Linux内核模块奇怪的行为

Question

I am studying a linux kernel, so I try to write a simple modules.

The following code is supposed to control how many times read() for the /proc/proc_test is called:

#include <linux/init.h>
#include <linux/module.h>
#include <linux/proc_fs.h>
#include <linux/kernel.h>
#include <asm/current.h>

static int __init init(void);
static void __exit stop(void);
static int proc_read(char *buffer, char **start, off_t offset, int size, int *eof, void *data);

static int counter = 0;

static int proc_read(char *buffer, char **start, off_t offset, int size, int *eof, void *data) {
    if (size < 256) return -ENOMEM;
    return sprintf(buffer, "counter = %d\n", counter++); <-- supposed to be incremented once
}

static int __init init() {
    if (create_proc_read_entry("proc_test", 0, NULL, proc_read, NULL) == 0) {
        printk(KERN_ERR "Can not creat entry\n");
        return -ENOMEM;
    }
    printk("Entry created!\n");
    return 0;
}

static void __exit stop() {
    remove_proc_entry("proc_test", NULL);
    printk("Entry removed!\n");
    return;
}

module_init(init);
module_exit(stop);

MODULE_LICENSE("GPL");

I am facing the issue that when I read from the /proc/proc_test/ using cat or tail , the counter is incremented by 3 instead of by 1.

Output:

cat /proc/proc_test 
counter = 0

cat /proc/proc_test 
counter = 3

cat /proc/proc_test 
counter = 6

What am I doing wrong?

Answer 1

try this:

static int proc_read(char *buffer, char **start, off_t offset, int size, int *eof, void *data) {
    if (size < 256) return -ENOMEM;
    int count= sprintf(buffer, "counter = %d\n", counter++);
    *eof = 1;
    return count;
}

setting *eof=1, your driver informs the kernel (and the application which want to read your proc file) that your driver reached EOF.

Answer 2

The only thing wrong is that you have an unreasonable expectation. What makes you think cat /proc/proc_test will only call read once?

$ strace cat /proc/self/stat | grep read
read(3, "17423 (cat) R 17420 17420 13224 "..., 32768) = 238
read(3, "", 32768)                      = 0
$

Answer 3

Even though this answer has a good hint, I have been having this issue too. Setting *eof = 1 theoretically should have solved the problem but somehow it doesn't.

My fix was to also add this on the top of the function:

if (offset > 0)    // even though I set *eof = 1, the kernel is still calling me, so return 0 for it to stop
    return 0;

The comment above is in fact what I had written in my own module.

What this does is to make sure only the first call to your function (in which offset is 0) does anything.