FUSE and pseudo dev nodes

ewaller · 2014-03-02 23:32:11

I have been playing around with FUSE. It is kind of cute, but I have run into a barrier. I am looking for either direction, or for someone to tell me I've my head up and locked.

What I want to do is create is a 'file' using FUSE that behaves like a device node -- kind of like /dev/urandom. The idea is to create an arbitrary function generator that can be piped through something like aplay, or a graphing program, or a FFT analyzer. I want to create a file that, when you read a value from it (byte, int, whatever) you get the next value in the sequence.

I can get this to work if I define the file size as being ludicrously large (like -1) but that is an ugly, ugly hack. Is it possible to define a file that behaves like a device node using FUSE?, or do I need to do that in kernel space. (Hint: For a weekend project, that aint gonna happen)

Rexilion · 2014-03-03 13:22:28

So you want something like urandom, but that outputs sequential data as a result from an arbitrary function?

Why not create a program that generates the data and pipe that into a (named) fifo (mkfifo) and read from that?

Or do you want a combination of urandom and a function? I'm not able to connect all the information to a consistent and coherent question. I might have interperted it incorrectly.

ewaller · 2014-03-04 05:36:48

Thanks. The real answer is because that was not where my brain had been

I learned a bit about named pipes today. I need a file that can be opened from inside a program. I realize now that that is something named pipes do very well.
What I did not realize was that named pipes don't buffer the input. What I thought they would do would be to act like a FIFO, where they would accept data from the source as long as the source belched it out, would buffer it indefinitely (or until some limit was reached), and would un-buffer them later as they were read. I see I was wrong. Named pipes will probably do what I need.

The other need for FUSE is a bit unrelated to Arch. It is more what FUSE was probably designed for. I am dealing with an embedded Linux that copies an image from flash to RAM, and then runs from a RAM disk -- there is no physical disk. There are some data I need to be persistent, I have an unused Flash segment I can use, but no good way to implement a Flash controller in my image. At this point, I plan to do raw I/O to the unused flash sector in user space, and implement a FUSE file system to allow file system access to that bulk space.

Going back to my original problem, it too will ultimately be implemented in this embedded system, but as I now realize that the buffering is not deep (a byte or two), named pipes should work just fine.

Arch Linux

#1 2014-03-02 23:32:11

FUSE and pseudo dev nodes

#2 2014-03-03 13:22:28

Re: FUSE and pseudo dev nodes

#3 2014-03-04 05:36:48

Re: FUSE and pseudo dev nodes

Board footer