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