Re: [Qemu-devel] [PATCH 2/2] [SLIRP] Delayed IP packets
On 27/07/2011 11:30, Jan Kiszka wrote: On 2011-07-26 18:21, Fabien Chouteau wrote: In the current implementation, if Slirp tries to send an IP packet to a client with an unknown hardware address, the packet is simply dropped and an ARP request is sent (if_encap in slirp/slirp.c). This patch adds a list of delayed IP packets to handle such cases. If the hardware address is unknown, Slirp inserts the packet in delayed list and sends an ARP request. Each time the ARP table is updated Slirp retries to send the packet. Haven't looked at details yet, just two general thoughts so far: We already have queues for outgoing packets, why can't we reuse that infrastructure? That would also avoid additional memory allocations. Delayed packets should be requeued at the end and only one attempt to send them should be performed per queue flush. Sure, I didn't know about these queues. Where are they implemented? I think we need some timeout for the delayed packets. If invalid or non-reactive clients are addressed, we will otherwise pile them up until qemu terminates, right? Yes that's a good idea. Regards, -- Fabien Chouteau
Re: [Qemu-devel] [PATCH 2/2] [SLIRP] Delayed IP packets
On 2011-07-27 12:14, Fabien Chouteau wrote: On 27/07/2011 11:30, Jan Kiszka wrote: On 2011-07-26 18:21, Fabien Chouteau wrote: In the current implementation, if Slirp tries to send an IP packet to a client with an unknown hardware address, the packet is simply dropped and an ARP request is sent (if_encap in slirp/slirp.c). This patch adds a list of delayed IP packets to handle such cases. If the hardware address is unknown, Slirp inserts the packet in delayed list and sends an ARP request. Each time the ARP table is updated Slirp retries to send the packet. Haven't looked at details yet, just two general thoughts so far: We already have queues for outgoing packets, why can't we reuse that infrastructure? That would also avoid additional memory allocations. Delayed packets should be requeued at the end and only one attempt to send them should be performed per queue flush. Sure, I didn't know about these queues. Where are they implemented? Check e.g. what happens in and is documented for if_start(). Jan -- Siemens AG, Corporate Technology, CT T DE IT 1 Corporate Competence Center Embedded Linux
