LinuxQuestions.org - Why does the value of ret change out of nowhere in do_sync

Hi!

I'm having some fun creating a linux block device driver. This device driver will have to read from files and so I've been trying to solve all the problems involved. While doing my research I've hit a very weird behavior in a function defined in fs/read_write.c. It's do_sync_read. I have added some debug messages to it on my UML kernel (the one I'm developing my device driver on) to check all vaules involved while running the function cause I was getting a -EFAULT when I ran vfs_read. Following the functions involved I arrived at do_sync_read.

If you take a look at the function, you will see that there is one _seemingly_ endless for, then a check, then change of the ppos variable and finally a return of ret.

I have found that there is only one cycle of the for run, because there is a break that is hit if everything is OK inside the for.

First, ret is assigned like this:

Code:

ret = filp->f_op->aio_read(&kiocb, &iov, 1, kiocb.ki_pos);

After this point, the value of the variable ret is set to 0 meaning the operation ran normally (or so I like to think).

Then there is a check that is evaluated to true which means that ret != -EIOCBRETRY, great... but before I call break I print again the value of ret and now the value has been changed to -14 (-EFAULT). Why is that happening? Cause I dont see anything that could be going on that could be changing the value of ret from 0 to -14 between those two points.

Code:

for (;;) {

        ret = filp->f_op->aio_read(&kiocb, &iov, 1, kiocb.ki_pos);

        if (filp == debugFile) printk (KERN_NOTICE "Con la posicion %d resultado fue %d\n", kiocb.ki_pos, ret);

        if (ret != -EIOCBRETRY) {

                if (filp == debugFile) printk (KERN_NOTICE "Saliendo del ciclo con ret valiendo %d\n", ret);

                break;

        }

        wait_on_retry_sync_kiocb(&kiocb);

}

Output:

Code:

Con la posicion 2000 resultado fue 0

Saliendo del ciclo con ret valiendo -14