> atomic_add(size, &sk->wmem_alloc);
> mem = kmalloc(size, priority);
> - if (mem) {
> - /* Recheck because kmalloc might sleep */
> - if (atomic_read(&sk->wmem_alloc)+size < sk->sndbuf)
> - return mem;
> - kfree_s(mem, size);
> - }
> + if (mem)
> + return mem;
> atomic_sub(size, &sk->wmem_alloc);
+> sk->write_space(sk);
> }
> return mem;
The code was wrong, but the fix is wrong too.
We cannot leave atomic_read(&sk->wmem_alloc) >= sk->sndbuf here
(it can occur, if sk->wmem_alloc was increased by another process while sleep),
because it would block subsequent sock_alloc_send_skb forever.
Besides, sk->write_space(sk) after subtraction is missing.
Seems, the code should look like:
(Andy, please, check it. I believe you meaned exactly this thing)
/* Main idea:
Total wmem_alloc consumed by sock_kmalloc cannot exceed
sk->sndbuf minus 1 byte.
In this case, we always can transmit at least one skb
of data, and hence we avoid dead lock in sock_alloc_send_skb.
*/
void *sock_kmalloc(struct sock *sk, int size, int priority)
{
void *mem = NULL;
if (atomic_read(&sk->wmem_alloc)+size < sk->sndbuf) {
/* First do the add, to avoid the race if kmalloc
* might sleep.
*/
atomic_add(size, &sk->wmem_alloc);
mem = kmalloc(size, priority);
if (mem) {
/* Recheck because kmalloc might sleep */
if (atomic_read(&sk->wmem_alloc) < sk->sndbuf)
return mem;
sock_kfree_s(sk, mem, size);
return NULL;
}
}
return mem;
}
Alexey Kuznetsov
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu