datapath/actions.c

   1 /*
   2  * Distributed under the terms of the GNU GPL version 2.
   3  * Copyright (c) 2007, 2008, 2009, 2010, 2011 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_arp.h>
  18 #include <linux/if_vlan.h>
  19 #include <net/inet_ecn.h>
  20 #include <net/ip.h>
  21 #include <net/checksum.h>
  22
  23 #include "actions.h"
  24 #include "checksum.h"
  25 #include "datapath.h"
  26 #include "loop_counter.h"
  27 #include "openvswitch/datapath-protocol.h"
  28 #include "vlan.h"
  29 #include "vport.h"
  30
  31 static int do_execute_actions(struct datapath *, struct sk_buff *,
  32                               struct sw_flow_actions *acts);
  33
  34 static struct sk_buff *make_writable(struct sk_buff *skb, unsigned min_headroom)
  35 {
  36         if (skb_cloned(skb)) {
  37                 struct sk_buff *nskb;
  38                 unsigned headroom = max(min_headroom, skb_headroom(skb));
  39
  40                 nskb = skb_copy_expand(skb, headroom, skb_tailroom(skb), GFP_ATOMIC);
  41                 if (nskb) {
  42                         set_skb_csum_bits(skb, nskb);
  43                         kfree_skb(skb);
  44                         return nskb;
  45                 }
  46         } else {
  47                 unsigned int hdr_len = (skb_transport_offset(skb)
  48                                         + sizeof(struct tcphdr));
  49                 if (pskb_may_pull(skb, min(hdr_len, skb->len)))
  50                         return skb;
  51         }
  52         kfree_skb(skb);
  53         return NULL;
  54 }
  55
  56 static struct sk_buff *strip_vlan(struct sk_buff *skb)
  57 {
  58         struct ethhdr *eh;
  59
  60         if (vlan_tx_tag_present(skb)) {
  61                 vlan_set_tci(skb, 0);
  62                 return skb;
  63         }
  64
  65         if (unlikely(vlan_eth_hdr(skb)->h_vlan_proto != htons(ETH_P_8021Q) ||
  66             skb->len < VLAN_ETH_HLEN))
  67                 return skb;
  68
  69         skb = make_writable(skb, 0);
  70         if (unlikely(!skb))
  71                 return NULL;
  72
  73         if (get_ip_summed(skb) == OVS_CSUM_COMPLETE)
  74                 skb->csum = csum_sub(skb->csum, csum_partial(skb->data
  75                                         + ETH_HLEN, VLAN_HLEN, 0));
  76
  77         memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN);
  78
  79         eh = (struct ethhdr *)skb_pull(skb, VLAN_HLEN);
  80
  81         skb->protocol = eh->h_proto;
  82         skb->mac_header += VLAN_HLEN;
  83
  84         return skb;
  85 }
  86
  87 static struct sk_buff *modify_vlan_tci(struct sk_buff *skb, __be16 tci)
  88 {
  89         struct vlan_ethhdr *vh;
  90         __be16 old_tci;
  91
  92         if (vlan_tx_tag_present(skb) || skb->protocol != htons(ETH_P_8021Q))
  93                 return __vlan_hwaccel_put_tag(skb, ntohs(tci));
  94
  95         skb = make_writable(skb, 0);
  96         if (unlikely(!skb))
  97                 return NULL;
  98
  99         if (unlikely(skb->len < VLAN_ETH_HLEN))
 100                 return skb;
 101
 102         vh = vlan_eth_hdr(skb);
 103
 104         old_tci = vh->h_vlan_TCI;
 105         vh->h_vlan_TCI = tci;
 106
 107         if (get_ip_summed(skb) == OVS_CSUM_COMPLETE) {
 108                 __be16 diff[] = { ~old_tci, vh->h_vlan_TCI };
 109                 skb->csum = ~csum_partial((char *)diff, sizeof(diff), ~skb->csum);
 110         }
 111
 112         return skb;
 113 }
 114
 115 static bool is_ip(struct sk_buff *skb)
 116 {
 117         return (OVS_CB(skb)->flow->key.dl_type == htons(ETH_P_IP) &&
 118                 skb->transport_header > skb->network_header);
 119 }
 120
 121 static __sum16 *get_l4_checksum(struct sk_buff *skb)
 122 {
 123         u8 nw_proto = OVS_CB(skb)->flow->key.nw_proto;
 124         int transport_len = skb->len - skb_transport_offset(skb);
 125         if (nw_proto == IPPROTO_TCP) {
 126                 if (likely(transport_len >= sizeof(struct tcphdr)))
 127                         return &tcp_hdr(skb)->check;
 128         } else if (nw_proto == IPPROTO_UDP) {
 129                 if (likely(transport_len >= sizeof(struct udphdr)))
 130                         return &udp_hdr(skb)->check;
 131         }
 132         return NULL;
 133 }
 134
 135 static struct sk_buff *set_nw_addr(struct sk_buff *skb, const struct nlattr *a)
 136 {
 137         __be32 new_nwaddr = nla_get_be32(a);
 138         struct iphdr *nh;
 139         __sum16 *check;
 140         __be32 *nwaddr;
 141
 142         if (unlikely(!is_ip(skb)))
 143                 return skb;
 144
 145         skb = make_writable(skb, 0);
 146         if (unlikely(!skb))
 147                 return NULL;
 148
 149         nh = ip_hdr(skb);
 150         nwaddr = nla_type(a) == ODP_ACTION_ATTR_SET_NW_SRC ? &nh->saddr : &nh->daddr;
 151
 152         check = get_l4_checksum(skb);
 153         if (likely(check))
 154                 inet_proto_csum_replace4(check, skb, *nwaddr, new_nwaddr, 1);
 155         csum_replace4(&nh->check, *nwaddr, new_nwaddr);
 156
 157         skb_clear_rxhash(skb);
 158
 159         *nwaddr = new_nwaddr;
 160
 161         return skb;
 162 }
 163
 164 static struct sk_buff *set_nw_tos(struct sk_buff *skb, u8 nw_tos)
 165 {
 166         if (unlikely(!is_ip(skb)))
 167                 return skb;
 168
 169         skb = make_writable(skb, 0);
 170         if (skb) {
 171                 struct iphdr *nh = ip_hdr(skb);
 172                 u8 *f = &nh->tos;
 173                 u8 old = *f;
 174                 u8 new;
 175
 176                 /* Set the DSCP bits and preserve the ECN bits. */
 177                 new = nw_tos | (nh->tos & INET_ECN_MASK);
 178                 csum_replace4(&nh->check, (__force __be32)old,
 179                                           (__force __be32)new);
 180                 *f = new;
 181         }
 182         return skb;
 183 }
 184
 185 static struct sk_buff *set_tp_port(struct sk_buff *skb, const struct nlattr *a)
 186 {
 187         struct udphdr *th;
 188         __sum16 *check;
 189         __be16 *port;
 190
 191         if (unlikely(!is_ip(skb)))
 192                 return skb;
 193
 194         skb = make_writable(skb, 0);
 195         if (unlikely(!skb))
 196                 return NULL;
 197
 198         /* Must follow make_writable() since that can move the skb data. */
 199         check = get_l4_checksum(skb);
 200         if (unlikely(!check))
 201                 return skb;
 202
 203         /*
 204          * Update port and checksum.
 205          *
 206          * This is OK because source and destination port numbers are at the
 207          * same offsets in both UDP and TCP headers, and get_l4_checksum() only
 208          * supports those protocols.
 209          */
 210         th = udp_hdr(skb);
 211         port = nla_type(a) == ODP_ACTION_ATTR_SET_TP_SRC ? &th->source : &th->dest;
 212         inet_proto_csum_replace2(check, skb, *port, nla_get_be16(a), 0);
 213         *port = nla_get_be16(a);
 214         skb_clear_rxhash(skb);
 215
 216         return skb;
 217 }
 218
 219 /**
 220  * is_spoofed_arp - check for invalid ARP packet
 221  *
 222  * @skb: skbuff containing an Ethernet packet, with network header pointing
 223  * just past the Ethernet and optional 802.1Q header.
 224  *
 225  * Returns true if @skb is an invalid Ethernet+IPv4 ARP packet: one with screwy
 226  * or truncated header fields or one whose inner and outer Ethernet address
 227  * differ.
 228  */
 229 static bool is_spoofed_arp(struct sk_buff *skb)
 230 {
 231         struct arp_eth_header *arp;
 232
 233         if (OVS_CB(skb)->flow->key.dl_type != htons(ETH_P_ARP))
 234                 return false;
 235
 236         if (skb_network_offset(skb) + sizeof(struct arp_eth_header) > skb->len)
 237                 return true;
 238
 239         arp = (struct arp_eth_header *)skb_network_header(skb);
 240         return (arp->ar_hrd != htons(ARPHRD_ETHER) ||
 241                 arp->ar_pro != htons(ETH_P_IP) ||
 242                 arp->ar_hln != ETH_ALEN ||
 243                 arp->ar_pln != 4 ||
 244                 compare_ether_addr(arp->ar_sha, eth_hdr(skb)->h_source));
 245 }
 246
 247 static void do_output(struct datapath *dp, struct sk_buff *skb, int out_port)
 248 {
 249         struct vport *p;
 250
 251         if (!skb)
 252                 goto error;
 253
 254         p = rcu_dereference(dp->ports[out_port]);
 255         if (!p)
 256                 goto error;
 257
 258         vport_send(p, skb);
 259         return;
 260
 261 error:
 262         kfree_skb(skb);
 263 }
 264
 265 static int output_control(struct datapath *dp, struct sk_buff *skb, u64 arg)
 266 {
 267         struct dp_upcall_info upcall;
 268
 269         skb = skb_clone(skb, GFP_ATOMIC);
 270         if (!skb)
 271                 return -ENOMEM;
 272
 273         upcall.cmd = ODP_PACKET_CMD_ACTION;
 274         upcall.key = &OVS_CB(skb)->flow->key;
 275         upcall.userdata = arg;
 276         upcall.sample_pool = 0;
 277         upcall.actions = NULL;
 278         upcall.actions_len = 0;
 279         return dp_upcall(dp, skb, &upcall);
 280 }
 281
 282 /* Execute a list of actions against 'skb'. */
 283 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
 284                               struct sw_flow_actions *acts)
 285 {
 286         /* Every output action needs a separate clone of 'skb', but the common
 287          * case is just a single output action, so that doing a clone and
 288          * then freeing the original skbuff is wasteful.  So the following code
 289          * is slightly obscure just to avoid that. */
 290         int prev_port = -1;
 291         u32 priority = skb->priority;
 292         const struct nlattr *a;
 293         int rem, err;
 294
 295         for (a = acts->actions, rem = acts->actions_len; rem > 0;
 296              a = nla_next(a, &rem)) {
 297                 if (prev_port != -1) {
 298                         do_output(dp, skb_clone(skb, GFP_ATOMIC), prev_port);
 299                         prev_port = -1;
 300                 }
 301
 302                 switch (nla_type(a)) {
 303                 case ODP_ACTION_ATTR_OUTPUT:
 304                         prev_port = nla_get_u32(a);
 305                         break;
 306
 307                 case ODP_ACTION_ATTR_CONTROLLER:
 308                         err = output_control(dp, skb, nla_get_u64(a));
 309                         if (err) {
 310                                 kfree_skb(skb);
 311                                 return err;
 312                         }
 313                         break;
 314
 315                 case ODP_ACTION_ATTR_SET_TUNNEL:
 316                         OVS_CB(skb)->tun_id = nla_get_be64(a);
 317                         break;
 318
 319                 case ODP_ACTION_ATTR_SET_DL_TCI:
 320                         skb = modify_vlan_tci(skb, nla_get_be16(a));
 321                         break;
 322
 323                 case ODP_ACTION_ATTR_STRIP_VLAN:
 324                         skb = strip_vlan(skb);
 325                         break;
 326
 327                 case ODP_ACTION_ATTR_SET_DL_SRC:
 328                         skb = make_writable(skb, 0);
 329                         if (!skb)
 330                                 return -ENOMEM;
 331                         memcpy(eth_hdr(skb)->h_source, nla_data(a), ETH_ALEN);
 332                         break;
 333
 334                 case ODP_ACTION_ATTR_SET_DL_DST:
 335                         skb = make_writable(skb, 0);
 336                         if (!skb)
 337                                 return -ENOMEM;
 338                         memcpy(eth_hdr(skb)->h_dest, nla_data(a), ETH_ALEN);
 339                         break;
 340
 341                 case ODP_ACTION_ATTR_SET_NW_SRC:
 342                 case ODP_ACTION_ATTR_SET_NW_DST:
 343                         skb = set_nw_addr(skb, a);
 344                         break;
 345
 346                 case ODP_ACTION_ATTR_SET_NW_TOS:
 347                         skb = set_nw_tos(skb, nla_get_u8(a));
 348                         break;
 349
 350                 case ODP_ACTION_ATTR_SET_TP_SRC:
 351                 case ODP_ACTION_ATTR_SET_TP_DST:
 352                         skb = set_tp_port(skb, a);
 353                         break;
 354
 355                 case ODP_ACTION_ATTR_SET_PRIORITY:
 356                         skb->priority = nla_get_u32(a);
 357                         break;
 358
 359                 case ODP_ACTION_ATTR_POP_PRIORITY:
 360                         skb->priority = priority;
 361                         break;
 362
 363                 case ODP_ACTION_ATTR_DROP_SPOOFED_ARP:
 364                         if (unlikely(is_spoofed_arp(skb)))
 365                                 goto exit;
 366                         break;
 367                 }
 368                 if (!skb)
 369                         return -ENOMEM;
 370         }
 371 exit:
 372         if (prev_port != -1)
 373                 do_output(dp, skb, prev_port);
 374         else
 375                 kfree_skb(skb);
 376         return 0;
 377 }
 378
 379 static void sflow_sample(struct datapath *dp, struct sk_buff *skb,
 380                          struct sw_flow_actions *acts)
 381 {
 382         struct sk_buff *nskb;
 383         struct vport *p = OVS_CB(skb)->vport;
 384         struct dp_upcall_info upcall;
 385
 386         if (unlikely(!p))
 387                 return;
 388
 389         atomic_inc(&p->sflow_pool);
 390         if (net_random() >= dp->sflow_probability)
 391                 return;
 392
 393         nskb = skb_clone(skb, GFP_ATOMIC);
 394         if (unlikely(!nskb))
 395                 return;
 396
 397         upcall.cmd = ODP_PACKET_CMD_SAMPLE;
 398         upcall.key = &OVS_CB(skb)->flow->key;
 399         upcall.userdata = 0;
 400         upcall.sample_pool = atomic_read(&p->sflow_pool);
 401         upcall.actions = acts->actions;
 402         upcall.actions_len = acts->actions_len;
 403         dp_upcall(dp, nskb, &upcall);
 404 }
 405
 406 /* Execute a list of actions against 'skb'. */
 407 int execute_actions(struct datapath *dp, struct sk_buff *skb)
 408 {
 409         struct sw_flow_actions *acts = rcu_dereference(OVS_CB(skb)->flow->sf_acts);
 410         struct loop_counter *loop;
 411         int error;
 412
 413         /* Check whether we've looped too much. */
 414         loop = loop_get_counter();
 415         if (unlikely(++loop->count > MAX_LOOPS))
 416                 loop->looping = true;
 417         if (unlikely(loop->looping)) {
 418                 error = loop_suppress(dp, acts);
 419                 kfree_skb(skb);
 420                 goto out_loop;
 421         }
 422
 423         /* Really execute actions. */
 424         if (dp->sflow_probability)
 425                 sflow_sample(dp, skb, acts);
 426         OVS_CB(skb)->tun_id = 0;
 427         error = do_execute_actions(dp, skb, acts);
 428
 429         /* Check whether sub-actions looped too much. */
 430         if (unlikely(loop->looping))
 431                 error = loop_suppress(dp, acts);
 432
 433 out_loop:
 434         /* Decrement loop counter. */
 435         if (!--loop->count)
 436                 loop->looping = false;
 437         loop_put_counter();
 438
 439         return error;
 440 }