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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  12
  13 #include <linux/skbuff.h>
  14 #include <linux/in.h>
  15 #include <linux/ip.h>
  16 #include <linux/tcp.h>
  17 #include <linux/udp.h>
  18 #include <linux/in6.h>
  19 #include <linux/if_arp.h>
  20 #include <linux/if_vlan.h>
  21 #include <net/inet_ecn.h>
  22 #include <net/ip.h>
  23 #include <net/checksum.h>
  24
  25 #include "actions.h"
  26 #include "checksum.h"
  27 #include "datapath.h"
  28 #include "openvswitch/datapath-protocol.h"
  29 #include "vlan.h"
  30 #include "vport.h"
  31
  32 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
  33                         const struct nlattr *attr, int len, bool keep_skb);
  34
  35 static int make_writable(struct sk_buff *skb, int write_len)
  36 {
  37         if (!skb_cloned(skb) || skb_clone_writable(skb, write_len))
  38                 return 0;
  39
  40         return pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
  41 }
  42
  43 /* remove VLAN header from packet and update csum accrodingly. */
  44 static int __pop_vlan_tci(struct sk_buff *skb, __be16 *current_tci)
  45 {
  46         struct ethhdr *eh;
  47         struct vlan_ethhdr *veth;
  48         int err;
  49
  50         err = make_writable(skb, VLAN_ETH_HLEN);
  51         if (unlikely(err))
  52                 return err;
  53
  54         if (get_ip_summed(skb) == OVS_CSUM_COMPLETE)
  55                 skb->csum = csum_sub(skb->csum, csum_partial(skb->data
  56                                         + ETH_HLEN, VLAN_HLEN, 0));
  57
  58         veth = (struct vlan_ethhdr *) skb->data;
  59         *current_tci = veth->h_vlan_TCI;
  60
  61         memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN);
  62
  63         eh = (struct ethhdr *)__skb_pull(skb, VLAN_HLEN);
  64
  65         skb->protocol = eh->h_proto;
  66         skb->mac_header += VLAN_HLEN;
  67
  68         return 0;
  69 }
  70
  71 static int pop_vlan(struct sk_buff *skb)
  72 {
  73         __be16 tci;
  74         int err;
  75
  76         if (likely(vlan_tx_tag_present(skb))) {
  77                 vlan_set_tci(skb, 0);
  78         } else {
  79                 if (unlikely(skb->protocol != htons(ETH_P_8021Q) ||
  80                     skb->len < VLAN_ETH_HLEN))
  81                         return 0;
  82
  83                 err = __pop_vlan_tci(skb, &tci);
  84                 if (err)
  85                         return err;
  86         }
  87         /* move next vlan tag to hw accel tag */
  88         if (likely(skb->protocol != htons(ETH_P_8021Q) ||
  89             skb->len < VLAN_ETH_HLEN))
  90                 return 0;
  91
  92         err = __pop_vlan_tci(skb, &tci);
  93         if (unlikely(err))
  94                 return err;
  95
  96         __vlan_hwaccel_put_tag(skb, ntohs(tci));
  97         return 0;
  98 }
  99
 100 static int push_vlan(struct sk_buff *skb, __be16 new_tci)
 101 {
 102         if (unlikely(vlan_tx_tag_present(skb))) {
 103                 u16 current_tag;
 104
 105                 /* push down current VLAN tag */
 106                 current_tag = vlan_tx_tag_get(skb);
 107
 108                 if (!__vlan_put_tag(skb, current_tag))
 109                         return -ENOMEM;
 110
 111                 if (get_ip_summed(skb) == OVS_CSUM_COMPLETE)
 112                         skb->csum = csum_add(skb->csum, csum_partial(skb->data
 113                                         + ETH_HLEN, VLAN_HLEN, 0));
 114
 115         }
 116         __vlan_hwaccel_put_tag(skb, ntohs(new_tci));
 117         return 0;
 118 }
 119
 120 static bool is_ip(struct sk_buff *skb)
 121 {
 122         return (OVS_CB(skb)->flow->key.eth.type == htons(ETH_P_IP) &&
 123                 skb->transport_header > skb->network_header);
 124 }
 125
 126 static __sum16 *get_l4_checksum(struct sk_buff *skb)
 127 {
 128         u8 nw_proto = OVS_CB(skb)->flow->key.ip.proto;
 129         int transport_len = skb->len - skb_transport_offset(skb);
 130         if (nw_proto == IPPROTO_TCP) {
 131                 if (likely(transport_len >= sizeof(struct tcphdr)))
 132                         return &tcp_hdr(skb)->check;
 133         } else if (nw_proto == IPPROTO_UDP) {
 134                 if (likely(transport_len >= sizeof(struct udphdr)))
 135                         return &udp_hdr(skb)->check;
 136         }
 137         return NULL;
 138 }
 139
 140 static int set_nw_addr(struct sk_buff *skb, const struct nlattr *a)
 141 {
 142         __be32 new_nwaddr = nla_get_be32(a);
 143         struct iphdr *nh;
 144         __sum16 *check;
 145         __be32 *nwaddr;
 146         int err;
 147
 148         if (unlikely(!is_ip(skb)))
 149                 return 0;
 150
 151         err = make_writable(skb, skb_network_offset(skb) +
 152                                  sizeof(struct iphdr));
 153         if (unlikely(err))
 154                 return err;
 155
 156         nh = ip_hdr(skb);
 157         nwaddr = nla_type(a) == OVS_ACTION_ATTR_SET_NW_SRC ? &nh->saddr : &nh->daddr;
 158
 159         check = get_l4_checksum(skb);
 160         if (likely(check))
 161                 inet_proto_csum_replace4(check, skb, *nwaddr, new_nwaddr, 1);
 162         csum_replace4(&nh->check, *nwaddr, new_nwaddr);
 163
 164         skb_clear_rxhash(skb);
 165
 166         *nwaddr = new_nwaddr;
 167
 168         return 0;
 169 }
 170
 171 static int set_nw_tos(struct sk_buff *skb, u8 nw_tos)
 172 {
 173         struct iphdr *nh = ip_hdr(skb);
 174         u8 old, new;
 175         int err;
 176
 177         if (unlikely(!is_ip(skb)))
 178                 return 0;
 179
 180         err = make_writable(skb, skb_network_offset(skb) +
 181                                  sizeof(struct iphdr));
 182         if (unlikely(err))
 183                 return err;
 184
 185         /* Set the DSCP bits and preserve the ECN bits. */
 186         old = nh->tos;
 187         new = nw_tos | (nh->tos & INET_ECN_MASK);
 188         csum_replace4(&nh->check, (__force __be32)old,
 189                                   (__force __be32)new);
 190         nh->tos = new;
 191
 192         return 0;
 193 }
 194
 195 static int set_tp_port(struct sk_buff *skb, const struct nlattr *a)
 196 {
 197         struct udphdr *th;
 198         __sum16 *check;
 199         __be16 *port;
 200         int err;
 201
 202         if (unlikely(!is_ip(skb)))
 203                 return 0;
 204
 205         err = make_writable(skb, skb_transport_offset(skb) +
 206                                  sizeof(struct tcphdr));
 207         if (unlikely(err))
 208                 return err;
 209
 210         /* Must follow make_writable() since that can move the skb data. */
 211         check = get_l4_checksum(skb);
 212         if (unlikely(!check))
 213                 return 0;
 214
 215         /*
 216          * Update port and checksum.
 217          *
 218          * This is OK because source and destination port numbers are at the
 219          * same offsets in both UDP and TCP headers, and get_l4_checksum() only
 220          * supports those protocols.
 221          */
 222         th = udp_hdr(skb);
 223         port = nla_type(a) == OVS_ACTION_ATTR_SET_TP_SRC ? &th->source : &th->dest;
 224         inet_proto_csum_replace2(check, skb, *port, nla_get_be16(a), 0);
 225         *port = nla_get_be16(a);
 226         skb_clear_rxhash(skb);
 227
 228         return 0;
 229 }
 230
 231 static int do_output(struct datapath *dp, struct sk_buff *skb, int out_port)
 232 {
 233         struct vport *vport;
 234
 235         if (unlikely(!skb))
 236                 return -ENOMEM;
 237
 238         vport = rcu_dereference(dp->ports[out_port]);
 239         if (unlikely(!vport)) {
 240                 kfree_skb(skb);
 241                 return -ENODEV;
 242         }
 243
 244         vport_send(vport, skb);
 245         return 0;
 246 }
 247
 248 static int output_userspace(struct datapath *dp, struct sk_buff *skb,
 249                             const struct nlattr *attr)
 250 {
 251         struct dp_upcall_info upcall;
 252         const struct nlattr *a;
 253         int rem;
 254
 255         upcall.cmd = OVS_PACKET_CMD_ACTION;
 256         upcall.key = &OVS_CB(skb)->flow->key;
 257         upcall.userdata = NULL;
 258         upcall.pid = 0;
 259
 260         for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
 261                  a = nla_next(a, &rem)) {
 262                 switch (nla_type(a)) {
 263                 case OVS_USERSPACE_ATTR_USERDATA:
 264                         upcall.userdata = a;
 265                         break;
 266
 267                 case OVS_USERSPACE_ATTR_PID:
 268                         upcall.pid = nla_get_u32(a);
 269                         break;
 270                 }
 271         }
 272
 273         return dp_upcall(dp, skb, &upcall);
 274 }
 275
 276 static int sample(struct datapath *dp, struct sk_buff *skb,
 277                   const struct nlattr *attr)
 278 {
 279         const struct nlattr *acts_list = NULL;
 280         const struct nlattr *a;
 281         int rem;
 282
 283         for (a = nla_data(attr), rem = nla_len(attr); rem > 0;
 284                  a = nla_next(a, &rem)) {
 285                 switch (nla_type(a)) {
 286                 case OVS_SAMPLE_ATTR_PROBABILITY:
 287                         if (net_random() >= nla_get_u32(a))
 288                                 return 0;
 289                         break;
 290
 291                 case OVS_SAMPLE_ATTR_ACTIONS:
 292                         acts_list = a;
 293                         break;
 294                 }
 295         }
 296
 297         return do_execute_actions(dp, skb, nla_data(acts_list),
 298                                                  nla_len(acts_list), true);
 299 }
 300
 301 /* Execute a list of actions against 'skb'. */
 302 static int do_execute_actions(struct datapath *dp, struct sk_buff *skb,
 303                         const struct nlattr *attr, int len, bool keep_skb)
 304 {
 305         /* Every output action needs a separate clone of 'skb', but the common
 306          * case is just a single output action, so that doing a clone and
 307          * then freeing the original skbuff is wasteful.  So the following code
 308          * is slightly obscure just to avoid that. */
 309         int prev_port = -1;
 310         u32 priority = skb->priority;
 311         const struct nlattr *a;
 312         int rem;
 313
 314         for (a = attr, rem = len; rem > 0;
 315              a = nla_next(a, &rem)) {
 316                 int err = 0;
 317
 318                 if (prev_port != -1) {
 319                         do_output(dp, skb_clone(skb, GFP_ATOMIC), prev_port);
 320                         prev_port = -1;
 321                 }
 322
 323                 switch (nla_type(a)) {
 324                 case OVS_ACTION_ATTR_OUTPUT:
 325                         prev_port = nla_get_u32(a);
 326                         break;
 327
 328                 case OVS_ACTION_ATTR_USERSPACE:
 329                         output_userspace(dp, skb, a);
 330                         break;
 331
 332                 case OVS_ACTION_ATTR_SET_TUNNEL:
 333                         OVS_CB(skb)->tun_id = nla_get_be64(a);
 334                         break;
 335
 336                 case OVS_ACTION_ATTR_PUSH_VLAN:
 337                         err = push_vlan(skb, nla_get_be16(a));
 338                         if (unlikely(err)) /* skb already freed */
 339                                 return err;
 340                         break;
 341
 342                 case OVS_ACTION_ATTR_POP_VLAN:
 343                         err = pop_vlan(skb);
 344                         break;
 345
 346                 case OVS_ACTION_ATTR_SET_DL_SRC:
 347                         err = make_writable(skb, ETH_HLEN);
 348                         if (likely(!err))
 349                                 memcpy(eth_hdr(skb)->h_source, nla_data(a), ETH_ALEN);
 350                         break;
 351
 352                 case OVS_ACTION_ATTR_SET_DL_DST:
 353                         err = make_writable(skb, ETH_HLEN);
 354                         if (likely(!err))
 355                                 memcpy(eth_hdr(skb)->h_dest, nla_data(a), ETH_ALEN);
 356                         break;
 357
 358                 case OVS_ACTION_ATTR_SET_NW_SRC:
 359                 case OVS_ACTION_ATTR_SET_NW_DST:
 360                         err = set_nw_addr(skb, a);
 361                         break;
 362
 363                 case OVS_ACTION_ATTR_SET_NW_TOS:
 364                         err = set_nw_tos(skb, nla_get_u8(a));
 365                         break;
 366
 367                 case OVS_ACTION_ATTR_SET_TP_SRC:
 368                 case OVS_ACTION_ATTR_SET_TP_DST:
 369                         err = set_tp_port(skb, a);
 370                         break;
 371
 372                 case OVS_ACTION_ATTR_SET_PRIORITY:
 373                         skb->priority = nla_get_u32(a);
 374                         break;
 375
 376                 case OVS_ACTION_ATTR_POP_PRIORITY:
 377                         skb->priority = priority;
 378                         break;
 379
 380                 case OVS_ACTION_ATTR_SAMPLE:
 381                         err = sample(dp, skb, a);
 382                         break;
 383
 384                 }
 385                 if (unlikely(err)) {
 386                         kfree_skb(skb);
 387                         return err;
 388                 }
 389         }
 390
 391         if (prev_port != -1) {
 392                 if (keep_skb)
 393                         skb = skb_clone(skb, GFP_ATOMIC);
 394
 395                 do_output(dp, skb, prev_port);
 396         } else if (!keep_skb)
 397                 consume_skb(skb);
 398
 399         return 0;
 400 }
 401
 402 /* We limit the number of times that we pass into execute_actions()
 403  * to avoid blowing out the stack in the event that we have a loop. */
 404 #define MAX_LOOPS 5
 405
 406 struct loop_counter {
 407         u8 count;               /* Count. */
 408         bool looping;           /* Loop detected? */
 409 };
 410
 411 static DEFINE_PER_CPU(struct loop_counter, loop_counters);
 412
 413 static int loop_suppress(struct datapath *dp, struct sw_flow_actions *actions)
 414 {
 415         if (net_ratelimit())
 416                 pr_warn("%s: flow looped %d times, dropping\n",
 417                                 dp_name(dp), MAX_LOOPS);
 418         actions->actions_len = 0;
 419         return -ELOOP;
 420 }
 421
 422 /* Execute a list of actions against 'skb'. */
 423 int execute_actions(struct datapath *dp, struct sk_buff *skb)
 424 {
 425         struct sw_flow_actions *acts = rcu_dereference(OVS_CB(skb)->flow->sf_acts);
 426         struct loop_counter *loop;
 427         int error;
 428
 429         /* Check whether we've looped too much. */
 430         loop = &__get_cpu_var(loop_counters);
 431         if (unlikely(++loop->count > MAX_LOOPS))
 432                 loop->looping = true;
 433         if (unlikely(loop->looping)) {
 434                 error = loop_suppress(dp, acts);
 435                 kfree_skb(skb);
 436                 goto out_loop;
 437         }
 438
 439         OVS_CB(skb)->tun_id = 0;
 440         error = do_execute_actions(dp, skb, acts->actions,
 441                                          acts->actions_len, false);
 442
 443         /* Check whether sub-actions looped too much. */
 444         if (unlikely(loop->looping))
 445                 error = loop_suppress(dp, acts);
 446
 447 out_loop:
 448         /* Decrement loop counter. */
 449         if (!--loop->count)
 450                 loop->looping = false;
 451
 452         return error;
 453 }