pintos-os.org Git - openvswitch/blob - datapath/actions.c

   1 /*
   2  * Distributed under the terms of the GNU GPL version 2.
   3  * Copyright (c) 2007, 2008, 2009 Nicira Networks.
   4  *
   5  * Significant portions of this file may be copied from parts of the Linux
   6  * kernel, by Linus Torvalds and others.
   7  */
   8
   9 /* Functions for executing flow actions. */
  10
  11 #include <linux/skbuff.h>
  12 #include <linux/in.h>
  13 #include <linux/ip.h>
  14 #include <linux/tcp.h>
  15 #include <linux/udp.h>
  16 #include <linux/in6.h>
  17 #include <linux/if_vlan.h>
  18 #include <net/ip.h>
  19 #include <net/checksum.h>
  20 #include "datapath.h"
  21 #include "dp_dev.h"
  22 #include "actions.h"
  23 #include "openvswitch/datapath-protocol.h"
  24
  25 struct sk_buff *
  26 make_writable(struct sk_buff *skb, gfp_t gfp)
  27 {
  28         if (skb_shared(skb) || skb_cloned(skb)) {
  29                 struct sk_buff *nskb = skb_copy(skb, gfp);
  30                 if (nskb) {
  31 #if defined(CONFIG_XEN) && LINUX_VERSION_CODE == KERNEL_VERSION(2,6,18)
  32                         /* These fields are copied in skb_clone but not in
  33                          * skb_copy or related functions.  We need to manually
  34                          * copy them over here. */
  35                         nskb->proto_data_valid = skb->proto_data_valid;
  36                         nskb->proto_csum_blank = skb->proto_csum_blank;
  37 #endif
  38                         kfree_skb(skb);
  39                         return nskb;
  40                 }
  41         } else {
  42                 unsigned int hdr_len = (skb_transport_offset(skb)
  43                                         + sizeof(struct tcphdr));
  44                 if (pskb_may_pull(skb, min(hdr_len, skb->len)))
  45                         return skb;
  46         }
  47         kfree_skb(skb);
  48         return NULL;
  49 }
  50
  51
  52 static struct sk_buff *
  53 vlan_pull_tag(struct sk_buff *skb)
  54 {
  55         struct vlan_ethhdr *vh = vlan_eth_hdr(skb);
  56         struct ethhdr *eh;
  57
  58
  59         /* Verify we were given a vlan packet */
  60         if (vh->h_vlan_proto != htons(ETH_P_8021Q))
  61                 return skb;
  62
  63         memmove(skb->data + VLAN_HLEN, skb->data, 2 * VLAN_ETH_ALEN);
  64
  65         eh = (struct ethhdr *)skb_pull(skb, VLAN_HLEN);
  66
  67         skb->protocol = eh->h_proto;
  68         skb->mac_header += VLAN_HLEN;
  69
  70         return skb;
  71 }
  72
  73
  74 static struct sk_buff *
  75 modify_vlan_tci(struct datapath *dp, struct sk_buff *skb,
  76                 struct odp_flow_key *key, const union odp_action *a,
  77                 int n_actions, gfp_t gfp)
  78 {
  79         u16 tci, mask;
  80
  81         if (a->type == ODPAT_SET_VLAN_VID) {
  82                 tci = ntohs(a->vlan_vid.vlan_vid);
  83                 mask = VLAN_VID_MASK;
  84                 key->dl_vlan = htons(tci & mask);
  85         } else {
  86                 tci = a->vlan_pcp.vlan_pcp << 13;
  87                 mask = VLAN_PCP_MASK;
  88         }
  89
  90         skb = make_writable(skb, gfp);
  91         if (!skb)
  92                 return ERR_PTR(-ENOMEM);
  93
  94         if (skb->protocol == htons(ETH_P_8021Q)) {
  95                 /* Modify vlan id, but maintain other TCI values */
  96                 struct vlan_ethhdr *vh = vlan_eth_hdr(skb);
  97                 vh->h_vlan_TCI = htons((ntohs(vh->h_vlan_TCI) & ~mask) | tci);
  98         } else {
  99                 /* Add vlan header */
 100
 101                 /* Set up checksumming pointers for checksum-deferred packets
 102                  * on Xen.  Otherwise, dev_queue_xmit() will try to do this
 103                  * when we send the packet out on the wire, and it will fail at
 104                  * that point because skb_checksum_setup() will not look inside
 105                  * an 802.1Q header. */
 106                 skb_checksum_setup(skb);
 107
 108                 /* GSO is not implemented for packets with an 802.1Q header, so
 109                  * we have to do segmentation before we add that header.
 110                  *
 111                  * GSO does work with hardware-accelerated VLAN tagging, but we
 112                  * can't use hardware-accelerated VLAN tagging since it
 113                  * requires the device to have a VLAN group configured (with
 114                  * e.g. vconfig(8)) and we don't do that.
 115                  *
 116                  * Having to do this here may be a performance loss, since we
 117                  * can't take advantage of TSO hardware support, although it
 118                  * does not make a measurable network performance difference
 119                  * for 1G Ethernet.  Fixing that would require patching the
 120                  * kernel (either to add GSO support to the VLAN protocol or to
 121                  * support hardware-accelerated VLAN tagging without VLAN
 122                  * groups configured). */
 123                 if (skb_is_gso(skb)) {
 124                         struct sk_buff *segs;
 125
 126                         segs = skb_gso_segment(skb, 0);
 127                         kfree_skb(skb);
 128                         if (unlikely(IS_ERR(segs)))
 129                                 return ERR_CAST(segs);
 130
 131                         do {
 132                                 struct sk_buff *nskb = segs->next;
 133                                 int err;
 134
 135                                 segs->next = NULL;
 136
 137                                 segs = __vlan_put_tag(segs, tci);
 138                                 err = -ENOMEM;
 139                                 if (segs) {
 140                                         struct odp_flow_key segkey = *key;
 141                                         err = execute_actions(dp, segs,
 142                                                               &segkey, a + 1,
 143                                                               n_actions - 1,
 144                                                               gfp);
 145                                 }
 146
 147                                 if (unlikely(err)) {
 148                                         while ((segs = nskb)) {
 149                                                 nskb = segs->next;
 150                                                 segs->next = NULL;
 151                                                 kfree_skb(segs);
 152                                         }
 153                                         return ERR_PTR(err);
 154                                 }
 155
 156                                 segs = nskb;
 157                         } while (segs->next);
 158
 159                         skb = segs;
 160                 }
 161
 162                 /* The hardware-accelerated version of vlan_put_tag() works
 163                  * only for a device that has a VLAN group configured (with
 164                  * e.g. vconfig(8)), so call the software-only version
 165                  * __vlan_put_tag() directly instead.
 166                  */
 167                 skb = __vlan_put_tag(skb, tci);
 168                 if (!skb)
 169                         return ERR_PTR(-ENOMEM);
 170         }
 171
 172         return skb;
 173 }
 174
 175 static struct sk_buff *strip_vlan(struct sk_buff *skb,
 176                                   struct odp_flow_key *key, gfp_t gfp)
 177 {
 178         skb = make_writable(skb, gfp);
 179         if (skb) {
 180                 vlan_pull_tag(skb);
 181                 key->dl_vlan = htons(ODP_VLAN_NONE);
 182         }
 183         return skb;
 184 }
 185
 186 static struct sk_buff *set_dl_addr(struct sk_buff *skb,
 187                                    const struct odp_action_dl_addr *a,
 188                                    gfp_t gfp)
 189 {
 190         skb = make_writable(skb, gfp);
 191         if (skb) {
 192                 struct ethhdr *eh = eth_hdr(skb);
 193                 memcpy(a->type == ODPAT_SET_DL_SRC ? eh->h_source : eh->h_dest,
 194                        a->dl_addr, ETH_ALEN);
 195         }
 196         return skb;
 197 }
 198
 199 /* Updates 'sum', which is a field in 'skb''s data, given that a 4-byte field
 200  * covered by the sum has been changed from 'from' to 'to'.  If set,
 201  * 'pseudohdr' indicates that the field is in the TCP or UDP pseudo-header.
 202  * Based on nf_proto_csum_replace4. */
 203 static void update_csum(__sum16 *sum, struct sk_buff *skb,
 204                         __be32 from, __be32 to, int pseudohdr)
 205 {
 206         __be32 diff[] = { ~from, to };
 207         if (skb->ip_summed != CHECKSUM_PARTIAL) {
 208                 *sum = csum_fold(csum_partial((char *)diff, sizeof(diff),
 209                                 ~csum_unfold(*sum)));
 210                 if (skb->ip_summed == CHECKSUM_COMPLETE && pseudohdr)
 211                         skb->csum = ~csum_partial((char *)diff, sizeof(diff),
 212                                                 ~skb->csum);
 213         } else if (pseudohdr)
 214                 *sum = ~csum_fold(csum_partial((char *)diff, sizeof(diff),
 215                                 csum_unfold(*sum)));
 216 }
 217
 218 static struct sk_buff *set_nw_addr(struct sk_buff *skb,
 219                                    struct odp_flow_key *key,
 220                                    const struct odp_action_nw_addr *a,
 221                                    gfp_t gfp)
 222 {
 223         if (key->dl_type != htons(ETH_P_IP))
 224                 return skb;
 225
 226         skb = make_writable(skb, gfp);
 227         if (skb) {
 228                 struct iphdr *nh = ip_hdr(skb);
 229                 u32 *f = a->type == ODPAT_SET_NW_SRC ? &nh->saddr : &nh->daddr;
 230                 u32 old = *f;
 231                 u32 new = a->nw_addr;
 232
 233                 if (key->nw_proto == IPPROTO_TCP) {
 234                         struct tcphdr *th = tcp_hdr(skb);
 235                         update_csum(&th->check, skb, old, new, 1);
 236                 } else if (key->nw_proto == IPPROTO_UDP) {
 237                         struct udphdr *th = udp_hdr(skb);
 238                         update_csum(&th->check, skb, old, new, 1);
 239                 }
 240                 update_csum(&nh->check, skb, old, new, 0);
 241                 *f = new;
 242         }
 243         return skb;
 244 }
 245
 246 static struct sk_buff *
 247 set_tp_port(struct sk_buff *skb, struct odp_flow_key *key,
 248             const struct odp_action_tp_port *a,
 249             gfp_t gfp)
 250 {
 251         int check_ofs;
 252
 253         if (key->dl_type != htons(ETH_P_IP))
 254                 return skb;
 255
 256         if (key->nw_proto == IPPROTO_TCP)
 257                 check_ofs = offsetof(struct tcphdr, check);
 258         else if (key->nw_proto == IPPROTO_UDP)
 259                 check_ofs = offsetof(struct udphdr, check);
 260         else
 261                 return skb;
 262
 263         skb = make_writable(skb, gfp);
 264         if (skb) {
 265                 struct udphdr *th = udp_hdr(skb);
 266                 u16 *f = a->type == ODPAT_SET_TP_SRC ? &th->source : &th->dest;
 267                 u16 old = *f;
 268                 u16 new = a->tp_port;
 269                 update_csum((u16*)((u8*)skb->data + check_ofs),
 270                             skb, old, new, 1);
 271                 *f = new;
 272         }
 273         return skb;
 274 }
 275
 276 static inline unsigned packet_length(const struct sk_buff *skb)
 277 {
 278         unsigned length = skb->len - ETH_HLEN;
 279         if (skb->protocol == htons(ETH_P_8021Q))
 280                 length -= VLAN_HLEN;
 281         return length;
 282 }
 283
 284 int dp_xmit_skb(struct sk_buff *skb)
 285 {
 286         struct datapath *dp = skb->dev->br_port->dp;
 287         int len = skb->len;
 288
 289         if (packet_length(skb) > skb->dev->mtu && !skb_is_gso(skb)) {
 290                 printk(KERN_WARNING "%s: dropped over-mtu packet: %d > %d\n",
 291                        dp_name(dp), packet_length(skb), skb->dev->mtu);
 292                 kfree_skb(skb);
 293                 return -E2BIG;
 294         }
 295
 296         dev_queue_xmit(skb);
 297
 298         return len;
 299 }
 300
 301 static void
 302 do_output(struct datapath *dp, struct sk_buff *skb, int out_port)
 303 {
 304         struct net_bridge_port *p;
 305         struct net_device *dev;
 306
 307         if (!skb)
 308                 goto error;
 309
 310         p = dp->ports[out_port];
 311         if (!p)
 312                 goto error;
 313
 314         dev = skb->dev = p->dev;
 315         if (is_dp_dev(dev))
 316                 dp_dev_recv(dev, skb);
 317         else
 318                 dp_xmit_skb(skb);
 319         return;
 320
 321 error:
 322         kfree_skb(skb);
 323 }
 324
 325 /* Never consumes 'skb'.  Returns a port that 'skb' should be sent to, -1 if
 326  * none.  */
 327 static int output_group(struct datapath *dp, __u16 group,
 328                         struct sk_buff *skb, gfp_t gfp)
 329 {
 330         struct dp_port_group *g = rcu_dereference(dp->groups[group]);
 331         int prev_port = -1;
 332         int i;
 333
 334         if (!g)
 335                 return -1;
 336         for (i = 0; i < g->n_ports; i++) {
 337                 struct net_bridge_port *p = dp->ports[g->ports[i]];
 338                 if (!p || skb->dev == p->dev)
 339                         continue;
 340                 if (prev_port != -1) {
 341                         struct sk_buff *clone = skb_clone(skb, gfp);
 342                         if (!clone)
 343                                 return -1;
 344                         do_output(dp, clone, prev_port);
 345                 }
 346                 prev_port = p->port_no;
 347         }
 348         return prev_port;
 349 }
 350
 351 static int
 352 output_control(struct datapath *dp, struct sk_buff *skb, u32 arg, gfp_t gfp)
 353 {
 354         skb = skb_clone(skb, gfp);
 355         if (!skb)
 356                 return -ENOMEM;
 357         return dp_output_control(dp, skb, _ODPL_ACTION_NR, arg);
 358 }
 359
 360 /* Execute a list of actions against 'skb'. */
 361 int execute_actions(struct datapath *dp, struct sk_buff *skb,
 362                     struct odp_flow_key *key,
 363                     const union odp_action *a, int n_actions,
 364                     gfp_t gfp)
 365 {
 366         /* Every output action needs a separate clone of 'skb', but the common
 367          * case is just a single output action, so that doing a clone and
 368          * then freeing the original skbuff is wasteful.  So the following code
 369          * is slightly obscure just to avoid that. */
 370         int prev_port = -1;
 371         int err = 0;
 372         for (; n_actions > 0; a++, n_actions--) {
 373                 WARN_ON_ONCE(skb_shared(skb));
 374                 if (prev_port != -1) {
 375                         do_output(dp, skb_clone(skb, gfp), prev_port);
 376                         prev_port = -1;
 377                 }
 378
 379                 switch (a->type) {
 380                 case ODPAT_OUTPUT:
 381                         prev_port = a->output.port;
 382                         break;
 383
 384                 case ODPAT_OUTPUT_GROUP:
 385                         prev_port = output_group(dp, a->output_group.group,
 386                                                  skb, gfp);
 387                         break;
 388
 389                 case ODPAT_CONTROLLER:
 390                         err = output_control(dp, skb, a->controller.arg, gfp);
 391                         if (err) {
 392                                 kfree_skb(skb);
 393                                 return err;
 394                         }
 395                         break;
 396
 397                 case ODPAT_SET_VLAN_VID:
 398                 case ODPAT_SET_VLAN_PCP:
 399                         skb = modify_vlan_tci(dp, skb, key, a, n_actions, gfp);
 400                         if (IS_ERR(skb))
 401                                 return PTR_ERR(skb);
 402                         break;
 403
 404                 case ODPAT_STRIP_VLAN:
 405                         skb = strip_vlan(skb, key, gfp);
 406                         break;
 407
 408                 case ODPAT_SET_DL_SRC:
 409                 case ODPAT_SET_DL_DST:
 410                         skb = set_dl_addr(skb, &a->dl_addr, gfp);
 411                         break;
 412
 413                 case ODPAT_SET_NW_SRC:
 414                 case ODPAT_SET_NW_DST:
 415                         skb = set_nw_addr(skb, key, &a->nw_addr, gfp);
 416                         break;
 417
 418                 case ODPAT_SET_TP_SRC:
 419                 case ODPAT_SET_TP_DST:
 420                         skb = set_tp_port(skb, key, &a->tp_port, gfp);
 421                         break;
 422                 }
 423                 if (!skb)
 424                         return -ENOMEM;
 425         }
 426         if (prev_port != -1)
 427                 do_output(dp, skb, prev_port);
 428         else
 429                 kfree_skb(skb);
 430         return err;
 431 }