2007-06-14 17:22:33

by Oleg Nesterov

Subject: [PATCH] nbd.c:sock_xmit: cleanup signal related code

sock_xmit() re-implements sigprocmask() and dequeue_signal_lock().

Note: I can't understand this dequeue_signal(), it can dequeue SIGKILL
for the user-space task doing nbd_ioctl() ?

Signed-off-by: Oleg Nesterov <[email protected]>

--- sig/drivers/block/nbd.c~1_nbd 2007-05-21 13:57:50.000000000 +0400
+++ sig/drivers/block/nbd.c 2007-06-14 19:39:55.000000000 +0400
@@ -122,17 +122,12 @@ static int sock_xmit(struct socket *sock
int result;
struct msghdr msg;
struct kvec iov;
- unsigned long flags;
- sigset_t oldset;
+ sigset_t blocked, oldset;

/* Allow interception of SIGKILL only
* Don't allow other signals to interrupt the transmission */
- spin_lock_irqsave(&current->sighand->siglock, flags);
- oldset = current->blocked;
- sigfillset(&current->blocked);
- sigdelsetmask(&current->blocked, sigmask(SIGKILL));
- recalc_sigpending();
- spin_unlock_irqrestore(&current->sighand->siglock, flags);
+ siginitsetinv(&blocked, sigmask(SIGKILL));
+ sigprocmask(SIG_SETMASK, &blocked, &oldset);

do {
sock->sk->sk_allocation = GFP_NOIO;
@@ -151,11 +146,9 @@ static int sock_xmit(struct socket *sock

if (signal_pending(current)) {
siginfo_t info;
- spin_lock_irqsave(&current->sighand->siglock, flags);
printk(KERN_WARNING "nbd (pid %d: %s) got signal %d\n",
- current->pid, current->comm,
- dequeue_signal(current, &current->blocked, &info));
- spin_unlock_irqrestore(&current->sighand->siglock, flags);
+ current->pid, current->comm,
+ dequeue_signal_lock(current, &current->blocked, &info));
result = -EINTR;
break;
}
@@ -169,10 +162,7 @@ static int sock_xmit(struct socket *sock
buf += result;
} while (size > 0);

- spin_lock_irqsave(&current->sighand->siglock, flags);
- current->blocked = oldset;
- recalc_sigpending();
- spin_unlock_irqrestore(&current->sighand->siglock, flags);
+ sigprocmask(SIG_SETMASK, &oldset, NULL);

return result;
}

2007-06-14 19:36:50

by Paul Clements

Subject: Re: [PATCH] nbd.c:sock_xmit: cleanup signal related code

Oleg Nesterov wrote:
> sock_xmit() re-implements sigprocmask() and dequeue_signal_lock().

Yeah, that code was written before those existed. Thanks for the clean up.

> Note: I can't understand this dequeue_signal(), it can dequeue SIGKILL
> for the user-space task doing nbd_ioctl() ?

So we can interrupt an nbd transmission without waiting for a TCP
timeout (when we know the network is down).

> Signed-off-by: Oleg Nesterov <[email protected]>

Patch looks good to me.

--
Paul

2007-06-14 20:11:33

by Oleg Nesterov

Subject: Re: [PATCH] nbd.c:sock_xmit: cleanup signal related code

On 06/14, Paul Clements wrote:
>
> Oleg Nesterov wrote:
>
> >Note: I can't understand this dequeue_signal(), it can dequeue SIGKILL
> >for the user-space task doing nbd_ioctl() ?
>
> So we can interrupt an nbd transmission without waiting for a TCP
> timeout (when we know the network is down).

Yes, but that task should not survive after SIGKILL ? Note that SIGNAL_GROUP_EXIT
is set, this is not good.

Oleg.