From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1751847AbaHAEcY (ORCPT ); Fri, 1 Aug 2014 00:32:24 -0400 Received: from shards.monkeyblade.net ([149.20.54.216]:53967 "EHLO shards.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1750770AbaHAEcW (ORCPT ); Fri, 1 Aug 2014 00:32:22 -0400 Date: Thu, 31 Jul 2014 21:32:20 -0700 (PDT) Message-Id: <20140731.213220.2290285957519220651.davem@davemloft.net> To: zoltan.kiss@citrix.com Cc: steffen.klassert@secunet.com, minipli@googlemail.com, dborkman@redhat.com, tgraf@suug.ch, joe@perches.com, netdev@vger.kernel.org, linux-kernel@vger.kernel.org, xen-devel@lists.xenproject.org Subject: Re: [PATCH net-next v3 4/4 RFC] pktgen: Allow sending IPv4 TCP packets From: David Miller In-Reply-To: <1406737212-23351-5-git-send-email-zoltan.kiss@citrix.com> References: <1406737212-23351-1-git-send-email-zoltan.kiss@citrix.com> <1406737212-23351-5-git-send-email-zoltan.kiss@citrix.com> X-Mailer: Mew version 6.5 on Emacs 24.1 / Mule 6.0 (HANACHIRUSATO) Mime-Version: 1.0 Content-Type: Text/Plain; charset=us-ascii Content-Transfer-Encoding: 7bit X-Greylist: Sender succeeded SMTP AUTH, not delayed by milter-greylist-4.5.7 (shards.monkeyblade.net [149.20.54.216]); Thu, 31 Jul 2014 21:32:21 -0700 (PDT) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org From: Zoltan Kiss Date: Wed, 30 Jul 2014 17:20:12 +0100 > This is a prototype patch to enable sending IPv4 TCP packets with pktgen. The > original motivation is to test TCP GSO with xen-netback/netfront, but I'm not > sure about how the checksum should be set up, and also someone should verify the > GSO settings I'm using. > > Signed-off-by: Zoltan Kiss > Cc: "David S. Miller" > Cc: Thomas Graf > Cc: Joe Perches > Cc: netdev@vger.kernel.org > Cc: linux-kernel@vger.kernel.org > Cc: xen-devel@lists.xenproject.org > --- > v3: > - mention explicitly that this for IPv4 > - memset the TCP header and set up doff > - rework of checksum handling and GSO setting in fill_packet_ipv4 > - bail out in pktgen_xmit if the device won't be able to handle GSO > > diff --git a/net/core/pktgen.c b/net/core/pktgen.c > index 0d0aaac..9d93bda 100644 > --- a/net/core/pktgen.c > +++ b/net/core/pktgen.c > @@ -162,6 +162,7 @@ > #include > #include > #include > +#include > #include > #include > #ifdef CONFIG_XFRM > @@ -203,6 +204,7 @@ > #define F_NODE (1<<15) /* Node memory alloc*/ > #define F_UDPCSUM (1<<16) /* Include UDP checksum */ > #define F_PATTERN (1<<17) /* Fill the payload with a pattern */ > +#define F_TCP (1<<18) /* Send TCP packet instead of UDP */ > > /* Thread control flag bits */ > #define T_STOP (1<<0) /* Stop run */ > @@ -664,6 +666,9 @@ static int pktgen_if_show(struct seq_file *seq, void *v) > if (pkt_dev->flags & F_PATTERN) > seq_puts(seq, "PATTERN "); > > + if (pkt_dev->flags & F_TCP) > + seq_puts(seq, "TCP "); > + > if (pkt_dev->flags & F_MPLS_RND) > seq_puts(seq, "MPLS_RND "); > > @@ -1342,6 +1347,12 @@ static ssize_t pktgen_if_write(struct file *file, > else if (strcmp(f, "!PATTERN") == 0) > pkt_dev->flags &= ~F_PATTERN; > > + else if (strcmp(f, "TCP") == 0) > + pkt_dev->flags |= F_TCP; > + > + else if (strcmp(f, "!TCP") == 0) > + pkt_dev->flags &= ~F_TCP; > + > else { > sprintf(pg_result, > "Flag -:%s:- unknown\nAvailable flags, (prepend ! to un-set flag):\n%s", > @@ -2955,7 +2966,8 @@ static struct sk_buff *fill_packet_ipv4(struct net_device *odev, > { > struct sk_buff *skb = NULL; > __u8 *eth; > - struct udphdr *udph; > + struct udphdr *udph = NULL; > + struct tcphdr *tcph; > int datalen, iplen; > struct iphdr *iph; > __be16 protocol = htons(ETH_P_IP); > @@ -3017,29 +3029,40 @@ static struct sk_buff *fill_packet_ipv4(struct net_device *odev, > iph = (struct iphdr *) skb_put(skb, sizeof(struct iphdr)); > > skb_set_transport_header(skb, skb->len); > - udph = (struct udphdr *) skb_put(skb, sizeof(struct udphdr)); > + > + if (pkt_dev->flags & F_TCP) { > + datalen = pkt_dev->cur_pkt_size - ETH_HLEN - 20 - > + sizeof(struct tcphdr) - pkt_dev->pkt_overhead; > + tcph = (struct tcphdr *)skb_put(skb, sizeof(struct tcphdr)); > + memset(tcph, 0, sizeof(*tcph)); > + tcph->source = htons(pkt_dev->cur_udp_src); > + tcph->dest = htons(pkt_dev->cur_udp_dst); > + tcph->doff = sizeof(struct tcphdr) >> 2; > + } else { > + datalen = pkt_dev->cur_pkt_size - ETH_HLEN - 20 - > + sizeof(struct udphdr) - pkt_dev->pkt_overhead; > + udph = (struct udphdr *)skb_put(skb, sizeof(struct udphdr)); > + udph->source = htons(pkt_dev->cur_udp_src); > + udph->dest = htons(pkt_dev->cur_udp_dst); > + udph->len = htons(datalen + sizeof(struct udphdr)); > + udph->check = 0; > + } > + As more protocols (SCTP, etc.) get supported, this is going to become completely unmanageable. Please use callbacks or something like that so this function doesn't turn into even more spaghetti. > + } else if (pkt_dev->flags & F_TCP) { > + struct inet_sock inet; > + > + inet.inet_saddr = iph->saddr; > + inet.inet_daddr = iph->daddr; > + skb->ip_summed = CHECKSUM_NONE; > + tcp_v4_send_check((struct sock *)&inet, skb); Please don't do things like this. Making fake sockets on the stack, don't do it. Do other non-socket contexts compute TCP checksums this way? Check netfilter or similar, see what they do. Worst case export __tcp_v4_send_check() or just duplicate it's contents in the tcp case here.