netdev - Re: [PATCH 2/3] caif: Add support for flow-control on device's tx-queue

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <1322836682.2762.8.camel@edumazet-laptop>
Date:	Fri, 02 Dec 2011 15:38:02 +0100
From:	Eric Dumazet <eric.dumazet@...il.com>
To:	Sjur Brændeland 
	<sjur.brandeland@...ricsson.com>
Cc:	netdev@...r.kernel.org, David Miller <davem@...emloft.net>,
	Alexey Orishko <alexey.orishko@...ricsson.com>
Subject: Re: [PATCH 2/3] caif: Add support for flow-control on device's
 tx-queue

Le vendredi 02 décembre 2011 à 15:06 +0100, Sjur Brændeland a écrit :
> Flow control is implemented by inspecting the qdisc queue length
> in order to detect potential overflow on the TX queue. When a threshold
> is reached flow-off is sent upwards in the CAIF stack. At the same time
> the skb->destructor is hi-jacked in order to detect when the last packet
> put on queue is consumed. When this "hi-jacked" packet is consumed, flow-on
> is sent upwards in the CAIF stack.
> 
> Signed-off-by: Sjur Brændeland <sjur.brandeland@...ricsson.com>
> ---
>  net/caif/caif_dev.c |   48 ++++++++++++++++++++++++++++++++++++++++++++++++
>  1 files changed, 48 insertions(+), 0 deletions(-)
> 
> diff --git a/net/caif/caif_dev.c b/net/caif/caif_dev.c
> index f7e8c70..415353e 100644
> --- a/net/caif/caif_dev.c
> +++ b/net/caif/caif_dev.c
> @@ -34,6 +34,7 @@ struct caif_device_entry {
>  	struct list_head list;
>  	struct net_device *netdev;
>  	int __percpu *pcpu_refcnt;
> +	bool xoff;
>  };
>  
>  struct caif_device_entry_list {
> @@ -48,6 +49,7 @@ struct caif_net {
>  };
>  
>  static int caif_net_id;
> +static int q_high = 50; /* Percent */
>  
>  struct cfcnfg *get_cfcnfg(struct net *net)
>  {
> @@ -126,9 +128,28 @@ static struct caif_device_entry *caif_get(struct net_device *dev)
>  	return NULL;
>  }
>  
> +void caif_flow_cb(struct sk_buff *skb)
> +{
> +	struct caif_device_entry *caifd;
> +	WARN_ON(skb->dev == NULL);
> +
> +	rcu_read_lock();
> +	caifd = caif_get(skb->dev);
> +	caifd->xoff = 0;
> +	caifd_hold(caifd);
> +	rcu_read_unlock();
> +
> +	caifd->layer.up->
> +		ctrlcmd(caifd->layer.up,
> +			_CAIF_CTRLCMD_PHYIF_FLOW_ON_IND,
> +			caifd->layer.id);
> +	caifd_put(caifd);
> +}
> +
>  static int transmit(struct cflayer *layer, struct cfpkt *pkt)
>  {
>  	int err;
> +	struct caif_dev_common *caifdev;
>  	struct caif_device_entry *caifd =
>  	    container_of(layer, struct caif_device_entry, layer);
>  	struct sk_buff *skb;
> @@ -137,6 +158,33 @@ static int transmit(struct cflayer *layer, struct cfpkt *pkt)
>  	skb->dev = caifd->netdev;
>  	skb_reset_network_header(skb);
>  	skb->protocol = htons(ETH_P_CAIF);
> +	caifdev = netdev_priv(caifd->netdev);
> +
> +	if (caifdev->flowctrl == NULL && caifd->netdev->tx_queue_len > 0 &&
> +			!caifd->xoff) {
> +		struct netdev_queue *txq;
> +		int high;
> +
> +		txq = netdev_get_tx_queue(skb->dev, 0);

Why queue 0 and not another one ?

> +		high = (caifd->netdev->tx_queue_len * q_high) / 100;
> +
> +		/* If we run with a TX queue, check if the queue is too long*/

Are you sure only this cpu can run here ? any lock is held ?

> +		if (netif_queue_stopped(caifd->netdev) ||
> +			qdisc_qlen(txq->qdisc) > high) {
> +
> +			pr_debug("queue stop(%d) or full (%d>%d) - XOFF\n",
> +					netif_queue_stopped(caifd->netdev),
> +					qdisc_qlen(txq->qdisc), high);
> +			caifd->xoff = 1;
> +			/* Hijack this skb free callback function. */
> +			skb_orphan(skb);
> +			skb->destructor = caif_flow_cb;
> +			caifd->layer.up->
> +				ctrlcmd(caifd->layer.up,
> +					_CAIF_CTRLCMD_PHYIF_FLOW_OFF_IND,
> +					caifd->layer.id);
> +		}
> +	}
>  
>  	err = dev_queue_xmit(skb);
>  	if (err > 0)

What prevents dev_queue_xmit() to early orphan skb ?



--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html