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 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,36)
  41 #define rt_dst(rt) (rt->dst)
  42 #else
  43 #define rt_dst(rt) (rt->u.dst)
  44 #endif
  45
  46 /**
  47  *      ovs_tnl_rcv - ingress point for generic tunnel code
  48  *
  49  * @vport: port this packet was received on
  50  * @skb: received packet
  51  * @tos: ToS from encapsulating IP packet, used to copy ECN bits
  52  *
  53  * Must be called with rcu_read_lock.
  54  *
  55  * Packets received by this function are in the following state:
  56  * - skb->data points to the inner Ethernet header.
  57  * - The inner Ethernet header is in the linear data area.
  58  * - skb->csum does not include the inner Ethernet header.
  59  * - The layer pointers are undefined.
  60  */
  61 void ovs_tnl_rcv(struct vport *vport, struct sk_buff *skb)
  62 {
  63         struct ethhdr *eh;
  64
  65         skb_reset_mac_header(skb);
  66         eh = eth_hdr(skb);
  67
  68         if (likely(ntohs(eh->h_proto) >= ETH_P_802_3_MIN))
  69                 skb->protocol = eh->h_proto;
  70         else
  71                 skb->protocol = htons(ETH_P_802_2);
  72
  73         skb_dst_drop(skb);
  74         nf_reset(skb);
  75         skb_clear_rxhash(skb);
  76         secpath_reset(skb);
  77         vlan_set_tci(skb, 0);
  78
  79         if (unlikely(compute_ip_summed(skb, false))) {
  80                 kfree_skb(skb);
  81                 return;
  82         }
  83
  84         ovs_vport_receive(vport, skb);
  85 }
  86
  87 static struct rtable *find_route(struct net *net,
  88                 __be32 *saddr, __be32 daddr, u8 ipproto,
  89                 u8 tos, u32 skb_mark)
  90 {
  91         struct rtable *rt;
  92         /* Tunnel configuration keeps DSCP part of TOS bits, But Linux
  93          * router expect RT_TOS bits only. */
  94
  95 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,39)
  96         struct flowi fl = { .nl_u = { .ip4_u = {
  97                                         .daddr = daddr,
  98                                         .saddr = *saddr,
  99 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,20)
 100                                         .fwmark = skb_mark,
 101 #endif
 102                                         .tos   = RT_TOS(tos) } },
 103 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,20)
 104                                         .mark = skb_mark,
 105 #endif
 106                                         .proto = ipproto };
 107
 108         if (unlikely(ip_route_output_key(net, &rt, &fl)))
 109                 return ERR_PTR(-EADDRNOTAVAIL);
 110         *saddr = fl.nl_u.ip4_u.saddr;
 111         return rt;
 112 #else
 113         struct flowi4 fl = { .daddr = daddr,
 114                              .saddr = *saddr,
 115                              .flowi4_tos = RT_TOS(tos),
 116                              .flowi4_mark = skb_mark,
 117                              .flowi4_proto = ipproto };
 118
 119         rt = ip_route_output_key(net, &fl);
 120         *saddr = fl.saddr;
 121         return rt;
 122 #endif
 123 }
 124
 125 static bool need_linearize(const struct sk_buff *skb)
 126 {
 127         int i;
 128
 129         if (unlikely(skb_shinfo(skb)->frag_list))
 130                 return true;
 131
 132         /*
 133          * Generally speaking we should linearize if there are paged frags.
 134          * However, if all of the refcounts are 1 we know nobody else can
 135          * change them from underneath us and we can skip the linearization.
 136          */
 137         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
 138                 if (unlikely(page_count(skb_frag_page(&skb_shinfo(skb)->frags[i])) > 1))
 139                         return true;
 140
 141         return false;
 142 }
 143
 144 static struct sk_buff *handle_offloads(struct sk_buff *skb)
 145 {
 146         int err;
 147
 148         forward_ip_summed(skb, true);
 149
 150         if (skb_is_gso(skb)) {
 151                 struct sk_buff *nskb;
 152
 153                 nskb = __skb_gso_segment(skb, 0, false);
 154                 if (IS_ERR(nskb)) {
 155                         err = PTR_ERR(nskb);
 156                         goto error;
 157                 }
 158
 159                 consume_skb(skb);
 160                 skb = nskb;
 161         } else if (get_ip_summed(skb) == OVS_CSUM_PARTIAL) {
 162                 /* Pages aren't locked and could change at any time.
 163                  * If this happens after we compute the checksum, the
 164                  * checksum will be wrong.  We linearize now to avoid
 165                  * this problem.
 166                  */
 167                 if (unlikely(need_linearize(skb))) {
 168                         err = __skb_linearize(skb);
 169                         if (unlikely(err))
 170                                 goto error;
 171                 }
 172
 173                 err = skb_checksum_help(skb);
 174                 if (unlikely(err))
 175                         goto error;
 176         }
 177
 178         set_ip_summed(skb, OVS_CSUM_NONE);
 179
 180         return skb;
 181
 182 error:
 183         return ERR_PTR(err);
 184 }
 185
 186 /* Compute source UDP port for outgoing packet.
 187  * Currently we use the flow hash.
 188  */
 189 u16 ovs_tnl_get_src_port(struct sk_buff *skb)
 190 {
 191         int low;
 192         int high;
 193         unsigned int range;
 194         u32 hash = OVS_CB(skb)->flow->hash;
 195
 196         inet_get_local_port_range(&low, &high);
 197         range = (high - low) + 1;
 198         return (((u64) hash * range) >> 32) + low;
 199 }
 200
 201 int ovs_tnl_send(struct vport *vport, struct sk_buff *skb,
 202                  u8 ipproto, int tunnel_hlen,
 203                  void (*build_header)(const struct vport *,
 204                                       struct sk_buff *,
 205                                       int tunnel_hlen))
 206 {
 207         int min_headroom;
 208         struct rtable *rt;
 209         __be32 saddr;
 210         int sent_len = 0;
 211         int err;
 212         struct sk_buff *nskb;
 213
 214         /* Route lookup */
 215         saddr = OVS_CB(skb)->tun_key->ipv4_src;
 216         rt = find_route(ovs_dp_get_net(vport->dp),
 217                         &saddr,
 218                         OVS_CB(skb)->tun_key->ipv4_dst,
 219                         ipproto,
 220                         OVS_CB(skb)->tun_key->ipv4_tos,
 221                         skb_get_mark(skb));
 222         if (IS_ERR(rt)) {
 223                 err = PTR_ERR(rt);
 224                 goto error;
 225         }
 226
 227         tunnel_hlen += sizeof(struct iphdr);
 228
 229         min_headroom = LL_RESERVED_SPACE(rt_dst(rt).dev) + rt_dst(rt).header_len
 230                         + tunnel_hlen
 231                         + (vlan_tx_tag_present(skb) ? VLAN_HLEN : 0);
 232
 233         if (skb_headroom(skb) < min_headroom || skb_header_cloned(skb)) {
 234                 int head_delta = SKB_DATA_ALIGN(min_headroom -
 235                                                 skb_headroom(skb) +
 236                                                 16);
 237
 238                 err = pskb_expand_head(skb, max_t(int, head_delta, 0),
 239                                         0, GFP_ATOMIC);
 240                 if (unlikely(err))
 241                         goto err_free_rt;
 242         }
 243
 244         /* Offloading */
 245         nskb = handle_offloads(skb);
 246         if (IS_ERR(nskb)) {
 247                 err = PTR_ERR(nskb);
 248                 goto err_free_rt;
 249         }
 250         skb = nskb;
 251
 252         /* Reset SKB */
 253         nf_reset(skb);
 254         secpath_reset(skb);
 255         skb_dst_drop(skb);
 256         skb_clear_rxhash(skb);
 257
 258         while (skb) {
 259                 struct sk_buff *next_skb = skb->next;
 260                 struct iphdr *iph;
 261                 int frag_len;
 262
 263                 skb->next = NULL;
 264
 265                 if (unlikely(vlan_deaccel_tag(skb)))
 266                         goto next;
 267
 268                 frag_len = skb->len;
 269                 skb_push(skb, tunnel_hlen);
 270                 skb_reset_network_header(skb);
 271                 skb_set_transport_header(skb, sizeof(struct iphdr));
 272
 273                 if (next_skb)
 274                         skb_dst_set(skb, dst_clone(&rt_dst(rt)));
 275                 else
 276                         skb_dst_set(skb, &rt_dst(rt));
 277
 278                 /* Push Tunnel header. */
 279                 build_header(vport, skb, tunnel_hlen);
 280
 281                 /* Push IP header. */
 282                 iph = ip_hdr(skb);
 283                 iph->version    = 4;
 284                 iph->ihl        = sizeof(struct iphdr) >> 2;
 285                 iph->protocol   = ipproto;
 286                 iph->daddr      = OVS_CB(skb)->tun_key->ipv4_dst;
 287                 iph->saddr      = saddr;
 288                 iph->tos        = OVS_CB(skb)->tun_key->ipv4_tos;
 289                 iph->ttl        = OVS_CB(skb)->tun_key->ipv4_ttl;
 290                 iph->frag_off   = OVS_CB(skb)->tun_key->tun_flags &
 291                                   OVS_TNL_F_DONT_FRAGMENT ?  htons(IP_DF) : 0;
 292                 /*
 293                  * Allow our local IP stack to fragment the outer packet even
 294                  * if the DF bit is set as a last resort.  We also need to
 295                  * force selection of an IP ID here with __ip_select_ident(),
 296                  * as ip_select_ident() assumes a proper ID is not needed when
 297                  * when the DF bit is set.
 298                  */
 299                 skb->local_df = 1;
 300                 __ip_select_ident(iph, skb_dst(skb), 0);
 301
 302                 memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
 303
 304                 err = ip_local_out(skb);
 305                 if (unlikely(net_xmit_eval(err)))
 306                         goto next;
 307
 308                 sent_len += frag_len;
 309
 310 next:
 311                 skb = next_skb;
 312         }
 313
 314         return sent_len;
 315
 316 err_free_rt:
 317         ip_rt_put(rt);
 318 error:
 319         return err;
 320 }