| Message ID | 1366284715-10107-1-git-send-email-s.fedorov@samsung.com |
|---|---|
| State | New |
| Headers | show |
On Thu, Apr 18, 2013 at 03:31:55PM +0400, Sergey Fedorov wrote: > Network hub should always receive incoming packets. Then forward them to > the appropriate port queue and let the qemu_send_packet() do the right > things. If the destination queue cannot receive the packet it will be > appended to the queue. When the receiver call > qemu_flush_queued_packets() later the queue will be really flushed and > no packets will be stalled in the sender network queue. > > Signed-off-by: Sergey Fedorov <s.fedorov@samsung.com> > --- > net/hub.c | 20 -------------------- > 1 file changed, 20 deletions(-) What is the point of this change? There is no semantic difference for well-behaved net clients. Does it fix a bug, if so, please include details? Stefan
On 04/22/2013 03:47 PM, Stefan Hajnoczi wrote: > On Thu, Apr 18, 2013 at 03:31:55PM +0400, Sergey Fedorov wrote: >> Network hub should always receive incoming packets. Then forward them to >> the appropriate port queue and let the qemu_send_packet() do the right >> things. If the destination queue cannot receive the packet it will be >> appended to the queue. When the receiver call >> qemu_flush_queued_packets() later the queue will be really flushed and >> no packets will be stalled in the sender network queue. >> >> Signed-off-by: Sergey Fedorov <s.fedorov@samsung.com> >> --- >> net/hub.c | 20 -------------------- >> 1 file changed, 20 deletions(-) > What is the point of this change? There is no semantic difference for > well-behaved net clients. > > Does it fix a bug, if so, please include details? > > Stefan > > Yes, this fixes a bug. There were packet stalls when using user-mode networking with USB network device. slirp_output() calls qemu_send_packet() which eventually calls qemu_net_queue_send(). qemu_net_queue_send() calls qemu_can_send_packet(), which calls can_receive() callback of network hub. Then net_hub_port_can_receive() also calls qemu_can_send_packet() for each port except packet source port. Sometimes USB network device is not able to receive packet and qemu_can_send_packet() returns false. In my case there is no more ports and net_hub_port_can_receive() returns false. So qemu_net_queue_send() call qemu_net_queue_append() instead of qemu_net_queue_deliver(). qemu_net_queue_append() appends the packet to the receiving port of the network hub which is not flushed when USB netork device calls qemu_flush_queued_packets(). It is flushed only when slirp resend the packet by timeout. Actually there is no need in net_hub_port_can_receive() as the network hub can always receive packets and pass it to its port network clients with qemu_send_packet(). And if the destination port network client cannot receive the packet it will be queued in the *destination* port network client queue. Queued packets from that queue will be delivered as soon as the network client call qemu_flush_queued_packets().
On Mon, Apr 22, 2013 at 04:26:16PM +0400, Fedorov Sergey wrote: > On 04/22/2013 03:47 PM, Stefan Hajnoczi wrote: > >On Thu, Apr 18, 2013 at 03:31:55PM +0400, Sergey Fedorov wrote: > >>Network hub should always receive incoming packets. Then forward them to > >>the appropriate port queue and let the qemu_send_packet() do the right > >>things. If the destination queue cannot receive the packet it will be > >>appended to the queue. When the receiver call > >>qemu_flush_queued_packets() later the queue will be really flushed and > >>no packets will be stalled in the sender network queue. > >> > >>Signed-off-by: Sergey Fedorov <s.fedorov@samsung.com> > >>--- > >> net/hub.c | 20 -------------------- > >> 1 file changed, 20 deletions(-) > >What is the point of this change? There is no semantic difference for > >well-behaved net clients. > > > >Does it fix a bug, if so, please include details? > > > >Stefan > > > > > > Yes, this fixes a bug. There were packet stalls when using user-mode > networking with USB network device. slirp_output() calls > qemu_send_packet() which eventually calls qemu_net_queue_send(). > qemu_net_queue_send() calls qemu_can_send_packet(), which calls > can_receive() callback of network hub. Then > net_hub_port_can_receive() also calls qemu_can_send_packet() for > each port except packet source port. > > Sometimes USB network device is not able to receive packet and > qemu_can_send_packet() returns false. In my case there is no more > ports and net_hub_port_can_receive() returns false. So > qemu_net_queue_send() call qemu_net_queue_append() instead of > qemu_net_queue_deliver(). qemu_net_queue_append() appends the packet > to the receiving port of the network hub which is not flushed when > USB netork device calls qemu_flush_queued_packets(). It is flushed > only when slirp resend the packet by timeout. > > Actually there is no need in net_hub_port_can_receive() as the > network hub can always receive packets and pass it to its port > network clients with qemu_send_packet(). And if the destination port > network client cannot receive the packet it will be queued in the > *destination* port network client queue. Queued packets from that > queue will be delivered as soon as the network client call > qemu_flush_queued_packets(). Please confirm the bug is still present in qemu.git/master. It should have been fixed by the following commit: commit 199ee608f0d08510b5c6c37f31a7fbff211d63c4 Author: Luigi Rizzo <rizzo@iet.unipi.it> Date: Tue Feb 5 17:53:31 2013 +0100 net: fix qemu_flush_queued_packets() in presence of a hub Stefan
On 04/22/2013 06:57 PM, Stefan Hajnoczi wrote: > On Mon, Apr 22, 2013 at 04:26:16PM +0400, Fedorov Sergey wrote: >> On 04/22/2013 03:47 PM, Stefan Hajnoczi wrote: >>> On Thu, Apr 18, 2013 at 03:31:55PM +0400, Sergey Fedorov wrote: >>>> Network hub should always receive incoming packets. Then forward them to >>>> the appropriate port queue and let the qemu_send_packet() do the right >>>> things. If the destination queue cannot receive the packet it will be >>>> appended to the queue. When the receiver call >>>> qemu_flush_queued_packets() later the queue will be really flushed and >>>> no packets will be stalled in the sender network queue. >>>> >>>> Signed-off-by: Sergey Fedorov <s.fedorov@samsung.com> >>>> --- >>>> net/hub.c | 20 -------------------- >>>> 1 file changed, 20 deletions(-) >>> What is the point of this change? There is no semantic difference for >>> well-behaved net clients. >>> >>> Does it fix a bug, if so, please include details? >>> >>> Stefan >>> >>> >> Yes, this fixes a bug. There were packet stalls when using user-mode >> networking with USB network device. slirp_output() calls >> qemu_send_packet() which eventually calls qemu_net_queue_send(). >> qemu_net_queue_send() calls qemu_can_send_packet(), which calls >> can_receive() callback of network hub. Then >> net_hub_port_can_receive() also calls qemu_can_send_packet() for >> each port except packet source port. >> >> Sometimes USB network device is not able to receive packet and >> qemu_can_send_packet() returns false. In my case there is no more >> ports and net_hub_port_can_receive() returns false. So >> qemu_net_queue_send() call qemu_net_queue_append() instead of >> qemu_net_queue_deliver(). qemu_net_queue_append() appends the packet >> to the receiving port of the network hub which is not flushed when >> USB netork device calls qemu_flush_queued_packets(). It is flushed >> only when slirp resend the packet by timeout. >> >> Actually there is no need in net_hub_port_can_receive() as the >> network hub can always receive packets and pass it to its port >> network clients with qemu_send_packet(). And if the destination port >> network client cannot receive the packet it will be queued in the >> *destination* port network client queue. Queued packets from that >> queue will be delivered as soon as the network client call >> qemu_flush_queued_packets(). > Please confirm the bug is still present in qemu.git/master. It should > have been fixed by the following commit: > > commit 199ee608f0d08510b5c6c37f31a7fbff211d63c4 > Author: Luigi Rizzo <rizzo@iet.unipi.it> > Date: Tue Feb 5 17:53:31 2013 +0100 > > net: fix qemu_flush_queued_packets() in presence of a hub > > Stefan > Yes, this commit fixes a bug but from other side. I think it's better to just let the qemu_send_packet() do the right things. E.g. network hub has 3 ports. Suppose when iterating through port list in net_hub_port_can_receive() a packet is successfully delivered to the first port, and then is queued in the source port queue because the second port cannot receive packets. Later net_hub_flush() will flush the packet from the source port queue and it will be delivered in every port. But it had been already delivered to one of them. So it will be delivered twice to some ports. Moreover there is less chance to dequeue the packet if several clients can't receive periodically. Anyway, actually there is no need in net_hub_port_can_receive() as the network hub can always receive packets and pass it to its port network clients with qemu_send_packet(). I think it's more natural solution. Sergey
Il 22/04/2013 17:27, Fedorov Sergey ha scritto: > > E.g. network hub has 3 ports. Suppose when iterating through port list > in net_hub_port_can_receive() a packet is successfully delivered to the > first port, and then is queued in the source port queue because the > second port cannot receive packets. Later net_hub_flush() will flush the > packet from the source port queue and it will be delivered in every > port. But it had been already delivered to one of them. So it will be > delivered twice to some ports. Moreover there is less chance to dequeue > the packet if several clients can't receive periodically. Perhaps it is indeed wrong to do this blocking in can_receive()... You're right that a hubport can always receive, but the hub itself should have a queue. If one port cannot receive, the packet should be appended to the hub's queue. And net_hub_flush will just go through the hub's queue. > Anyway, actually there is no need in net_hub_port_can_receive() as the > network hub can always receive packets and pass it to its port network > clients with qemu_send_packet(). I think it's more natural solution. I think the point was to keep dumps in sync with what actually happened on the other ports. Otherwise a "-net dump" port will show the packet immediately, even though it hasn't been delivered yet. Paolo
On Mon, Apr 22, 2013 at 07:27:21PM +0400, Fedorov Sergey wrote: > On 04/22/2013 06:57 PM, Stefan Hajnoczi wrote: > >On Mon, Apr 22, 2013 at 04:26:16PM +0400, Fedorov Sergey wrote: > >>On 04/22/2013 03:47 PM, Stefan Hajnoczi wrote: > >>>On Thu, Apr 18, 2013 at 03:31:55PM +0400, Sergey Fedorov wrote: > >>>>Network hub should always receive incoming packets. Then forward them to > >>>>the appropriate port queue and let the qemu_send_packet() do the right > >>>>things. If the destination queue cannot receive the packet it will be > >>>>appended to the queue. When the receiver call > >>>>qemu_flush_queued_packets() later the queue will be really flushed and > >>>>no packets will be stalled in the sender network queue. > >>>> > >>>>Signed-off-by: Sergey Fedorov <s.fedorov@samsung.com> > >>>>--- > >>>> net/hub.c | 20 -------------------- > >>>> 1 file changed, 20 deletions(-) > >>>What is the point of this change? There is no semantic difference for > >>>well-behaved net clients. > >>> > >>>Does it fix a bug, if so, please include details? > >>> > >>>Stefan > >>> > >>> > >>Yes, this fixes a bug. There were packet stalls when using user-mode > >>networking with USB network device. slirp_output() calls > >>qemu_send_packet() which eventually calls qemu_net_queue_send(). > >>qemu_net_queue_send() calls qemu_can_send_packet(), which calls > >>can_receive() callback of network hub. Then > >>net_hub_port_can_receive() also calls qemu_can_send_packet() for > >>each port except packet source port. > >> > >>Sometimes USB network device is not able to receive packet and > >>qemu_can_send_packet() returns false. In my case there is no more > >>ports and net_hub_port_can_receive() returns false. So > >>qemu_net_queue_send() call qemu_net_queue_append() instead of > >>qemu_net_queue_deliver(). qemu_net_queue_append() appends the packet > >>to the receiving port of the network hub which is not flushed when > >>USB netork device calls qemu_flush_queued_packets(). It is flushed > >>only when slirp resend the packet by timeout. > >> > >>Actually there is no need in net_hub_port_can_receive() as the > >>network hub can always receive packets and pass it to its port > >>network clients with qemu_send_packet(). And if the destination port > >>network client cannot receive the packet it will be queued in the > >>*destination* port network client queue. Queued packets from that > >>queue will be delivered as soon as the network client call > >>qemu_flush_queued_packets(). > >Please confirm the bug is still present in qemu.git/master. It should > >have been fixed by the following commit: > > > > commit 199ee608f0d08510b5c6c37f31a7fbff211d63c4 > > Author: Luigi Rizzo <rizzo@iet.unipi.it> > > Date: Tue Feb 5 17:53:31 2013 +0100 > > > > net: fix qemu_flush_queued_packets() in presence of a hub > > > >Stefan > > > Yes, this commit fixes a bug but from other side. I think it's > better to just let the qemu_send_packet() do the right things. > > E.g. network hub has 3 ports. Suppose when iterating through port > list in net_hub_port_can_receive() a packet is successfully > delivered to the first port, and then is queued in the source port > queue because the second port cannot receive packets. Later > net_hub_flush() will flush the packet from the source port queue and > it will be delivered in every port. But it had been already > delivered to one of them. So it will be delivered twice to some > ports. Moreover there is less chance to dequeue the packet if > several clients can't receive periodically. Did you mean "second port" instead of "source port" in the beginning? I don't see a scenario where the packet is delivered to the same port twice. If one port can receive then the packet will be sent and other ports will queue the packet. Note that the source port will not queue the packet. When the blocked ports can receive again their send queues will be flushed. Paolo mentioned that we want to provide a consistent view of packets, especially when -net dump is used. Beyond that, we also want to avoid growing net queues indefinitely. If the hub does not implement .can_receive() then it relies on growing queues (keeping packets buffered in memory). Stefan
On 04/22/2013 08:09 PM, Paolo Bonzini wrote: > Il 22/04/2013 17:27, Fedorov Sergey ha scritto: >> E.g. network hub has 3 ports. Suppose when iterating through port list >> in net_hub_port_can_receive() a packet is successfully delivered to the >> first port, and then is queued in the source port queue because the >> second port cannot receive packets. Later net_hub_flush() will flush the >> packet from the source port queue and it will be delivered in every >> port. But it had been already delivered to one of them. So it will be >> delivered twice to some ports. Moreover there is less chance to dequeue >> the packet if several clients can't receive periodically. > Perhaps it is indeed wrong to do this blocking in can_receive()... > You're right that a hubport can always receive, but the hub itself > should have a queue. If one port cannot receive, the packet should be > appended to the hub's queue. And net_hub_flush will just go through the > hub's queue. > >> Anyway, actually there is no need in net_hub_port_can_receive() as the >> network hub can always receive packets and pass it to its port network >> clients with qemu_send_packet(). I think it's more natural solution. > I think the point was to keep dumps in sync with what actually happened > on the other ports. Otherwise a "-net dump" port will show the packet > immediately, even though it hasn't been delivered yet. > > Paolo > The user documentation says that `-net dump' dumps network traffic on VLAN. There is nothing said about interfaces connected to VALN nor synchronization with them. In my view, VLAN hub is a device that forwards an incoming packet to all other ports as soon as possible, just like real hardware hub. I think of `-net dump' as a special network interface that dumps every packet received by VLAN hub itself, not delivered to its clients. If we want to dump packets actually delivered to network interfaces we need to do it in per-interface manner. Maybe I misunderstand?
On 04/23/2013 10:58 AM, Stefan Hajnoczi wrote: > On Mon, Apr 22, 2013 at 07:27:21PM +0400, Fedorov Sergey wrote: >> On 04/22/2013 06:57 PM, Stefan Hajnoczi wrote: >>> On Mon, Apr 22, 2013 at 04:26:16PM +0400, Fedorov Sergey wrote: >>>> On 04/22/2013 03:47 PM, Stefan Hajnoczi wrote: >>>>> On Thu, Apr 18, 2013 at 03:31:55PM +0400, Sergey Fedorov wrote: >>>>>> Network hub should always receive incoming packets. Then forward them to >>>>>> the appropriate port queue and let the qemu_send_packet() do the right >>>>>> things. If the destination queue cannot receive the packet it will be >>>>>> appended to the queue. When the receiver call >>>>>> qemu_flush_queued_packets() later the queue will be really flushed and >>>>>> no packets will be stalled in the sender network queue. >>>>>> >>>>>> Signed-off-by: Sergey Fedorov <s.fedorov@samsung.com> >>>>>> --- >>>>>> net/hub.c | 20 -------------------- >>>>>> 1 file changed, 20 deletions(-) >>>>> What is the point of this change? There is no semantic difference for >>>>> well-behaved net clients. >>>>> >>>>> Does it fix a bug, if so, please include details? >>>>> >>>>> Stefan >>>>> >>>>> >>>> Yes, this fixes a bug. There were packet stalls when using user-mode >>>> networking with USB network device. slirp_output() calls >>>> qemu_send_packet() which eventually calls qemu_net_queue_send(). >>>> qemu_net_queue_send() calls qemu_can_send_packet(), which calls >>>> can_receive() callback of network hub. Then >>>> net_hub_port_can_receive() also calls qemu_can_send_packet() for >>>> each port except packet source port. >>>> >>>> Sometimes USB network device is not able to receive packet and >>>> qemu_can_send_packet() returns false. In my case there is no more >>>> ports and net_hub_port_can_receive() returns false. So >>>> qemu_net_queue_send() call qemu_net_queue_append() instead of >>>> qemu_net_queue_deliver(). qemu_net_queue_append() appends the packet >>>> to the receiving port of the network hub which is not flushed when >>>> USB netork device calls qemu_flush_queued_packets(). It is flushed >>>> only when slirp resend the packet by timeout. >>>> >>>> Actually there is no need in net_hub_port_can_receive() as the >>>> network hub can always receive packets and pass it to its port >>>> network clients with qemu_send_packet(). And if the destination port >>>> network client cannot receive the packet it will be queued in the >>>> *destination* port network client queue. Queued packets from that >>>> queue will be delivered as soon as the network client call >>>> qemu_flush_queued_packets(). >>> Please confirm the bug is still present in qemu.git/master. It should >>> have been fixed by the following commit: >>> >>> commit 199ee608f0d08510b5c6c37f31a7fbff211d63c4 >>> Author: Luigi Rizzo <rizzo@iet.unipi.it> >>> Date: Tue Feb 5 17:53:31 2013 +0100 >>> >>> net: fix qemu_flush_queued_packets() in presence of a hub >>> >>> Stefan >>> >> Yes, this commit fixes a bug but from other side. I think it's >> better to just let the qemu_send_packet() do the right things. >> >> E.g. network hub has 3 ports. Suppose when iterating through port >> list in net_hub_port_can_receive() a packet is successfully >> delivered to the first port, and then is queued in the source port >> queue because the second port cannot receive packets. Later >> net_hub_flush() will flush the packet from the source port queue and >> it will be delivered in every port. But it had been already >> delivered to one of them. So it will be delivered twice to some >> ports. Moreover there is less chance to dequeue the packet if >> several clients can't receive periodically. > Did you mean "second port" instead of "source port" in the beginning? > > I don't see a scenario where the packet is delivered to the same port > twice. If one port can receive then the packet will be sent and other > ports will queue the packet. I agree. I was wrong with that. > Note that the source port will not queue the packet. The source port will queue the packet if no port, except the source one, can receive. > When the blocked ports can receive again their send queues > will be flushed. > > Paolo mentioned that we want to provide a consistent view of packets, > especially when -net dump is used. > > Beyond that, we also want to avoid growing net queues indefinitely. If > the hub does not implement .can_receive() then it relies on growing > queues (keeping packets buffered in memory). No, net_hub_receive() calls qemu_send_packet(). If the destination queue cannot receive the packet qemu_net_queue_append() will take care of queue->nq_maxlen. > > Stefan >
On 04/23/2013 10:58 AM, Stefan Hajnoczi wrote: > On Mon, Apr 22, 2013 at 07:27:21PM +0400, Fedorov Sergey wrote: >> On 04/22/2013 06:57 PM, Stefan Hajnoczi wrote: >>> On Mon, Apr 22, 2013 at 04:26:16PM +0400, Fedorov Sergey wrote: >>>> On 04/22/2013 03:47 PM, Stefan Hajnoczi wrote: >>>>> On Thu, Apr 18, 2013 at 03:31:55PM +0400, Sergey Fedorov wrote: >>>>>> Network hub should always receive incoming packets. Then forward them to >>>>>> the appropriate port queue and let the qemu_send_packet() do the right >>>>>> things. If the destination queue cannot receive the packet it will be >>>>>> appended to the queue. When the receiver call >>>>>> qemu_flush_queued_packets() later the queue will be really flushed and >>>>>> no packets will be stalled in the sender network queue. >>>>>> >>>>>> Signed-off-by: Sergey Fedorov <s.fedorov@samsung.com> >>>>>> --- >>>>>> net/hub.c | 20 -------------------- >>>>>> 1 file changed, 20 deletions(-) >>>>> What is the point of this change? There is no semantic difference for >>>>> well-behaved net clients. >>>>> >>>>> Does it fix a bug, if so, please include details? >>>>> >>>>> Stefan >>>>> >>>>> >>>> Yes, this fixes a bug. There were packet stalls when using user-mode >>>> networking with USB network device. slirp_output() calls >>>> qemu_send_packet() which eventually calls qemu_net_queue_send(). >>>> qemu_net_queue_send() calls qemu_can_send_packet(), which calls >>>> can_receive() callback of network hub. Then >>>> net_hub_port_can_receive() also calls qemu_can_send_packet() for >>>> each port except packet source port. >>>> >>>> Sometimes USB network device is not able to receive packet and >>>> qemu_can_send_packet() returns false. In my case there is no more >>>> ports and net_hub_port_can_receive() returns false. So >>>> qemu_net_queue_send() call qemu_net_queue_append() instead of >>>> qemu_net_queue_deliver(). qemu_net_queue_append() appends the packet >>>> to the receiving port of the network hub which is not flushed when >>>> USB netork device calls qemu_flush_queued_packets(). It is flushed >>>> only when slirp resend the packet by timeout. >>>> >>>> Actually there is no need in net_hub_port_can_receive() as the >>>> network hub can always receive packets and pass it to its port >>>> network clients with qemu_send_packet(). And if the destination port >>>> network client cannot receive the packet it will be queued in the >>>> *destination* port network client queue. Queued packets from that >>>> queue will be delivered as soon as the network client call >>>> qemu_flush_queued_packets(). >>> Please confirm the bug is still present in qemu.git/master. It should >>> have been fixed by the following commit: >>> >>> commit 199ee608f0d08510b5c6c37f31a7fbff211d63c4 >>> Author: Luigi Rizzo <rizzo@iet.unipi.it> >>> Date: Tue Feb 5 17:53:31 2013 +0100 >>> >>> net: fix qemu_flush_queued_packets() in presence of a hub >>> >>> Stefan >>> >> Yes, this commit fixes a bug but from other side. I think it's >> better to just let the qemu_send_packet() do the right things. >> >> E.g. network hub has 3 ports. Suppose when iterating through port >> list in net_hub_port_can_receive() a packet is successfully >> delivered to the first port, and then is queued in the source port >> queue because the second port cannot receive packets. Later >> net_hub_flush() will flush the packet from the source port queue and >> it will be delivered in every port. But it had been already >> delivered to one of them. So it will be delivered twice to some >> ports. Moreover there is less chance to dequeue the packet if >> several clients can't receive periodically. > Did you mean "second port" instead of "source port" in the beginning? > > I don't see a scenario where the packet is delivered to the same port > twice. If one port can receive then the packet will be sent and other > ports will queue the packet. Whether dump network client can always receive? If so, then the packet will always be sent to the dump client regardless of whether the other clients can receive. Is there actually any synchronization? > Note that the source port will not queue > the packet. When the blocked ports can receive again their send queues > will be flushed. > > Paolo mentioned that we want to provide a consistent view of packets, > especially when -net dump is used. > > Beyond that, we also want to avoid growing net queues indefinitely. If > the hub does not implement .can_receive() then it relies on growing > queues (keeping packets buffered in memory). > > Stefan >
On Tue, Apr 23, 2013 at 01:32:11PM +0400, Fedorov Sergey wrote: > > On 04/23/2013 10:58 AM, Stefan Hajnoczi wrote: > >On Mon, Apr 22, 2013 at 07:27:21PM +0400, Fedorov Sergey wrote: > >>On 04/22/2013 06:57 PM, Stefan Hajnoczi wrote: > >>>On Mon, Apr 22, 2013 at 04:26:16PM +0400, Fedorov Sergey wrote: > >>>>On 04/22/2013 03:47 PM, Stefan Hajnoczi wrote: > >>>>>On Thu, Apr 18, 2013 at 03:31:55PM +0400, Sergey Fedorov wrote: > >>>>>>Network hub should always receive incoming packets. Then forward them to > >>>>>>the appropriate port queue and let the qemu_send_packet() do the right > >>>>>>things. If the destination queue cannot receive the packet it will be > >>>>>>appended to the queue. When the receiver call > >>>>>>qemu_flush_queued_packets() later the queue will be really flushed and > >>>>>>no packets will be stalled in the sender network queue. > >>>>>> > >>>>>>Signed-off-by: Sergey Fedorov <s.fedorov@samsung.com> > >>>>>>--- > >>>>>> net/hub.c | 20 -------------------- > >>>>>> 1 file changed, 20 deletions(-) > >>>>>What is the point of this change? There is no semantic difference for > >>>>>well-behaved net clients. > >>>>> > >>>>>Does it fix a bug, if so, please include details? > >>>>> > >>>>>Stefan > >>>>> > >>>>> > >>>>Yes, this fixes a bug. There were packet stalls when using user-mode > >>>>networking with USB network device. slirp_output() calls > >>>>qemu_send_packet() which eventually calls qemu_net_queue_send(). > >>>>qemu_net_queue_send() calls qemu_can_send_packet(), which calls > >>>>can_receive() callback of network hub. Then > >>>>net_hub_port_can_receive() also calls qemu_can_send_packet() for > >>>>each port except packet source port. > >>>> > >>>>Sometimes USB network device is not able to receive packet and > >>>>qemu_can_send_packet() returns false. In my case there is no more > >>>>ports and net_hub_port_can_receive() returns false. So > >>>>qemu_net_queue_send() call qemu_net_queue_append() instead of > >>>>qemu_net_queue_deliver(). qemu_net_queue_append() appends the packet > >>>>to the receiving port of the network hub which is not flushed when > >>>>USB netork device calls qemu_flush_queued_packets(). It is flushed > >>>>only when slirp resend the packet by timeout. > >>>> > >>>>Actually there is no need in net_hub_port_can_receive() as the > >>>>network hub can always receive packets and pass it to its port > >>>>network clients with qemu_send_packet(). And if the destination port > >>>>network client cannot receive the packet it will be queued in the > >>>>*destination* port network client queue. Queued packets from that > >>>>queue will be delivered as soon as the network client call > >>>>qemu_flush_queued_packets(). > >>>Please confirm the bug is still present in qemu.git/master. It should > >>>have been fixed by the following commit: > >>> > >>> commit 199ee608f0d08510b5c6c37f31a7fbff211d63c4 > >>> Author: Luigi Rizzo <rizzo@iet.unipi.it> > >>> Date: Tue Feb 5 17:53:31 2013 +0100 > >>> > >>> net: fix qemu_flush_queued_packets() in presence of a hub > >>> > >>>Stefan > >>> > >>Yes, this commit fixes a bug but from other side. I think it's > >>better to just let the qemu_send_packet() do the right things. > >> > >>E.g. network hub has 3 ports. Suppose when iterating through port > >>list in net_hub_port_can_receive() a packet is successfully > >>delivered to the first port, and then is queued in the source port > >>queue because the second port cannot receive packets. Later > >>net_hub_flush() will flush the packet from the source port queue and > >>it will be delivered in every port. But it had been already > >>delivered to one of them. So it will be delivered twice to some > >>ports. Moreover there is less chance to dequeue the packet if > >>several clients can't receive periodically. > >Did you mean "second port" instead of "source port" in the beginning? > > > >I don't see a scenario where the packet is delivered to the same port > >twice. If one port can receive then the packet will be sent and other > >ports will queue the packet. > Whether dump network client can always receive? If so, then the > packet will always be sent to the dump client regardless of whether > the other clients can receive. Is there actually any > synchronization? Yes, the dump net client can always receive. Other net clients may not. There is no explicit synchronization between send queues on the hub. Stefan
On 04/23/2013 03:48 PM, Stefan Hajnoczi wrote: > On Tue, Apr 23, 2013 at 01:32:11PM +0400, Fedorov Sergey wrote: >> On 04/23/2013 10:58 AM, Stefan Hajnoczi wrote: >>> On Mon, Apr 22, 2013 at 07:27:21PM +0400, Fedorov Sergey wrote: >>>> On 04/22/2013 06:57 PM, Stefan Hajnoczi wrote: >>>>> On Mon, Apr 22, 2013 at 04:26:16PM +0400, Fedorov Sergey wrote: >>>>>> On 04/22/2013 03:47 PM, Stefan Hajnoczi wrote: >>>>>>> On Thu, Apr 18, 2013 at 03:31:55PM +0400, Sergey Fedorov wrote: >>>>>>>> Network hub should always receive incoming packets. Then forward them to >>>>>>>> the appropriate port queue and let the qemu_send_packet() do the right >>>>>>>> things. If the destination queue cannot receive the packet it will be >>>>>>>> appended to the queue. When the receiver call >>>>>>>> qemu_flush_queued_packets() later the queue will be really flushed and >>>>>>>> no packets will be stalled in the sender network queue. >>>>>>>> >>>>>>>> Signed-off-by: Sergey Fedorov <s.fedorov@samsung.com> >>>>>>>> --- >>>>>>>> net/hub.c | 20 -------------------- >>>>>>>> 1 file changed, 20 deletions(-) >>>>>>> What is the point of this change? There is no semantic difference for >>>>>>> well-behaved net clients. >>>>>>> >>>>>>> Does it fix a bug, if so, please include details? >>>>>>> >>>>>>> Stefan >>>>>>> >>>>>>> >>>>>> Yes, this fixes a bug. There were packet stalls when using user-mode >>>>>> networking with USB network device. slirp_output() calls >>>>>> qemu_send_packet() which eventually calls qemu_net_queue_send(). >>>>>> qemu_net_queue_send() calls qemu_can_send_packet(), which calls >>>>>> can_receive() callback of network hub. Then >>>>>> net_hub_port_can_receive() also calls qemu_can_send_packet() for >>>>>> each port except packet source port. >>>>>> >>>>>> Sometimes USB network device is not able to receive packet and >>>>>> qemu_can_send_packet() returns false. In my case there is no more >>>>>> ports and net_hub_port_can_receive() returns false. So >>>>>> qemu_net_queue_send() call qemu_net_queue_append() instead of >>>>>> qemu_net_queue_deliver(). qemu_net_queue_append() appends the packet >>>>>> to the receiving port of the network hub which is not flushed when >>>>>> USB netork device calls qemu_flush_queued_packets(). It is flushed >>>>>> only when slirp resend the packet by timeout. >>>>>> >>>>>> Actually there is no need in net_hub_port_can_receive() as the >>>>>> network hub can always receive packets and pass it to its port >>>>>> network clients with qemu_send_packet(). And if the destination port >>>>>> network client cannot receive the packet it will be queued in the >>>>>> *destination* port network client queue. Queued packets from that >>>>>> queue will be delivered as soon as the network client call >>>>>> qemu_flush_queued_packets(). >>>>> Please confirm the bug is still present in qemu.git/master. It should >>>>> have been fixed by the following commit: >>>>> >>>>> commit 199ee608f0d08510b5c6c37f31a7fbff211d63c4 >>>>> Author: Luigi Rizzo <rizzo@iet.unipi.it> >>>>> Date: Tue Feb 5 17:53:31 2013 +0100 >>>>> >>>>> net: fix qemu_flush_queued_packets() in presence of a hub >>>>> >>>>> Stefan >>>>> >>>> Yes, this commit fixes a bug but from other side. I think it's >>>> better to just let the qemu_send_packet() do the right things. >>>> >>>> E.g. network hub has 3 ports. Suppose when iterating through port >>>> list in net_hub_port_can_receive() a packet is successfully >>>> delivered to the first port, and then is queued in the source port >>>> queue because the second port cannot receive packets. Later >>>> net_hub_flush() will flush the packet from the source port queue and >>>> it will be delivered in every port. But it had been already >>>> delivered to one of them. So it will be delivered twice to some >>>> ports. Moreover there is less chance to dequeue the packet if >>>> several clients can't receive periodically. >>> Did you mean "second port" instead of "source port" in the beginning? >>> >>> I don't see a scenario where the packet is delivered to the same port >>> twice. If one port can receive then the packet will be sent and other >>> ports will queue the packet. >> Whether dump network client can always receive? If so, then the >> packet will always be sent to the dump client regardless of whether >> the other clients can receive. Is there actually any >> synchronization? > Yes, the dump net client can always receive. Other net clients may not. > > There is no explicit synchronization between send queues on the hub. > > Stefan > So why don't we apply this patch?
On Tue, Apr 23, 2013 at 11:41:42AM +0400, Fedorov Sergey wrote: > >Beyond that, we also want to avoid growing net queues indefinitely. If > >the hub does not implement .can_receive() then it relies on growing > >queues (keeping packets buffered in memory). > No, net_hub_receive() calls qemu_send_packet(). If the destination > queue cannot receive the packet qemu_net_queue_append() will take > care of queue->nq_maxlen. You are right, sorry. We do discard packets at nq_maxlen. The problem with ignoring .can_receive() on the hub is that it breaks flow control. For example, net/tap.c is designed to avoid reading more packets if its peer cannot receive (see tap_can_send()). If the hub claims it can always receive we waste cycles reading packets from the tap device only to discard them. Since qemu.git already has a fix which preserves flow control, I am not going to merge your patch. Stefan
On 04/23/2013 04:00 PM, Stefan Hajnoczi wrote: > On Tue, Apr 23, 2013 at 11:41:42AM +0400, Fedorov Sergey wrote: >>> Beyond that, we also want to avoid growing net queues indefinitely. If >>> the hub does not implement .can_receive() then it relies on growing >>> queues (keeping packets buffered in memory). >> No, net_hub_receive() calls qemu_send_packet(). If the destination >> queue cannot receive the packet qemu_net_queue_append() will take >> care of queue->nq_maxlen. > You are right, sorry. We do discard packets at nq_maxlen. > > The problem with ignoring .can_receive() on the hub is that it breaks > flow control. For example, net/tap.c is designed to avoid reading more > packets if its peer cannot receive (see tap_can_send()). > > If the hub claims it can always receive we waste cycles reading packets > from the tap device only to discard them. > > Since qemu.git already has a fix which preserves flow control, I am not > going to merge your patch. > > Stefan > OK, I see. Thanks for attention.
On 04/23/2013 04:00 PM, Stefan Hajnoczi wrote: > On Tue, Apr 23, 2013 at 11:41:42AM +0400, Fedorov Sergey wrote: >>> Beyond that, we also want to avoid growing net queues indefinitely. If >>> the hub does not implement .can_receive() then it relies on growing >>> queues (keeping packets buffered in memory). >> No, net_hub_receive() calls qemu_send_packet(). If the destination >> queue cannot receive the packet qemu_net_queue_append() will take >> care of queue->nq_maxlen. > You are right, sorry. We do discard packets at nq_maxlen. > > The problem with ignoring .can_receive() on the hub is that it breaks > flow control. For example, net/tap.c is designed to avoid reading more > packets if its peer cannot receive (see tap_can_send()). > > If the hub claims it can always receive we waste cycles reading packets > from the tap device only to discard them. > > Since qemu.git already has a fix which preserves flow control, I am not > going to merge your patch. > > Stefan > > Dear, Stefan Hajnoczi, After our discussion about this patch I decided to keep my patch in our branch until rebase onto a new release. Recently I have rebased our branch onto v1.5.3 and reverted my patch. Then I face an issue when using user-mode networking with USB network device for mounting root file system through NFS. Fragmented UDP packets from host to guest does not handled properly. Seems that some fragments is lost or somehow stalled. See guest tcpdump log below. 03:16:52.259690 IP (tos 0x0, ttl 64, id 0, offset 0, flags [DF], proto UDP (17), length 164) 10.0.2.15.3369105030 > 10.0.2.2.nfs: 136 readdirplus fh Unknown/0100070004001200000000002873593C9B3C43388E23748B0BAD870C00000000 512 bytes @ 0 max 4096 verf 0000000000000000 03:16:52.262323 IP (tos 0x0, ttl 64, id 16, offset 0, flags [+], proto UDP (17), length 1500) 10.0.2.2.nfs > 10.0.2.15.3369105030: reply ok 1472 readdirplus POST: DIR 40777 ids 0/0 sz 4096 verf 0000000000000000 03:16:52.264592 IP (tos 0x0, ttl 64, id 16, offset 1480, flags [+], proto UDP (17), length 1500) 10.0.2.2 > 10.0.2.15: udp 03:16:54.462961 IP (tos 0x0, ttl 64, id 0, offset 0, flags [DF], proto UDP (17), length 164) 10.0.2.15.3369105030 > 10.0.2.2.nfs: 136 readdirplus fh Unknown/0100070004001200000000002873593C9B3C43388E23748B0BAD870C00000000 512 bytes @ 0 max 4096 verf 0000000000000000 03:16:54.466300 IP (tos 0x0, ttl 64, id 17, offset 0, flags [+], proto UDP (17), length 1500) 10.0.2.2.nfs > 10.0.2.15.3369105030: reply ok 1472 readdirplus POST: DIR 40777 ids 0/0 sz 4096 verf 0000000000000000 03:16:54.467084 IP (tos 0x0, ttl 64, id 17, offset 1480, flags [+], proto UDP (17), length 1500) 10.0.2.2 > 10.0.2.15: udp ... I didn't investigate the cause of the problem in detail. I just reverted commit 199ee608f0d08510b5c6c37f31a7fbff211d63c4 Author: Luigi Rizzo <rizzo@iet.unipi.it> Date: Tue Feb 5 17:53:31 2013 +0100 net: fix qemu_flush_queued_packets() in presence of a hub And then applied my patch. After that everything works fine for me. See guest tcpdump log below. 04:45:15.897245 IP (tos 0x0, ttl 64, id 0, offset 0, flags [DF], proto UDP (17), length 164) 10.0.2.15.3642011847 > 10.0.2.2.nfs: 136 readdirplus fh Unknown/0100070004001200000000002873593C9B3C43388E23748B0BAD870C00000000 512 bytes @ 0 max 4096 verf 0000000000000000 04:45:15.899686 IP (tos 0x0, ttl 64, id 15, offset 0, flags [+], proto UDP (17), length 1500) 10.0.2.2.nfs > 10.0.2.15.3642011847: reply ok 1472 readdirplus POST: DIR 40777 ids 0/0 sz 4096 verf 0000000000000000 04:45:15.906253 IP (tos 0x0, ttl 64, id 15, offset 1480, flags [+], proto UDP (17), length 1500) 10.0.2.2 > 10.0.2.15: udp 04:45:15.906687 IP (tos 0x0, ttl 64, id 15, offset 2960, flags [none], proto UDP (17), length 240) 10.0.2.2 > 10.0.2.15: udp So there must be something wrong with already applied patch. What could you suggest?
On 10/21/2013 03:44 PM, Fedorov Sergey wrote: > On 04/23/2013 04:00 PM, Stefan Hajnoczi wrote: >> On Tue, Apr 23, 2013 at 11:41:42AM +0400, Fedorov Sergey wrote: >>>> Beyond that, we also want to avoid growing net queues >>>> indefinitely. If >>>> the hub does not implement .can_receive() then it relies on growing >>>> queues (keeping packets buffered in memory). >>> No, net_hub_receive() calls qemu_send_packet(). If the destination >>> queue cannot receive the packet qemu_net_queue_append() will take >>> care of queue->nq_maxlen. >> You are right, sorry. We do discard packets at nq_maxlen. >> >> The problem with ignoring .can_receive() on the hub is that it breaks >> flow control. For example, net/tap.c is designed to avoid reading more >> packets if its peer cannot receive (see tap_can_send()). >> >> If the hub claims it can always receive we waste cycles reading packets >> from the tap device only to discard them. >> >> Since qemu.git already has a fix which preserves flow control, I am not >> going to merge your patch. >> >> Stefan >> >> > > Dear, Stefan Hajnoczi, > > After our discussion about this patch I decided to keep my patch in > our branch until rebase onto a new release. Recently I have rebased > our branch onto v1.5.3 and reverted my patch. Then I face an issue > when using user-mode networking with USB network device for mounting > root file system through NFS. Fragmented UDP packets from host to > guest does not handled properly. Seems that some fragments is lost or > somehow stalled. See guest tcpdump log below. > > 03:16:52.259690 IP (tos 0x0, ttl 64, id 0, offset 0, flags [DF], proto > UDP (17), length 164) > 10.0.2.15.3369105030 > 10.0.2.2.nfs: 136 readdirplus fh > Unknown/0100070004001200000000002873593C9B3C43388E23748B0BAD870C00000000 > 512 bytes @ 0 max 4096 verf 0000000000000000 > 03:16:52.262323 IP (tos 0x0, ttl 64, id 16, offset 0, flags [+], proto > UDP (17), length 1500) > 10.0.2.2.nfs > 10.0.2.15.3369105030: reply ok 1472 readdirplus > POST: DIR 40777 ids 0/0 sz 4096 verf 0000000000000000 > 03:16:52.264592 IP (tos 0x0, ttl 64, id 16, offset 1480, flags [+], > proto UDP (17), length 1500) > 10.0.2.2 > 10.0.2.15: udp > 03:16:54.462961 IP (tos 0x0, ttl 64, id 0, offset 0, flags [DF], proto > UDP (17), length 164) > 10.0.2.15.3369105030 > 10.0.2.2.nfs: 136 readdirplus fh > Unknown/0100070004001200000000002873593C9B3C43388E23748B0BAD870C00000000 > 512 bytes @ 0 max 4096 verf 0000000000000000 > 03:16:54.466300 IP (tos 0x0, ttl 64, id 17, offset 0, flags [+], proto > UDP (17), length 1500) > 10.0.2.2.nfs > 10.0.2.15.3369105030: reply ok 1472 readdirplus > POST: DIR 40777 ids 0/0 sz 4096 verf 0000000000000000 > 03:16:54.467084 IP (tos 0x0, ttl 64, id 17, offset 1480, flags [+], > proto UDP (17), length 1500) > 10.0.2.2 > 10.0.2.15: udp > ... > > I didn't investigate the cause of the problem in detail. I just reverted > > commit 199ee608f0d08510b5c6c37f31a7fbff211d63c4 > Author: Luigi Rizzo <rizzo@iet.unipi.it> > Date: Tue Feb 5 17:53:31 2013 +0100 > > net: fix qemu_flush_queued_packets() in presence of a hub > > And then applied my patch. After that everything works fine for me. > See guest tcpdump log below. > > 04:45:15.897245 IP (tos 0x0, ttl 64, id 0, offset 0, flags [DF], proto > UDP (17), length 164) > 10.0.2.15.3642011847 > 10.0.2.2.nfs: 136 readdirplus fh > Unknown/0100070004001200000000002873593C9B3C43388E23748B0BAD870C00000000 > 512 bytes @ 0 max 4096 verf 0000000000000000 > 04:45:15.899686 IP (tos 0x0, ttl 64, id 15, offset 0, flags [+], proto > UDP (17), length 1500) > 10.0.2.2.nfs > 10.0.2.15.3642011847: reply ok 1472 readdirplus > POST: DIR 40777 ids 0/0 sz 4096 verf 0000000000000000 > 04:45:15.906253 IP (tos 0x0, ttl 64, id 15, offset 1480, flags [+], > proto UDP (17), length 1500) > 10.0.2.2 > 10.0.2.15: udp > 04:45:15.906687 IP (tos 0x0, ttl 64, id 15, offset 2960, flags [none], > proto UDP (17), length 240) > 10.0.2.2 > 10.0.2.15: udp > > So there must be something wrong with already applied patch. What > could you suggest? > Sorry, I missed that Anthony Liguori email address has been changed. So I resend the email.
On 10/21/2013 03:52 PM, Fedorov Sergey wrote: > > On 10/21/2013 03:44 PM, Fedorov Sergey wrote: >> On 04/23/2013 04:00 PM, Stefan Hajnoczi wrote: >>> On Tue, Apr 23, 2013 at 11:41:42AM +0400, Fedorov Sergey wrote: >>>>> Beyond that, we also want to avoid growing net queues >>>>> indefinitely. If >>>>> the hub does not implement .can_receive() then it relies on growing >>>>> queues (keeping packets buffered in memory). >>>> No, net_hub_receive() calls qemu_send_packet(). If the destination >>>> queue cannot receive the packet qemu_net_queue_append() will take >>>> care of queue->nq_maxlen. >>> You are right, sorry. We do discard packets at nq_maxlen. >>> >>> The problem with ignoring .can_receive() on the hub is that it breaks >>> flow control. For example, net/tap.c is designed to avoid reading more >>> packets if its peer cannot receive (see tap_can_send()). >>> >>> If the hub claims it can always receive we waste cycles reading packets >>> from the tap device only to discard them. >>> >>> Since qemu.git already has a fix which preserves flow control, I am not >>> going to merge your patch. >>> >>> Stefan >>> >>> >> >> Dear, Stefan Hajnoczi, >> >> After our discussion about this patch I decided to keep my patch in >> our branch until rebase onto a new release. Recently I have rebased >> our branch onto v1.5.3 and reverted my patch. Then I face an issue >> when using user-mode networking with USB network device for mounting >> root file system through NFS. Fragmented UDP packets from host to >> guest does not handled properly. Seems that some fragments is lost or >> somehow stalled. See guest tcpdump log below. >> >> 03:16:52.259690 IP (tos 0x0, ttl 64, id 0, offset 0, flags [DF], >> proto UDP (17), length 164) >> 10.0.2.15.3369105030 > 10.0.2.2.nfs: 136 readdirplus fh >> Unknown/0100070004001200000000002873593C9B3C43388E23748B0BAD870C00000000 >> 512 bytes @ 0 max 4096 verf 0000000000000000 >> 03:16:52.262323 IP (tos 0x0, ttl 64, id 16, offset 0, flags [+], >> proto UDP (17), length 1500) >> 10.0.2.2.nfs > 10.0.2.15.3369105030: reply ok 1472 readdirplus >> POST: DIR 40777 ids 0/0 sz 4096 verf 0000000000000000 >> 03:16:52.264592 IP (tos 0x0, ttl 64, id 16, offset 1480, flags [+], >> proto UDP (17), length 1500) >> 10.0.2.2 > 10.0.2.15: udp >> 03:16:54.462961 IP (tos 0x0, ttl 64, id 0, offset 0, flags [DF], >> proto UDP (17), length 164) >> 10.0.2.15.3369105030 > 10.0.2.2.nfs: 136 readdirplus fh >> Unknown/0100070004001200000000002873593C9B3C43388E23748B0BAD870C00000000 >> 512 bytes @ 0 max 4096 verf 0000000000000000 >> 03:16:54.466300 IP (tos 0x0, ttl 64, id 17, offset 0, flags [+], >> proto UDP (17), length 1500) >> 10.0.2.2.nfs > 10.0.2.15.3369105030: reply ok 1472 readdirplus >> POST: DIR 40777 ids 0/0 sz 4096 verf 0000000000000000 >> 03:16:54.467084 IP (tos 0x0, ttl 64, id 17, offset 1480, flags [+], >> proto UDP (17), length 1500) >> 10.0.2.2 > 10.0.2.15: udp >> ... >> >> I didn't investigate the cause of the problem in detail. I just reverted >> >> commit 199ee608f0d08510b5c6c37f31a7fbff211d63c4 >> Author: Luigi Rizzo <rizzo@iet.unipi.it> >> Date: Tue Feb 5 17:53:31 2013 +0100 >> >> net: fix qemu_flush_queued_packets() in presence of a hub >> >> And then applied my patch. After that everything works fine for me. >> See guest tcpdump log below. >> >> 04:45:15.897245 IP (tos 0x0, ttl 64, id 0, offset 0, flags [DF], >> proto UDP (17), length 164) >> 10.0.2.15.3642011847 > 10.0.2.2.nfs: 136 readdirplus fh >> Unknown/0100070004001200000000002873593C9B3C43388E23748B0BAD870C00000000 >> 512 bytes @ 0 max 4096 verf 0000000000000000 >> 04:45:15.899686 IP (tos 0x0, ttl 64, id 15, offset 0, flags [+], >> proto UDP (17), length 1500) >> 10.0.2.2.nfs > 10.0.2.15.3642011847: reply ok 1472 readdirplus >> POST: DIR 40777 ids 0/0 sz 4096 verf 0000000000000000 >> 04:45:15.906253 IP (tos 0x0, ttl 64, id 15, offset 1480, flags [+], >> proto UDP (17), length 1500) >> 10.0.2.2 > 10.0.2.15: udp >> 04:45:15.906687 IP (tos 0x0, ttl 64, id 15, offset 2960, flags >> [none], proto UDP (17), length 240) >> 10.0.2.2 > 10.0.2.15: udp >> >> So there must be something wrong with already applied patch. What >> could you suggest? >> > > Sorry, I missed that Anthony Liguori email address has been changed. > So I resend the email. > Ping.
On Mon, Oct 21, 2013 at 03:44:46PM +0400, Fedorov Sergey wrote: > After our discussion about this patch I decided to keep my patch in > our branch until rebase onto a new release. Recently I have rebased > our branch onto v1.5.3 and reverted my patch. Then I face an issue > when using user-mode networking with USB network device for mounting > root file system through NFS. Fragmented UDP packets from host to > guest does not handled properly. Seems that some fragments is lost > or somehow stalled. See guest tcpdump log below. > > 03:16:52.259690 IP (tos 0x0, ttl 64, id 0, offset 0, flags [DF], > proto UDP (17), length 164) > 10.0.2.15.3369105030 > 10.0.2.2.nfs: 136 readdirplus fh Unknown/0100070004001200000000002873593C9B3C43388E23748B0BAD870C00000000 > 512 bytes @ 0 max 4096 verf 0000000000000000 > 03:16:52.262323 IP (tos 0x0, ttl 64, id 16, offset 0, flags [+], > proto UDP (17), length 1500) > 10.0.2.2.nfs > 10.0.2.15.3369105030: reply ok 1472 readdirplus > POST: DIR 40777 ids 0/0 sz 4096 verf 0000000000000000 > 03:16:52.264592 IP (tos 0x0, ttl 64, id 16, offset 1480, flags [+], > proto UDP (17), length 1500) > 10.0.2.2 > 10.0.2.15: udp > 03:16:54.462961 IP (tos 0x0, ttl 64, id 0, offset 0, flags [DF], > proto UDP (17), length 164) > 10.0.2.15.3369105030 > 10.0.2.2.nfs: 136 readdirplus fh Unknown/0100070004001200000000002873593C9B3C43388E23748B0BAD870C00000000 > 512 bytes @ 0 max 4096 verf 0000000000000000 > 03:16:54.466300 IP (tos 0x0, ttl 64, id 17, offset 0, flags [+], > proto UDP (17), length 1500) > 10.0.2.2.nfs > 10.0.2.15.3369105030: reply ok 1472 readdirplus > POST: DIR 40777 ids 0/0 sz 4096 verf 0000000000000000 > 03:16:54.467084 IP (tos 0x0, ttl 64, id 17, offset 1480, flags [+], > proto UDP (17), length 1500) > 10.0.2.2 > 10.0.2.15: udp > ... > > I didn't investigate the cause of the problem in detail. I just reverted > > commit 199ee608f0d08510b5c6c37f31a7fbff211d63c4 > Author: Luigi Rizzo <rizzo@iet.unipi.it> > Date: Tue Feb 5 17:53:31 2013 +0100 > > net: fix qemu_flush_queued_packets() in presence of a hub > > And then applied my patch. After that everything works fine for me. > See guest tcpdump log below. > > 04:45:15.897245 IP (tos 0x0, ttl 64, id 0, offset 0, flags [DF], > proto UDP (17), length 164) > 10.0.2.15.3642011847 > 10.0.2.2.nfs: 136 readdirplus fh Unknown/0100070004001200000000002873593C9B3C43388E23748B0BAD870C00000000 > 512 bytes @ 0 max 4096 verf 0000000000000000 > 04:45:15.899686 IP (tos 0x0, ttl 64, id 15, offset 0, flags [+], > proto UDP (17), length 1500) > 10.0.2.2.nfs > 10.0.2.15.3642011847: reply ok 1472 readdirplus > POST: DIR 40777 ids 0/0 sz 4096 verf 0000000000000000 > 04:45:15.906253 IP (tos 0x0, ttl 64, id 15, offset 1480, flags [+], > proto UDP (17), length 1500) > 10.0.2.2 > 10.0.2.15: udp > 04:45:15.906687 IP (tos 0x0, ttl 64, id 15, offset 2960, flags > [none], proto UDP (17), length 240) > 10.0.2.2 > 10.0.2.15: udp > > So there must be something wrong with already applied patch. What > could you suggest? The next step is to investigate the cause. Perhaps hw/usb/dev-network.c:usb_net_handle_datain() is not calling qemu_flush_queued_packets() every time in_buf[] is read completely. This if statement looks strange to me: if (s->in_ptr >= s->in_len && (is_rndis(s) || (s->in_len & (64 - 1)) || !len)) { /* no short packet necessary */ usb_net_reset_in_buf(s); } Try placing printfs to find out whether qemu_flush_queued_packets() is getting called when you see packet loss. Stefan
On 10/29/2013 06:55 PM, Stefan Hajnoczi wrote: > On Mon, Oct 21, 2013 at 03:44:46PM +0400, Fedorov Sergey wrote: >> After our discussion about this patch I decided to keep my patch in >> our branch until rebase onto a new release. Recently I have rebased >> our branch onto v1.5.3 and reverted my patch. Then I face an issue >> when using user-mode networking with USB network device for mounting >> root file system through NFS. Fragmented UDP packets from host to >> guest does not handled properly. Seems that some fragments is lost >> or somehow stalled. See guest tcpdump log below. >> >> 03:16:52.259690 IP (tos 0x0, ttl 64, id 0, offset 0, flags [DF], >> proto UDP (17), length 164) >> 10.0.2.15.3369105030 > 10.0.2.2.nfs: 136 readdirplus fh Unknown/0100070004001200000000002873593C9B3C43388E23748B0BAD870C00000000 >> 512 bytes @ 0 max 4096 verf 0000000000000000 >> 03:16:52.262323 IP (tos 0x0, ttl 64, id 16, offset 0, flags [+], >> proto UDP (17), length 1500) >> 10.0.2.2.nfs > 10.0.2.15.3369105030: reply ok 1472 readdirplus >> POST: DIR 40777 ids 0/0 sz 4096 verf 0000000000000000 >> 03:16:52.264592 IP (tos 0x0, ttl 64, id 16, offset 1480, flags [+], >> proto UDP (17), length 1500) >> 10.0.2.2 > 10.0.2.15: udp >> 03:16:54.462961 IP (tos 0x0, ttl 64, id 0, offset 0, flags [DF], >> proto UDP (17), length 164) >> 10.0.2.15.3369105030 > 10.0.2.2.nfs: 136 readdirplus fh Unknown/0100070004001200000000002873593C9B3C43388E23748B0BAD870C00000000 >> 512 bytes @ 0 max 4096 verf 0000000000000000 >> 03:16:54.466300 IP (tos 0x0, ttl 64, id 17, offset 0, flags [+], >> proto UDP (17), length 1500) >> 10.0.2.2.nfs > 10.0.2.15.3369105030: reply ok 1472 readdirplus >> POST: DIR 40777 ids 0/0 sz 4096 verf 0000000000000000 >> 03:16:54.467084 IP (tos 0x0, ttl 64, id 17, offset 1480, flags [+], >> proto UDP (17), length 1500) >> 10.0.2.2 > 10.0.2.15: udp >> ... >> >> I didn't investigate the cause of the problem in detail. I just reverted >> >> commit 199ee608f0d08510b5c6c37f31a7fbff211d63c4 >> Author: Luigi Rizzo <rizzo@iet.unipi.it> >> Date: Tue Feb 5 17:53:31 2013 +0100 >> >> net: fix qemu_flush_queued_packets() in presence of a hub >> >> And then applied my patch. After that everything works fine for me. >> See guest tcpdump log below. >> >> 04:45:15.897245 IP (tos 0x0, ttl 64, id 0, offset 0, flags [DF], >> proto UDP (17), length 164) >> 10.0.2.15.3642011847 > 10.0.2.2.nfs: 136 readdirplus fh Unknown/0100070004001200000000002873593C9B3C43388E23748B0BAD870C00000000 >> 512 bytes @ 0 max 4096 verf 0000000000000000 >> 04:45:15.899686 IP (tos 0x0, ttl 64, id 15, offset 0, flags [+], >> proto UDP (17), length 1500) >> 10.0.2.2.nfs > 10.0.2.15.3642011847: reply ok 1472 readdirplus >> POST: DIR 40777 ids 0/0 sz 4096 verf 0000000000000000 >> 04:45:15.906253 IP (tos 0x0, ttl 64, id 15, offset 1480, flags [+], >> proto UDP (17), length 1500) >> 10.0.2.2 > 10.0.2.15: udp >> 04:45:15.906687 IP (tos 0x0, ttl 64, id 15, offset 2960, flags >> [none], proto UDP (17), length 240) >> 10.0.2.2 > 10.0.2.15: udp >> >> So there must be something wrong with already applied patch. What >> could you suggest? > The next step is to investigate the cause. > > Perhaps hw/usb/dev-network.c:usb_net_handle_datain() is not calling > qemu_flush_queued_packets() every time in_buf[] is read completely. > This if statement looks strange to me: > > if (s->in_ptr >= s->in_len && > (is_rndis(s) || (s->in_len & (64 - 1)) || !len)) { > /* no short packet necessary */ > usb_net_reset_in_buf(s); > } > > Try placing printfs to find out whether qemu_flush_queued_packets() is > getting called when you see packet loss. > > Stefan > Seems that I have figured out the problem. net_hub_flush() does not flush source port. And qemu_flush_queued_packets() also returns after calling net_hub_flush(). So I think the problem is that neither qemu_flush_queued_packets() nor net_hub_flush() call qemu_net_queue_flush() for the source port. So I think it sould be fixed in qemu_flush_queued_packets() by removing the return statement after calling net_hub_flush(). That fix does work for me. So I could submit that patch after getting permission for that.
On Wed, Oct 30, 2013 at 02:29:00PM +0400, Fedorov Sergey wrote: > On 10/29/2013 06:55 PM, Stefan Hajnoczi wrote: > >On Mon, Oct 21, 2013 at 03:44:46PM +0400, Fedorov Sergey wrote: > >>After our discussion about this patch I decided to keep my patch in > >>our branch until rebase onto a new release. Recently I have rebased > >>our branch onto v1.5.3 and reverted my patch. Then I face an issue > >>when using user-mode networking with USB network device for mounting > >>root file system through NFS. Fragmented UDP packets from host to > >>guest does not handled properly. Seems that some fragments is lost > >>or somehow stalled. See guest tcpdump log below. > >> > >>03:16:52.259690 IP (tos 0x0, ttl 64, id 0, offset 0, flags [DF], > >>proto UDP (17), length 164) > >> 10.0.2.15.3369105030 > 10.0.2.2.nfs: 136 readdirplus fh Unknown/0100070004001200000000002873593C9B3C43388E23748B0BAD870C00000000 > >>512 bytes @ 0 max 4096 verf 0000000000000000 > >>03:16:52.262323 IP (tos 0x0, ttl 64, id 16, offset 0, flags [+], > >>proto UDP (17), length 1500) > >> 10.0.2.2.nfs > 10.0.2.15.3369105030: reply ok 1472 readdirplus > >>POST: DIR 40777 ids 0/0 sz 4096 verf 0000000000000000 > >>03:16:52.264592 IP (tos 0x0, ttl 64, id 16, offset 1480, flags [+], > >>proto UDP (17), length 1500) > >> 10.0.2.2 > 10.0.2.15: udp > >>03:16:54.462961 IP (tos 0x0, ttl 64, id 0, offset 0, flags [DF], > >>proto UDP (17), length 164) > >> 10.0.2.15.3369105030 > 10.0.2.2.nfs: 136 readdirplus fh Unknown/0100070004001200000000002873593C9B3C43388E23748B0BAD870C00000000 > >>512 bytes @ 0 max 4096 verf 0000000000000000 > >>03:16:54.466300 IP (tos 0x0, ttl 64, id 17, offset 0, flags [+], > >>proto UDP (17), length 1500) > >> 10.0.2.2.nfs > 10.0.2.15.3369105030: reply ok 1472 readdirplus > >>POST: DIR 40777 ids 0/0 sz 4096 verf 0000000000000000 > >>03:16:54.467084 IP (tos 0x0, ttl 64, id 17, offset 1480, flags [+], > >>proto UDP (17), length 1500) > >> 10.0.2.2 > 10.0.2.15: udp > >>... > >> > >>I didn't investigate the cause of the problem in detail. I just reverted > >> > >>commit 199ee608f0d08510b5c6c37f31a7fbff211d63c4 > >>Author: Luigi Rizzo <rizzo@iet.unipi.it> > >>Date: Tue Feb 5 17:53:31 2013 +0100 > >> > >> net: fix qemu_flush_queued_packets() in presence of a hub > >> > >>And then applied my patch. After that everything works fine for me. > >>See guest tcpdump log below. > >> > >>04:45:15.897245 IP (tos 0x0, ttl 64, id 0, offset 0, flags [DF], > >>proto UDP (17), length 164) > >> 10.0.2.15.3642011847 > 10.0.2.2.nfs: 136 readdirplus fh Unknown/0100070004001200000000002873593C9B3C43388E23748B0BAD870C00000000 > >>512 bytes @ 0 max 4096 verf 0000000000000000 > >>04:45:15.899686 IP (tos 0x0, ttl 64, id 15, offset 0, flags [+], > >>proto UDP (17), length 1500) > >> 10.0.2.2.nfs > 10.0.2.15.3642011847: reply ok 1472 readdirplus > >>POST: DIR 40777 ids 0/0 sz 4096 verf 0000000000000000 > >>04:45:15.906253 IP (tos 0x0, ttl 64, id 15, offset 1480, flags [+], > >>proto UDP (17), length 1500) > >> 10.0.2.2 > 10.0.2.15: udp > >>04:45:15.906687 IP (tos 0x0, ttl 64, id 15, offset 2960, flags > >>[none], proto UDP (17), length 240) > >> 10.0.2.2 > 10.0.2.15: udp > >> > >>So there must be something wrong with already applied patch. What > >>could you suggest? > >The next step is to investigate the cause. > > > >Perhaps hw/usb/dev-network.c:usb_net_handle_datain() is not calling > >qemu_flush_queued_packets() every time in_buf[] is read completely. > >This if statement looks strange to me: > > > >if (s->in_ptr >= s->in_len && > > (is_rndis(s) || (s->in_len & (64 - 1)) || !len)) { > > /* no short packet necessary */ > > usb_net_reset_in_buf(s); > >} > > > >Try placing printfs to find out whether qemu_flush_queued_packets() is > >getting called when you see packet loss. > > > >Stefan > > > > Seems that I have figured out the problem. net_hub_flush() does not > flush source port. And qemu_flush_queued_packets() also returns > after calling net_hub_flush(). So I think the problem is that > neither qemu_flush_queued_packets() nor net_hub_flush() call > qemu_net_queue_flush() for the source port. So I think it sould be > fixed in qemu_flush_queued_packets() by removing the return > statement after calling net_hub_flush(). That fix does work for me. > So I could submit that patch after getting permission for that. Sounds good to me. I have CCed Luigi in case he wants to comment. Stefan
diff --git a/net/hub.c b/net/hub.c index df32074..552e970 100644 --- a/net/hub.c +++ b/net/hub.c @@ -87,25 +87,6 @@ static NetHub *net_hub_new(int id) return hub; } -static int net_hub_port_can_receive(NetClientState *nc) -{ - NetHubPort *port; - NetHubPort *src_port = DO_UPCAST(NetHubPort, nc, nc); - NetHub *hub = src_port->hub; - - QLIST_FOREACH(port, &hub->ports, next) { - if (port == src_port) { - continue; - } - - if (qemu_can_send_packet(&port->nc)) { - return 1; - } - } - - return 0; -} - static ssize_t net_hub_port_receive(NetClientState *nc, const uint8_t *buf, size_t len) { @@ -132,7 +113,6 @@ static void net_hub_port_cleanup(NetClientState *nc) static NetClientInfo net_hub_port_info = { .type = NET_CLIENT_OPTIONS_KIND_HUBPORT, .size = sizeof(NetHubPort), - .can_receive = net_hub_port_can_receive, .receive = net_hub_port_receive, .receive_iov = net_hub_port_receive_iov, .cleanup = net_hub_port_cleanup,
Network hub should always receive incoming packets. Then forward them to the appropriate port queue and let the qemu_send_packet() do the right things. If the destination queue cannot receive the packet it will be appended to the queue. When the receiver call qemu_flush_queued_packets() later the queue will be really flushed and no packets will be stalled in the sender network queue. Signed-off-by: Sergey Fedorov <s.fedorov@samsung.com> --- net/hub.c | 20 -------------------- 1 file changed, 20 deletions(-)