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         WARN_ON_ONCE(skb_shared(skb));
  72         WARN_ON_ONCE(skb->destructor);
  73         if (run_flow_through_tables(chain, skb, p))
  74                 dp_output_control(chain->dp, skb, fwd_save_skb(skb),
  75                                   chain->dp->miss_send_len,
  76                                   OFPR_NO_MATCH);
  77 }
  78
  79 static int
  80 recv_hello(struct sw_chain *chain, const struct sender *sender,
  81            const void *msg)
  82 {
  83         return dp_send_hello(chain->dp, sender, msg);
  84 }
  85
  86 static int
  87 recv_features_request(struct sw_chain *chain, const struct sender *sender,
  88                       const void *msg)
  89 {
  90         return dp_send_features_reply(chain->dp, sender);
  91 }
  92
  93 static int
  94 recv_get_config_request(struct sw_chain *chain, const struct sender *sender,
  95                         const void *msg)
  96 {
  97         return dp_send_config_reply(chain->dp, sender);
  98 }
  99
 100 static int
 101 recv_set_config(struct sw_chain *chain, const struct sender *sender,
 102                 const void *msg)
 103 {
 104         const struct ofp_switch_config *osc = msg;
 105         int flags;
 106
 107         flags = ntohs(osc->flags) & (OFPC_SEND_FLOW_EXP | OFPC_FRAG_MASK);
 108         if ((flags & OFPC_FRAG_MASK) != OFPC_FRAG_NORMAL
 109             && (flags & OFPC_FRAG_MASK) != OFPC_FRAG_DROP) {
 110                 flags = (flags & ~OFPC_FRAG_MASK) | OFPC_FRAG_DROP;
 111         }
 112         chain->dp->flags = flags;
 113
 114         chain->dp->miss_send_len = ntohs(osc->miss_send_len);
 115
 116         return 0;
 117 }
 118
 119 static int
 120 recv_packet_out(struct sw_chain *chain, const struct sender *sender,
 121                 const void *msg)
 122 {
 123         const struct ofp_packet_out *opo = msg;
 124         struct sk_buff *skb;
 125         struct vlan_ethhdr *mac;
 126         int nh_ofs;
 127         uint16_t v_code;
 128         struct sw_flow_key key;
 129         size_t actions_len = ntohs(opo->actions_len);
 130
 131         if (actions_len > (ntohs(opo->header.length) - sizeof *opo)) {
 132                 if (net_ratelimit())
 133                         printk("message too short for number of actions\n");
 134                 return -EINVAL;
 135         }
 136
 137         if (ntohl(opo->buffer_id) == (uint32_t) -1) {
 138                 int data_len = ntohs(opo->header.length) - sizeof *opo - actions_len;
 139
 140                 /* FIXME: there is likely a way to reuse the data in msg. */
 141                 skb = alloc_skb(data_len, GFP_ATOMIC);
 142                 if (!skb)
 143                         return -ENOMEM;
 144
 145                 /* FIXME?  We don't reserve NET_IP_ALIGN or NET_SKB_PAD since
 146                  * we're just transmitting this raw without examining anything
 147                  * at those layers. */
 148                 memcpy(skb_put(skb, data_len), (uint8_t *)opo->actions + actions_len,
 149                                 data_len);
 150
 151                 skb_set_mac_header(skb, 0);
 152                 mac = vlan_eth_hdr(skb);
 153                 if (likely(mac->h_vlan_proto != htons(ETH_P_8021Q)))
 154                         nh_ofs = sizeof(struct ethhdr);
 155                 else
 156                         nh_ofs = sizeof(struct vlan_ethhdr);
 157                 skb_set_network_header(skb, nh_ofs);
 158         } else {
 159                 skb = retrieve_skb(ntohl(opo->buffer_id));
 160                 if (!skb)
 161                         return -ESRCH;
 162         }
 163
 164         dp_set_origin(chain->dp, ntohs(opo->in_port), skb);
 165
 166         flow_extract(skb, ntohs(opo->in_port), &key);
 167
 168         v_code = validate_actions(chain->dp, &key, opo->actions, actions_len);
 169         if (v_code != ACT_VALIDATION_OK) {
 170                 dp_send_error_msg(chain->dp, sender, OFPET_BAD_ACTION, v_code,
 171                                   msg, ntohs(opo->header.length));
 172                 goto error;
 173         }
 174
 175         execute_actions(chain->dp, skb, &key, opo->actions, actions_len, 1);
 176
 177         return 0;
 178
 179 error:
 180         kfree_skb(skb);
 181         return -EINVAL;
 182 }
 183
 184 static int
 185 recv_port_mod(struct sw_chain *chain, const struct sender *sender,
 186               const void *msg)
 187 {
 188         const struct ofp_port_mod *opm = msg;
 189
 190         dp_update_port_flags(chain->dp, opm);
 191
 192         return 0;
 193 }
 194
 195 static int
 196 recv_echo_request(struct sw_chain *chain, const struct sender *sender,
 197                   const void *msg)
 198 {
 199         return dp_send_echo_reply(chain->dp, sender, msg);
 200 }
 201
 202 static int
 203 recv_echo_reply(struct sw_chain *chain, const struct sender *sender,
 204                   const void *msg)
 205 {
 206         return 0;
 207 }
 208
 209 static int
 210 add_flow(struct sw_chain *chain, const struct sender *sender,
 211                 const struct ofp_flow_mod *ofm)
 212 {
 213         int error = -ENOMEM;
 214         uint16_t v_code;
 215         struct sw_flow *flow;
 216         size_t actions_len = ntohs(ofm->header.length) - sizeof *ofm;
 217
 218         /* Allocate memory. */
 219         flow = flow_alloc(actions_len, GFP_ATOMIC);
 220         if (flow == NULL)
 221                 goto error;
 222
 223         flow_extract_match(&flow->key, &ofm->match);
 224
 225         v_code = validate_actions(chain->dp, &flow->key, ofm->actions, actions_len);
 226         if (v_code != ACT_VALIDATION_OK) {
 227                 dp_send_error_msg(chain->dp, sender, OFPET_BAD_ACTION, v_code,
 228                                   ofm, ntohs(ofm->header.length));
 229                 goto error_free_flow;
 230         }
 231
 232         /* Fill out flow. */
 233         flow->priority = flow->key.wildcards ? ntohs(ofm->priority) : -1;
 234         flow->idle_timeout = ntohs(ofm->idle_timeout);
 235         flow->hard_timeout = ntohs(ofm->hard_timeout);
 236         flow->used = jiffies;
 237         flow->init_time = jiffies;
 238         flow->byte_count = 0;
 239         flow->packet_count = 0;
 240         spin_lock_init(&flow->lock);
 241         memcpy(flow->sf_acts->actions, ofm->actions, actions_len);
 242
 243         /* Act. */
 244         error = chain_insert(chain, flow);
 245         if (error == -ENOBUFS) {
 246                 dp_send_error_msg(chain->dp, sender, OFPET_FLOW_MOD_FAILED,
 247                                 OFPFMFC_ALL_TABLES_FULL, ofm, ntohs(ofm->header.length));
 248                 goto error_free_flow;
 249         } else if (error)
 250                 goto error_free_flow;
 251         error = 0;
 252         if (ntohl(ofm->buffer_id) != (uint32_t) -1) {
 253                 struct sk_buff *skb = retrieve_skb(ntohl(ofm->buffer_id));
 254                 if (skb) {
 255                         struct sw_flow_key key;
 256                         flow_used(flow, skb);
 257                         dp_set_origin(chain->dp, ntohs(ofm->match.in_port), skb);
 258                         flow_extract(skb, ntohs(ofm->match.in_port), &key);
 259                         execute_actions(chain->dp, skb, &key, ofm->actions, actions_len, 0);
 260                 }
 261                 else
 262                         error = -ESRCH;
 263         }
 264         return error;
 265
 266 error_free_flow:
 267         flow_free(flow);
 268 error:
 269         if (ntohl(ofm->buffer_id) != (uint32_t) -1)
 270                 discard_skb(ntohl(ofm->buffer_id));
 271         return error;
 272 }
 273
 274 static int
 275 mod_flow(struct sw_chain *chain, const struct sender *sender,
 276                 const struct ofp_flow_mod *ofm)
 277 {
 278         int error = -ENOMEM;
 279         uint16_t v_code;
 280         size_t actions_len;
 281         struct sw_flow_key key;
 282         uint16_t priority;
 283         int strict;
 284
 285         flow_extract_match(&key, &ofm->match);
 286
 287         actions_len = ntohs(ofm->header.length) - sizeof *ofm;
 288
 289         v_code = validate_actions(chain->dp, &key, ofm->actions, actions_len);
 290         if (v_code != ACT_VALIDATION_OK) {
 291                 dp_send_error_msg(chain->dp, sender, OFPET_BAD_ACTION, v_code,
 292                                   ofm, ntohs(ofm->header.length));
 293                 goto error;
 294         }
 295
 296         priority = key.wildcards ? ntohs(ofm->priority) : -1;
 297         strict = (ofm->command == htons(OFPFC_MODIFY_STRICT)) ? 1 : 0;
 298         chain_modify(chain, &key, priority, strict, ofm->actions, actions_len);
 299
 300         if (ntohl(ofm->buffer_id) != (uint32_t) -1) {
 301                 struct sk_buff *skb = retrieve_skb(ntohl(ofm->buffer_id));
 302                 if (skb) {
 303                         struct sw_flow_key skb_key;
 304                         flow_extract(skb, ntohs(ofm->match.in_port), &skb_key);
 305                         execute_actions(chain->dp, skb, &skb_key,
 306                                         ofm->actions, actions_len, 0);
 307                 }
 308                 else
 309                         error = -ESRCH;
 310         }
 311         return error;
 312
 313 error:
 314         if (ntohl(ofm->buffer_id) != (uint32_t) -1)
 315                 discard_skb(ntohl(ofm->buffer_id));
 316         return error;
 317 }
 318
 319 static int
 320 recv_flow(struct sw_chain *chain, const struct sender *sender, const void *msg)
 321 {
 322         const struct ofp_flow_mod *ofm = msg;
 323         uint16_t command = ntohs(ofm->command);
 324
 325         if (command == OFPFC_ADD) {
 326                 return add_flow(chain, sender, ofm);
 327         } else if ((command == OFPFC_MODIFY) || (command == OFPFC_MODIFY_STRICT)) {
 328                 return mod_flow(chain, sender, ofm);
 329         }  else if (command == OFPFC_DELETE) {
 330                 struct sw_flow_key key;
 331                 flow_extract_match(&key, &ofm->match);
 332                 return chain_delete(chain, &key, 0, 0) ? 0 : -ESRCH;
 333         } else if (command == OFPFC_DELETE_STRICT) {
 334                 struct sw_flow_key key;
 335                 uint16_t priority;
 336                 flow_extract_match(&key, &ofm->match);
 337                 priority = key.wildcards ? ntohs(ofm->priority) : -1;
 338                 return chain_delete(chain, &key, priority, 1) ? 0 : -ESRCH;
 339         } else {
 340                 return -ENOTSUPP;
 341         }
 342 }
 343
 344 static int
 345 recv_vendor(struct sw_chain *chain, const struct sender *sender,
 346                 const void *msg)
 347 {
 348         const struct ofp_vendor_header *ovh = msg;
 349
 350         switch(ntohl(ovh->vendor))
 351         {
 352         case NX_VENDOR_ID:
 353                 return nx_recv_msg(chain, sender, msg);
 354         default:
 355                 if (net_ratelimit())
 356                         printk("Uknown vendor: %#x\n", ntohl(ovh->vendor));
 357                 dp_send_error_msg(chain->dp, sender, OFPET_BAD_REQUEST,
 358                                   OFPBRC_BAD_VENDOR, msg, ntohs(ovh->header.length));
 359                 return -EINVAL;
 360         }
 361 }
 362
 363 /* 'msg', which is 'length' bytes long, was received across Netlink from
 364  * 'sender'.  Apply it to 'chain'. */
 365 int
 366 fwd_control_input(struct sw_chain *chain, const struct sender *sender,
 367                   const void *msg, size_t length)
 368 {
 369
 370         struct openflow_packet {
 371                 size_t min_size;
 372                 int (*handler)(struct sw_chain *, const struct sender *,
 373                                const void *);
 374         };
 375
 376         static const struct openflow_packet packets[] = {
 377                 [OFPT_HELLO] = {
 378                         sizeof (struct ofp_header),
 379                         recv_hello,
 380                 },
 381                 [OFPT_ECHO_REQUEST] = {
 382                         sizeof (struct ofp_header),
 383                         recv_echo_request,
 384                 },
 385                 [OFPT_ECHO_REPLY] = {
 386                         sizeof (struct ofp_header),
 387                         recv_echo_reply,
 388                 },
 389                 [OFPT_VENDOR] = {
 390                         sizeof (struct ofp_vendor_header),
 391                         recv_vendor,
 392                 },
 393                 [OFPT_FEATURES_REQUEST] = {
 394                         sizeof (struct ofp_header),
 395                         recv_features_request,
 396                 },
 397                 [OFPT_GET_CONFIG_REQUEST] = {
 398                         sizeof (struct ofp_header),
 399                         recv_get_config_request,
 400                 },
 401                 [OFPT_SET_CONFIG] = {
 402                         sizeof (struct ofp_switch_config),
 403                         recv_set_config,
 404                 },
 405                 [OFPT_PACKET_OUT] = {
 406                         sizeof (struct ofp_packet_out),
 407                         recv_packet_out,
 408                 },
 409                 [OFPT_FLOW_MOD] = {
 410                         sizeof (struct ofp_flow_mod),
 411                         recv_flow,
 412                 },
 413                 [OFPT_PORT_MOD] = {
 414                         sizeof (struct ofp_port_mod),
 415                         recv_port_mod,
 416                 }
 417         };
 418
 419         struct ofp_header *oh;
 420
 421         oh = (struct ofp_header *) msg;
 422         if (oh->version != OFP_VERSION
 423             && oh->type != OFPT_HELLO
 424             && oh->type != OFPT_ERROR
 425             && oh->type != OFPT_ECHO_REQUEST
 426             && oh->type != OFPT_ECHO_REPLY
 427             && oh->type != OFPT_VENDOR)
 428         {
 429                 dp_send_error_msg(chain->dp, sender, OFPET_BAD_REQUEST,
 430                                   OFPBRC_BAD_VERSION, msg, length);
 431                 return -EINVAL;
 432         }
 433         if (ntohs(oh->length) != length) {
 434                 if (net_ratelimit())
 435                         printk("received message length wrong: %d/%d\n",
 436                                 ntohs(oh->length), length);
 437                 return -EINVAL;
 438         }
 439
 440         if (oh->type < ARRAY_SIZE(packets)) {
 441                 const struct openflow_packet *pkt = &packets[oh->type];
 442                 if (pkt->handler) {
 443                         if (length < pkt->min_size)
 444                                 return -EFAULT;
 445                         return pkt->handler(chain, sender, msg);
 446                 }
 447         }
 448         dp_send_error_msg(chain->dp, sender, OFPET_BAD_REQUEST,
 449                           OFPBRC_BAD_TYPE, msg, length);
 450         return -EINVAL;
 451 }
 452
 453 /* Packet buffering. */
 454
 455 #define OVERWRITE_SECS  1
 456 #define OVERWRITE_JIFFIES (OVERWRITE_SECS * HZ)
 457
 458 struct packet_buffer {
 459         struct sk_buff *skb;
 460         uint32_t cookie;
 461         unsigned long exp_jiffies;
 462 };
 463
 464 static struct packet_buffer buffers[N_PKT_BUFFERS];
 465 static unsigned int buffer_idx;
 466 static DEFINE_SPINLOCK(buffer_lock);
 467
 468 uint32_t fwd_save_skb(struct sk_buff *skb)
 469 {
 470         struct sk_buff *old_skb = NULL;
 471         struct packet_buffer *p;
 472         unsigned long int flags;
 473         uint32_t id;
 474
 475         spin_lock_irqsave(&buffer_lock, flags);
 476         buffer_idx = (buffer_idx + 1) & PKT_BUFFER_MASK;
 477         p = &buffers[buffer_idx];
 478         if (p->skb) {
 479                 /* Don't buffer packet if existing entry is less than
 480                  * OVERWRITE_SECS old. */
 481                 if (time_before(jiffies, p->exp_jiffies)) {
 482                         spin_unlock_irqrestore(&buffer_lock, flags);
 483                         return -1;
 484                 } else {
 485                         /* Defer kfree_skb() until interrupts re-enabled. */
 486                         old_skb = p->skb;
 487                 }
 488         }
 489         /* Don't use maximum cookie value since the all-bits-1 id is
 490          * special. */
 491         if (++p->cookie >= (1u << PKT_COOKIE_BITS) - 1)
 492                 p->cookie = 0;
 493         skb_get(skb);
 494         p->skb = skb;
 495         p->exp_jiffies = jiffies + OVERWRITE_JIFFIES;
 496         id = buffer_idx | (p->cookie << PKT_BUFFER_BITS);
 497         spin_unlock_irqrestore(&buffer_lock, flags);
 498
 499         if (old_skb)
 500                 kfree_skb(old_skb);
 501
 502         return id;
 503 }
 504
 505 static struct sk_buff *retrieve_skb(uint32_t id)
 506 {
 507         unsigned long int flags;
 508         struct sk_buff *skb = NULL;
 509         struct packet_buffer *p;
 510
 511         spin_lock_irqsave(&buffer_lock, flags);
 512         p = &buffers[id & PKT_BUFFER_MASK];
 513         if (p->cookie == id >> PKT_BUFFER_BITS) {
 514                 skb = p->skb;
 515                 p->skb = NULL;
 516         } else {
 517                 printk("cookie mismatch: %x != %x\n",
 518                                 id >> PKT_BUFFER_BITS, p->cookie);
 519         }
 520         spin_unlock_irqrestore(&buffer_lock, flags);
 521
 522         return skb;
 523 }
 524
 525 void fwd_discard_all(void)
 526 {
 527         int i;
 528
 529         for (i = 0; i < N_PKT_BUFFERS; i++) {
 530                 struct sk_buff *skb;
 531                 unsigned long int flags;
 532
 533                 /* Defer kfree_skb() until interrupts re-enabled. */
 534                 spin_lock_irqsave(&buffer_lock, flags);
 535                 skb = buffers[i].skb;
 536                 buffers[i].skb = NULL;
 537                 spin_unlock_irqrestore(&buffer_lock, flags);
 538
 539                 kfree_skb(skb);
 540         }
 541 }
 542
 543 static void discard_skb(uint32_t id)
 544 {
 545         struct sk_buff *old_skb = NULL;
 546         unsigned long int flags;
 547         struct packet_buffer *p;
 548
 549         spin_lock_irqsave(&buffer_lock, flags);
 550         p = &buffers[id & PKT_BUFFER_MASK];
 551         if (p->cookie == id >> PKT_BUFFER_BITS) {
 552                 /* Defer kfree_skb() until interrupts re-enabled. */
 553                 old_skb = p->skb;
 554                 p->skb = NULL;
 555         }
 556         spin_unlock_irqrestore(&buffer_lock, flags);
 557
 558         if (old_skb)
 559                 kfree_skb(old_skb);
 560 }
 561
 562 void fwd_exit(void)
 563 {
 564         fwd_discard_all();
 565 }