[ENBD] Kernel oops (nbd-2.4.31) or failed connections

[Anders Blomdell]

> I do not believe the window will be so common that it merits closing.
> It will be rare. And the open, if it succeeds, will have to find the
> device still there, and then have it be killed by the other thread, for
> harm to be done ... looks very unlikely.
If this is your position on how to write server code, I'll not use enbd 
8^(

> I'll study that section of code a bit to see if there's a short section
> that can be protected. Essentially, we don't want to start protecting
> all accesses to the device with a semaphore!
Nope, but the following is a must:

open:
   down(enbd_sem);
   if (enbd_dev[nbd] == NULL) {
     enbd_dev[nbd] = kmalloc(...);
     ...
     enbd_dev[nbd]->refcnt++;
   }
   up(enbd_sem);

release:
   to_be_destroyed = NULL;
   down(enbd_sem);
   // can enbd_dev[nbd] be NULL??
   enbd_dev[nbd]->refcnt--;
   if (enbd_dev[nbd]->refcnt <= 0) {
	to_be_destroyed = enbd_dev[nbd];
     enbd_dev[nbd] = NULL;
   }
   up(enbd_sem);
   if (to_be_destroyed) {
     enbd_destroy(to_be_destroyed);
   }


Regards

Anders Blomdell