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  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]
Date:	Fri, 11 Apr 2014 12:41:42 -0400
From:	Brian Adamson <>
To:	Steven Galgano <>
Cc:	"Michael S. Tsirkin" <>,,,,,,,,,,
	Joseph Giovatto <>
Subject: Re: [PATCH] tuntap: add flow control to support back pressure

To weigh in on the desire to have support (at least as an optional behavior) for the legacy flow control behavior, there are many existing uses of it.  Many these are related to experimental purposes where the tuntap driver can be used (with a little user space code) as a surrogate for a network interface type that may not even yet exist.  And in some cases these experimental purposes have had utility for actual deployment (e.g. disaster relief wireless networks where  the TAP device has provided some intermediate assistance for routing or other protocols, even an underwater acoustic sensor network proposed for reef monitoring, etc where a TAP device provides a network interface and the sound card is used as a modem on an embedded system).  Some of these networks have low data rates or packet loss and delays that make TCP (which provides flow control as part of its usual reliable transport for more typical networking purpose) not an ideal protocol to use and so UDP or other alternatives or used.  To keep this short, I'll list a few use cases here I know (and was involved with the implementation of some) with some links (where I know them):

1) CORE network emulation tool  (

2) EMANE network emulation tool (

(likely other network emulation tools exist that have used tuntap as surrogates for real physical interfaces and expect the same backpressure to sockets and queues that physical interfaces provide)

3) I don't have a link to it but I implemented an experimental IP interface/ MAC protocol called SLIDE (serial-link internet daemon) that implemented a user-space CSMA MAC protocol where an underwater acoustic modem was connected to the serial port and TAP was used to present a virtual network interface to the IP stack.  Because of the low data rates involved, the back pressure flow control to application sockets (and protocol daemons and qdiscs applied)  was important.

4)  User space implementation of Simplified Multicast Forwarding (SMF) of RFC 6621 has a "device" option that establishes TAP interfaces to perform distributed "backpressure" based flow control (and potentially routing) for MANET wireless networks.  (

There are probably some more, among the more esoteric wireless and other special networking communities, where host (or routing/gateway/proxy non-host), e.g. special embedded system devices based on Linux such as sensors, etc) have a first hop network attachment that is _not_ the typical Ethernet or something and may be using tuntap along with a sort of user-space "driver" to present an IP interface to the network stack. some of this stuff, especially embedded systems, tend to lag behind with respect to kernel versions and this behavior change in Linux may be yet undiscovered so far even though the change was put in a couple years ago.

Several of these are implemented across multiple platforms, and, for example, BSD-based systems tuntap provides the same flow control behavior.  Even if it was never formally documented, I think this behavior was fairly well known (at least for these sorts of experimental purposes) and used.  I understand the concern that a single bad behaving flow can possibly block the flow of others unless traffic control queuing disciplines (as done for other network interfaces).  For the purposes of which I'm aware, I think having this behavior as _optional_ is probably OK … If accepted, and something is implemented here, it may be a good opportunity to have it documented (and the pros and cons of its use) for the more general Linux community.

BTW, in my initial noticing this issue, it _seemed_ that even the default interface pfifo_fast priority bands were not being properly enforced for the tap interface without the old flow control behavior?.  I need to do a little more "old vs new" comparison testing on this regard.

best regards,


On Apr 10, 2014, at 9:42 PM, Steven Galgano <> wrote:

> On 04/10/2014 06:29 AM, Michael S. Tsirkin wrote:
>> On Wed, Apr 09, 2014 at 10:19:40PM -0400, Steven Galgano wrote:
>>> Add tuntap flow control support for use by back pressure routing protocols. Setting the new TUNSETIFF flag IFF_FLOW_CONTROL, will signal resources as unavailable when the tx queue limit is reached by issuing a netif_tx_stop_all_queues() rather than discarding frames. A netif_tx_wake_all_queues() is issued after reading a frame from the queue to signal resource availability.
>>> Back pressure capability was previously supported by the legacy tun default mode. This change restores that functionality, which was last present in v3.7.
>>> Reported-by: Brian Adamson <>
>>> Tested-by: Joseph Giovatto <>
>>> Signed-off-by: Steven Galgano <>
>> I don't think it's a good idea.
>> This trivial flow control really created more problems than it was worth.
>> In particular this blocks all flows so it's trivially easy for one flow
>> to block and starve all others: just send a bunch of packets to loopback
>> destinations that get queued all over the place.
>> Luckily it was never documented so we changed the default and nothing
>> seems to break, but we won't be so lucky if we add an explicit API.
>> One way to implement this would be with ubuf_info callback this is
>> already invoked in most places where a packet might get stuck for a long
>> time.  It's still incomplete though: this will prevent head of queue
>> blocking literally forever, but a single bad flow can still degrade
>> performance significantly.
>> Another alternative is to try and isolate the flows that we
>> can handle and throttle them.
>> It's all fixable but we really need to fix the issues *before*
>> exposing the interface to userspace.
> It was only after recent upgrades that we picked up a newer kernel and
> discovered the change to the default tun mode.
> The new default behavior has broken applications that depend on the
> legacy behavior. Although not documented, the legacy behavior was well
> known at least to those working in the back pressure research community.
> The default legacy mode was/is a valid use case although I am not sure
> it fits with recent multiqueue support.
> When back pressure protocols are running over a tun interface they
> require legacy flow control in order to communicate congestion detected
> on the physical media they are using. Multiqueues do not apply here.
> These protocols only use one queue, so netif_tx_stop_all_queues() is the
> necessary behavior.
> I'm not tied to the idea of IFF_FLOW_CONTROL. I was aiming for middle
> ground where an application controlling the tun interface can state it
> wants the legacy flow control behavior understanding its limitations
> concerning multiple queues.
> What if we resurrect IFF_ONE_QUEUE and use that as a mechanism to
> indicate legacy flow control. A tun instance initially configured with
> IFF_ONE_QUEUE would not be allowed to attach or detach queues with
> TUNSETQUEUE and any additional opens with the same device name would
> fail. This mode would use the
> netif_tx_stop_all_queues()/netif_tx_wake_all_queues() flow control
> mechanism.
> If a tun application wants the current default behavior with only a
> single queue, it would not set the IFF_ONE_QUEUE flag. Not setting
> IFF_MULTI_QUEUE would not imply IFF_ONE_QUEUE.
> I'd be happy to implement this change if it is an acceptable solution.
> This would allow multiqueue tun development to advance while still
> supporting use cases dependent on legacy flow control.
> -steve
>>> ---
>>> diff --git a/drivers/net/tun.c b/drivers/net/tun.c
>>> index ee328ba..268130c 100644
>>> --- a/drivers/net/tun.c
>>> +++ b/drivers/net/tun.c
>>> @@ -783,8 +783,19 @@ static netdev_tx_t tun_net_xmit(struct sk_buff *skb, struct net_device *dev)
>>> 	 * number of queues.
>>> 	 */
>>> 	if (skb_queue_len(&tfile->>sk_receive_queue) * numqueues
>>> -			  >= dev->tx_queue_len)
>>> -		goto drop;
>>> +			>= dev->tx_queue_len) {
>>> +		if (tun->flags & TUN_FLOW_CONTROL) {
>>> +			/* Resources unavailable stop transmissions */
>>> +			netif_tx_stop_all_queues(dev);
>>> +
>>> +			/* We won't see all dropped packets individually, so
>>> +			 * over run error is more appropriate.
>>> +			 */
>>> +			dev->stats.tx_fifo_errors++;
>>> +		} else {
>>> +			goto drop;
>>> +		}
>>> +	}
>>> 	if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
>>> 		goto drop;
>>> @@ -1362,6 +1373,9 @@ static ssize_t tun_do_read(struct tun_struct *tun, struct tun_file *tfile,
>>> 			continue;
>>> 		}
>>> +		/* Wake in case resources previously signaled unavailable */
>>> +		netif_tx_wake_all_queues(tun->dev);
>>> +
>>> 		ret = tun_put_user(tun, tfile, skb, iv, len);
>>> 		kfree_skb(skb);
>>> 		break;
>>> @@ -1550,6 +1564,9 @@ static int tun_flags(struct tun_struct *tun)
>>> 	if (tun->flags & TUN_PERSIST)
>>> 		flags |= IFF_PERSIST;
>>> +	if (tun->flags & TUN_FLOW_CONTROL)
>>> +		flags |= IFF_FLOW_CONTROL;
>>> +
>>> 	return flags;
>>> }
>>> @@ -1732,6 +1749,11 @@ static int tun_set_iff(struct net *net, struct file *file, struct ifreq *ifr)
>>> 	else
>>> 		tun->flags &= ~TUN_TAP_MQ;
>>> +	if (ifr->ifr_flags & IFF_FLOW_CONTROL)
>>> +		tun->flags |= TUN_FLOW_CONTROL;
>>> +	else
>>> +		tun->flags &= ~TUN_FLOW_CONTROL;
>>> +
>>> 	/* Make sure persistent devices do not get stuck in
>>> 	 * xoff state.
>>> 	 */
>>> @@ -1900,7 +1922,8 @@ static long __tun_chr_ioctl(struct file *file, unsigned int cmd,
>>> 		 * This is needed because we never checked for invalid flags on
>>> 		 * TUNSETIFF. */
>>> 		return put_user(IFF_TUN | IFF_TAP | IFF_NO_PI | IFF_ONE_QUEUE |
>>> 				(unsigned int __user*)argp);
>>> 	} else if (cmd == TUNSETQUEUE)
>>> 		return tun_set_queue(file, &ifr);
>>> diff --git a/include/uapi/linux/if_tun.h b/include/uapi/linux/if_tun.h
>>> index e9502dd..bcf2790 100644
>>> --- a/include/uapi/linux/if_tun.h
>>> +++ b/include/uapi/linux/if_tun.h
>>> @@ -36,6 +36,7 @@
>>> #define TUN_PERSIST 	0x0100	
>>> #define TUN_VNET_HDR 	0x0200
>>> #define TUN_TAP_MQ      0x0400
>>> +#define TUN_FLOW_CONTROL 0x0800
>>> /* Ioctl defines */
>>> #define TUNSETNOCSUM  _IOW('T', 200, int) 
>>> @@ -70,6 +71,7 @@
>>> #define IFF_MULTI_QUEUE 0x0100
>>> #define IFF_ATTACH_QUEUE 0x0200
>>> #define IFF_DETACH_QUEUE 0x0400
>>> +#define IFF_FLOW_CONTROL 0x0010
>>> /* read-only flag */
>>> #define IFF_PERSIST	0x0800
>>> #define IFF_NOFILTER	0x1000

To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to
More majordomo info at

Powered by blists - more mailing lists