/dev/mydevice doesn't work!

I've gotten a little further into learning about device drivers and made a simple char device module that allocates 4k of contiguous memory and supposedly attaches it for universal use to a device node:

Code:

---

#include <linux/module.h>
#include <linux/kernel.h>
#include <linux/fs.h> 
#include <linux/cdev.h>
#include <linux/types.h>

MODULE_LICENSE("GPL");

#define DEVSIZE 4096

typedef struct {
        char data[DEVSIZE];
        struct cdev cdev;
} OneDevice;



void cleanup(void);
int init(void);
ssize_t read_onedev(struct file *file, char __user *buf, size_t count, loff_t *pos);
ssize_t write_onedev(struct file *file, const char __user *buf, size_t count, loff_t *pos);
//int open_onedev(struct inode *inode, struct file *file);
 
int Major;
int Minor = 0;
OneDevice TestOne;
char *Current=TestOne.data; /* position pointer */

void cleanup(void) {
        memset(TestOne.data,0,4096);
        cdev_del(&TestOne.cdev);
        printk(KERN_ALERT "Unloaded Test One\n");
} 


int init(void) {
        int err;
        dev_t devMM = 0;
        struct file_operations fops = {
                .owner = THIS_MODULE,
                .read = read_onedev,
                .write = write_onedev,
//                .open = open_onedev, 
        };

        printk("Loading Test One...");
        if ((err=alloc_chrdev_region(&devMM,Minor,1,"testone"))<0) {
                printk(KERN_WARNING "alloc_chrdev failed!\n");
                return err; }
        Major=MAJOR(devMM);
        printk("registered MAJOR %d MINOR 0\n",Major);               
        memset(TestOne.data,0,DEVSIZE);
        MKDEV(Major,Minor);
        cdev_init(&TestOne.cdev,&fops);
       
        return 0;
}


ssize_t read_onedev(struct file *file, char __user *buf, size_t count, loff_t *pos) {
        const void *ptr=Current+(long)*pos;
        if ((long)ptr-(long)TestOne.data>=DEVSIZE) { return 0; }
        if (DEVSIZE-((long)ptr-(long)TestOne.data)<count) count=DEVSIZE-((long)ptr-(long)TestOne.data);
        memcpy((void*)buf,ptr,count);
        *pos+=count;
        Current=TestOne.data+(long)*pos;
        return count;       
}


ssize_t write_onedev(struct file *file, const char __user *buf, size_t count, loff_t *pos) {
        void *ptr=Current+(long)*pos;
        if (DEVSIZE-((long)ptr-(long)TestOne.data)<count) count=DEVSIZE-((long)ptr-(long)TestOne.data);
        if (count<=0) { return 0; }
        memcpy(ptr,(const void*)buf,count);
        *pos+=count;
        Current=TestOne.data+(long)*pos;
        return count;
}

module_init(init);
module_exit(cleanup);

---

It compiles, loads, and registers fine. Then I get the actual device node number from /var/messages and use it as an argument to this shell script:

Code:

---

#!/bin/bash

if [ -c /dev/testone ]
then
        rm -v -f /dev/testone       
fi
mknod /dev/testone c $1 0
chgrp wheel /dev/testone
chmod 664 /dev/testone

---

Now, supposedly I should be able to do basic normal things like "cat somefile > /dev/testone" but when I try, I get "bash: /dev/testone: No such device or address" altho stat says it exists and is a character special file.

So I was just wondering if anyone had done this stuff before and might have an insight.

I wasn't actually registering the major number:

Code:

---

cdev_add(&TestOne.cdev,devMM,1);

---

There's some work to do tho, I managed to produce my first Oops (Message from syslogd@lhost at Apr 29 16:36:42 ...) by overwriting the 4k, which I guess means just saying "no" ain't good enough.

Anyway the whole time I have been thinking of a thread by sean from a while ago:
http://cboard.cprogramming.com/linux...m-i-right.html

This idea (ie, a character memory device) seems perfect for that (it's not my idea, nb), since it produces a block which can be accessed by any process via a device node, eg.

Code:

---

cat some.data > /dev/testone
cp /dev/testone data.copy

---

The whole thing (I only used 4k) is right there in kernel space! This could be coded to allocate dynamically.

In other words, sean could have built his linked list there and freely accessed it from his other processes.

I would guess that the reason this is not in widespread use is because if everyone wanted to use it that way, ZONE_NORMAL memory would be gone quick. Anyone got a better reason on why not to do this?