined to be used in
> userspace) . Will using netfilter or other stuffs work ?Maybe crafting
> the packet directly would help :-)? Any innovations are welcome

have a look on the tun/tap driver, I think that's what you need.
http://lxr.free-electrons.com/source/drivers/net/tun.c

>
> 3)How do i configure the driver , like how do i let it know that it
> should connect to a specific host >in user space programs we usually do
> this by config files , but to change the connection settings of a driver
> how do you manage it ? (sysfs ?? Frankly i never worked with sysfs or
> proc/ )
>
> 4)Now as the design specifies all the transfer of files is in kernel
> space . When i transfer huge chunks of data through kernel space , will
> it slow down the system ?. The design shouldnt overload the system ,
> doing it in a pre-empitble kernel will help, but are there other
> charecterstics i should be aware of to make the design as efficient as
> possible (i'm doing it on the latest linux kernel) ?
>
> 5)And i dont want to leave a huge memory imprint so is there any way i
> can reduce it while tranferring of data , may be cacheing data on disk ?
>
> 6)I will have to design a user space programme to co ordinate the driver
> wont i ? I cannot let the driver alone do the whole task , tat wud be
> tedious
>
> Comments Welcome
>
>
> I've also posted the same question at stackoverflow too
> http://stackoverflow.com/questions/11167424/writing-a-new-device-driver
>
> Regards
> Malice
>
>
>
> _______________________________________________
> Kernelnewbies mailing list
> Kernelnewbies at kernelnewbies.org
> http://lists.kernelnewbies.org/mailman/listinfo/kernelnewbies
>