How can I allocate memory in the Linux kernel for a char* type string?

Question

I'm trying to allocate some memory for a char* as follows.

static ssize_t memo_write(struct file *filp, const char __user *buf, 
   size_t count, loff_t *f_pos){
    ssize_t retval = -ENOMEM;
    printk("write function\n");

    if((data = kmalloc(strlen(buf), GFP_KERNEL)) == NULL)
        printk("kmalloc fail\n");

    if(copy_from_user(data, buf, strlen(buf))){
        retval = -EFAULT;
        goto out;
    }
    *f_pos += strlen(buf);
    retval = strlen(buf);

    out:
        return retval;
}

'data' is declared in a header file as

char *data;

When I call the write function, the 'kmalloc fail' line isn't reached, which leads me to believe the kmalloc succeeded, however the data isn't displayed when I try to read from the 'data' variable again.

More confusingly, if I get rid of the kmalloc bit altogether, the data can be read from the driver. Although the problem then is it is followed by a load of other data because i don't have the opportunity to memset() it.

Am I using kmalloc correctly? Presumably not. How should I be doing this?

Additionally, my read function is as follows.

static ssize_t memo_read(struct file *f, char __user *buf, 
    size_t count, loff_t *f_pos){
    ssize_t retval = 0;

    printk("read function\n");
    printk("data = %s\n", data);

    if(*f_pos >= strlen(data)){
        printk("EOF\n");
        goto out;
    }

    if(copy_to_user(buf, data, strlen(data))){
        retval = -EFAULT;
        goto out;
    }
    printk("copy_to_user success\n");
    *f_pos += strlen(data);
    retval = strlen(data);
    out:
        return retval;
}

Thanks.

Answer 1

Don't forget to kfree(data) in your error cases...

In any case, buf is a pointer to user memory, so DON'T call strlen(buf). You must copy_from_user first. Why not

data = kmalloc(count);
copy_from_user(data, buf, count);

?

---

Your read handler assumes that data is a NUL-terminated string. When you were using an array, this may have been true by accident, but you never actually ensure this in your write handler. My guess is that copy_to_user fails.

Here's a working example of a "memo" module that I wrote up just now, using kmalloc:

#include <linux/fs.h>
#include <linux/miscdevice.h>
#include <linux/module.h>
#include <linux/types.h>
#include <linux/uaccess.h>

static char *data;
static size_t len;

static ssize_t
memo_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
{
        ssize_t copy_len = min(len - min(len, *ppos), count);
        ssize_t retval;

        if (copy_to_user(buf, data + *ppos, copy_len)) {
                retval = -EFAULT;
                goto out;
        }

        *ppos += copy_len;
        retval = copy_len;

out:
        return retval;
}

static ssize_t
memo_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
{
        ssize_t retval;
        char *newdata;

        newdata = kmalloc(count, GFP_KERNEL);
        if (!newdata) {
                retval = -ENOMEM;
                goto out;
        }

        if (copy_from_user(newdata, buf, count)) {
                retval = -EFAULT;
                goto out;
        }

        kfree(data);
        data = newdata;
        newdata = NULL;
        retval = len = count;

out:
        kfree(newdata);
        return retval;
}

static const struct file_operations memo_fops = {
        .owner = THIS_MODULE,
        .llseek = no_llseek,
        .read = memo_read,
        .write = memo_write,
};

static struct miscdevice memo_misc = { MISC_DYNAMIC_MINOR, "memo", &memo_fops };

static int __init memo_init(void)
{
        int result;

        result = misc_register(&memo_misc);
        if (result < 0)
                return -ENODEV;

        return 0;
}

static void __exit memo_exit(void)
{
        misc_deregister(&memo_misc);
        kfree(data);
        return;
}

module_init(memo_init);
module_exit(memo_exit);
MODULE_AUTHOR("ephemient");
MODULE_LICENSE("GPL");

Of course this is missing locking and other safety precautions, but I hope this helps.

Answer 2

You should be using strlen_user() on the userspace pointer, instead of strlen() - and you should only call it once, and keep the result around (otherwise, you have a potential kernel exploit, because a second userspace thread could change the buffer while you're working on it).

Alternatively, you could use strncpy_from_user().

Apart from that, the kmalloc looks OK.

---

(But really, as ephemient says, you should rethink your whole approach and use the count argument instead of treating the input as a string).

---

Since you can't rely on the data written to a file being nul-terminated strings, you'll need to keep a data_len length parameter around alongside the data. Then your read/write implementations would be along these lines:

static char *data = NULL;
static size_t data_len;
static DEFINE_MUTEX(data_mutex);

static ssize_t memo_read(struct file *f, char __user *buf, size_t count, loff_t *f_pos
{
    ssize_t retval = 0;
    char *start;

    mutex_lock(&data_mutex);

    if (!data)
    {
        retval = -EINVAL; /* Or whatever you want to do here... */
        goto out;
    }

    if (*f_pos >= data_len)
        goto out; /* EOF */

    start = data + *f_pos;
    retval = data_len - *f_pos;

    if (retval > count)
        retval = count;

    if (copy_to_user(buf, start, retval))
    {
        retval = -EFAULT;
        goto out;
    }

    *f_pos += retval;

out:
    mutex_unlock(&data_mutex);
    return retval;
}

static ssize_t memo_write(struct file *filp, const char __user *buf, size_t count, loff_t *f_pos)
{
    ssize_t retval = -ENOMEM;

    mutex_lock(&data_mutex);

    if (data)
        kfree(data);

    data = kmalloc(count, GFP_KERNEL);

    if (!data)
        goto out;

    if (copy_from_user(data, buf, count))
    {
        kfree(data);
        retval = -EFAULT;
        goto out;
    }

    *f_pos = count;
    retval = count;
    data_len = count;

out:
    mutex_unlock(&data_mutex);
    return retval;
}