Re: + signalfd-retrieve-multiple-signals-with-one-read-call.patch added to -mm tree

Re: + signalfd-retrieve-multiple-signals-with-one-read-call.patch added to -mm tree

[Oleg Nesterov]

A couple of very minor nits,

> +static ssize_t signalfd_dequeue(struct signalfd_ctx *ctx, siginfo_t *info,
> +				int nonblock)
> +{
> +	int locked;
> +	ssize_t ret;
> +	struct signalfd_lockctx lk;
> +	DECLARE_WAITQUEUE(wait, current);
> +
> +	locked = signalfd_lock(ctx, &lk);
> +	if (!locked)
> +		return 0;
> +
> +	ret = dequeue_signal(lk.tsk, &ctx->sigmask, info);
> +	switch (ret) {
> +	case 0:
> +		if (!nonblock)
> +			break;
> +		ret = -EAGAIN;
> +	default:
> +		signalfd_unlock(&lk);
> +		return ret;
> +	}
> +
> +	add_wait_queue(&ctx->wqh, &wait);
> +	for (;;) {
> +		set_current_state(TASK_INTERRUPTIBLE);
> +		ret = dequeue_signal(lk.tsk, &ctx->sigmask, info);
> +		if (ret != 0)
> +			break;
> +		if (signal_pending(current)) {
> +			ret = -ERESTARTSYS;
> +			break;
> +		}
> +		signalfd_unlock(&lk);

The locking looks a bit overcomplicated. We don't need signalfd_lock() to
check ret != 0 or signal_pending(), we can drop it earlier. This way we
always leave the loop in "unlocked" state.

> +		schedule();
> +		locked = signalfd_lock(ctx, &lk);
> +		if (unlikely(!locked)) {
> +			/*
> +			 * Let the caller read zero byte, ala socket
> +			 * recv() when the peer disconnect. This test
> +			 * must be done before doing a dequeue_signal(),
> +			 * because if the sighand has been orphaned,
> +			 * the dequeue_signal() call is going to crash.
> +			 */

Imho, the comment is a bit confusing. dequeue_signal() needs ->siglock
even if signalfd_ctx is not orphaned.

> +			 ret = 0;
> +			 break;
> +		}
> +	}
> +
> +	remove_wait_queue(&ctx->wqh, &wait);
> +	__set_current_state(TASK_RUNNING);
> +
> +	if (likely(locked))
> +		signalfd_unlock(&lk);
> +
> +	return ret;
> +}

IOW, how about this?

	static ssize_t signalfd_dequeue(struct signalfd_ctx *ctx, siginfo_t *info,
					int nonblock)
	{
		ssize_t ret;
		struct signalfd_lockctx lk;
		DECLARE_WAITQUEUE(wait, current);

		if (!signalfd_lock(ctx, &lk))
			return 0;

		ret = dequeue_signal(lk.tsk, &ctx->sigmask, info);
		switch (ret) {
		case 0:
			if (!nonblock)
				break;
			ret = -EAGAIN;
		default:
			signalfd_unlock(&lk);
			return ret;
		}

		add_wait_queue(&ctx->wqh, &wait);
		for (;;) {
			set_current_state(TASK_INTERRUPTIBLE);
			ret = dequeue_signal(lk.tsk, &ctx->sigmask, info);
			signalfd_unlock(&lk);

			if (ret != 0)
				break;

			ret = -ERESTARTSYS;
			if (signal_pending(current))
				break;

			schedule();

			ret = 0;
			if (!signalfd_lock(ctx, &lk))
				break;
		}

		remove_wait_queue(&ctx->wqh, &wait);
		__set_current_state(TASK_RUNNING);

		return ret;
	}

Oleg.

Re: + signalfd-retrieve-multiple-signals-with-one-read-call.patch added to -mm tree

[Davide Libenzi]

On Mon, 21 May 2007, Oleg Nesterov wrote:

> > +		schedule();
> > +		locked = signalfd_lock(ctx, &lk);
> > +		if (unlikely(!locked)) {
> > +			/*
> > +			 * Let the caller read zero byte, ala socket
> > +			 * recv() when the peer disconnect. This test
> > +			 * must be done before doing a dequeue_signal(),
> > +			 * because if the sighand has been orphaned,
> > +			 * the dequeue_signal() call is going to crash.
> > +			 */
> 
> Imho, the comment is a bit confusing. dequeue_signal() needs ->siglock
> even if signalfd_ctx is not orphaned.

The comment looks clear to me. It states:

1) The policy of returning 0 when the sighand has been detached

2) That we _must_not_ call dequeue_signal() in case signalfd_lock() fails

#ACK on the code mod below.



- Davide

Re: + signalfd-retrieve-multiple-signals-with-one-read-call.patch added to -mm tree

[Davi Arnaut]

Davide Libenzi wrote:
> On Mon, 21 May 2007, Oleg Nesterov wrote:
> 
>>> +		schedule();
>>> +		locked = signalfd_lock(ctx, &lk);
>>> +		if (unlikely(!locked)) {
>>> +			/*
>>> +			 * Let the caller read zero byte, ala socket
>>> +			 * recv() when the peer disconnect. This test
>>> +			 * must be done before doing a dequeue_signal(),
>>> +			 * because if the sighand has been orphaned,
>>> +			 * the dequeue_signal() call is going to crash.
>>> +			 */
>> Imho, the comment is a bit confusing. dequeue_signal() needs ->siglock
>> even if signalfd_ctx is not orphaned.
> 
> The comment looks clear to me. It states:
> 
> 1) The policy of returning 0 when the sighand has been detached
> 
> 2) That we _must_not_ call dequeue_signal() in case signalfd_lock() fails
> 
> #ACK on the code mod below.
>
> - Davide
> 

Andrew, please apply on top.

Simplify signalfd locking following suggestions by Oleg Nesterov.

Signed-off-by: Davi E. M. Arnaut <davi@haxent.com.br>

Index: linux-2.6/fs/signalfd.c
===================================================================
--- linux-2.6.orig/fs/signalfd.c
+++ linux-2.6/fs/signalfd.c
@@ -211,13 +211,11 @@ static int signalfd_copyinfo(struct sign
 static ssize_t signalfd_dequeue(struct signalfd_ctx *ctx, siginfo_t *info,
 				int nonblock)
 {
-	int locked;
 	ssize_t ret;
 	struct signalfd_lockctx lk;
 	DECLARE_WAITQUEUE(wait, current);
 
-	locked = signalfd_lock(ctx, &lk);
-	if (!locked)
+	if (!signalfd_lock(ctx, &lk))
 		return 0;
 
 	ret = dequeue_signal(lk.tsk, &ctx->sigmask, info);
@@ -235,24 +233,24 @@ static ssize_t signalfd_dequeue(struct s
 	for (;;) {
 		set_current_state(TASK_INTERRUPTIBLE);
 		ret = dequeue_signal(lk.tsk, &ctx->sigmask, info);
+		signalfd_unlock(&lk);
 		if (ret != 0)
 			break;
 		if (signal_pending(current)) {
 			ret = -ERESTARTSYS;
 			break;
 		}
-		signalfd_unlock(&lk);
 		schedule();
-		locked = signalfd_lock(ctx, &lk);
-		if (unlikely(!locked)) {
+		ret = signalfd_lock(ctx, &lk);
+		if (unlikely(!ret)) {
 			/*
 			 * Let the caller read zero byte, ala socket
 			 * recv() when the peer disconnect. This test
 			 * must be done before doing a dequeue_signal(),
 			 * because if the sighand has been orphaned,
-			 * the dequeue_signal() call is going to crash.
+			 * the dequeue_signal() call is going to crash
+			 * because ->sighand will be long gone.
 			 */
-			 ret = 0;
 			 break;
 		}
 	}
@@ -260,9 +258,6 @@ static ssize_t signalfd_dequeue(struct s
 	remove_wait_queue(&ctx->wqh, &wait);
 	__set_current_state(TASK_RUNNING);
 
-	if (likely(locked))
-		signalfd_unlock(&lk);
-
 	return ret;
 }