datapath/forward.c

   1 /*
   2  * Distributed under the terms of the GNU GPL version 2.
   3  * Copyright (c) 2007, 2008 The Board of Trustees of The Leland
   4  * Stanford Junior University
   5  */
   6
   7 #include <linux/netdevice.h>
   8 #include <linux/etherdevice.h>
   9 #include <linux/if_ether.h>
  10 #include <linux/if_vlan.h>
  11 #include <asm/uaccess.h>
  12 #include <linux/types.h>
  13 #include "forward.h"
  14 #include "datapath.h"
  15 #include "openflow/nicira-ext.h"
  16 #include "dp_act.h"
  17 #include "nx_msg.h"
  18 #include "chain.h"
  19 #include "flow.h"
  20
  21 /* FIXME: do we need to use GFP_ATOMIC everywhere here? */
  22
  23
  24 static struct sk_buff *retrieve_skb(uint32_t id);
  25 static void discard_skb(uint32_t id);
  26
  27 /* 'skb' was received on port 'p', which may be a physical switch port, the
  28  * local port, or a null pointer.  Process it according to 'chain'.  Returns 0
  29  * if successful, in which case 'skb' is destroyed, or -ESRCH if there is no
  30  * matching flow, in which case 'skb' still belongs to the caller. */
  31 int run_flow_through_tables(struct sw_chain *chain, struct sk_buff *skb,
  32                             struct net_bridge_port *p)
  33 {
  34         /* Ethernet address used as the destination for STP frames. */
  35         static const uint8_t stp_eth_addr[ETH_ALEN]
  36                 = { 0x01, 0x80, 0xC2, 0x00, 0x00, 0x01 };
  37         struct sw_flow_key key;
  38         struct sw_flow *flow;
  39
  40         if (flow_extract(skb, p ? p->port_no : OFPP_NONE, &key)
  41             && (chain->dp->flags & OFPC_FRAG_MASK) == OFPC_FRAG_DROP) {
  42                 /* Drop fragment. */
  43                 kfree_skb(skb);
  44                 return 0;
  45         }
  46         if (p && p->config & (OFPPC_NO_RECV | OFPPC_NO_RECV_STP) &&
  47             p->config & (compare_ether_addr(key.dl_dst, stp_eth_addr)
  48                         ? OFPPC_NO_RECV : OFPPC_NO_RECV_STP)) {
  49                 kfree_skb(skb);
  50                 return 0;
  51         }
  52
  53         flow = chain_lookup(chain, &key);
  54         if (likely(flow != NULL)) {
  55                 struct sw_flow_actions *sf_acts = rcu_dereference(flow->sf_acts);
  56                 flow_used(flow, skb);
  57                 execute_actions(chain->dp, skb, &key,
  58                                 sf_acts->actions, sf_acts->actions_len, 0);
  59                 return 0;
  60         } else {
  61                 return -ESRCH;
  62         }
  63 }
  64
  65 /* 'skb' was received on port 'p', which may be a physical switch port, the
  66  * local port, or a null pointer.  Process it according to 'chain', sending it
  67  * up to the controller if no flow matches.  Takes ownership of 'skb'. */
  68 void fwd_port_input(struct sw_chain *chain, struct sk_buff *skb,
  69                     struct net_bridge_port *p)
  70 {
  71         if (run_flow_through_tables(chain, skb, p))
  72                 dp_output_control(chain->dp, skb, fwd_save_skb(skb),
  73                                   chain->dp->miss_send_len,
  74                                   OFPR_NO_MATCH);
  75 }
  76
  77 static int
  78 recv_hello(struct sw_chain *chain, const struct sender *sender,
  79            const void *msg)
  80 {
  81         return dp_send_hello(chain->dp, sender, msg);
  82 }
  83
  84 static int
  85 recv_features_request(struct sw_chain *chain, const struct sender *sender,
  86                       const void *msg)
  87 {
  88         return dp_send_features_reply(chain->dp, sender);
  89 }
  90
  91 static int
  92 recv_get_config_request(struct sw_chain *chain, const struct sender *sender,
  93                         const void *msg)
  94 {
  95         return dp_send_config_reply(chain->dp, sender);
  96 }
  97
  98 static int
  99 recv_set_config(struct sw_chain *chain, const struct sender *sender,
 100                 const void *msg)
 101 {
 102         const struct ofp_switch_config *osc = msg;
 103         int flags;
 104
 105         flags = ntohs(osc->flags) & (OFPC_SEND_FLOW_EXP | OFPC_FRAG_MASK);
 106         if ((flags & OFPC_FRAG_MASK) != OFPC_FRAG_NORMAL
 107             && (flags & OFPC_FRAG_MASK) != OFPC_FRAG_DROP) {
 108                 flags = (flags & ~OFPC_FRAG_MASK) | OFPC_FRAG_DROP;
 109         }
 110         chain->dp->flags = flags;
 111
 112         chain->dp->miss_send_len = ntohs(osc->miss_send_len);
 113
 114         return 0;
 115 }
 116
 117 static int
 118 recv_packet_out(struct sw_chain *chain, const struct sender *sender,
 119                 const void *msg)
 120 {
 121         const struct ofp_packet_out *opo = msg;
 122         struct sk_buff *skb;
 123         struct vlan_ethhdr *mac;
 124         int nh_ofs;
 125         uint16_t v_code;
 126         struct sw_flow_key key;
 127         size_t actions_len = ntohs(opo->actions_len);
 128
 129         if (actions_len > (ntohs(opo->header.length) - sizeof *opo)) {
 130                 if (net_ratelimit())
 131                         printk("message too short for number of actions\n");
 132                 return -EINVAL;
 133         }
 134
 135         if (ntohl(opo->buffer_id) == (uint32_t) -1) {
 136                 int data_len = ntohs(opo->header.length) - sizeof *opo - actions_len;
 137
 138                 /* FIXME: there is likely a way to reuse the data in msg. */
 139                 skb = alloc_skb(data_len, GFP_ATOMIC);
 140                 if (!skb)
 141                         return -ENOMEM;
 142
 143                 /* FIXME?  We don't reserve NET_IP_ALIGN or NET_SKB_PAD since
 144                  * we're just transmitting this raw without examining anything
 145                  * at those layers. */
 146                 memcpy(skb_put(skb, data_len), (uint8_t *)opo->actions + actions_len,
 147                                 data_len);
 148
 149                 skb_set_mac_header(skb, 0);
 150                 mac = vlan_eth_hdr(skb);
 151                 if (likely(mac->h_vlan_proto != htons(ETH_P_8021Q)))
 152                         nh_ofs = sizeof(struct ethhdr);
 153                 else
 154                         nh_ofs = sizeof(struct vlan_ethhdr);
 155                 skb_set_network_header(skb, nh_ofs);
 156         } else {
 157                 skb = retrieve_skb(ntohl(opo->buffer_id));
 158                 if (!skb)
 159                         return -ESRCH;
 160         }
 161
 162         dp_set_origin(chain->dp, ntohs(opo->in_port), skb);
 163
 164         flow_extract(skb, ntohs(opo->in_port), &key);
 165
 166         v_code = validate_actions(chain->dp, &key, opo->actions, actions_len);
 167         if (v_code != ACT_VALIDATION_OK) {
 168                 dp_send_error_msg(chain->dp, sender, OFPET_BAD_ACTION, v_code,
 169                                   msg, ntohs(opo->header.length));
 170                 goto error;
 171         }
 172
 173         execute_actions(chain->dp, skb, &key, opo->actions, actions_len, 1);
 174
 175         return 0;
 176
 177 error:
 178         kfree_skb(skb);
 179         return -EINVAL;
 180 }
 181
 182 static int
 183 recv_port_mod(struct sw_chain *chain, const struct sender *sender,
 184               const void *msg)
 185 {
 186         const struct ofp_port_mod *opm = msg;
 187
 188         dp_update_port_flags(chain->dp, opm);
 189
 190         return 0;
 191 }
 192
 193 static int
 194 recv_echo_request(struct sw_chain *chain, const struct sender *sender,
 195                   const void *msg)
 196 {
 197         return dp_send_echo_reply(chain->dp, sender, msg);
 198 }
 199
 200 static int
 201 recv_echo_reply(struct sw_chain *chain, const struct sender *sender,
 202                   const void *msg)
 203 {
 204         return 0;
 205 }
 206
 207 static int
 208 add_flow(struct sw_chain *chain, const struct sender *sender,
 209                 const struct ofp_flow_mod *ofm)
 210 {
 211         int error = -ENOMEM;
 212         uint16_t v_code;
 213         struct sw_flow *flow;
 214         size_t actions_len = ntohs(ofm->header.length) - sizeof *ofm;
 215
 216         /* Allocate memory. */
 217         flow = flow_alloc(actions_len, GFP_ATOMIC);
 218         if (flow == NULL)
 219                 goto error;
 220
 221         flow_extract_match(&flow->key, &ofm->match);
 222
 223         v_code = validate_actions(chain->dp, &flow->key, ofm->actions, actions_len);
 224         if (v_code != ACT_VALIDATION_OK) {
 225                 dp_send_error_msg(chain->dp, sender, OFPET_BAD_ACTION, v_code,
 226                                   ofm, ntohs(ofm->header.length));
 227                 goto error_free_flow;
 228         }
 229
 230         /* Fill out flow. */
 231         flow->priority = flow->key.wildcards ? ntohs(ofm->priority) : -1;
 232         flow->idle_timeout = ntohs(ofm->idle_timeout);
 233         flow->hard_timeout = ntohs(ofm->hard_timeout);
 234         flow->used = jiffies;
 235         flow->init_time = jiffies;
 236         flow->byte_count = 0;
 237         flow->packet_count = 0;
 238         spin_lock_init(&flow->lock);
 239         memcpy(flow->sf_acts->actions, ofm->actions, actions_len);
 240
 241         /* Act. */
 242         error = chain_insert(chain, flow);
 243         if (error == -ENOBUFS) {
 244                 dp_send_error_msg(chain->dp, sender, OFPET_FLOW_MOD_FAILED,
 245                                 OFPFMFC_ALL_TABLES_FULL, ofm, ntohs(ofm->header.length));
 246                 goto error_free_flow;
 247         } else if (error)
 248                 goto error_free_flow;
 249         error = 0;
 250         if (ntohl(ofm->buffer_id) != (uint32_t) -1) {
 251                 struct sk_buff *skb = retrieve_skb(ntohl(ofm->buffer_id));
 252                 if (skb) {
 253                         struct sw_flow_key key;
 254                         flow_used(flow, skb);
 255                         dp_set_origin(chain->dp, ntohs(ofm->match.in_port), skb);
 256                         flow_extract(skb, ntohs(ofm->match.in_port), &key);
 257                         execute_actions(chain->dp, skb, &key, ofm->actions, actions_len, 0);
 258                 }
 259                 else
 260                         error = -ESRCH;
 261         }
 262         return error;
 263
 264 error_free_flow:
 265         flow_free(flow);
 266 error:
 267         if (ntohl(ofm->buffer_id) != (uint32_t) -1)
 268                 discard_skb(ntohl(ofm->buffer_id));
 269         return error;
 270 }
 271
 272 static int
 273 mod_flow(struct sw_chain *chain, const struct sender *sender,
 274                 const struct ofp_flow_mod *ofm)
 275 {
 276         int error = -ENOMEM;
 277         uint16_t v_code;
 278         size_t actions_len;
 279         struct sw_flow_key key;
 280         uint16_t priority;
 281         int strict;
 282
 283         flow_extract_match(&key, &ofm->match);
 284
 285         actions_len = ntohs(ofm->header.length) - sizeof *ofm;
 286
 287         v_code = validate_actions(chain->dp, &key, ofm->actions, actions_len);
 288         if (v_code != ACT_VALIDATION_OK) {
 289                 dp_send_error_msg(chain->dp, sender, OFPET_BAD_ACTION, v_code,
 290                                   ofm, ntohs(ofm->header.length));
 291                 goto error;
 292         }
 293
 294         priority = key.wildcards ? ntohs(ofm->priority) : -1;
 295         strict = (ofm->command == htons(OFPFC_MODIFY_STRICT)) ? 1 : 0;
 296         chain_modify(chain, &key, priority, strict, ofm->actions, actions_len);
 297
 298         if (ntohl(ofm->buffer_id) != (uint32_t) -1) {
 299                 struct sk_buff *skb = retrieve_skb(ntohl(ofm->buffer_id));
 300                 if (skb) {
 301                         struct sw_flow_key skb_key;
 302                         flow_extract(skb, ntohs(ofm->match.in_port), &skb_key);
 303                         execute_actions(chain->dp, skb, &skb_key,
 304                                         ofm->actions, actions_len, 0);
 305                 }
 306                 else
 307                         error = -ESRCH;
 308         }
 309         return error;
 310
 311 error:
 312         if (ntohl(ofm->buffer_id) != (uint32_t) -1)
 313                 discard_skb(ntohl(ofm->buffer_id));
 314         return error;
 315 }
 316
 317 static int
 318 recv_flow(struct sw_chain *chain, const struct sender *sender, const void *msg)
 319 {
 320         const struct ofp_flow_mod *ofm = msg;
 321         uint16_t command = ntohs(ofm->command);
 322
 323         if (command == OFPFC_ADD) {
 324                 return add_flow(chain, sender, ofm);
 325         } else if ((command == OFPFC_MODIFY) || (command == OFPFC_MODIFY_STRICT)) {
 326                 return mod_flow(chain, sender, ofm);
 327         }  else if (command == OFPFC_DELETE) {
 328                 struct sw_flow_key key;
 329                 flow_extract_match(&key, &ofm->match);
 330                 return chain_delete(chain, &key, 0, 0) ? 0 : -ESRCH;
 331         } else if (command == OFPFC_DELETE_STRICT) {
 332                 struct sw_flow_key key;
 333                 uint16_t priority;
 334                 flow_extract_match(&key, &ofm->match);
 335                 priority = key.wildcards ? ntohs(ofm->priority) : -1;
 336                 return chain_delete(chain, &key, priority, 1) ? 0 : -ESRCH;
 337         } else {
 338                 return -ENOTSUPP;
 339         }
 340 }
 341
 342 static int
 343 recv_vendor(struct sw_chain *chain, const struct sender *sender,
 344                 const void *msg)
 345 {
 346         const struct ofp_vendor_header *ovh = msg;
 347
 348         switch(ntohl(ovh->vendor))
 349         {
 350         case NX_VENDOR_ID:
 351                 return nx_recv_msg(chain, sender, msg);
 352         default:
 353                 if (net_ratelimit())
 354                         printk("Uknown vendor: %#x\n", ntohl(ovh->vendor));
 355                 dp_send_error_msg(chain->dp, sender, OFPET_BAD_REQUEST,
 356                                   OFPBRC_BAD_VENDOR, msg, ntohs(ovh->header.length));
 357                 return -EINVAL;
 358         }
 359 }
 360
 361 /* 'msg', which is 'length' bytes long, was received across Netlink from
 362  * 'sender'.  Apply it to 'chain'. */
 363 int
 364 fwd_control_input(struct sw_chain *chain, const struct sender *sender,
 365                   const void *msg, size_t length)
 366 {
 367
 368         struct openflow_packet {
 369                 size_t min_size;
 370                 int (*handler)(struct sw_chain *, const struct sender *,
 371                                const void *);
 372         };
 373
 374         static const struct openflow_packet packets[] = {
 375                 [OFPT_HELLO] = {
 376                         sizeof (struct ofp_header),
 377                         recv_hello,
 378                 },
 379                 [OFPT_ECHO_REQUEST] = {
 380                         sizeof (struct ofp_header),
 381                         recv_echo_request,
 382                 },
 383                 [OFPT_ECHO_REPLY] = {
 384                         sizeof (struct ofp_header),
 385                         recv_echo_reply,
 386                 },
 387                 [OFPT_VENDOR] = {
 388                         sizeof (struct ofp_vendor_header),
 389                         recv_vendor,
 390                 },
 391                 [OFPT_FEATURES_REQUEST] = {
 392                         sizeof (struct ofp_header),
 393                         recv_features_request,
 394                 },
 395                 [OFPT_GET_CONFIG_REQUEST] = {
 396                         sizeof (struct ofp_header),
 397                         recv_get_config_request,
 398                 },
 399                 [OFPT_SET_CONFIG] = {
 400                         sizeof (struct ofp_switch_config),
 401                         recv_set_config,
 402                 },
 403                 [OFPT_PACKET_OUT] = {
 404                         sizeof (struct ofp_packet_out),
 405                         recv_packet_out,
 406                 },
 407                 [OFPT_FLOW_MOD] = {
 408                         sizeof (struct ofp_flow_mod),
 409                         recv_flow,
 410                 },
 411                 [OFPT_PORT_MOD] = {
 412                         sizeof (struct ofp_port_mod),
 413                         recv_port_mod,
 414                 }
 415         };
 416
 417         struct ofp_header *oh;
 418
 419         oh = (struct ofp_header *) msg;
 420         if (oh->version != OFP_VERSION
 421             && oh->type != OFPT_HELLO
 422             && oh->type != OFPT_ERROR
 423             && oh->type != OFPT_ECHO_REQUEST
 424             && oh->type != OFPT_ECHO_REPLY
 425             && oh->type != OFPT_VENDOR)
 426         {
 427                 dp_send_error_msg(chain->dp, sender, OFPET_BAD_REQUEST,
 428                                   OFPBRC_BAD_VERSION, msg, length);
 429                 return -EINVAL;
 430         }
 431         if (ntohs(oh->length) != length) {
 432                 if (net_ratelimit())
 433                         printk("received message length wrong: %d/%d\n",
 434                                 ntohs(oh->length), length);
 435                 return -EINVAL;
 436         }
 437
 438         if (oh->type < ARRAY_SIZE(packets)) {
 439                 const struct openflow_packet *pkt = &packets[oh->type];
 440                 if (pkt->handler) {
 441                         if (length < pkt->min_size)
 442                                 return -EFAULT;
 443                         return pkt->handler(chain, sender, msg);
 444                 }
 445         }
 446         dp_send_error_msg(chain->dp, sender, OFPET_BAD_REQUEST,
 447                           OFPBRC_BAD_TYPE, msg, length);
 448         return -EINVAL;
 449 }
 450
 451 /* Packet buffering. */
 452
 453 #define OVERWRITE_SECS  1
 454 #define OVERWRITE_JIFFIES (OVERWRITE_SECS * HZ)
 455
 456 struct packet_buffer {
 457         struct sk_buff *skb;
 458         uint32_t cookie;
 459         unsigned long exp_jiffies;
 460 };
 461
 462 static struct packet_buffer buffers[N_PKT_BUFFERS];
 463 static unsigned int buffer_idx;
 464 static DEFINE_SPINLOCK(buffer_lock);
 465
 466 uint32_t fwd_save_skb(struct sk_buff *skb)
 467 {
 468         struct sk_buff *old_skb = NULL;
 469         struct packet_buffer *p;
 470         unsigned long int flags;
 471         uint32_t id;
 472
 473         spin_lock_irqsave(&buffer_lock, flags);
 474         buffer_idx = (buffer_idx + 1) & PKT_BUFFER_MASK;
 475         p = &buffers[buffer_idx];
 476         if (p->skb) {
 477                 /* Don't buffer packet if existing entry is less than
 478                  * OVERWRITE_SECS old. */
 479                 if (time_before(jiffies, p->exp_jiffies)) {
 480                         spin_unlock_irqrestore(&buffer_lock, flags);
 481                         return -1;
 482                 } else {
 483                         /* Defer kfree_skb() until interrupts re-enabled. */
 484                         old_skb = p->skb;
 485                 }
 486         }
 487         /* Don't use maximum cookie value since the all-bits-1 id is
 488          * special. */
 489         if (++p->cookie >= (1u << PKT_COOKIE_BITS) - 1)
 490                 p->cookie = 0;
 491         skb_get(skb);
 492         p->skb = skb;
 493         p->exp_jiffies = jiffies + OVERWRITE_JIFFIES;
 494         id = buffer_idx | (p->cookie << PKT_BUFFER_BITS);
 495         spin_unlock_irqrestore(&buffer_lock, flags);
 496
 497         if (old_skb)
 498                 kfree_skb(old_skb);
 499
 500         return id;
 501 }
 502
 503 static struct sk_buff *retrieve_skb(uint32_t id)
 504 {
 505         unsigned long int flags;
 506         struct sk_buff *skb = NULL;
 507         struct packet_buffer *p;
 508
 509         spin_lock_irqsave(&buffer_lock, flags);
 510         p = &buffers[id & PKT_BUFFER_MASK];
 511         if (p->cookie == id >> PKT_BUFFER_BITS) {
 512                 skb = p->skb;
 513                 p->skb = NULL;
 514         } else {
 515                 printk("cookie mismatch: %x != %x\n",
 516                                 id >> PKT_BUFFER_BITS, p->cookie);
 517         }
 518         spin_unlock_irqrestore(&buffer_lock, flags);
 519
 520         return skb;
 521 }
 522
 523 void fwd_discard_all(void)
 524 {
 525         int i;
 526
 527         for (i = 0; i < N_PKT_BUFFERS; i++) {
 528                 struct sk_buff *skb;
 529                 unsigned long int flags;
 530
 531                 /* Defer kfree_skb() until interrupts re-enabled. */
 532                 spin_lock_irqsave(&buffer_lock, flags);
 533                 skb = buffers[i].skb;
 534                 buffers[i].skb = NULL;
 535                 spin_unlock_irqrestore(&buffer_lock, flags);
 536
 537                 kfree_skb(skb);
 538         }
 539 }
 540
 541 static void discard_skb(uint32_t id)
 542 {
 543         struct sk_buff *old_skb = NULL;
 544         unsigned long int flags;
 545         struct packet_buffer *p;
 546
 547         spin_lock_irqsave(&buffer_lock, flags);
 548         p = &buffers[id & PKT_BUFFER_MASK];
 549         if (p->cookie == id >> PKT_BUFFER_BITS) {
 550                 /* Defer kfree_skb() until interrupts re-enabled. */
 551                 old_skb = p->skb;
 552                 p->skb = NULL;
 553         }
 554         spin_unlock_irqrestore(&buffer_lock, flags);
 555
 556         if (old_skb)
 557                 kfree_skb(old_skb);
 558 }
 559
 560 void fwd_exit(void)
 561 {
 562         fwd_discard_all();
 563 }