Hi Sven, Yes even i saw netlink_unicast is called for the packet after building the packet, but in this case what if the socket receive buffer is full. are the packets which cannot be written onto the netlink socket receive buffer, because the buffer is full kept in the ip_queue's queue or are they discarded, and if they are discarded what is the significance of ip_queue_maxlen field ? (/proc/sys/net/ipv4/ip_queue_maxlen) ? Amit Sven Schuster <schuster.sven@xxxxxx> on 05/26/2004 01:07:47 PM To: Amit Kumar Singh/HSS@HSS cc: netfilter@xxxxxxxxxxxxxxxxxxx Subject: Re: ip_queue.o and libipq Hi Amit, On Wed, May 26, 2004 at 10:45:23AM +0530, aksingh@xxxxxxxxxxx told us: > > Hi, > > Does anybody have an idea which function of ip_queue.c is invoked when > a userland process does a ipq_read ? ipqueue works via netlink. When a packet is queued ipq_enqueue_packet is called with the skb as one of its arguments. This function then prepares the netlink packet and sends it to userspace via netlink_unicast. netlink_unicast itself will enqueue the netlink packet on the socket receive queue of the socket (see net/netlink/af_netlink.c:: netlink_sendskb which is called from netlink_unicast). When userspace (ipq_read) reads from the netlink socket, when there's data on this socket receive queue the first packet will be dequeued and returned to userspace. (I hope this explanation is understandable, I'm still in the learning process with regard to how the kernel works, so should I be wrong anybody please correct me :) > Also, once a NF_QUEUE is returned by a netfilter hook and till the packet > is copied to userland by ipq_read how many copies are involved ?. > > I believe there is just one copy when the packet is copied from sk_buff in > the queue maintained by ip_queue is copied to the buffer supplied by > ipq_read ? or are there more involved ? > > also --- is netfilter server down ? havent received any mails since the > last two days, there seems to be no problem at my mail server ? Seems like it's down, I didn't receive mails since may 23rd... > > thanks > Amit > > "DISCLAIMER: This message is proprietary to Hughes Software Systems Limited Do you think such a disclaimer makes sense when posting to a mailing list?? Sven -- Linux zion 2.6.6-tcp #4 Thu May 13 20:39:05 CEST 2004 i686 athlon i386 GNU/Linux 09:27:29 up 12 days, 11:30, 1 user, load average: 0.10, 0.03, 0.01
Attachment:
C.DTF
Description: Binary data
