datapath/datapath.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 /* Functions for managing the dp interface/device. */
   8
   9 #include <linux/init.h>
  10 #include <linux/module.h>
  11 #include <linux/if_arp.h>
  12 #include <linux/if_bridge.h>
  13 #include <linux/if_vlan.h>
  14 #include <linux/in.h>
  15 #include <net/genetlink.h>
  16 #include <linux/ip.h>
  17 #include <linux/delay.h>
  18 #include <linux/time.h>
  19 #include <linux/etherdevice.h>
  20 #include <linux/kernel.h>
  21 #include <linux/kthread.h>
  22 #include <linux/mutex.h>
  23 #include <linux/rtnetlink.h>
  24 #include <linux/rcupdate.h>
  25 #include <linux/version.h>
  26 #include <linux/ethtool.h>
  27 #include <linux/random.h>
  28 #include <asm/system.h>
  29 #include <asm/div64.h>
  30 #include <linux/netfilter_bridge.h>
  31 #include <linux/netfilter_ipv4.h>
  32 #include <linux/inetdevice.h>
  33 #include <linux/list.h>
  34 #include <linux/rculist.h>
  35 #include <linux/workqueue.h>
  36 #include <linux/dmi.h>
  37
  38 #include "openflow/nicira-ext.h"
  39 #include "openflow/openflow-netlink.h"
  40 #include "datapath.h"
  41 #include "nx_act_snat.h"
  42 #include "table.h"
  43 #include "chain.h"
  44 #include "dp_dev.h"
  45 #include "forward.h"
  46 #include "flow.h"
  47
  48 #include "compat.h"
  49
  50
  51 /* Strings to describe the manufacturer, hardware, and software.  This data
  52  * is queriable through the switch description stats message. */
  53 static char mfr_desc[DESC_STR_LEN] = "Nicira Networks, Inc.";
  54 static char hw_desc[DESC_STR_LEN] = "Reference Linux Kernel Module";
  55 static char sw_desc[DESC_STR_LEN] = VERSION BUILDNR;
  56 static char serial_num[SERIAL_NUM_LEN] = "None";
  57
  58 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)
  59 module_param_string(mfr_desc, mfr_desc, sizeof mfr_desc, 0444);
  60 module_param_string(hw_desc, hw_desc, sizeof hw_desc, 0444);
  61 module_param_string(sw_desc, sw_desc, sizeof sw_desc, 0444);
  62 module_param_string(serial_num, serial_num, sizeof serial_num, 0444);
  63 #else
  64 MODULE_PARM(mfr_desc, "s");
  65 MODULE_PARM(hw_desc, "s");
  66 MODULE_PARM(sw_desc, "s");
  67 MODULE_PARM(serial_num, "s");
  68 #endif
  69
  70
  71 /* Number of milliseconds between runs of the maintenance thread. */
  72 #define MAINT_SLEEP_MSECS 1000
  73
  74 #define UINT32_MAX                        4294967295U
  75 #define UINT16_MAX                        65535
  76 #define MAX(X, Y) ((X) > (Y) ? (X) : (Y))
  77
  78 static struct genl_family dp_genl_family;
  79 static struct genl_multicast_group mc_group;
  80
  81 /* It's hard to imagine wanting more than one datapath, but... */
  82 #define DP_MAX 32
  83
  84 /* Datapaths.  Protected on the read side by rcu_read_lock, on the write side
  85  * by dp_mutex.  dp_mutex is almost completely redundant with genl_mutex
  86  * maintained by the Generic Netlink code, but the timeout path needs mutual
  87  * exclusion too.
  88  *
  89  * It is safe to access the datapath and net_bridge_port structures with just
  90  * dp_mutex.
  91  */
  92 static struct datapath *dps[DP_MAX];
  93 DEFINE_MUTEX(dp_mutex);
  94 EXPORT_SYMBOL(dp_mutex);
  95
  96 static int dp_maint_func(void *data);
  97 static void init_port_status(struct net_bridge_port *p);
  98 static int dp_genl_openflow_done(struct netlink_callback *);
  99 static struct net_bridge_port *new_nbp(struct datapath *,
 100                                        struct net_device *, int port_no);
 101
 102 /* nla_shrink - reduce amount of space reserved by nla_reserve
 103  * @skb: socket buffer from which to recover room
 104  * @nla: netlink attribute to adjust
 105  * @len: new length of attribute payload
 106  *
 107  * Reduces amount of space reserved by a call to nla_reserve.
 108  *
 109  * No other attributes may be added between calling nla_reserve and this
 110  * function, since it will create a hole in the message.
 111  */
 112 void nla_shrink(struct sk_buff *skb, struct nlattr *nla, int len)
 113 {
 114         int delta = nla_total_size(len) - nla_total_size(nla_len(nla));
 115         BUG_ON(delta > 0);
 116         skb->tail += delta;
 117         skb->len  += delta;
 118         nla->nla_len = nla_attr_size(len);
 119 }
 120
 121 /* Puts a set of openflow headers for a message of the given 'type' into 'skb'.
 122  * If 'sender' is nonnull, then it is used as the message's destination.  'dp'
 123  * must specify the datapath to use.
 124  *
 125  * '*max_openflow_len' receives the maximum number of bytes that are available
 126  * for the embedded OpenFlow message.  The caller must call
 127  * resize_openflow_skb() to set the actual size of the message to this number
 128  * of bytes or less.
 129  *
 130  * Returns the openflow header if successful, otherwise (if 'skb' is too small)
 131  * an error code. */
 132 static void *
 133 put_openflow_headers(struct datapath *dp, struct sk_buff *skb, uint8_t type,
 134                      const struct sender *sender, int *max_openflow_len)
 135 {
 136         struct ofp_header *oh;
 137         struct nlattr *attr;
 138         int openflow_len;
 139
 140         /* Assemble the Generic Netlink wrapper. */
 141         if (!genlmsg_put(skb,
 142                          sender ? sender->pid : 0,
 143                          sender ? sender->seq : 0,
 144                          &dp_genl_family, 0, DP_GENL_C_OPENFLOW))
 145                 return ERR_PTR(-ENOBUFS);
 146         if (nla_put_u32(skb, DP_GENL_A_DP_IDX, dp->dp_idx) < 0)
 147                 return ERR_PTR(-ENOBUFS);
 148         openflow_len = (skb_tailroom(skb) - NLA_HDRLEN) & ~(NLA_ALIGNTO - 1);
 149         if (openflow_len < sizeof *oh)
 150                 return ERR_PTR(-ENOBUFS);
 151         *max_openflow_len = openflow_len;
 152         attr = nla_reserve(skb, DP_GENL_A_OPENFLOW, openflow_len);
 153         BUG_ON(!attr);
 154
 155         /* Fill in the header.  The caller is responsible for the length. */
 156         oh = nla_data(attr);
 157         oh->version = OFP_VERSION;
 158         oh->type = type;
 159         oh->xid = sender ? sender->xid : 0;
 160
 161         return oh;
 162 }
 163
 164 /* Resizes OpenFlow header 'oh', which must be at the tail end of 'skb', to new
 165  * length 'new_length' (in bytes), adjusting pointers and size values as
 166  * necessary. */
 167 static void
 168 resize_openflow_skb(struct sk_buff *skb,
 169                     struct ofp_header *oh, size_t new_length)
 170 {
 171         struct nlattr *attr = ((void *) oh) - NLA_HDRLEN;
 172         nla_shrink(skb, attr, new_length);
 173         oh->length = htons(new_length);
 174         nlmsg_end(skb, (struct nlmsghdr *) skb->data);
 175 }
 176
 177 /* Allocates a new skb to contain an OpenFlow message 'openflow_len' bytes in
 178  * length.  Returns a null pointer if memory is unavailable, otherwise returns
 179  * the OpenFlow header and stores a pointer to the skb in '*pskb'.
 180  *
 181  * 'type' is the OpenFlow message type.  If 'sender' is nonnull, then it is
 182  * used as the message's destination.  'dp' must specify the datapath to
 183  * use.  */
 184 static void *
 185 alloc_openflow_skb(struct datapath *dp, size_t openflow_len, uint8_t type,
 186                    const struct sender *sender, struct sk_buff **pskb)
 187 {
 188         struct ofp_header *oh;
 189         size_t genl_len;
 190         struct sk_buff *skb;
 191         int max_openflow_len;
 192
 193         if ((openflow_len + sizeof(struct ofp_header)) > UINT16_MAX) {
 194                 if (net_ratelimit())
 195                         printk("alloc_openflow_skb: openflow message too large: %zu\n",
 196                                         openflow_len);
 197                 return NULL;
 198         }
 199
 200         genl_len = nlmsg_total_size(GENL_HDRLEN + dp_genl_family.hdrsize);
 201         genl_len += nla_total_size(sizeof(uint32_t)); /* DP_GENL_A_DP_IDX */
 202         genl_len += nla_total_size(openflow_len);    /* DP_GENL_A_OPENFLOW */
 203         skb = *pskb = genlmsg_new(genl_len, GFP_ATOMIC);
 204         if (!skb) {
 205                 if (net_ratelimit())
 206                         printk("alloc_openflow_skb: genlmsg_new failed\n");
 207                 return NULL;
 208         }
 209
 210         oh = put_openflow_headers(dp, skb, type, sender, &max_openflow_len);
 211         BUG_ON(!oh || IS_ERR(oh));
 212         resize_openflow_skb(skb, oh, openflow_len);
 213
 214         return oh;
 215 }
 216
 217 /* Sends 'skb' to 'sender' if it is nonnull, otherwise multicasts 'skb' to all
 218  * listeners. */
 219 static int
 220 send_openflow_skb(struct sk_buff *skb, const struct sender *sender)
 221 {
 222         return (sender
 223                 ? genlmsg_unicast(skb, sender->pid)
 224                 : genlmsg_multicast(skb, 0, mc_group.id, GFP_ATOMIC));
 225 }
 226
 227 /* Retrieves the datapath id, which is the MAC address of the "of" device. */
 228 static
 229 uint64_t get_datapath_id(struct net_device *dev)
 230 {
 231         uint64_t id = 0;
 232         int i;
 233
 234         for (i=0; i<ETH_ALEN; i++)
 235                 id |= (uint64_t)dev->dev_addr[i] << (8*(ETH_ALEN-1 - i));
 236
 237         return id;
 238 }
 239
 240 /* Creates a new datapath numbered 'dp_idx'.  Returns 0 for success or a
 241  * negative error code. */
 242 static int new_dp(int dp_idx)
 243 {
 244         struct datapath *dp;
 245         int err;
 246
 247         if (dp_idx < 0 || dp_idx >= DP_MAX)
 248                 return -EINVAL;
 249
 250         if (!try_module_get(THIS_MODULE))
 251                 return -ENODEV;
 252
 253         /* Exit early if a datapath with that number already exists. */
 254         if (dps[dp_idx]) {
 255                 err = -EEXIST;
 256                 goto err_unlock;
 257         }
 258
 259         err = -ENOMEM;
 260         dp = kzalloc(sizeof *dp, GFP_KERNEL);
 261         if (dp == NULL)
 262                 goto err_unlock;
 263
 264         dp->dp_idx = dp_idx;
 265
 266         /* Setup our "of" device */
 267         err = dp_dev_setup(dp);
 268         if (err)
 269                 goto err_free_dp;
 270
 271         dp->chain = chain_create(dp);
 272         if (dp->chain == NULL)
 273                 goto err_destroy_dp_dev;
 274         INIT_LIST_HEAD(&dp->port_list);
 275
 276         dp->local_port = new_nbp(dp, dp->netdev, OFPP_LOCAL);
 277         if (IS_ERR(dp->local_port)) {
 278                 err = PTR_ERR(dp->local_port);
 279                 goto err_destroy_local_port;
 280         }
 281
 282         dp->flags = 0;
 283         dp->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
 284
 285         dp->dp_task = kthread_run(dp_maint_func, dp, "dp%d", dp_idx);
 286         if (IS_ERR(dp->dp_task))
 287                 goto err_destroy_chain;
 288
 289         dps[dp_idx] = dp;
 290
 291         return 0;
 292
 293 err_destroy_local_port:
 294         dp_del_switch_port(dp->local_port);
 295 err_destroy_chain:
 296         chain_destroy(dp->chain);
 297 err_destroy_dp_dev:
 298         dp_dev_destroy(dp);
 299 err_free_dp:
 300         kfree(dp);
 301 err_unlock:
 302         module_put(THIS_MODULE);
 303                 return err;
 304 }
 305
 306 /* Find and return a free port number under 'dp'. */
 307 static int find_portno(struct datapath *dp)
 308 {
 309         int i;
 310         for (i = 0; i < DP_MAX_PORTS; i++)
 311                 if (dp->ports[i] == NULL)
 312                         return i;
 313         return -EXFULL;
 314 }
 315
 316 static struct net_bridge_port *new_nbp(struct datapath *dp,
 317                                        struct net_device *dev, int port_no)
 318 {
 319         struct net_bridge_port *p;
 320
 321         if (dev->br_port != NULL)
 322                 return ERR_PTR(-EBUSY);
 323
 324         p = kzalloc(sizeof(*p), GFP_KERNEL);
 325         if (p == NULL)
 326                 return ERR_PTR(-ENOMEM);
 327
 328         rtnl_lock();
 329         dev_set_promiscuity(dev, 1);
 330         rtnl_unlock();
 331         dev_hold(dev);
 332         p->dp = dp;
 333         p->dev = dev;
 334         p->port_no = port_no;
 335         spin_lock_init(&p->lock);
 336         INIT_WORK(&p->port_task, NULL);
 337         if (port_no != OFPP_LOCAL)
 338                 rcu_assign_pointer(dev->br_port, p);
 339         if (port_no < DP_MAX_PORTS)
 340                 rcu_assign_pointer(dp->ports[port_no], p);
 341         list_add_rcu(&p->node, &dp->port_list);
 342
 343         return p;
 344 }
 345
 346 int add_switch_port(struct datapath *dp, struct net_device *dev)
 347 {
 348         struct net_bridge_port *p;
 349         int port_no;
 350
 351         if (dev->flags & IFF_LOOPBACK || dev->type != ARPHRD_ETHER
 352             || is_dp_dev(dev))
 353                 return -EINVAL;
 354
 355         port_no = find_portno(dp);
 356         if (port_no < 0)
 357                 return port_no;
 358
 359         p = new_nbp(dp, dev, port_no);
 360         if (IS_ERR(p))
 361                 return PTR_ERR(p);
 362
 363         init_port_status(p);
 364
 365         /* Notify the ctlpath that this port has been added */
 366         dp_send_port_status(p, OFPPR_ADD);
 367
 368         return 0;
 369 }
 370
 371 /* Delete 'p' from switch. */
 372 int dp_del_switch_port(struct net_bridge_port *p)
 373 {
 374 #ifdef SUPPORT_SNAT
 375         unsigned long flags;
 376 #endif
 377
 378         /* First drop references to device. */
 379         cancel_work_sync(&p->port_task);
 380         rtnl_lock();
 381         dev_set_promiscuity(p->dev, -1);
 382         rtnl_unlock();
 383         list_del_rcu(&p->node);
 384         if (p->port_no != OFPP_LOCAL)
 385                 rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
 386         rcu_assign_pointer(p->dev->br_port, NULL);
 387
 388         /* Then wait until no one is still using it, and destroy it. */
 389         synchronize_rcu();
 390
 391 #ifdef SUPPORT_SNAT
 392         /* Free any SNAT configuration on the port. */
 393         spin_lock_irqsave(&p->lock, flags);
 394         snat_free_conf(p);
 395         spin_unlock_irqrestore(&p->lock, flags);
 396 #endif
 397
 398         /* Notify the ctlpath that this port no longer exists */
 399         dp_send_port_status(p, OFPPR_DELETE);
 400
 401         dev_put(p->dev);
 402         kfree(p);
 403
 404         return 0;
 405 }
 406
 407 static void del_dp(struct datapath *dp)
 408 {
 409         struct net_bridge_port *p, *n;
 410
 411         kthread_stop(dp->dp_task);
 412
 413         /* Drop references to DP. */
 414         list_for_each_entry_safe (p, n, &dp->port_list, node)
 415                 dp_del_switch_port(p);
 416         rcu_assign_pointer(dps[dp->dp_idx], NULL);
 417
 418         /* Kill off local_port dev references from buffered packets that have
 419          * associated dst entries. */
 420         synchronize_rcu();
 421         fwd_discard_all();
 422
 423         /* Destroy dp->netdev.  (Must follow deleting switch ports since
 424          * dp->local_port has a reference to it.) */
 425         dp_dev_destroy(dp);
 426
 427         /* Wait until no longer in use, then destroy it. */
 428         synchronize_rcu();
 429         chain_destroy(dp->chain);
 430         kfree(dp);
 431         module_put(THIS_MODULE);
 432 }
 433
 434 static int dp_maint_func(void *data)
 435 {
 436         struct datapath *dp = (struct datapath *) data;
 437
 438         while (!kthread_should_stop()) {
 439 #ifdef SUPPORT_SNAT
 440                 struct net_bridge_port *p;
 441
 442                 /* Expire old SNAT entries */
 443                 rcu_read_lock();
 444                 list_for_each_entry_rcu (p, &dp->port_list, node)
 445                         snat_maint(p);
 446                 rcu_read_unlock();
 447 #endif
 448
 449                 /* Timeout old entries */
 450                 chain_timeout(dp->chain);
 451                 msleep_interruptible(MAINT_SLEEP_MSECS);
 452         }
 453
 454         return 0;
 455 }
 456
 457 static void
 458 do_port_input(struct net_bridge_port *p, struct sk_buff *skb)
 459 {
 460         /* Make our own copy of the packet.  Otherwise we will mangle the
 461          * packet for anyone who came before us (e.g. tcpdump via AF_PACKET).
 462          * (No one comes after us, since we tell handle_bridge() that we took
 463          * the packet.) */
 464         skb = skb_share_check(skb, GFP_ATOMIC);
 465         if (!skb)
 466                 return;
 467
 468 #ifdef SUPPORT_SNAT
 469         /* Check if this packet needs early SNAT processing. */
 470         if (snat_pre_route(skb)) {
 471                 return;
 472         }
 473 #endif
 474
 475         /* Push the Ethernet header back on. */
 476         skb_push(skb, ETH_HLEN);
 477         skb_reset_mac_header(skb);
 478         fwd_port_input(p->dp->chain, skb, p);
 479 }
 480
 481 /*
 482  * Used as br_handle_frame_hook.  (Cannot run bridge at the same time, even on
 483  * different set of devices!)
 484  */
 485 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,22)
 486 /* Called with rcu_read_lock. */
 487 static struct sk_buff *dp_frame_hook(struct net_bridge_port *p,
 488                                          struct sk_buff *skb)
 489 {
 490         do_port_input(p, skb);
 491         return NULL;
 492 }
 493 #elif LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)
 494 static int dp_frame_hook(struct net_bridge_port *p, struct sk_buff **pskb)
 495 {
 496         do_port_input(p, *pskb);
 497         return 1;
 498 }
 499 #else
 500 /* NB: This has only been tested on 2.4.35 */
 501 static void dp_frame_hook(struct sk_buff *skb)
 502 {
 503         struct net_bridge_port *p = skb->dev->br_port;
 504         if (p) {
 505                 rcu_read_lock();
 506                 do_port_input(p, skb);
 507                 rcu_read_unlock();
 508         } else
 509                 kfree_skb(skb);
 510 }
 511 #endif
 512
 513 /* Forwarding output path.
 514  * Based on net/bridge/br_forward.c. */
 515
 516 static inline unsigned packet_length(const struct sk_buff *skb)
 517 {
 518         int length = skb->len - ETH_HLEN;
 519         if (skb->protocol == htons(ETH_P_8021Q))
 520                 length -= VLAN_HLEN;
 521         return length;
 522 }
 523
 524 /* Send packets out all the ports except the originating one.  If the
 525  * "flood" argument is set, only send along the minimum spanning tree.
 526  */
 527 static int
 528 output_all(struct datapath *dp, struct sk_buff *skb, int flood)
 529 {
 530         u32 disable = flood ? OFPPC_NO_FLOOD : 0;
 531         struct net_bridge_port *p;
 532         int prev_port = -1;
 533
 534         list_for_each_entry_rcu (p, &dp->port_list, node) {
 535                 if (skb->dev == p->dev || p->config & disable)
 536                         continue;
 537                 if (prev_port != -1) {
 538                         struct sk_buff *clone = skb_clone(skb, GFP_ATOMIC);
 539                         if (!clone) {
 540                                 kfree_skb(skb);
 541                                 return -ENOMEM;
 542                         }
 543                         dp_output_port(dp, clone, prev_port, 0);
 544                 }
 545                 prev_port = p->port_no;
 546         }
 547         if (prev_port != -1)
 548                 dp_output_port(dp, skb, prev_port, 0);
 549         else
 550                 kfree_skb(skb);
 551
 552         return 0;
 553 }
 554
 555 /* Marks 'skb' as having originated from 'in_port' in 'dp'.
 556    FIXME: how are devices reference counted? */
 557 void dp_set_origin(struct datapath *dp, uint16_t in_port,
 558                            struct sk_buff *skb)
 559 {
 560         struct net_bridge_port *p;
 561         p = (in_port < DP_MAX_PORTS ? dp->ports[in_port]
 562              : in_port == OFPP_LOCAL ? dp->local_port
 563              : NULL);
 564         if (p)
 565                 skb->dev = p->dev;
 566          else
 567                 skb->dev = NULL;
 568 }
 569
 570 int
 571 dp_xmit_skb(struct sk_buff *skb)
 572 {
 573         int len = skb->len;
 574         if (packet_length(skb) > skb->dev->mtu) {
 575                 printk("dropped over-mtu packet: %d > %d\n",
 576                            packet_length(skb), skb->dev->mtu);
 577                 kfree_skb(skb);
 578                 return -E2BIG;
 579         }
 580
 581         dev_queue_xmit(skb);
 582
 583         return len;
 584 }
 585
 586 /* Takes ownership of 'skb' and transmits it to 'out_port' on 'dp'.
 587  */
 588 int dp_output_port(struct datapath *dp, struct sk_buff *skb, int out_port,
 589                    int ignore_no_fwd)
 590 {
 591         BUG_ON(!skb);
 592         switch (out_port){
 593         case OFPP_IN_PORT:
 594                 /* Send it out the port it came in on, which is already set in
 595                  * the skb. */
 596                 if (!skb->dev) {
 597                         if (net_ratelimit())
 598                                 printk("skb device not set forwarding to in_port\n");
 599                         kfree_skb(skb);
 600                         return -ESRCH;
 601                 }
 602                 return dp_xmit_skb(skb);
 603
 604         case OFPP_TABLE: {
 605                 int retval = run_flow_through_tables(dp->chain, skb,
 606                                                      skb->dev->br_port);
 607                 if (retval)
 608                         kfree_skb(skb);
 609                 return retval;
 610         }
 611
 612         case OFPP_FLOOD:
 613                 return output_all(dp, skb, 1);
 614
 615         case OFPP_ALL:
 616                 return output_all(dp, skb, 0);
 617
 618         case OFPP_CONTROLLER:
 619                 return dp_output_control(dp, skb, fwd_save_skb(skb), 0,
 620                                                   OFPR_ACTION);
 621
 622         case OFPP_LOCAL: {
 623                 struct net_device *dev = dp->netdev;
 624 #ifdef SUPPORT_SNAT
 625                 snat_local_in(skb);
 626 #endif
 627                 return dev ? dp_dev_recv(dev, skb) : -ESRCH;
 628         }
 629
 630         case 0 ... DP_MAX_PORTS - 1: {
 631                 struct net_bridge_port *p = dp->ports[out_port];
 632                 if (p == NULL)
 633                         goto bad_port;
 634                 if (p->dev == skb->dev) {
 635                         /* To send to the input port, must use OFPP_IN_PORT */
 636                         kfree_skb(skb);
 637                         if (net_ratelimit())
 638                                 printk("can't directly forward to input port\n");
 639                         return -EINVAL;
 640                 }
 641                 if (p->config & OFPPC_NO_FWD && !ignore_no_fwd) {
 642                         kfree_skb(skb);
 643                         return 0;
 644                 }
 645                 skb->dev = p->dev;
 646                 return dp_xmit_skb(skb);
 647         }
 648
 649         default:
 650                 goto bad_port;
 651         }
 652
 653 bad_port:
 654         kfree_skb(skb);
 655         if (net_ratelimit())
 656                 printk("can't forward to bad port %d\n", out_port);
 657         return -ENOENT;
 658 }
 659
 660 /* Takes ownership of 'skb' and transmits it to 'dp''s control path.  If
 661  * 'buffer_id' != -1, then only the first 64 bytes of 'skb' are sent;
 662  * otherwise, all of 'skb' is sent.  'reason' indicates why 'skb' is being
 663  * sent. 'max_len' sets the maximum number of bytes that the caller
 664  * wants to be sent; a value of 0 indicates the entire packet should be
 665  * sent. */
 666 int
 667 dp_output_control(struct datapath *dp, struct sk_buff *skb,
 668                            uint32_t buffer_id, size_t max_len, int reason)
 669 {
 670         /* FIXME?  Can we avoid creating a new skbuff in the case where we
 671          * forward the whole packet? */
 672         struct sk_buff *f_skb;
 673         struct ofp_packet_in *opi;
 674         size_t fwd_len, opi_len;
 675         int err;
 676
 677         fwd_len = skb->len;
 678         if ((buffer_id != (uint32_t) -1) && max_len)
 679                 fwd_len = min(fwd_len, max_len);
 680
 681         opi_len = offsetof(struct ofp_packet_in, data) + fwd_len;
 682         opi = alloc_openflow_skb(dp, opi_len, OFPT_PACKET_IN, NULL, &f_skb);
 683         if (!opi) {
 684                 err = -ENOMEM;
 685                 goto out;
 686         }
 687         opi->buffer_id      = htonl(buffer_id);
 688         opi->total_len      = htons(skb->len);
 689         opi->in_port        = htons(skb->dev && skb->dev->br_port
 690                                     ? skb->dev->br_port->port_no
 691                                     : OFPP_LOCAL);
 692         opi->reason         = reason;
 693         opi->pad            = 0;
 694         skb_copy_bits(skb, 0, opi->data, fwd_len);
 695         err = send_openflow_skb(f_skb, NULL);
 696
 697 out:
 698         kfree_skb(skb);
 699         return err;
 700 }
 701
 702 static void fill_port_desc(struct net_bridge_port *p, struct ofp_phy_port *desc)
 703 {
 704         unsigned long flags;
 705         desc->port_no = htons(p->port_no);
 706         strncpy(desc->name, p->dev->name, OFP_MAX_PORT_NAME_LEN);
 707         desc->name[OFP_MAX_PORT_NAME_LEN-1] = '\0';
 708         memcpy(desc->hw_addr, p->dev->dev_addr, ETH_ALEN);
 709         desc->curr = 0;
 710         desc->supported = 0;
 711         desc->advertised = 0;
 712         desc->peer = 0;
 713
 714         spin_lock_irqsave(&p->lock, flags);
 715         desc->config = htonl(p->config);
 716         desc->state = htonl(p->state);
 717         spin_unlock_irqrestore(&p->lock, flags);
 718
 719 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,4,24)
 720         if (p->dev->ethtool_ops && p->dev->ethtool_ops->get_settings) {
 721                 struct ethtool_cmd ecmd = { .cmd = ETHTOOL_GSET };
 722
 723                 if (!p->dev->ethtool_ops->get_settings(p->dev, &ecmd)) {
 724                         /* Set the supported features */
 725                         if (ecmd.supported & SUPPORTED_10baseT_Half)
 726                                 desc->supported |= OFPPF_10MB_HD;
 727                         if (ecmd.supported & SUPPORTED_10baseT_Full)
 728                                 desc->supported |= OFPPF_10MB_FD;
 729                         if (ecmd.supported & SUPPORTED_100baseT_Half)
 730                                 desc->supported |= OFPPF_100MB_HD;
 731                         if (ecmd.supported & SUPPORTED_100baseT_Full)
 732                                 desc->supported |= OFPPF_100MB_FD;
 733                         if (ecmd.supported & SUPPORTED_1000baseT_Half)
 734                                 desc->supported |= OFPPF_1GB_HD;
 735                         if (ecmd.supported & SUPPORTED_1000baseT_Full)
 736                                 desc->supported |= OFPPF_1GB_FD;
 737                         if (ecmd.supported & SUPPORTED_10000baseT_Full)
 738                                 desc->supported |= OFPPF_10GB_FD;
 739                         if (ecmd.supported & SUPPORTED_TP)
 740                                 desc->supported |= OFPPF_COPPER;
 741                         if (ecmd.supported & SUPPORTED_FIBRE)
 742                                 desc->supported |= OFPPF_FIBER;
 743                         if (ecmd.supported & SUPPORTED_Autoneg)
 744                                 desc->supported |= OFPPF_AUTONEG;
 745 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,14)
 746                         if (ecmd.supported & SUPPORTED_Pause)
 747                                 desc->supported |= OFPPF_PAUSE;
 748                         if (ecmd.supported & SUPPORTED_Asym_Pause)
 749                                 desc->supported |= OFPPF_PAUSE_ASYM;
 750 #endif /* kernel >= 2.6.14 */
 751
 752                         /* Set the advertised features */
 753                         if (ecmd.advertising & ADVERTISED_10baseT_Half)
 754                                 desc->advertised |= OFPPF_10MB_HD;
 755                         if (ecmd.advertising & ADVERTISED_10baseT_Full)
 756                                 desc->advertised |= OFPPF_10MB_FD;
 757                         if (ecmd.advertising & ADVERTISED_100baseT_Half)
 758                                 desc->advertised |= OFPPF_100MB_HD;
 759                         if (ecmd.advertising & ADVERTISED_100baseT_Full)
 760                                 desc->advertised |= OFPPF_100MB_FD;
 761                         if (ecmd.advertising & ADVERTISED_1000baseT_Half)
 762                                 desc->advertised |= OFPPF_1GB_HD;
 763                         if (ecmd.advertising & ADVERTISED_1000baseT_Full)
 764                                 desc->advertised |= OFPPF_1GB_FD;
 765                         if (ecmd.advertising & ADVERTISED_10000baseT_Full)
 766                                 desc->advertised |= OFPPF_10GB_FD;
 767                         if (ecmd.advertising & ADVERTISED_TP)
 768                                 desc->advertised |= OFPPF_COPPER;
 769                         if (ecmd.advertising & ADVERTISED_FIBRE)
 770                                 desc->advertised |= OFPPF_FIBER;
 771                         if (ecmd.advertising & ADVERTISED_Autoneg)
 772                                 desc->advertised |= OFPPF_AUTONEG;
 773 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,14)
 774                         if (ecmd.advertising & ADVERTISED_Pause)
 775                                 desc->advertised |= OFPPF_PAUSE;
 776                         if (ecmd.advertising & ADVERTISED_Asym_Pause)
 777                                 desc->advertised |= OFPPF_PAUSE_ASYM;
 778 #endif /* kernel >= 2.6.14 */
 779
 780                         /* Set the current features */
 781                         if (ecmd.speed == SPEED_10)
 782                                 desc->curr = (ecmd.duplex) ? OFPPF_10MB_FD : OFPPF_10MB_HD;
 783                         else if (ecmd.speed == SPEED_100)
 784                                 desc->curr = (ecmd.duplex) ? OFPPF_100MB_FD : OFPPF_100MB_HD;
 785                         else if (ecmd.speed == SPEED_1000)
 786                                 desc->curr = (ecmd.duplex) ? OFPPF_1GB_FD : OFPPF_1GB_HD;
 787                         else if (ecmd.speed == SPEED_10000)
 788                                 desc->curr = OFPPF_10GB_FD;
 789
 790                         if (ecmd.port == PORT_TP)
 791                                 desc->curr |= OFPPF_COPPER;
 792                         else if (ecmd.port == PORT_FIBRE)
 793                                 desc->curr |= OFPPF_FIBER;
 794
 795                         if (ecmd.autoneg)
 796                                 desc->curr |= OFPPF_AUTONEG;
 797                 }
 798         }
 799 #endif
 800         desc->curr = htonl(desc->curr);
 801         desc->supported = htonl(desc->supported);
 802         desc->advertised = htonl(desc->advertised);
 803         desc->peer = htonl(desc->peer);
 804 }
 805
 806 static int
 807 fill_features_reply(struct datapath *dp, struct ofp_switch_features *ofr)
 808 {
 809         struct net_bridge_port *p;
 810         uint64_t dpid = get_datapath_id(dp->netdev);
 811         int port_count = 0;
 812
 813         ofr->datapath_id  = cpu_to_be64(dpid);
 814
 815         ofr->n_buffers    = htonl(N_PKT_BUFFERS);
 816         ofr->n_tables     = dp->chain->n_tables;
 817         ofr->capabilities = htonl(OFP_SUPPORTED_CAPABILITIES);
 818         ofr->actions      = htonl(OFP_SUPPORTED_ACTIONS);
 819         memset(ofr->pad, 0, sizeof ofr->pad);
 820
 821         list_for_each_entry_rcu (p, &dp->port_list, node) {
 822                 fill_port_desc(p, &ofr->ports[port_count]);
 823                 port_count++;
 824         }
 825
 826         return port_count;
 827 }
 828
 829 int
 830 dp_send_features_reply(struct datapath *dp, const struct sender *sender)
 831 {
 832         struct sk_buff *skb;
 833         struct ofp_switch_features *ofr;
 834         size_t ofr_len, port_max_len;
 835         int port_count;
 836
 837         /* Overallocate. */
 838         port_max_len = sizeof(struct ofp_phy_port) * DP_MAX_PORTS;
 839         ofr = alloc_openflow_skb(dp, sizeof(*ofr) + port_max_len,
 840                                  OFPT_FEATURES_REPLY, sender, &skb);
 841         if (!ofr)
 842                 return -ENOMEM;
 843
 844         /* Fill. */
 845         port_count = fill_features_reply(dp, ofr);
 846
 847         /* Shrink to fit. */
 848         ofr_len = sizeof(*ofr) + (sizeof(struct ofp_phy_port) * port_count);
 849         resize_openflow_skb(skb, &ofr->header, ofr_len);
 850         return send_openflow_skb(skb, sender);
 851 }
 852
 853 int
 854 dp_send_config_reply(struct datapath *dp, const struct sender *sender)
 855 {
 856         struct sk_buff *skb;
 857         struct ofp_switch_config *osc;
 858
 859         osc = alloc_openflow_skb(dp, sizeof *osc, OFPT_GET_CONFIG_REPLY, sender,
 860                                  &skb);
 861         if (!osc)
 862                 return -ENOMEM;
 863
 864         osc->flags = htons(dp->flags);
 865         osc->miss_send_len = htons(dp->miss_send_len);
 866
 867         return send_openflow_skb(skb, sender);
 868 }
 869
 870 int
 871 dp_send_hello(struct datapath *dp, const struct sender *sender,
 872               const struct ofp_header *request)
 873 {
 874         if (request->version < OFP_VERSION) {
 875                 char err[64];
 876                 sprintf(err, "Only version 0x%02x supported", OFP_VERSION);
 877                 dp_send_error_msg(dp, sender, OFPET_HELLO_FAILED,
 878                                   OFPHFC_INCOMPATIBLE, err, strlen(err));
 879                 return -EINVAL;
 880         } else {
 881                 struct sk_buff *skb;
 882                 struct ofp_header *reply;
 883
 884                 reply = alloc_openflow_skb(dp, sizeof *reply,
 885                                            OFPT_HELLO, sender, &skb);
 886                 if (!reply)
 887                         return -ENOMEM;
 888
 889                 return send_openflow_skb(skb, sender);
 890         }
 891 }
 892
 893 /* Callback function for a workqueue to disable an interface */
 894 static void
 895 down_port_cb(struct work_struct *work)
 896 {
 897         struct net_bridge_port *p = container_of(work, struct net_bridge_port,
 898                         port_task);
 899
 900         rtnl_lock();
 901         if (dev_change_flags(p->dev, p->dev->flags & ~IFF_UP) < 0)
 902                 if (net_ratelimit())
 903                         printk("problem bringing up port %s\n", p->dev->name);
 904         rtnl_unlock();
 905         p->config |= OFPPC_PORT_DOWN;
 906 }
 907
 908 /* Callback function for a workqueue to enable an interface */
 909 static void
 910 up_port_cb(struct work_struct *work)
 911 {
 912         struct net_bridge_port *p = container_of(work, struct net_bridge_port,
 913                         port_task);
 914
 915         rtnl_lock();
 916         if (dev_change_flags(p->dev, p->dev->flags | IFF_UP) < 0)
 917                 if (net_ratelimit())
 918                         printk("problem bringing down port %s\n", p->dev->name);
 919         rtnl_unlock();
 920         p->config &= ~OFPPC_PORT_DOWN;
 921 }
 922
 923 int
 924 dp_update_port_flags(struct datapath *dp, const struct ofp_port_mod *opm)
 925 {
 926         unsigned long int flags;
 927         int port_no = ntohs(opm->port_no);
 928         struct net_bridge_port *p;
 929         p = (port_no < DP_MAX_PORTS ? dp->ports[port_no]
 930              : port_no == OFPP_LOCAL ? dp->local_port
 931              : NULL);
 932
 933         /* Make sure the port id hasn't changed since this was sent */
 934         if (!p || memcmp(opm->hw_addr, p->dev->dev_addr, ETH_ALEN))
 935                 return -1;
 936
 937         spin_lock_irqsave(&p->lock, flags);
 938         if (opm->mask) {
 939                 uint32_t config_mask = ntohl(opm->mask);
 940                 p->config &= ~config_mask;
 941                 p->config |= ntohl(opm->config) & config_mask;
 942         }
 943
 944         /* Modifying the status of an interface requires taking a lock
 945          * that cannot be done from here.  For this reason, we use a shared
 946          * workqueue, which will cause it to be executed from a safer
 947          * context. */
 948         if (opm->mask & htonl(OFPPC_PORT_DOWN)) {
 949                 if ((opm->config & htonl(OFPPC_PORT_DOWN))
 950                     && (p->config & OFPPC_PORT_DOWN) == 0) {
 951                         PREPARE_WORK(&p->port_task, down_port_cb);
 952                         schedule_work(&p->port_task);
 953                 } else if ((opm->config & htonl(OFPPC_PORT_DOWN)) == 0
 954                            && (p->config & OFPPC_PORT_DOWN)) {
 955                         PREPARE_WORK(&p->port_task, up_port_cb);
 956                         schedule_work(&p->port_task);
 957                 }
 958         }
 959         spin_unlock_irqrestore(&p->lock, flags);
 960
 961         return 0;
 962 }
 963
 964 /* Initialize the port status field of the bridge port. */
 965 static void
 966 init_port_status(struct net_bridge_port *p)
 967 {
 968         unsigned long int flags;
 969
 970         spin_lock_irqsave(&p->lock, flags);
 971
 972         if (p->dev->flags & IFF_UP)
 973                 p->config &= ~OFPPC_PORT_DOWN;
 974         else
 975                 p->config |= OFPPC_PORT_DOWN;
 976
 977         if (netif_carrier_ok(p->dev))
 978                 p->state &= ~OFPPS_LINK_DOWN;
 979         else
 980                 p->state |= OFPPS_LINK_DOWN;
 981
 982         spin_unlock_irqrestore(&p->lock, flags);
 983 }
 984
 985 int
 986 dp_send_port_status(struct net_bridge_port *p, uint8_t status)
 987 {
 988         struct sk_buff *skb;
 989         struct ofp_port_status *ops;
 990
 991         ops = alloc_openflow_skb(p->dp, sizeof *ops, OFPT_PORT_STATUS, NULL,
 992                                  &skb);
 993         if (!ops)
 994                 return -ENOMEM;
 995         ops->reason = status;
 996         memset(ops->pad, 0, sizeof ops->pad);
 997         fill_port_desc(p, &ops->desc);
 998
 999         return send_openflow_skb(skb, NULL);
1000 }
1001
1002 /* Convert jiffies_64 to milliseconds. */
1003 static u64 inline jiffies_64_to_msecs(const u64 j)
1004 {
1005 #if HZ <= MSEC_PER_SEC && !(MSEC_PER_SEC % HZ)
1006                 return (MSEC_PER_SEC / HZ) * j;
1007 #elif HZ > MSEC_PER_SEC && !(HZ % MSEC_PER_SEC)
1008                 return (j + (HZ / MSEC_PER_SEC) - 1)/(HZ / MSEC_PER_SEC);
1009 #else
1010                 return (j * MSEC_PER_SEC) / HZ;
1011 #endif
1012 }
1013
1014 int
1015 dp_send_flow_end(struct datapath *dp, struct sw_flow *flow,
1016                      enum nx_flow_end_reason reason)
1017 {
1018         struct sk_buff *skb;
1019         struct nx_flow_end *nfe;
1020
1021         if (!dp->send_flow_end)
1022                 return 0;
1023
1024         nfe = alloc_openflow_skb(dp, sizeof *nfe, OFPT_VENDOR, 0, &skb);
1025         if (!nfe)
1026                 return -ENOMEM;
1027
1028         nfe->header.vendor = htonl(NX_VENDOR_ID);
1029         nfe->header.subtype = htonl(NXT_FLOW_END);
1030
1031         flow_fill_match(&nfe->match, &flow->key);
1032
1033         nfe->priority = htons(flow->priority);
1034         nfe->reason = reason;
1035
1036         nfe->tcp_flags = flow->tcp_flags;
1037         nfe->ip_tos = flow->ip_tos;
1038
1039         memset(nfe->pad, 0, sizeof nfe->pad);
1040
1041         nfe->init_time = cpu_to_be64(jiffies_64_to_msecs(flow->created));
1042         nfe->used_time = cpu_to_be64(jiffies_64_to_msecs(flow->used));
1043         nfe->end_time = cpu_to_be64(jiffies_64_to_msecs(get_jiffies_64()));
1044
1045         nfe->packet_count = cpu_to_be64(flow->packet_count);
1046         nfe->byte_count   = cpu_to_be64(flow->byte_count);
1047
1048         return send_openflow_skb(skb, NULL);
1049 }
1050 EXPORT_SYMBOL(dp_send_flow_end);
1051
1052 int
1053 dp_send_error_msg(struct datapath *dp, const struct sender *sender,
1054                 uint16_t type, uint16_t code, const void *data, size_t len)
1055 {
1056         struct sk_buff *skb;
1057         struct ofp_error_msg *oem;
1058
1059
1060         oem = alloc_openflow_skb(dp, sizeof(*oem)+len, OFPT_ERROR,
1061                         sender, &skb);
1062         if (!oem)
1063                 return -ENOMEM;
1064
1065         oem->type = htons(type);
1066         oem->code = htons(code);
1067         memcpy(oem->data, data, len);
1068
1069         return send_openflow_skb(skb, sender);
1070 }
1071
1072 int
1073 dp_send_echo_reply(struct datapath *dp, const struct sender *sender,
1074                    const struct ofp_header *rq)
1075 {
1076         struct sk_buff *skb;
1077         struct ofp_header *reply;
1078
1079         reply = alloc_openflow_skb(dp, ntohs(rq->length), OFPT_ECHO_REPLY,
1080                                    sender, &skb);
1081         if (!reply)
1082                 return -ENOMEM;
1083
1084         memcpy(reply + 1, rq + 1, ntohs(rq->length) - sizeof *rq);
1085         return send_openflow_skb(skb, sender);
1086 }
1087
1088 /* Generic Netlink interface.
1089  *
1090  * See netlink(7) for an introduction to netlink.  See
1091  * http://linux-net.osdl.org/index.php/Netlink for more information and
1092  * pointers on how to work with netlink and Generic Netlink in the kernel and
1093  * in userspace. */
1094
1095 static struct genl_family dp_genl_family = {
1096         .id = GENL_ID_GENERATE,
1097         .hdrsize = 0,
1098         .name = DP_GENL_FAMILY_NAME,
1099         .version = 1,
1100         .maxattr = DP_GENL_A_MAX,
1101 };
1102
1103 /* Attribute policy: what each attribute may contain.  */
1104 static struct nla_policy dp_genl_policy[DP_GENL_A_MAX + 1] = {
1105         [DP_GENL_A_DP_IDX] = { .type = NLA_U32 },
1106         [DP_GENL_A_MC_GROUP] = { .type = NLA_U32 },
1107         [DP_GENL_A_PORTNAME] = { .type = NLA_STRING }
1108 };
1109
1110 static int dp_genl_add(struct sk_buff *skb, struct genl_info *info)
1111 {
1112         if (!info->attrs[DP_GENL_A_DP_IDX])
1113                 return -EINVAL;
1114
1115         return new_dp(nla_get_u32(info->attrs[DP_GENL_A_DP_IDX]));
1116 }
1117
1118 static struct genl_ops dp_genl_ops_add_dp = {
1119         .cmd = DP_GENL_C_ADD_DP,
1120         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1121         .policy = dp_genl_policy,
1122         .doit = dp_genl_add,
1123         .dumpit = NULL,
1124 };
1125
1126 struct datapath *dp_get(int dp_idx)
1127 {
1128         if (dp_idx < 0 || dp_idx >= DP_MAX)
1129                 return NULL;
1130         return rcu_dereference(dps[dp_idx]);
1131 }
1132
1133 static int dp_genl_del(struct sk_buff *skb, struct genl_info *info)
1134 {
1135         struct datapath *dp;
1136         int err;
1137
1138         if (!info->attrs[DP_GENL_A_DP_IDX])
1139                 return -EINVAL;
1140
1141         dp = dp_get(nla_get_u32(info->attrs[DP_GENL_A_DP_IDX]));
1142         if (!dp)
1143                 err = -ENOENT;
1144         else {
1145                 del_dp(dp);
1146                 err = 0;
1147         }
1148         return err;
1149 }
1150
1151 static struct genl_ops dp_genl_ops_del_dp = {
1152         .cmd = DP_GENL_C_DEL_DP,
1153         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1154         .policy = dp_genl_policy,
1155         .doit = dp_genl_del,
1156         .dumpit = NULL,
1157 };
1158
1159 /* Queries a datapath for related information.  Currently the only relevant
1160  * information is the datapath's multicast group ID.  Really we want one
1161  * multicast group per datapath, but because of locking issues[*] we can't
1162  * easily get one.  Thus, every datapath will currently return the same
1163  * global multicast group ID, but in the future it would be nice to fix that.
1164  *
1165  * [*] dp_genl_add, to add a new datapath, is called under the genl_lock
1166  *       mutex, and genl_register_mc_group, called to acquire a new multicast
1167  *       group ID, also acquires genl_lock, thus deadlock.
1168  */
1169 static int dp_genl_query(struct sk_buff *skb, struct genl_info *info)
1170 {
1171         struct datapath *dp;
1172         struct sk_buff *ans_skb = NULL;
1173         int dp_idx;
1174         int err = -ENOMEM;
1175
1176         if (!info->attrs[DP_GENL_A_DP_IDX])
1177                 return -EINVAL;
1178
1179         rcu_read_lock();
1180         dp_idx = nla_get_u32((info->attrs[DP_GENL_A_DP_IDX]));
1181         dp = dp_get(dp_idx);
1182         if (!dp)
1183                 err = -ENOENT;
1184         else {
1185                 void *data;
1186                 ans_skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1187                 if (!ans_skb) {
1188                         err = -ENOMEM;
1189                         goto err;
1190                 }
1191                 data = genlmsg_put_reply(ans_skb, info, &dp_genl_family,
1192                                          0, DP_GENL_C_QUERY_DP);
1193                 if (data == NULL) {
1194                         err = -ENOMEM;
1195                         goto err;
1196                 }
1197                 NLA_PUT_U32(ans_skb, DP_GENL_A_DP_IDX, dp_idx);
1198                 NLA_PUT_U32(ans_skb, DP_GENL_A_MC_GROUP, mc_group.id);
1199
1200                 genlmsg_end(ans_skb, data);
1201                 err = genlmsg_reply(ans_skb, info);
1202                 ans_skb = NULL;
1203         }
1204 err:
1205 nla_put_failure:
1206         kfree_skb(ans_skb);
1207         rcu_read_unlock();
1208         return err;
1209 }
1210
1211 static struct genl_ops dp_genl_ops_query_dp = {
1212         .cmd = DP_GENL_C_QUERY_DP,
1213         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1214         .policy = dp_genl_policy,
1215         .doit = dp_genl_query,
1216         .dumpit = NULL,
1217 };
1218
1219 static int dp_genl_add_del_port(struct sk_buff *skb, struct genl_info *info)
1220 {
1221         struct datapath *dp;
1222         struct net_device *port;
1223         int err;
1224
1225         if (!info->attrs[DP_GENL_A_DP_IDX] || !info->attrs[DP_GENL_A_PORTNAME])
1226                 return -EINVAL;
1227
1228         /* Get datapath. */
1229         dp = dp_get(nla_get_u32(info->attrs[DP_GENL_A_DP_IDX]));
1230         if (!dp) {
1231                 err = -ENOENT;
1232                 goto out;
1233         }
1234
1235         /* Get interface to add/remove. */
1236         port = dev_get_by_name(&init_net,
1237                         nla_data(info->attrs[DP_GENL_A_PORTNAME]));
1238         if (!port) {
1239                 err = -ENOENT;
1240                 goto out;
1241         }
1242
1243         /* Execute operation. */
1244         if (info->genlhdr->cmd == DP_GENL_C_ADD_PORT)
1245                 err = add_switch_port(dp, port);
1246         else {
1247                 if (port->br_port == NULL || port->br_port->dp != dp) {
1248                         err = -ENOENT;
1249                         goto out_put;
1250                 }
1251                 err = dp_del_switch_port(port->br_port);
1252         }
1253
1254 out_put:
1255         dev_put(port);
1256 out:
1257         return err;
1258 }
1259
1260 static struct genl_ops dp_genl_ops_add_port = {
1261         .cmd = DP_GENL_C_ADD_PORT,
1262         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1263         .policy = dp_genl_policy,
1264         .doit = dp_genl_add_del_port,
1265         .dumpit = NULL,
1266 };
1267
1268 static struct genl_ops dp_genl_ops_del_port = {
1269         .cmd = DP_GENL_C_DEL_PORT,
1270         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1271         .policy = dp_genl_policy,
1272         .doit = dp_genl_add_del_port,
1273         .dumpit = NULL,
1274 };
1275
1276 static int dp_genl_openflow(struct sk_buff *skb, struct genl_info *info)
1277 {
1278         struct nlattr *va = info->attrs[DP_GENL_A_OPENFLOW];
1279         struct datapath *dp;
1280         struct ofp_header *oh;
1281         struct sender sender;
1282         int err;
1283
1284         if (!info->attrs[DP_GENL_A_DP_IDX] || !va)
1285                 return -EINVAL;
1286
1287         dp = dp_get(nla_get_u32(info->attrs[DP_GENL_A_DP_IDX]));
1288         if (!dp)
1289                 return -ENOENT;
1290
1291         if (nla_len(va) < sizeof(struct ofp_header))
1292                 return -EINVAL;
1293         oh = nla_data(va);
1294
1295         sender.xid = oh->xid;
1296         sender.pid = info->snd_pid;
1297         sender.seq = info->snd_seq;
1298
1299         mutex_lock(&dp_mutex);
1300         err = fwd_control_input(dp->chain, &sender,
1301                                 nla_data(va), nla_len(va));
1302         mutex_unlock(&dp_mutex);
1303         return err;
1304 }
1305
1306 static struct nla_policy dp_genl_openflow_policy[DP_GENL_A_MAX + 1] = {
1307         [DP_GENL_A_DP_IDX] = { .type = NLA_U32 },
1308 };
1309
1310 static int desc_stats_dump(struct datapath *dp, void *state,
1311                             void *body, int *body_len)
1312 {
1313         struct ofp_desc_stats *ods = body;
1314         int n_bytes = sizeof *ods;
1315
1316         if (n_bytes > *body_len) {
1317                 return -ENOBUFS;
1318         }
1319         *body_len = n_bytes;
1320
1321         strncpy(ods->mfr_desc, mfr_desc, sizeof ods->mfr_desc);
1322         strncpy(ods->hw_desc, hw_desc, sizeof ods->hw_desc);
1323         strncpy(ods->sw_desc, sw_desc, sizeof ods->sw_desc);
1324         strncpy(ods->serial_num, serial_num, sizeof ods->serial_num);
1325
1326         return 0;
1327 }
1328
1329 struct flow_stats_state {
1330         int table_idx;
1331         struct sw_table_position position;
1332         const struct ofp_flow_stats_request *rq;
1333
1334         void *body;
1335         int bytes_used, bytes_allocated;
1336 };
1337
1338 static int flow_stats_init(struct datapath *dp, const void *body, int body_len,
1339                            void **state)
1340 {
1341         const struct ofp_flow_stats_request *fsr = body;
1342         struct flow_stats_state *s = kmalloc(sizeof *s, GFP_ATOMIC);
1343         if (!s)
1344                 return -ENOMEM;
1345         s->table_idx = fsr->table_id == 0xff ? 0 : fsr->table_id;
1346         memset(&s->position, 0, sizeof s->position);
1347         s->rq = fsr;
1348         *state = s;
1349         return 0;
1350 }
1351
1352 static int flow_stats_dump_callback(struct sw_flow *flow, void *private)
1353 {
1354         struct sw_flow_actions *sf_acts = rcu_dereference(flow->sf_acts);
1355         struct flow_stats_state *s = private;
1356         struct ofp_flow_stats *ofs;
1357         int length;
1358         uint64_t duration;
1359
1360         length = sizeof *ofs + sf_acts->actions_len;
1361         if (length + s->bytes_used > s->bytes_allocated)
1362                 return 1;
1363
1364         ofs = s->body + s->bytes_used;
1365         ofs->length          = htons(length);
1366         ofs->table_id        = s->table_idx;
1367         ofs->pad             = 0;
1368         ofs->match.wildcards = htonl(flow->key.wildcards);
1369         ofs->match.in_port   = flow->key.in_port;
1370         memcpy(ofs->match.dl_src, flow->key.dl_src, ETH_ALEN);
1371         memcpy(ofs->match.dl_dst, flow->key.dl_dst, ETH_ALEN);
1372         ofs->match.dl_vlan   = flow->key.dl_vlan;
1373         ofs->match.dl_type   = flow->key.dl_type;
1374         ofs->match.nw_src    = flow->key.nw_src;
1375         ofs->match.nw_dst    = flow->key.nw_dst;
1376         ofs->match.nw_proto  = flow->key.nw_proto;
1377         ofs->match.pad       = 0;
1378         ofs->match.tp_src    = flow->key.tp_src;
1379         ofs->match.tp_dst    = flow->key.tp_dst;
1380
1381         /* The kernel doesn't support 64-bit division, so use the 'do_div'
1382          * macro instead.  The first argument is replaced with the quotient,
1383          * while the remainder is the return value. */
1384         duration = get_jiffies_64() - flow->created;
1385         do_div(duration, HZ);
1386         ofs->duration        = htonl(duration);
1387
1388         ofs->priority        = htons(flow->priority);
1389         ofs->idle_timeout    = htons(flow->idle_timeout);
1390         ofs->hard_timeout    = htons(flow->hard_timeout);
1391         memset(ofs->pad2, 0, sizeof ofs->pad2);
1392         ofs->packet_count    = cpu_to_be64(flow->packet_count);
1393         ofs->byte_count      = cpu_to_be64(flow->byte_count);
1394         memcpy(ofs->actions, sf_acts->actions, sf_acts->actions_len);
1395
1396         s->bytes_used += length;
1397         return 0;
1398 }
1399
1400 static int flow_stats_dump(struct datapath *dp, void *state,
1401                            void *body, int *body_len)
1402 {
1403         struct flow_stats_state *s = state;
1404         struct sw_flow_key match_key;
1405         int error = 0;
1406
1407         s->bytes_used = 0;
1408         s->bytes_allocated = *body_len;
1409         s->body = body;
1410
1411         flow_extract_match(&match_key, &s->rq->match);
1412         while (s->table_idx < dp->chain->n_tables
1413                && (s->rq->table_id == 0xff || s->rq->table_id == s->table_idx))
1414         {
1415                 struct sw_table *table = dp->chain->tables[s->table_idx];
1416
1417                 error = table->iterate(table, &match_key, s->rq->out_port,
1418                                 &s->position, flow_stats_dump_callback, s);
1419                 if (error)
1420                         break;
1421
1422                 s->table_idx++;
1423                 memset(&s->position, 0, sizeof s->position);
1424         }
1425         *body_len = s->bytes_used;
1426
1427         /* If error is 0, we're done.
1428          * Otherwise, if some bytes were used, there are more flows to come.
1429          * Otherwise, we were not able to fit even a single flow in the body,
1430          * which indicates that we have a single flow with too many actions to
1431          * fit.  We won't ever make any progress at that rate, so give up. */
1432         return !error ? 0 : s->bytes_used ? 1 : -ENOMEM;
1433 }
1434
1435 static void flow_stats_done(void *state)
1436 {
1437         kfree(state);
1438 }
1439
1440 static int aggregate_stats_init(struct datapath *dp,
1441                                 const void *body, int body_len,
1442                                 void **state)
1443 {
1444         *state = (void *)body;
1445         return 0;
1446 }
1447
1448 static int aggregate_stats_dump_callback(struct sw_flow *flow, void *private)
1449 {
1450         struct ofp_aggregate_stats_reply *rpy = private;
1451         rpy->packet_count += flow->packet_count;
1452         rpy->byte_count += flow->byte_count;
1453         rpy->flow_count++;
1454         return 0;
1455 }
1456
1457 static int aggregate_stats_dump(struct datapath *dp, void *state,
1458                                 void *body, int *body_len)
1459 {
1460         struct ofp_aggregate_stats_request *rq = state;
1461         struct ofp_aggregate_stats_reply *rpy;
1462         struct sw_table_position position;
1463         struct sw_flow_key match_key;
1464         int table_idx;
1465
1466         if (*body_len < sizeof *rpy)
1467                 return -ENOBUFS;
1468         rpy = body;
1469         *body_len = sizeof *rpy;
1470
1471         memset(rpy, 0, sizeof *rpy);
1472
1473         flow_extract_match(&match_key, &rq->match);
1474         table_idx = rq->table_id == 0xff ? 0 : rq->table_id;
1475         memset(&position, 0, sizeof position);
1476         while (table_idx < dp->chain->n_tables
1477                && (rq->table_id == 0xff || rq->table_id == table_idx))
1478         {
1479                 struct sw_table *table = dp->chain->tables[table_idx];
1480                 int error;
1481
1482                 error = table->iterate(table, &match_key, rq->out_port, &position,
1483                                        aggregate_stats_dump_callback, rpy);
1484                 if (error)
1485                         return error;
1486
1487                 table_idx++;
1488                 memset(&position, 0, sizeof position);
1489         }
1490
1491         rpy->packet_count = cpu_to_be64(rpy->packet_count);
1492         rpy->byte_count = cpu_to_be64(rpy->byte_count);
1493         rpy->flow_count = htonl(rpy->flow_count);
1494         return 0;
1495 }
1496
1497 static int table_stats_dump(struct datapath *dp, void *state,
1498                             void *body, int *body_len)
1499 {
1500         struct ofp_table_stats *ots;
1501         int n_bytes = dp->chain->n_tables * sizeof *ots;
1502         int i;
1503         if (n_bytes > *body_len)
1504                 return -ENOBUFS;
1505         *body_len = n_bytes;
1506         for (i = 0, ots = body; i < dp->chain->n_tables; i++, ots++) {
1507                 struct sw_table_stats stats;
1508                 dp->chain->tables[i]->stats(dp->chain->tables[i], &stats);
1509                 strncpy(ots->name, stats.name, sizeof ots->name);
1510                 ots->table_id = i;
1511                 ots->wildcards = htonl(stats.wildcards);
1512                 memset(ots->pad, 0, sizeof ots->pad);
1513                 ots->max_entries = htonl(stats.max_flows);
1514                 ots->active_count = htonl(stats.n_flows);
1515                 ots->lookup_count = cpu_to_be64(stats.n_lookup);
1516                 ots->matched_count = cpu_to_be64(stats.n_matched);
1517         }
1518         return 0;
1519 }
1520
1521 struct port_stats_state {
1522         int port;
1523 };
1524
1525 static int port_stats_init(struct datapath *dp, const void *body, int body_len,
1526                            void **state)
1527 {
1528         struct port_stats_state *s = kmalloc(sizeof *s, GFP_ATOMIC);
1529         if (!s)
1530                 return -ENOMEM;
1531         s->port = 0;
1532         *state = s;
1533         return 0;
1534 }
1535
1536 static int port_stats_dump(struct datapath *dp, void *state,
1537                            void *body, int *body_len)
1538 {
1539         struct port_stats_state *s = state;
1540         struct ofp_port_stats *ops;
1541         int n_ports, max_ports;
1542         int i;
1543
1544         max_ports = *body_len / sizeof *ops;
1545         if (!max_ports)
1546                 return -ENOMEM;
1547         ops = body;
1548
1549         n_ports = 0;
1550         for (i = s->port; i < DP_MAX_PORTS && n_ports < max_ports; i++) {
1551                 struct net_bridge_port *p = dp->ports[i];
1552                 struct net_device_stats *stats;
1553                 if (!p)
1554                         continue;
1555                 stats = p->dev->get_stats(p->dev);
1556                 ops->port_no = htons(p->port_no);
1557                 memset(ops->pad, 0, sizeof ops->pad);
1558                 ops->rx_packets   = cpu_to_be64(stats->rx_packets);
1559                 ops->tx_packets   = cpu_to_be64(stats->tx_packets);
1560                 ops->rx_bytes     = cpu_to_be64(stats->rx_bytes);
1561                 ops->tx_bytes     = cpu_to_be64(stats->tx_bytes);
1562                 ops->rx_dropped   = cpu_to_be64(stats->rx_dropped);
1563                 ops->tx_dropped   = cpu_to_be64(stats->tx_dropped);
1564                 ops->rx_errors    = cpu_to_be64(stats->rx_errors);
1565                 ops->tx_errors    = cpu_to_be64(stats->tx_errors);
1566                 ops->rx_frame_err = cpu_to_be64(stats->rx_frame_errors);
1567                 ops->rx_over_err  = cpu_to_be64(stats->rx_over_errors);
1568                 ops->rx_crc_err   = cpu_to_be64(stats->rx_crc_errors);
1569                 ops->collisions   = cpu_to_be64(stats->collisions);
1570                 n_ports++;
1571                 ops++;
1572         }
1573         s->port = i;
1574         *body_len = n_ports * sizeof *ops;
1575         return n_ports >= max_ports;
1576 }
1577
1578 static void port_stats_done(void *state)
1579 {
1580         kfree(state);
1581 }
1582
1583 struct stats_type {
1584         /* Minimum and maximum acceptable number of bytes in body member of
1585          * struct ofp_stats_request. */
1586         size_t min_body, max_body;
1587
1588         /* Prepares to dump some kind of statistics on 'dp'.  'body' and
1589          * 'body_len' are the 'body' member of the struct ofp_stats_request.
1590          * Returns zero if successful, otherwise a negative error code.
1591          * May initialize '*state' to state information.  May be null if no
1592          * initialization is required.*/
1593         int (*init)(struct datapath *dp, const void *body, int body_len,
1594                     void **state);
1595
1596         /* Dumps statistics for 'dp' into the '*body_len' bytes at 'body', and
1597          * modifies '*body_len' to reflect the number of bytes actually used.
1598          * ('body' will be transmitted as the 'body' member of struct
1599          * ofp_stats_reply.) */
1600         int (*dump)(struct datapath *dp, void *state,
1601                     void *body, int *body_len);
1602
1603         /* Cleans any state created by the init or dump functions.  May be null
1604          * if no cleanup is required. */
1605         void (*done)(void *state);
1606 };
1607
1608 static const struct stats_type stats[] = {
1609         [OFPST_DESC] = {
1610                 0,
1611                 0,
1612                 NULL,
1613                 desc_stats_dump,
1614                 NULL
1615         },
1616         [OFPST_FLOW] = {
1617                 sizeof(struct ofp_flow_stats_request),
1618                 sizeof(struct ofp_flow_stats_request),
1619                 flow_stats_init,
1620                 flow_stats_dump,
1621                 flow_stats_done
1622         },
1623         [OFPST_AGGREGATE] = {
1624                 sizeof(struct ofp_aggregate_stats_request),
1625                 sizeof(struct ofp_aggregate_stats_request),
1626                 aggregate_stats_init,
1627                 aggregate_stats_dump,
1628                 NULL
1629         },
1630         [OFPST_TABLE] = {
1631                 0,
1632                 0,
1633                 NULL,
1634                 table_stats_dump,
1635                 NULL
1636         },
1637         [OFPST_PORT] = {
1638                 0,
1639                 0,
1640                 port_stats_init,
1641                 port_stats_dump,
1642                 port_stats_done
1643         },
1644 };
1645
1646 static int
1647 dp_genl_openflow_dumpit(struct sk_buff *skb, struct netlink_callback *cb)
1648 {
1649         struct datapath *dp;
1650         struct sender sender;
1651         const struct stats_type *s;
1652         struct ofp_stats_reply *osr;
1653         int dp_idx;
1654         int max_openflow_len, body_len;
1655         void *body;
1656         int err;
1657
1658         /* Set up the cleanup function for this dump.  Linux 2.6.20 and later
1659          * support setting up cleanup functions via the .doneit member of
1660          * struct genl_ops.  This kluge supports earlier versions also. */
1661         cb->done = dp_genl_openflow_done;
1662
1663         sender.pid = NETLINK_CB(cb->skb).pid;
1664         sender.seq = cb->nlh->nlmsg_seq;
1665         if (!cb->args[0]) {
1666                 struct nlattr *attrs[DP_GENL_A_MAX + 1];
1667                 struct ofp_stats_request *rq;
1668                 struct nlattr *va;
1669                 size_t len, body_len;
1670                 int type;
1671
1672                 err = nlmsg_parse(cb->nlh, GENL_HDRLEN, attrs, DP_GENL_A_MAX,
1673                                   dp_genl_openflow_policy);
1674                 if (err < 0)
1675                         return err;
1676
1677                 if (!attrs[DP_GENL_A_DP_IDX])
1678                         return -EINVAL;
1679                 dp_idx = nla_get_u16(attrs[DP_GENL_A_DP_IDX]);
1680                 dp = dp_get(dp_idx);
1681                 if (!dp)
1682                         return -ENOENT;
1683
1684                 va = attrs[DP_GENL_A_OPENFLOW];
1685                 len = nla_len(va);
1686                 if (!va || len < sizeof *rq)
1687                         return -EINVAL;
1688
1689                 rq = nla_data(va);
1690                 sender.xid = rq->header.xid;
1691                 type = ntohs(rq->type);
1692                 if (rq->header.version != OFP_VERSION) {
1693                         dp_send_error_msg(dp, &sender, OFPET_BAD_REQUEST,
1694                                           OFPBRC_BAD_VERSION, rq, len);
1695                         return -EINVAL;
1696                 }
1697                 if (rq->header.type != OFPT_STATS_REQUEST
1698                     || ntohs(rq->header.length) != len)
1699                         return -EINVAL;
1700
1701                 if (type >= ARRAY_SIZE(stats) || !stats[type].dump) {
1702                         dp_send_error_msg(dp, &sender, OFPET_BAD_REQUEST,
1703                                           OFPBRC_BAD_STAT, rq, len);
1704                         return -EINVAL;
1705                 }
1706
1707                 s = &stats[type];
1708                 body_len = len - offsetof(struct ofp_stats_request, body);
1709                 if (body_len < s->min_body || body_len > s->max_body)
1710                         return -EINVAL;
1711
1712                 cb->args[0] = 1;
1713                 cb->args[1] = dp_idx;
1714                 cb->args[2] = type;
1715                 cb->args[3] = rq->header.xid;
1716                 if (s->init) {
1717                         void *state;
1718                         err = s->init(dp, rq->body, body_len, &state);
1719                         if (err)
1720                                 return err;
1721                         cb->args[4] = (long) state;
1722                 }
1723         } else if (cb->args[0] == 1) {
1724                 sender.xid = cb->args[3];
1725                 dp_idx = cb->args[1];
1726                 s = &stats[cb->args[2]];
1727
1728                 dp = dp_get(dp_idx);
1729                 if (!dp)
1730                         return -ENOENT;
1731         } else {
1732                 return 0;
1733         }
1734
1735         osr = put_openflow_headers(dp, skb, OFPT_STATS_REPLY, &sender,
1736                                    &max_openflow_len);
1737         if (IS_ERR(osr))
1738                 return PTR_ERR(osr);
1739         osr->type = htons(s - stats);
1740         osr->flags = 0;
1741         resize_openflow_skb(skb, &osr->header, max_openflow_len);
1742         body = osr->body;
1743         body_len = max_openflow_len - offsetof(struct ofp_stats_reply, body);
1744
1745         err = s->dump(dp, (void *) cb->args[4], body, &body_len);
1746         if (err >= 0) {
1747                 if (!err)
1748                         cb->args[0] = 2;
1749                 else
1750                         osr->flags = ntohs(OFPSF_REPLY_MORE);
1751                 resize_openflow_skb(skb, &osr->header,
1752                                     (offsetof(struct ofp_stats_reply, body)
1753                                      + body_len));
1754                 err = skb->len;
1755         }
1756
1757         return err;
1758 }
1759
1760 static int
1761 dp_genl_openflow_done(struct netlink_callback *cb)
1762 {
1763         if (cb->args[0]) {
1764                 const struct stats_type *s = &stats[cb->args[2]];
1765                 if (s->done)
1766                         s->done((void *) cb->args[4]);
1767         }
1768         return 0;
1769 }
1770
1771 static struct genl_ops dp_genl_ops_openflow = {
1772         .cmd = DP_GENL_C_OPENFLOW,
1773         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1774         .policy = dp_genl_openflow_policy,
1775         .doit = dp_genl_openflow,
1776         .dumpit = dp_genl_openflow_dumpit,
1777 };
1778
1779 static struct genl_ops *dp_genl_all_ops[] = {
1780         /* Keep this operation first.  Generic Netlink dispatching
1781          * looks up operations with linear search, so we want it at the
1782          * front. */
1783         &dp_genl_ops_openflow,
1784
1785         &dp_genl_ops_add_dp,
1786         &dp_genl_ops_del_dp,
1787         &dp_genl_ops_query_dp,
1788         &dp_genl_ops_add_port,
1789         &dp_genl_ops_del_port,
1790 };
1791
1792 static int dp_init_netlink(void)
1793 {
1794         int err;
1795         int i;
1796
1797         err = genl_register_family(&dp_genl_family);
1798         if (err)
1799                 return err;
1800
1801         for (i = 0; i < ARRAY_SIZE(dp_genl_all_ops); i++) {
1802                 err = genl_register_ops(&dp_genl_family, dp_genl_all_ops[i]);
1803                 if (err)
1804                         goto err_unregister;
1805         }
1806
1807         strcpy(mc_group.name, "openflow");
1808         err = genl_register_mc_group(&dp_genl_family, &mc_group);
1809         if (err < 0)
1810                 goto err_unregister;
1811
1812         return 0;
1813
1814 err_unregister:
1815         genl_unregister_family(&dp_genl_family);
1816                 return err;
1817 }
1818
1819 static void dp_uninit_netlink(void)
1820 {
1821         genl_unregister_family(&dp_genl_family);
1822 }
1823
1824 /* Set the description strings if appropriate values are available from
1825  * the DMI. */
1826 static void set_desc(void)
1827 {
1828         const char *uuid = dmi_get_system_info(DMI_PRODUCT_UUID);
1829         const char *uptr = uuid + 24;
1830
1831         if (!uuid || *uuid == '\0' || strlen(uuid) != 36)
1832                 return;
1833
1834         /* We are only interested version 1 UUIDs, since the last six bytes
1835          * are an IEEE 802 MAC address. */
1836         if (uuid[14] != '1')
1837                 return;
1838
1839         /* Only set if the UUID is from Nicira. */
1840         if (strncmp(uptr, NICIRA_OUI_STR, strlen(NICIRA_OUI_STR)))
1841                 return;
1842
1843         strlcpy(mfr_desc, dmi_get_system_info(DMI_SYS_VENDOR), sizeof(mfr_desc));
1844         snprintf(hw_desc, sizeof(hw_desc), "%s %s",
1845                         dmi_get_system_info(DMI_PRODUCT_NAME),
1846                         dmi_get_system_info(DMI_PRODUCT_VERSION));
1847         strlcpy(serial_num, dmi_get_system_info(DMI_PRODUCT_SERIAL),
1848                         sizeof(serial_num));
1849 }
1850
1851 static int __init dp_init(void)
1852 {
1853         int err;
1854
1855         printk("OpenFlow %s, built "__DATE__" "__TIME__", "
1856                "protocol 0x%02x\n", VERSION BUILDNR, OFP_VERSION);
1857
1858         err = flow_init();
1859         if (err)
1860                 goto error;
1861
1862         err = register_netdevice_notifier(&dp_device_notifier);
1863         if (err)
1864                 goto error_flow_exit;
1865
1866         err = dp_init_netlink();
1867         if (err)
1868                 goto error_unreg_notifier;
1869
1870         /* Check if better descriptions of the switch are available than the
1871          * defaults. */
1872         set_desc();
1873
1874         /* Hook into callback used by the bridge to intercept packets.
1875          * Parasites we are. */
1876         if (br_handle_frame_hook)
1877                 printk("openflow: hijacking bridge hook\n");
1878         br_handle_frame_hook = dp_frame_hook;
1879
1880         return 0;
1881
1882 error_unreg_notifier:
1883         unregister_netdevice_notifier(&dp_device_notifier);
1884 error_flow_exit:
1885         flow_exit();
1886 error:
1887         printk(KERN_EMERG "openflow: failed to install!");
1888         return err;
1889 }
1890
1891 static void dp_cleanup(void)
1892 {
1893         fwd_exit();
1894         dp_uninit_netlink();
1895         unregister_netdevice_notifier(&dp_device_notifier);
1896         flow_exit();
1897         br_handle_frame_hook = NULL;
1898 }
1899
1900 module_init(dp_init);
1901 module_exit(dp_cleanup);
1902
1903 MODULE_DESCRIPTION("OpenFlow switching datapath");
1904 MODULE_AUTHOR("Copyright (c) 2007, 2008 The Board of Trustees of The Leland Stanford Junior University");
1905 MODULE_LICENSE("GPL");