[TCP]: rare bad TCP checksum with 2.6.19
[linux-2.6.git] / net / ipv4 / ip_output.c
index c72fc87..f071f84 100644 (file)
@@ -53,7 +53,7 @@
 #include <linux/mm.h>
 #include <linux/string.h>
 #include <linux/errno.h>
-#include <linux/config.h>
+#include <linux/highmem.h>
 
 #include <linux/socket.h>
 #include <linux/sockios.h>
@@ -69,6 +69,7 @@
 #include <net/ip.h>
 #include <net/protocol.h>
 #include <net/route.h>
+#include <net/xfrm.h>
 #include <linux/skbuff.h>
 #include <net/sock.h>
 #include <net/arp.h>
 #include <linux/netlink.h>
 #include <linux/tcp.h>
 
-/*
- *      Shall we try to damage output packets if routing dev changes?
- */
-
-int sysctl_ip_dynaddr;
-int sysctl_ip_default_ttl = IPDEFTTL;
+int sysctl_ip_default_ttl __read_mostly = IPDEFTTL;
 
 /* Generate a checksum for an outgoing IP datagram. */
 __inline__ void ip_send_check(struct iphdr *iph)
@@ -123,7 +119,7 @@ static inline int ip_select_ttl(struct inet_sock *inet, struct dst_entry *dst)
  *
  */
 int ip_build_and_send_pkt(struct sk_buff *skb, struct sock *sk,
-                         u32 saddr, u32 daddr, struct ip_options *opt)
+                         __be32 saddr, __be32 daddr, struct ip_options *opt)
 {
        struct inet_sock *inet = inet_sk(sk);
        struct rtable *rt = (struct rtable *)skb->dst;
@@ -163,10 +159,11 @@ int ip_build_and_send_pkt(struct sk_buff *skb, struct sock *sk,
                       dst_output);
 }
 
+EXPORT_SYMBOL_GPL(ip_build_and_send_pkt);
+
 static inline int ip_finish_output2(struct sk_buff *skb)
 {
        struct dst_entry *dst = skb->dst;
-       struct hh_cache *hh = dst->hh;
        struct net_device *dev = dst->dev;
        int hh_len = LL_RESERVED_SPACE(dev);
 
@@ -185,16 +182,9 @@ static inline int ip_finish_output2(struct sk_buff *skb)
                skb = skb2;
        }
 
-       if (hh) {
-               int hh_alen;
-
-               read_lock_bh(&hh->hh_lock);
-               hh_alen = HH_DATA_ALIGN(hh->hh_len);
-               memcpy(skb->data - hh_alen, hh->hh_data, hh_alen);
-               read_unlock_bh(&hh->hh_lock);
-               skb_push(skb, hh->hh_len);
-               return hh->hh_output(skb);
-       } else if (dst->neighbour)
+       if (dst->hh)
+               return neigh_hh_output(dst->hh, skb);
+       else if (dst->neighbour)
                return dst->neighbour->output(skb);
 
        if (net_ratelimit())
@@ -203,15 +193,19 @@ static inline int ip_finish_output2(struct sk_buff *skb)
        return -EINVAL;
 }
 
-static int ip_finish_output(struct sk_buff *skb)
+static inline int ip_finish_output(struct sk_buff *skb)
 {
-       struct net_device *dev = skb->dst->dev;
-
-       skb->dev = dev;
-       skb->protocol = htons(ETH_P_IP);
-
-       return NF_HOOK(PF_INET, NF_IP_POST_ROUTING, skb, NULL, dev,
-                      ip_finish_output2);
+#if defined(CONFIG_NETFILTER) && defined(CONFIG_XFRM)
+       /* Policy lookup after SNAT yielded a new policy */
+       if (skb->dst->xfrm != NULL) {
+               IPCB(skb)->flags |= IPSKB_REROUTED;
+               return dst_output(skb);
+       }
+#endif
+       if (skb->len > dst_mtu(skb->dst) && !skb_is_gso(skb))
+               return ip_fragment(skb, ip_finish_output2);
+       else
+               return ip_finish_output2(skb);
 }
 
 int ip_mc_output(struct sk_buff *skb)
@@ -268,25 +262,27 @@ int ip_mc_output(struct sk_buff *skb)
                                newskb->dev, ip_dev_loopback_xmit);
        }
 
-       if (skb->len > dst_mtu(&rt->u.dst))
-               return ip_fragment(skb, ip_finish_output);
-       else
-               return ip_finish_output(skb);
+       return NF_HOOK_COND(PF_INET, NF_IP_POST_ROUTING, skb, NULL, skb->dev,
+                           ip_finish_output,
+                           !(IPCB(skb)->flags & IPSKB_REROUTED));
 }
 
 int ip_output(struct sk_buff *skb)
 {
+       struct net_device *dev = skb->dst->dev;
+
        IP_INC_STATS(IPSTATS_MIB_OUTREQUESTS);
 
-       if (skb->len > dst_mtu(skb->dst) && !skb_shinfo(skb)->tso_size)
-               return ip_fragment(skb, ip_finish_output);
-       else
-               return ip_finish_output(skb);
+       skb->dev = dev;
+       skb->protocol = htons(ETH_P_IP);
+
+       return NF_HOOK_COND(PF_INET, NF_IP_POST_ROUTING, skb, NULL, dev,
+                           ip_finish_output,
+                           !(IPCB(skb)->flags & IPSKB_REROUTED));
 }
 
-int ip_queue_xmit(struct sk_buff *skb, int ipfragok)
+int ip_queue_xmit(struct sk_buff *skb, struct sock *sk, int ipfragok)
 {
-       struct sock *sk = skb->sk;
        struct inet_sock *inet = inet_sk(sk);
        struct ip_options *opt = inet->opt;
        struct rtable *rt;
@@ -302,7 +298,7 @@ int ip_queue_xmit(struct sk_buff *skb, int ipfragok)
        /* Make sure we can route this packet. */
        rt = (struct rtable *)__sk_dst_check(sk, 0);
        if (rt == NULL) {
-               u32 daddr;
+               __be32 daddr;
 
                /* Use correct destination address if we have options. */
                daddr = inet->daddr;
@@ -324,6 +320,7 @@ int ip_queue_xmit(struct sk_buff *skb, int ipfragok)
                         * keep trying until route appears or the connection times
                         * itself out.
                         */
+                       security_sk_classify_flow(sk, &fl);
                        if (ip_route_output_flow(&rt, &fl, sk, 0))
                                goto no_route;
                }
@@ -337,7 +334,7 @@ packet_routed:
 
        /* OK, we know where to send it, allocate and build IP header. */
        iph = (struct iphdr *) skb_push(skb, sizeof(struct iphdr) + (opt ? opt->optlen : 0));
-       *((__u16 *)iph) = htons((4 << 12) | (5 << 8) | (inet->tos & 0xff));
+       *((__be16 *)iph) = htons((4 << 12) | (5 << 8) | (inet->tos & 0xff));
        iph->tot_len = htons(skb->len);
        if (ip_dont_fragment(sk, &rt->u.dst) && !ipfragok)
                iph->frag_off = htons(IP_DF);
@@ -355,7 +352,8 @@ packet_routed:
                ip_options_build(skb, opt, inet->daddr, rt, 0);
        }
 
-       ip_select_ident_more(iph, &rt->u.dst, sk, skb_shinfo(skb)->tso_segs);
+       ip_select_ident_more(iph, &rt->u.dst, sk,
+                            (skb_shinfo(skb)->gso_segs ?: 1) - 1);
 
        /* Add an IP checksum. */
        ip_send_check(iph);
@@ -380,6 +378,7 @@ static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from)
        dst_release(to->dst);
        to->dst = dst_clone(from->dst);
        to->dev = from->dev;
+       to->mark = from->mark;
 
        /* Copy the flags to each fragment. */
        IPCB(to)->flags = IPCB(from)->flags;
@@ -388,18 +387,21 @@ static void ip_copy_metadata(struct sk_buff *to, struct sk_buff *from)
        to->tc_index = from->tc_index;
 #endif
 #ifdef CONFIG_NETFILTER
-       to->nfmark = from->nfmark;
        /* Connection association is same as pre-frag packet */
        nf_conntrack_put(to->nfct);
        to->nfct = from->nfct;
        nf_conntrack_get(to->nfct);
        to->nfctinfo = from->nfctinfo;
+#if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
+       to->ipvs_property = from->ipvs_property;
+#endif
 #ifdef CONFIG_BRIDGE_NETFILTER
        nf_bridge_put(to->nf_bridge);
        to->nf_bridge = from->nf_bridge;
        nf_bridge_get(to->nf_bridge);
 #endif
 #endif
+       skb_copy_secmark(to, from);
 }
 
 /*
@@ -416,9 +418,9 @@ int ip_fragment(struct sk_buff *skb, int (*output)(struct sk_buff*))
        int ptr;
        struct net_device *dev;
        struct sk_buff *skb2;
-       unsigned int mtu, hlen, left, len, ll_rs;
+       unsigned int mtu, hlen, left, len, ll_rs, pad;
        int offset;
-       int not_last_frag;
+       __be16 not_last_frag;
        struct rtable *rt = (struct rtable*)skb->dst;
        int err = 0;
 
@@ -431,6 +433,7 @@ int ip_fragment(struct sk_buff *skb, int (*output)(struct sk_buff*))
        iph = skb->nh.iph;
 
        if (unlikely((iph->frag_off & htons(IP_DF)) && !skb->local_df)) {
+               IP_INC_STATS(IPSTATS_MIB_FRAGFAILS);
                icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
                          htonl(dst_mtu(&rt->u.dst)));
                kfree_skb(skb);
@@ -443,6 +446,7 @@ int ip_fragment(struct sk_buff *skb, int (*output)(struct sk_buff*))
 
        hlen = iph->ihl * 4;
        mtu = dst_mtu(&rt->u.dst) - hlen;       /* Size of data space */
+       IPCB(skb)->flags |= IPSKB_FRAG_COMPLETE;
 
        /* When frag_list is given, use it. First, check its validity:
         * some transformers could create wrong frag_list or break existing
@@ -516,6 +520,8 @@ int ip_fragment(struct sk_buff *skb, int (*output)(struct sk_buff*))
 
                        err = output(skb);
 
+                       if (!err)
+                               IP_INC_STATS(IPSTATS_MIB_FRAGCREATES);
                        if (err || !frag)
                                break;
 
@@ -542,14 +548,13 @@ slow_path:
        left = skb->len - hlen;         /* Space per frame */
        ptr = raw + hlen;               /* Where to start from */
 
-#ifdef CONFIG_BRIDGE_NETFILTER
        /* for bridged IP traffic encapsulated inside f.e. a vlan header,
-        * we need to make room for the encapsulating header */
-       ll_rs = LL_RESERVED_SPACE_EXTRA(rt->u.dst.dev, nf_bridge_pad(skb));
-       mtu -= nf_bridge_pad(skb);
-#else
-       ll_rs = LL_RESERVED_SPACE(rt->u.dst.dev);
-#endif
+        * we need to make room for the encapsulating header
+        */
+       pad = nf_bridge_pad(skb);
+       ll_rs = LL_RESERVED_SPACE_EXTRA(rt->u.dst.dev, pad);
+       mtu -= pad;
+
        /*
         *      Fragment the datagram.
         */
@@ -576,7 +581,7 @@ slow_path:
                 */
 
                if ((skb2 = alloc_skb(len+hlen+ll_rs, GFP_ATOMIC)) == NULL) {
-                       NETDEBUG(printk(KERN_INFO "IP: frag: no memory for new fragment!\n"));
+                       NETDEBUG(KERN_INFO "IP: frag: no memory for new fragment!\n");
                        err = -ENOMEM;
                        goto fail;
                }
@@ -639,9 +644,6 @@ slow_path:
                /*
                 *      Put this fragment into the sending queue.
                 */
-
-               IP_INC_STATS(IPSTATS_MIB_FRAGCREATES);
-
                iph->tot_len = htons(len + hlen);
 
                ip_send_check(iph);
@@ -649,6 +651,8 @@ slow_path:
                err = output(skb2);
                if (err)
                        goto fail;
+
+               IP_INC_STATS(IPSTATS_MIB_FRAGCREATES);
        }
        kfree_skb(skb);
        IP_INC_STATS(IPSTATS_MIB_FRAGOKS);
@@ -660,16 +664,18 @@ fail:
        return err;
 }
 
+EXPORT_SYMBOL(ip_fragment);
+
 int
 ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk_buff *skb)
 {
        struct iovec *iov = from;
 
-       if (skb->ip_summed == CHECKSUM_HW) {
+       if (skb->ip_summed == CHECKSUM_PARTIAL) {
                if (memcpy_fromiovecend(to, iov, offset, len) < 0)
                        return -EFAULT;
        } else {
-               unsigned int csum = 0;
+               __wsum csum = 0;
                if (csum_partial_copy_fromiovecend(to, iov, offset, len, &csum) < 0)
                        return -EFAULT;
                skb->csum = csum_block_add(skb->csum, csum, odd);
@@ -677,17 +683,72 @@ ip_generic_getfrag(void *from, char *to, int offset, int len, int odd, struct sk
        return 0;
 }
 
-static inline unsigned int
+static inline __wsum
 csum_page(struct page *page, int offset, int copy)
 {
        char *kaddr;
-       unsigned int csum;
+       __wsum csum;
        kaddr = kmap(page);
        csum = csum_partial(kaddr + offset, copy, 0);
        kunmap(page);
        return csum;
 }
 
+static inline int ip_ufo_append_data(struct sock *sk,
+                       int getfrag(void *from, char *to, int offset, int len,
+                              int odd, struct sk_buff *skb),
+                       void *from, int length, int hh_len, int fragheaderlen,
+                       int transhdrlen, int mtu,unsigned int flags)
+{
+       struct sk_buff *skb;
+       int err;
+
+       /* There is support for UDP fragmentation offload by network
+        * device, so create one single skb packet containing complete
+        * udp datagram
+        */
+       if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL) {
+               skb = sock_alloc_send_skb(sk,
+                       hh_len + fragheaderlen + transhdrlen + 20,
+                       (flags & MSG_DONTWAIT), &err);
+
+               if (skb == NULL)
+                       return err;
+
+               /* reserve space for Hardware header */
+               skb_reserve(skb, hh_len);
+
+               /* create space for UDP/IP header */
+               skb_put(skb,fragheaderlen + transhdrlen);
+
+               /* initialize network header pointer */
+               skb->nh.raw = skb->data;
+
+               /* initialize protocol header pointer */
+               skb->h.raw = skb->data + fragheaderlen;
+
+               skb->ip_summed = CHECKSUM_PARTIAL;
+               skb->csum = 0;
+               sk->sk_sndmsg_off = 0;
+       }
+
+       err = skb_append_datato_frags(sk,skb, getfrag, from,
+                              (length - transhdrlen));
+       if (!err) {
+               /* specify the length of each IP datagram fragment*/
+               skb_shinfo(skb)->gso_size = mtu - fragheaderlen;
+               skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
+               __skb_queue_tail(&sk->sk_write_queue, skb);
+
+               return 0;
+       }
+       /* There is not enough support do UFO ,
+        * so follow normal path
+        */
+       kfree_skb(skb);
+       return err;
+}
+
 /*
  *     ip_append_data() and ip_append_page() can make one large IP datagram
  *     from many pieces of data. Each pieces will be holded on the socket
@@ -772,11 +833,21 @@ int ip_append_data(struct sock *sk,
         */
        if (transhdrlen &&
            length + fragheaderlen <= mtu &&
-           rt->u.dst.dev->features&(NETIF_F_IP_CSUM|NETIF_F_NO_CSUM|NETIF_F_HW_CSUM) &&
+           rt->u.dst.dev->features & NETIF_F_ALL_CSUM &&
            !exthdrlen)
-               csummode = CHECKSUM_HW;
+               csummode = CHECKSUM_PARTIAL;
 
        inet->cork.length += length;
+       if (((length > mtu) && (sk->sk_protocol == IPPROTO_UDP)) &&
+                       (rt->u.dst.dev->features & NETIF_F_UFO)) {
+
+               err = ip_ufo_append_data(sk, getfrag, from, length, hh_len,
+                                        fragheaderlen, transhdrlen, mtu,
+                                        flags);
+               if (err)
+                       goto error;
+               return 0;
+       }
 
        /* So, what's going on in the loop below?
         *
@@ -827,7 +898,7 @@ alloc_new_skb:
                         * because we have no idea what fragment will be
                         * the last.
                         */
-                       if (datalen == length)
+                       if (datalen == length + fraggap)
                                alloclen += rt->u.dst.trailer_len;
 
                        if (transhdrlen) {
@@ -869,7 +940,7 @@ alloc_new_skb:
                                skb_prev->csum = csum_sub(skb_prev->csum,
                                                          skb->csum);
                                data += fraggap;
-                               skb_trim(skb_prev, maxfraglen);
+                               pskb_trim_unique(skb_prev, maxfraglen);
                        }
 
                        copy = datalen - transhdrlen - fraggap;
@@ -1008,14 +1079,25 @@ ssize_t ip_append_page(struct sock *sk, struct page *page,
                return -EINVAL;
 
        inet->cork.length += size;
+       if ((sk->sk_protocol == IPPROTO_UDP) &&
+           (rt->u.dst.dev->features & NETIF_F_UFO)) {
+               skb_shinfo(skb)->gso_size = mtu - fragheaderlen;
+               skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
+       }
+
 
        while (size > 0) {
                int i;
 
-               /* Check if the remaining data fits into current packet. */
-               len = mtu - skb->len;
-               if (len < size)
-                       len = maxfraglen - skb->len;
+               if (skb_is_gso(skb))
+                       len = size;
+               else {
+
+                       /* Check if the remaining data fits into current packet. */
+                       len = mtu - skb->len;
+                       if (len < size)
+                               len = maxfraglen - skb->len;
+               }
                if (len <= 0) {
                        struct sk_buff *skb_prev;
                        char *data;
@@ -1023,10 +1105,7 @@ ssize_t  ip_append_page(struct sock *sk, struct page *page,
                        int alloclen;
 
                        skb_prev = skb;
-                       if (skb_prev)
-                               fraggap = skb_prev->len - maxfraglen;
-                       else
-                               fraggap = 0;
+                       fraggap = skb_prev->len - maxfraglen;
 
                        alloclen = fragheaderlen + hh_len + fraggap + 15;
                        skb = sock_wmalloc(sk, alloclen, 1, sk->sk_allocation);
@@ -1056,7 +1135,7 @@ ssize_t   ip_append_page(struct sock *sk, struct page *page,
                                        data, fraggap, 0);
                                skb_prev->csum = csum_sub(skb_prev->csum,
                                                          skb->csum);
-                               skb_trim(skb_prev, maxfraglen);
+                               pskb_trim_unique(skb_prev, maxfraglen);
                        }
 
                        /*
@@ -1080,7 +1159,7 @@ ssize_t   ip_append_page(struct sock *sk, struct page *page,
                }
 
                if (skb->ip_summed == CHECKSUM_NONE) {
-                       unsigned int csum;
+                       __wsum csum;
                        csum = csum_page(page, offset, len);
                        skb->csum = csum_block_add(skb->csum, csum, skb->len);
                }
@@ -1110,7 +1189,7 @@ int ip_push_pending_frames(struct sock *sk)
        struct ip_options *opt = NULL;
        struct rtable *rt = inet->cork.rt;
        struct iphdr *iph;
-       int df = 0;
+       __be16 df = 0;
        __u8 ttl;
        int err = 0;
 
@@ -1166,11 +1245,7 @@ int ip_push_pending_frames(struct sock *sk)
        iph->tos = inet->tos;
        iph->tot_len = htons(skb->len);
        iph->frag_off = df;
-       if (!df) {
-               __ip_select_ident(iph, &rt->u.dst, 0);
-       } else {
-               iph->id = htons(inet->id++);
-       }
+       ip_select_ident(iph, &rt->u.dst, sk);
        iph->ttl = ttl;
        iph->protocol = sk->sk_protocol;
        iph->saddr = rt->rt_src;
@@ -1192,10 +1267,8 @@ int ip_push_pending_frames(struct sock *sk)
 
 out:
        inet->cork.flags &= ~IPCORK_OPT;
-       if (inet->cork.opt) {
-               kfree(inet->cork.opt);
-               inet->cork.opt = NULL;
-       }
+       kfree(inet->cork.opt);
+       inet->cork.opt = NULL;
        if (inet->cork.rt) {
                ip_rt_put(inet->cork.rt);
                inet->cork.rt = NULL;
@@ -1219,10 +1292,8 @@ void ip_flush_pending_frames(struct sock *sk)
                kfree_skb(skb);
 
        inet->cork.flags &= ~IPCORK_OPT;
-       if (inet->cork.opt) {
-               kfree(inet->cork.opt);
-               inet->cork.opt = NULL;
-       }
+       kfree(inet->cork.opt);
+       inet->cork.opt = NULL;
        if (inet->cork.rt) {
                ip_rt_put(inet->cork.rt);
                inet->cork.rt = NULL;
@@ -1236,7 +1307,7 @@ void ip_flush_pending_frames(struct sock *sk)
 static int ip_reply_glue_bits(void *dptr, char *to, int offset, 
                              int len, int odd, struct sk_buff *skb)
 {
-       unsigned int csum;
+       __wsum csum;
 
        csum = csum_partial_copy_nocheck(dptr+offset, to, len, 0);
        skb->csum = csum_block_add(skb->csum, csum, odd);
@@ -1261,7 +1332,7 @@ void ip_send_reply(struct sock *sk, struct sk_buff *skb, struct ip_reply_arg *ar
                char                    data[40];
        } replyopts;
        struct ipcm_cookie ipc;
-       u32 daddr;
+       __be32 daddr;
        struct rtable *rt = (struct rtable*)skb->dst;
 
        if (ip_options_echo(&replyopts.opt, skb))
@@ -1287,6 +1358,7 @@ void ip_send_reply(struct sock *sk, struct sk_buff *skb, struct ip_reply_arg *ar
                                               { .sport = skb->h.th->dest,
                                                 .dport = skb->h.th->source } },
                                    .proto = sk->sk_protocol };
+               security_skb_classify_flow(skb, &fl);
                if (ip_route_output_key(&rt, &fl))
                        return;
        }
@@ -1305,7 +1377,7 @@ void ip_send_reply(struct sock *sk, struct sk_buff *skb, struct ip_reply_arg *ar
                       &ipc, rt, MSG_DONTWAIT);
        if ((skb = skb_peek(&sk->sk_write_queue)) != NULL) {
                if (arg->csumoffset >= 0)
-                       *((u16 *)skb->h.raw + arg->csumoffset) = csum_fold(csum_add(skb->csum, arg->csum));
+                       *((__sum16 *)skb->h.raw + arg->csumoffset) = csum_fold(csum_add(skb->csum, arg->csum));
                skb->ip_summed = CHECKSUM_NONE;
                ip_push_pending_frames(sk);
        }
@@ -1325,7 +1397,6 @@ void __init ip_init(void)
 #endif
 }
 
-EXPORT_SYMBOL(ip_fragment);
 EXPORT_SYMBOL(ip_generic_getfrag);
 EXPORT_SYMBOL(ip_queue_xmit);
 EXPORT_SYMBOL(ip_send_check);