datapath/tunnel.c

   1 /*
   2  * Copyright (c) 2007-2012 Nicira, Inc.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of version 2 of the GNU General Public
   6  * License as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program; if not, write to the Free Software
  15  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
  16  * 02110-1301, USA
  17  */
  18
  19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  20
  21 #include <linux/in.h>
  22 #include <linux/in_route.h>
  23 #include <linux/inetdevice.h>
  24 #include <linux/jhash.h>
  25 #include <linux/list.h>
  26 #include <linux/kernel.h>
  27 #include <linux/version.h>
  28 #include <linux/workqueue.h>
  29 #include <linux/rculist.h>
  30 #include <net/route.h>
  31 #include <net/xfrm.h>
  32
  33 #include "checksum.h"
  34 #include "compat.h"
  35 #include "datapath.h"
  36 #include "tunnel.h"
  37 #include "vlan.h"
  38 #include "vport.h"
  39
  40 /**
  41  *      ovs_tnl_rcv - ingress point for generic tunnel code
  42  *
  43  * @vport: port this packet was received on
  44  * @skb: received packet
  45  * @tos: ToS from encapsulating IP packet, used to copy ECN bits
  46  *
  47  * Must be called with rcu_read_lock.
  48  *
  49  * Packets received by this function are in the following state:
  50  * - skb->data points to the inner Ethernet header.
  51  * - The inner Ethernet header is in the linear data area.
  52  * - skb->csum does not include the inner Ethernet header.
  53  * - The layer pointers are undefined.
  54  */
  55 void ovs_tnl_rcv(struct vport *vport, struct sk_buff *skb,
  56                  struct ovs_key_ipv4_tunnel *tun_key)
  57 {
  58         struct ethhdr *eh;
  59
  60         skb_reset_mac_header(skb);
  61         eh = eth_hdr(skb);
  62
  63         if (likely(ntohs(eh->h_proto) >= ETH_P_802_3_MIN))
  64                 skb->protocol = eh->h_proto;
  65         else
  66                 skb->protocol = htons(ETH_P_802_2);
  67
  68         skb_dst_drop(skb);
  69         nf_reset(skb);
  70         skb_clear_rxhash(skb);
  71         secpath_reset(skb);
  72         vlan_set_tci(skb, 0);
  73
  74         if (unlikely(compute_ip_summed(skb, false))) {
  75                 kfree_skb(skb);
  76                 return;
  77         }
  78
  79         ovs_vport_receive(vport, skb, tun_key);
  80 }
  81
  82 struct rtable *find_route(struct net *net,
  83                           __be32 *saddr, __be32 daddr, u8 ipproto,
  84                           u8 tos, u32 skb_mark)
  85 {
  86         struct rtable *rt;
  87         /* Tunnel configuration keeps DSCP part of TOS bits, But Linux
  88          * router expect RT_TOS bits only. */
  89
  90 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,39)
  91         struct flowi fl = { .nl_u = { .ip4_u = {
  92                                         .daddr = daddr,
  93                                         .saddr = *saddr,
  94 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,20)
  95                                         .fwmark = skb_mark,
  96 #endif
  97                                         .tos   = RT_TOS(tos) } },
  98 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,20)
  99                                         .mark = skb_mark,
 100 #endif
 101                                         .proto = ipproto };
 102
 103         if (unlikely(ip_route_output_key(net, &rt, &fl)))
 104                 return ERR_PTR(-EADDRNOTAVAIL);
 105         *saddr = fl.nl_u.ip4_u.saddr;
 106         return rt;
 107 #else
 108         struct flowi4 fl = { .daddr = daddr,
 109                              .saddr = *saddr,
 110                              .flowi4_tos = RT_TOS(tos),
 111                              .flowi4_mark = skb_mark,
 112                              .flowi4_proto = ipproto };
 113
 114         rt = ip_route_output_key(net, &fl);
 115         *saddr = fl.saddr;
 116         return rt;
 117 #endif
 118 }
 119
 120 static bool need_linearize(const struct sk_buff *skb)
 121 {
 122         int i;
 123
 124         if (unlikely(skb_shinfo(skb)->frag_list))
 125                 return true;
 126
 127         /*
 128          * Generally speaking we should linearize if there are paged frags.
 129          * However, if all of the refcounts are 1 we know nobody else can
 130          * change them from underneath us and we can skip the linearization.
 131          */
 132         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
 133                 if (unlikely(page_count(skb_frag_page(&skb_shinfo(skb)->frags[i])) > 1))
 134                         return true;
 135
 136         return false;
 137 }
 138
 139 static struct sk_buff *handle_offloads(struct sk_buff *skb)
 140 {
 141         int err;
 142
 143         forward_ip_summed(skb, true);
 144
 145         if (skb_is_gso(skb)) {
 146                 struct sk_buff *nskb;
 147
 148                 nskb = __skb_gso_segment(skb, 0, false);
 149                 if (IS_ERR(nskb)) {
 150                         err = PTR_ERR(nskb);
 151                         goto error;
 152                 }
 153
 154                 consume_skb(skb);
 155                 skb = nskb;
 156         } else if (get_ip_summed(skb) == OVS_CSUM_PARTIAL) {
 157                 /* Pages aren't locked and could change at any time.
 158                  * If this happens after we compute the checksum, the
 159                  * checksum will be wrong.  We linearize now to avoid
 160                  * this problem.
 161                  */
 162                 if (unlikely(need_linearize(skb))) {
 163                         err = __skb_linearize(skb);
 164                         if (unlikely(err))
 165                                 goto error;
 166                 }
 167
 168                 err = skb_checksum_help(skb);
 169                 if (unlikely(err))
 170                         goto error;
 171         }
 172
 173         set_ip_summed(skb, OVS_CSUM_NONE);
 174
 175         return skb;
 176
 177 error:
 178         return ERR_PTR(err);
 179 }
 180
 181 /* Compute source UDP port for outgoing packet.
 182  * Currently we use the flow hash.
 183  */
 184 u16 ovs_tnl_get_src_port(struct sk_buff *skb)
 185 {
 186         int low;
 187         int high;
 188         unsigned int range;
 189         struct sw_flow_key *pkt_key = OVS_CB(skb)->pkt_key;
 190         u32 hash = jhash2((const u32 *)pkt_key,
 191                           sizeof(*pkt_key) / sizeof(u32), 0);
 192
 193         inet_get_local_port_range(&low, &high);
 194         range = (high - low) + 1;
 195         return (((u64) hash * range) >> 32) + low;
 196 }
 197
 198 int ovs_tnl_send(struct vport *vport, struct sk_buff *skb,
 199                  u8 ipproto, int tunnel_hlen,
 200                  void (*build_header)(const struct vport *,
 201                                       struct sk_buff *,
 202                                       int tunnel_hlen))
 203 {
 204         int min_headroom;
 205         struct rtable *rt;
 206         __be32 saddr;
 207         int sent_len = 0;
 208         int err;
 209         struct sk_buff *nskb;
 210
 211         /* Route lookup */
 212         saddr = OVS_CB(skb)->tun_key->ipv4_src;
 213         rt = find_route(ovs_dp_get_net(vport->dp),
 214                         &saddr,
 215                         OVS_CB(skb)->tun_key->ipv4_dst,
 216                         ipproto,
 217                         OVS_CB(skb)->tun_key->ipv4_tos,
 218                         skb_get_mark(skb));
 219         if (IS_ERR(rt)) {
 220                 err = PTR_ERR(rt);
 221                 goto error;
 222         }
 223
 224         tunnel_hlen += sizeof(struct iphdr);
 225
 226         min_headroom = LL_RESERVED_SPACE(rt_dst(rt).dev) + rt_dst(rt).header_len
 227                         + tunnel_hlen
 228                         + (vlan_tx_tag_present(skb) ? VLAN_HLEN : 0);
 229
 230         if (skb_headroom(skb) < min_headroom || skb_header_cloned(skb)) {
 231                 int head_delta = SKB_DATA_ALIGN(min_headroom -
 232                                                 skb_headroom(skb) +
 233                                                 16);
 234
 235                 err = pskb_expand_head(skb, max_t(int, head_delta, 0),
 236                                         0, GFP_ATOMIC);
 237                 if (unlikely(err))
 238                         goto err_free_rt;
 239         }
 240
 241         /* Offloading */
 242         nskb = handle_offloads(skb);
 243         if (IS_ERR(nskb)) {
 244                 err = PTR_ERR(nskb);
 245                 goto err_free_rt;
 246         }
 247         skb = nskb;
 248
 249         /* Reset SKB */
 250         nf_reset(skb);
 251         secpath_reset(skb);
 252         skb_dst_drop(skb);
 253         skb_clear_rxhash(skb);
 254
 255         while (skb) {
 256                 struct sk_buff *next_skb = skb->next;
 257                 struct iphdr *iph;
 258                 int frag_len;
 259
 260                 skb->next = NULL;
 261
 262                 if (unlikely(vlan_deaccel_tag(skb)))
 263                         goto next;
 264
 265                 frag_len = skb->len;
 266                 skb_push(skb, tunnel_hlen);
 267                 skb_reset_network_header(skb);
 268                 skb_set_transport_header(skb, sizeof(struct iphdr));
 269
 270                 if (next_skb)
 271                         skb_dst_set(skb, dst_clone(&rt_dst(rt)));
 272                 else
 273                         skb_dst_set(skb, &rt_dst(rt));
 274
 275                 /* Push Tunnel header. */
 276                 build_header(vport, skb, tunnel_hlen);
 277
 278                 /* Push IP header. */
 279                 iph = ip_hdr(skb);
 280                 iph->version    = 4;
 281                 iph->ihl        = sizeof(struct iphdr) >> 2;
 282                 iph->protocol   = ipproto;
 283                 iph->daddr      = OVS_CB(skb)->tun_key->ipv4_dst;
 284                 iph->saddr      = saddr;
 285                 iph->tos        = OVS_CB(skb)->tun_key->ipv4_tos;
 286                 iph->ttl        = OVS_CB(skb)->tun_key->ipv4_ttl;
 287                 iph->frag_off   = OVS_CB(skb)->tun_key->tun_flags &
 288                                   TUNNEL_DONT_FRAGMENT ?  htons(IP_DF) : 0;
 289                 /*
 290                  * Allow our local IP stack to fragment the outer packet even
 291                  * if the DF bit is set as a last resort.  We also need to
 292                  * force selection of an IP ID here with __ip_select_ident(),
 293                  * as ip_select_ident() assumes a proper ID is not needed when
 294                  * when the DF bit is set.
 295                  */
 296                 skb->local_df = 1;
 297                 __ip_select_ident(iph, skb_dst(skb), 0);
 298
 299                 memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
 300
 301                 err = ip_local_out(skb);
 302                 if (unlikely(net_xmit_eval(err)))
 303                         goto next;
 304
 305                 sent_len += frag_len;
 306
 307 next:
 308                 skb = next_skb;
 309         }
 310
 311         return sent_len;
 312
 313 err_free_rt:
 314         ip_rt_put(rt);
 315 error:
 316         return err;
 317 }