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/etherdevice.h>
  19 #include <linux/kernel.h>
  20 #include <linux/kthread.h>
  21 #include <linux/mutex.h>
  22 #include <linux/rtnetlink.h>
  23 #include <linux/rcupdate.h>
  24 #include <linux/version.h>
  25 #include <linux/ethtool.h>
  26 #include <linux/random.h>
  27 #include <asm/system.h>
  28 #include <linux/netfilter_bridge.h>
  29 #include <linux/netfilter_ipv4.h>
  30 #include <linux/inetdevice.h>
  31 #include <linux/list.h>
  32 #include <linux/rculist.h>
  33 #include <linux/workqueue.h>
  34
  35 #include "openflow-netlink.h"
  36 #include "datapath.h"
  37 #include "nx_act_snat.h"
  38 #include "table.h"
  39 #include "chain.h"
  40 #include "dp_dev.h"
  41 #include "forward.h"
  42 #include "flow.h"
  43
  44 #include "compat.h"
  45
  46
  47 /* Strings to describe the manufacturer, hardware, and software.  This data
  48  * is queriable through the switch description stats message. */
  49 static char mfr_desc[DESC_STR_LEN] = "Nicira Networks";
  50 static char hw_desc[DESC_STR_LEN] = "Reference Linux Kernel Module";
  51 static char sw_desc[DESC_STR_LEN] = VERSION;
  52 static char serial_num[SERIAL_NUM_LEN] = "None";
  53
  54 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)
  55 module_param_string(mfr_desc, mfr_desc, sizeof mfr_desc, 0444);
  56 module_param_string(hw_desc, hw_desc, sizeof hw_desc, 0444);
  57 module_param_string(sw_desc, sw_desc, sizeof sw_desc, 0444);
  58 module_param_string(serial_num, serial_num, sizeof serial_num, 0444);
  59 #else
  60 MODULE_PARM(mfr_desc, "s");
  61 MODULE_PARM(hw_desc, "s");
  62 MODULE_PARM(sw_desc, "s");
  63 MODULE_PARM(serial_num, "s");
  64 #endif
  65
  66
  67 /* Number of milliseconds between runs of the maintenance thread. */
  68 #define MAINT_SLEEP_MSECS 1000
  69
  70 #define UINT32_MAX                        4294967295U
  71 #define UINT16_MAX                        65535
  72 #define MAX(X, Y) ((X) > (Y) ? (X) : (Y))
  73
  74 static struct genl_family dp_genl_family;
  75 static struct genl_multicast_group mc_group;
  76
  77 /* It's hard to imagine wanting more than one datapath, but... */
  78 #define DP_MAX 32
  79
  80 /* Datapaths.  Protected on the read side by rcu_read_lock, on the write side
  81  * by dp_mutex.  dp_mutex is almost completely redundant with genl_mutex
  82  * maintained by the Generic Netlink code, but the timeout path needs mutual
  83  * exclusion too.
  84  *
  85  * It is safe to access the datapath and net_bridge_port structures with just
  86  * dp_mutex.
  87  */
  88 static struct datapath *dps[DP_MAX];
  89 DEFINE_MUTEX(dp_mutex);
  90 EXPORT_SYMBOL(dp_mutex);
  91
  92 static int dp_maint_func(void *data);
  93 static void init_port_status(struct net_bridge_port *p);
  94 static int dp_genl_openflow_done(struct netlink_callback *);
  95 static struct net_bridge_port *new_nbp(struct datapath *,
  96                                        struct net_device *, int port_no);
  97
  98 /* nla_shrink - reduce amount of space reserved by nla_reserve
  99  * @skb: socket buffer from which to recover room
 100  * @nla: netlink attribute to adjust
 101  * @len: new length of attribute payload
 102  *
 103  * Reduces amount of space reserved by a call to nla_reserve.
 104  *
 105  * No other attributes may be added between calling nla_reserve and this
 106  * function, since it will create a hole in the message.
 107  */
 108 void nla_shrink(struct sk_buff *skb, struct nlattr *nla, int len)
 109 {
 110         int delta = nla_total_size(len) - nla_total_size(nla_len(nla));
 111         BUG_ON(delta > 0);
 112         skb->tail += delta;
 113         skb->len  += delta;
 114         nla->nla_len = nla_attr_size(len);
 115 }
 116
 117 /* Puts a set of openflow headers for a message of the given 'type' into 'skb'.
 118  * If 'sender' is nonnull, then it is used as the message's destination.  'dp'
 119  * must specify the datapath to use.
 120  *
 121  * '*max_openflow_len' receives the maximum number of bytes that are available
 122  * for the embedded OpenFlow message.  The caller must call
 123  * resize_openflow_skb() to set the actual size of the message to this number
 124  * of bytes or less.
 125  *
 126  * Returns the openflow header if successful, otherwise (if 'skb' is too small)
 127  * an error code. */
 128 static void *
 129 put_openflow_headers(struct datapath *dp, struct sk_buff *skb, uint8_t type,
 130                      const struct sender *sender, int *max_openflow_len)
 131 {
 132         struct ofp_header *oh;
 133         struct nlattr *attr;
 134         int openflow_len;
 135
 136         /* Assemble the Generic Netlink wrapper. */
 137         if (!genlmsg_put(skb,
 138                          sender ? sender->pid : 0,
 139                          sender ? sender->seq : 0,
 140                          &dp_genl_family, 0, DP_GENL_C_OPENFLOW))
 141                 return ERR_PTR(-ENOBUFS);
 142         if (nla_put_u32(skb, DP_GENL_A_DP_IDX, dp->dp_idx) < 0)
 143                 return ERR_PTR(-ENOBUFS);
 144         openflow_len = (skb_tailroom(skb) - NLA_HDRLEN) & ~(NLA_ALIGNTO - 1);
 145         if (openflow_len < sizeof *oh)
 146                 return ERR_PTR(-ENOBUFS);
 147         *max_openflow_len = openflow_len;
 148         attr = nla_reserve(skb, DP_GENL_A_OPENFLOW, openflow_len);
 149         BUG_ON(!attr);
 150
 151         /* Fill in the header.  The caller is responsible for the length. */
 152         oh = nla_data(attr);
 153         oh->version = OFP_VERSION;
 154         oh->type = type;
 155         oh->xid = sender ? sender->xid : 0;
 156
 157         return oh;
 158 }
 159
 160 /* Resizes OpenFlow header 'oh', which must be at the tail end of 'skb', to new
 161  * length 'new_length' (in bytes), adjusting pointers and size values as
 162  * necessary. */
 163 static void
 164 resize_openflow_skb(struct sk_buff *skb,
 165                     struct ofp_header *oh, size_t new_length)
 166 {
 167         struct nlattr *attr = ((void *) oh) - NLA_HDRLEN;
 168         nla_shrink(skb, attr, new_length);
 169         oh->length = htons(new_length);
 170         nlmsg_end(skb, (struct nlmsghdr *) skb->data);
 171 }
 172
 173 /* Allocates a new skb to contain an OpenFlow message 'openflow_len' bytes in
 174  * length.  Returns a null pointer if memory is unavailable, otherwise returns
 175  * the OpenFlow header and stores a pointer to the skb in '*pskb'.
 176  *
 177  * 'type' is the OpenFlow message type.  If 'sender' is nonnull, then it is
 178  * used as the message's destination.  'dp' must specify the datapath to
 179  * use.  */
 180 static void *
 181 alloc_openflow_skb(struct datapath *dp, size_t openflow_len, uint8_t type,
 182                    const struct sender *sender, struct sk_buff **pskb)
 183 {
 184         struct ofp_header *oh;
 185         size_t genl_len;
 186         struct sk_buff *skb;
 187         int max_openflow_len;
 188
 189         if ((openflow_len + sizeof(struct ofp_header)) > UINT16_MAX) {
 190                 if (net_ratelimit())
 191                         printk("alloc_openflow_skb: openflow message too large: %zu\n",
 192                                         openflow_len);
 193                 return NULL;
 194         }
 195
 196         genl_len = nlmsg_total_size(GENL_HDRLEN + dp_genl_family.hdrsize);
 197         genl_len += nla_total_size(sizeof(uint32_t)); /* DP_GENL_A_DP_IDX */
 198         genl_len += nla_total_size(openflow_len);    /* DP_GENL_A_OPENFLOW */
 199         skb = *pskb = genlmsg_new(genl_len, GFP_ATOMIC);
 200         if (!skb) {
 201                 if (net_ratelimit())
 202                         printk("alloc_openflow_skb: genlmsg_new failed\n");
 203                 return NULL;
 204         }
 205
 206         oh = put_openflow_headers(dp, skb, type, sender, &max_openflow_len);
 207         BUG_ON(!oh || IS_ERR(oh));
 208         resize_openflow_skb(skb, oh, openflow_len);
 209
 210         return oh;
 211 }
 212
 213 /* Sends 'skb' to 'sender' if it is nonnull, otherwise multicasts 'skb' to all
 214  * listeners. */
 215 static int
 216 send_openflow_skb(struct sk_buff *skb, const struct sender *sender)
 217 {
 218         return (sender
 219                 ? genlmsg_unicast(skb, sender->pid)
 220                 : genlmsg_multicast(skb, 0, mc_group.id, GFP_ATOMIC));
 221 }
 222
 223 /* Generates a unique datapath id.  It incorporates the datapath index
 224  * and a hardware address, if available.  If not, it generates a random
 225  * one.
 226  */
 227 static
 228 uint64_t gen_datapath_id(uint16_t dp_idx)
 229 {
 230         uint64_t id;
 231         int i;
 232         struct net_device *dev;
 233
 234         /* The top 16 bits are used to identify the datapath.  The lower 48 bits
 235          * use an interface address.  */
 236         id = (uint64_t)dp_idx << 48;
 237         if ((dev = dev_get_by_name(&init_net, "ctl0"))
 238                         || (dev = dev_get_by_name(&init_net, "eth0"))) {
 239                 for (i=0; i<ETH_ALEN; i++) {
 240                         id |= (uint64_t)dev->dev_addr[i] << (8*(ETH_ALEN-1 - i));
 241                 }
 242                 dev_put(dev);
 243         } else {
 244                 /* Randomly choose the lower 48 bits if we cannot find an
 245                  * address and mark the most significant bit to indicate that
 246                  * this was randomly generated. */
 247                 uint8_t rand[ETH_ALEN];
 248                 get_random_bytes(rand, ETH_ALEN);
 249                 id |= (uint64_t)1 << 63;
 250                 for (i=0; i<ETH_ALEN; i++) {
 251                         id |= (uint64_t)rand[i] << (8*(ETH_ALEN-1 - i));
 252                 }
 253         }
 254
 255         return id;
 256 }
 257
 258 /* Creates a new datapath numbered 'dp_idx'.  Returns 0 for success or a
 259  * negative error code. */
 260 static int new_dp(int dp_idx)
 261 {
 262         struct datapath *dp;
 263         int err;
 264
 265         if (dp_idx < 0 || dp_idx >= DP_MAX)
 266                 return -EINVAL;
 267
 268         if (!try_module_get(THIS_MODULE))
 269                 return -ENODEV;
 270
 271         /* Exit early if a datapath with that number already exists. */
 272         if (dps[dp_idx]) {
 273                 err = -EEXIST;
 274                 goto err_unlock;
 275         }
 276
 277         err = -ENOMEM;
 278         dp = kzalloc(sizeof *dp, GFP_KERNEL);
 279         if (dp == NULL)
 280                 goto err_unlock;
 281
 282         /* Setup our "of" device */
 283         err = dp_dev_setup(dp);
 284         if (err)
 285                 goto err_free_dp;
 286
 287         dp->dp_idx = dp_idx;
 288         dp->id = gen_datapath_id(dp_idx);
 289         dp->chain = chain_create(dp);
 290         if (dp->chain == NULL)
 291                 goto err_destroy_dp_dev;
 292         INIT_LIST_HEAD(&dp->port_list);
 293
 294         dp->local_port = new_nbp(dp, dp->netdev, OFPP_LOCAL);
 295         if (IS_ERR(dp->local_port)) {
 296                 err = PTR_ERR(dp->local_port);
 297                 goto err_destroy_local_port;
 298         }
 299
 300         dp->flags = 0;
 301         dp->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
 302
 303         dp->dp_task = kthread_run(dp_maint_func, dp, "dp%d", dp_idx);
 304         if (IS_ERR(dp->dp_task))
 305                 goto err_destroy_chain;
 306
 307         dps[dp_idx] = dp;
 308
 309         return 0;
 310
 311 err_destroy_local_port:
 312         dp_del_switch_port(dp->local_port);
 313 err_destroy_chain:
 314         chain_destroy(dp->chain);
 315 err_destroy_dp_dev:
 316         dp_dev_destroy(dp);
 317 err_free_dp:
 318         kfree(dp);
 319 err_unlock:
 320         module_put(THIS_MODULE);
 321                 return err;
 322 }
 323
 324 /* Find and return a free port number under 'dp'. */
 325 static int find_portno(struct datapath *dp)
 326 {
 327         int i;
 328         for (i = 0; i < OFPP_MAX; i++)
 329                 if (dp->ports[i] == NULL)
 330                         return i;
 331         return -EXFULL;
 332 }
 333
 334 static struct net_bridge_port *new_nbp(struct datapath *dp,
 335                                        struct net_device *dev, int port_no)
 336 {
 337         struct net_bridge_port *p;
 338
 339         if (dev->br_port != NULL)
 340                 return ERR_PTR(-EBUSY);
 341
 342         p = kzalloc(sizeof(*p), GFP_KERNEL);
 343         if (p == NULL)
 344                 return ERR_PTR(-ENOMEM);
 345
 346         rtnl_lock();
 347         dev_set_promiscuity(dev, 1);
 348         rtnl_unlock();
 349         dev_hold(dev);
 350         p->dp = dp;
 351         p->dev = dev;
 352         p->port_no = port_no;
 353         spin_lock_init(&p->lock);
 354         INIT_WORK(&p->port_task, NULL);
 355         if (port_no != OFPP_LOCAL)
 356                 rcu_assign_pointer(dev->br_port, p);
 357         if (port_no < OFPP_MAX)
 358                 rcu_assign_pointer(dp->ports[port_no], p);
 359         list_add_rcu(&p->node, &dp->port_list);
 360
 361         return p;
 362 }
 363
 364 int add_switch_port(struct datapath *dp, struct net_device *dev)
 365 {
 366         struct net_bridge_port *p;
 367         int port_no;
 368
 369         if (dev->flags & IFF_LOOPBACK || dev->type != ARPHRD_ETHER
 370             || is_dp_dev(dev))
 371                 return -EINVAL;
 372
 373         port_no = find_portno(dp);
 374         if (port_no < 0)
 375                 return port_no;
 376
 377         p = new_nbp(dp, dev, port_no);
 378         if (IS_ERR(p))
 379                 return PTR_ERR(p);
 380
 381         init_port_status(p);
 382
 383         /* Notify the ctlpath that this port has been added */
 384         dp_send_port_status(p, OFPPR_ADD);
 385
 386         return 0;
 387 }
 388
 389 /* Delete 'p' from switch. */
 390 int dp_del_switch_port(struct net_bridge_port *p)
 391 {
 392 #ifdef SUPPORT_SNAT
 393         unsigned long flags;
 394 #endif
 395
 396         /* First drop references to device. */
 397         cancel_work_sync(&p->port_task);
 398         rtnl_lock();
 399         dev_set_promiscuity(p->dev, -1);
 400         rtnl_unlock();
 401         list_del_rcu(&p->node);
 402         if (p->port_no != OFPP_LOCAL)
 403                 rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
 404         rcu_assign_pointer(p->dev->br_port, NULL);
 405
 406         /* Then wait until no one is still using it, and destroy it. */
 407         synchronize_rcu();
 408
 409 #ifdef SUPPORT_SNAT
 410         /* Free any SNAT configuration on the port. */
 411         spin_lock_irqsave(&p->lock, flags);
 412         snat_free_conf(p);
 413         spin_unlock_irqrestore(&p->lock, flags);
 414 #endif
 415
 416         /* Notify the ctlpath that this port no longer exists */
 417         dp_send_port_status(p, OFPPR_DELETE);
 418
 419         dev_put(p->dev);
 420         kfree(p);
 421
 422         return 0;
 423 }
 424
 425 static void del_dp(struct datapath *dp)
 426 {
 427         struct net_bridge_port *p, *n;
 428
 429         kthread_stop(dp->dp_task);
 430
 431         /* Drop references to DP. */
 432         list_for_each_entry_safe (p, n, &dp->port_list, node)
 433                 dp_del_switch_port(p);
 434         rcu_assign_pointer(dps[dp->dp_idx], NULL);
 435
 436         /* Kill off local_port dev references from buffered packets that have
 437          * associated dst entries. */
 438         synchronize_rcu();
 439         fwd_discard_all();
 440
 441         /* Destroy dp->netdev.  (Must follow deleting switch ports since
 442          * dp->local_port has a reference to it.) */
 443         dp_dev_destroy(dp);
 444
 445         /* Wait until no longer in use, then destroy it. */
 446         synchronize_rcu();
 447         chain_destroy(dp->chain);
 448         kfree(dp);
 449         module_put(THIS_MODULE);
 450 }
 451
 452 static int dp_maint_func(void *data)
 453 {
 454         struct datapath *dp = (struct datapath *) data;
 455
 456         while (!kthread_should_stop()) {
 457 #ifdef SUPPORT_SNAT
 458                 struct net_bridge_port *p;
 459
 460                 /* Expire old SNAT entries */
 461                 rcu_read_lock();
 462                 list_for_each_entry_rcu (p, &dp->port_list, node)
 463                         snat_maint(p);
 464                 rcu_read_unlock();
 465 #endif
 466
 467                 /* Timeout old entries */
 468                 chain_timeout(dp->chain);
 469                 msleep_interruptible(MAINT_SLEEP_MSECS);
 470         }
 471
 472         return 0;
 473 }
 474
 475 static void
 476 do_port_input(struct net_bridge_port *p, struct sk_buff *skb)
 477 {
 478 #ifdef SUPPORT_SNAT
 479         /* Check if this packet needs early SNAT processing. */
 480         if (snat_pre_route(skb)) {
 481                 kfree_skb(skb);
 482                 return;
 483         }
 484 #endif
 485
 486         /* Push the Ethernet header back on. */
 487         skb_push(skb, ETH_HLEN);
 488         fwd_port_input(p->dp->chain, skb, p);
 489 }
 490
 491 /*
 492  * Used as br_handle_frame_hook.  (Cannot run bridge at the same time, even on
 493  * different set of devices!)
 494  */
 495 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,22)
 496 /* Called with rcu_read_lock. */
 497 static struct sk_buff *dp_frame_hook(struct net_bridge_port *p,
 498                                          struct sk_buff *skb)
 499 {
 500         do_port_input(p, skb);
 501         return NULL;
 502 }
 503 #elif LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)
 504 static int dp_frame_hook(struct net_bridge_port *p, struct sk_buff **pskb)
 505 {
 506         do_port_input(p, *pskb);
 507         return 1;
 508 }
 509 #else
 510 /* NB: This has only been tested on 2.4.35 */
 511 static void dp_frame_hook(struct sk_buff *skb)
 512 {
 513         struct net_bridge_port *p = skb->dev->br_port;
 514         if (p) {
 515                 rcu_read_lock();
 516                 do_port_input(p, skb);
 517                 rcu_read_unlock();
 518         } else
 519                 kfree_skb(skb);
 520 }
 521 #endif
 522
 523 /* Forwarding output path.
 524  * Based on net/bridge/br_forward.c. */
 525
 526 static inline unsigned packet_length(const struct sk_buff *skb)
 527 {
 528         int length = skb->len - ETH_HLEN;
 529         if (skb->protocol == htons(ETH_P_8021Q))
 530                 length -= VLAN_HLEN;
 531         return length;
 532 }
 533
 534 /* Send packets out all the ports except the originating one.  If the
 535  * "flood" argument is set, only send along the minimum spanning tree.
 536  */
 537 static int
 538 output_all(struct datapath *dp, struct sk_buff *skb, int flood)
 539 {
 540         u32 disable = flood ? OFPPC_NO_FLOOD : 0;
 541         struct net_bridge_port *p;
 542         int prev_port = -1;
 543
 544         list_for_each_entry_rcu (p, &dp->port_list, node) {
 545                 if (skb->dev == p->dev || p->config & disable)
 546                         continue;
 547                 if (prev_port != -1) {
 548                         struct sk_buff *clone = skb_clone(skb, GFP_ATOMIC);
 549                         if (!clone) {
 550                                 kfree_skb(skb);
 551                                 return -ENOMEM;
 552                         }
 553                         dp_output_port(dp, clone, prev_port, 0);
 554                 }
 555                 prev_port = p->port_no;
 556         }
 557         if (prev_port != -1)
 558                 dp_output_port(dp, skb, prev_port, 0);
 559         else
 560                 kfree_skb(skb);
 561
 562         return 0;
 563 }
 564
 565 /* Marks 'skb' as having originated from 'in_port' in 'dp'.
 566    FIXME: how are devices reference counted? */
 567 void dp_set_origin(struct datapath *dp, uint16_t in_port,
 568                            struct sk_buff *skb)
 569 {
 570         struct net_bridge_port *p = (in_port < OFPP_MAX ? dp->ports[in_port]
 571                                      : in_port == OFPP_LOCAL ? dp->local_port
 572                                      : NULL);
 573         if (p)
 574                 skb->dev = p->dev;
 575          else
 576                 skb->dev = NULL;
 577 }
 578
 579 int
 580 dp_xmit_skb(struct sk_buff *skb)
 581 {
 582         int len = skb->len;
 583         if (packet_length(skb) > skb->dev->mtu) {
 584                 printk("dropped over-mtu packet: %d > %d\n",
 585                            packet_length(skb), skb->dev->mtu);
 586                 kfree_skb(skb);
 587                 return -E2BIG;
 588         }
 589
 590         dev_queue_xmit(skb);
 591
 592         return len;
 593 }
 594
 595 /* Takes ownership of 'skb' and transmits it to 'out_port' on 'dp'.
 596  */
 597 int dp_output_port(struct datapath *dp, struct sk_buff *skb, int out_port,
 598                    int ignore_no_fwd)
 599 {
 600         BUG_ON(!skb);
 601         switch (out_port){
 602         case OFPP_IN_PORT:
 603                 /* Send it out the port it came in on, which is already set in
 604                  * the skb. */
 605                 if (!skb->dev) {
 606                         if (net_ratelimit())
 607                                 printk("skb device not set forwarding to in_port\n");
 608                         kfree_skb(skb);
 609                         return -ESRCH;
 610                 }
 611                 return dp_xmit_skb(skb);
 612
 613         case OFPP_TABLE: {
 614                 int retval = run_flow_through_tables(dp->chain, skb,
 615                                                      skb->dev->br_port);
 616                 if (retval)
 617                         kfree_skb(skb);
 618                 return retval;
 619         }
 620
 621         case OFPP_FLOOD:
 622                 return output_all(dp, skb, 1);
 623
 624         case OFPP_ALL:
 625                 return output_all(dp, skb, 0);
 626
 627         case OFPP_CONTROLLER:
 628                 return dp_output_control(dp, skb, fwd_save_skb(skb), 0,
 629                                                   OFPR_ACTION);
 630
 631         case OFPP_LOCAL: {
 632                 struct net_device *dev = dp->netdev;
 633                 return dev ? dp_dev_recv(dev, skb) : -ESRCH;
 634         }
 635
 636         case 0 ... OFPP_MAX-1: {
 637                 struct net_bridge_port *p = dp->ports[out_port];
 638                 if (p == NULL)
 639                         goto bad_port;
 640                 if (p->dev == skb->dev) {
 641                         /* To send to the input port, must use OFPP_IN_PORT */
 642                         kfree_skb(skb);
 643                         if (net_ratelimit())
 644                                 printk("can't directly forward to input port\n");
 645                         return -EINVAL;
 646                 }
 647                 if (p->config & OFPPC_NO_FWD && !ignore_no_fwd) {
 648                         kfree_skb(skb);
 649                         return 0;
 650                 }
 651                 skb->dev = p->dev;
 652                 return dp_xmit_skb(skb);
 653         }
 654
 655         default:
 656                 goto bad_port;
 657         }
 658
 659 bad_port:
 660         kfree_skb(skb);
 661         if (net_ratelimit())
 662                 printk("can't forward to bad port %d\n", out_port);
 663         return -ENOENT;
 664 }
 665
 666 /* Takes ownership of 'skb' and transmits it to 'dp''s control path.  If
 667  * 'buffer_id' != -1, then only the first 64 bytes of 'skb' are sent;
 668  * otherwise, all of 'skb' is sent.  'reason' indicates why 'skb' is being
 669  * sent. 'max_len' sets the maximum number of bytes that the caller
 670  * wants to be sent; a value of 0 indicates the entire packet should be
 671  * sent. */
 672 int
 673 dp_output_control(struct datapath *dp, struct sk_buff *skb,
 674                            uint32_t buffer_id, size_t max_len, int reason)
 675 {
 676         /* FIXME?  Can we avoid creating a new skbuff in the case where we
 677          * forward the whole packet? */
 678         struct sk_buff *f_skb;
 679         struct ofp_packet_in *opi;
 680         struct net_bridge_port *p;
 681         size_t fwd_len, opi_len;
 682         int err;
 683
 684         fwd_len = skb->len;
 685         if ((buffer_id != (uint32_t) -1) && max_len)
 686                 fwd_len = min(fwd_len, max_len);
 687
 688         opi_len = offsetof(struct ofp_packet_in, data) + fwd_len;
 689         opi = alloc_openflow_skb(dp, opi_len, OFPT_PACKET_IN, NULL, &f_skb);
 690         if (!opi) {
 691                 err = -ENOMEM;
 692                 goto out;
 693         }
 694         opi->buffer_id      = htonl(buffer_id);
 695         opi->total_len      = htons(skb->len);
 696         p = skb->dev->br_port;
 697         opi->in_port        = htons(p ? p->port_no : OFPP_LOCAL);
 698         opi->reason         = reason;
 699         opi->pad            = 0;
 700         memcpy(opi->data, skb_mac_header(skb), fwd_len);
 701         err = send_openflow_skb(f_skb, NULL);
 702
 703 out:
 704         kfree_skb(skb);
 705         return err;
 706 }
 707
 708 static void fill_port_desc(struct net_bridge_port *p, struct ofp_phy_port *desc)
 709 {
 710         unsigned long flags;
 711         desc->port_no = htons(p->port_no);
 712         strncpy(desc->name, p->dev->name, OFP_MAX_PORT_NAME_LEN);
 713         desc->name[OFP_MAX_PORT_NAME_LEN-1] = '\0';
 714         memcpy(desc->hw_addr, p->dev->dev_addr, ETH_ALEN);
 715         desc->curr = 0;
 716         desc->supported = 0;
 717         desc->advertised = 0;
 718         desc->peer = 0;
 719
 720         spin_lock_irqsave(&p->lock, flags);
 721         desc->config = htonl(p->config);
 722         desc->state = htonl(p->state);
 723         spin_unlock_irqrestore(&p->lock, flags);
 724
 725 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,4,24)
 726         if (p->dev->ethtool_ops && p->dev->ethtool_ops->get_settings) {
 727                 struct ethtool_cmd ecmd = { .cmd = ETHTOOL_GSET };
 728
 729                 if (!p->dev->ethtool_ops->get_settings(p->dev, &ecmd)) {
 730                         /* Set the supported features */
 731                         if (ecmd.supported & SUPPORTED_10baseT_Half)
 732                                 desc->supported |= OFPPF_10MB_HD;
 733                         if (ecmd.supported & SUPPORTED_10baseT_Full)
 734                                 desc->supported |= OFPPF_10MB_FD;
 735                         if (ecmd.supported & SUPPORTED_100baseT_Half)
 736                                 desc->supported |= OFPPF_100MB_HD;
 737                         if (ecmd.supported & SUPPORTED_100baseT_Full)
 738                                 desc->supported |= OFPPF_100MB_FD;
 739                         if (ecmd.supported & SUPPORTED_1000baseT_Half)
 740                                 desc->supported |= OFPPF_1GB_HD;
 741                         if (ecmd.supported & SUPPORTED_1000baseT_Full)
 742                                 desc->supported |= OFPPF_1GB_FD;
 743                         if (ecmd.supported & SUPPORTED_10000baseT_Full)
 744                                 desc->supported |= OFPPF_10GB_FD;
 745                         if (ecmd.supported & SUPPORTED_TP)
 746                                 desc->supported |= OFPPF_COPPER;
 747                         if (ecmd.supported & SUPPORTED_FIBRE)
 748                                 desc->supported |= OFPPF_FIBER;
 749                         if (ecmd.supported & SUPPORTED_Autoneg)
 750                                 desc->supported |= OFPPF_AUTONEG;
 751 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,14)
 752                         if (ecmd.supported & SUPPORTED_Pause)
 753                                 desc->supported |= OFPPF_PAUSE;
 754                         if (ecmd.supported & SUPPORTED_Asym_Pause)
 755                                 desc->supported |= OFPPF_PAUSE_ASYM;
 756 #endif /* kernel >= 2.6.14 */
 757
 758                         /* Set the advertised features */
 759                         if (ecmd.advertising & ADVERTISED_10baseT_Half)
 760                                 desc->advertised |= OFPPF_10MB_HD;
 761                         if (ecmd.advertising & ADVERTISED_10baseT_Full)
 762                                 desc->advertised |= OFPPF_10MB_FD;
 763                         if (ecmd.advertising & ADVERTISED_100baseT_Half)
 764                                 desc->advertised |= OFPPF_100MB_HD;
 765                         if (ecmd.advertising & ADVERTISED_100baseT_Full)
 766                                 desc->advertised |= OFPPF_100MB_FD;
 767                         if (ecmd.advertising & ADVERTISED_1000baseT_Half)
 768                                 desc->advertised |= OFPPF_1GB_HD;
 769                         if (ecmd.advertising & ADVERTISED_1000baseT_Full)
 770                                 desc->advertised |= OFPPF_1GB_FD;
 771                         if (ecmd.advertising & ADVERTISED_10000baseT_Full)
 772                                 desc->advertised |= OFPPF_10GB_FD;
 773                         if (ecmd.advertising & ADVERTISED_TP)
 774                                 desc->advertised |= OFPPF_COPPER;
 775                         if (ecmd.advertising & ADVERTISED_FIBRE)
 776                                 desc->advertised |= OFPPF_FIBER;
 777                         if (ecmd.advertising & ADVERTISED_Autoneg)
 778                                 desc->advertised |= OFPPF_AUTONEG;
 779 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,14)
 780                         if (ecmd.advertising & ADVERTISED_Pause)
 781                                 desc->advertised |= OFPPF_PAUSE;
 782                         if (ecmd.advertising & ADVERTISED_Asym_Pause)
 783                                 desc->advertised |= OFPPF_PAUSE_ASYM;
 784 #endif /* kernel >= 2.6.14 */
 785
 786                         /* Set the current features */
 787                         if (ecmd.speed == SPEED_10)
 788                                 desc->curr = (ecmd.duplex) ? OFPPF_10MB_FD : OFPPF_10MB_HD;
 789                         else if (ecmd.speed == SPEED_100)
 790                                 desc->curr = (ecmd.duplex) ? OFPPF_100MB_FD : OFPPF_100MB_HD;
 791                         else if (ecmd.speed == SPEED_1000)
 792                                 desc->curr = (ecmd.duplex) ? OFPPF_1GB_FD : OFPPF_1GB_HD;
 793                         else if (ecmd.speed == SPEED_10000)
 794                                 desc->curr = OFPPF_10GB_FD;
 795
 796                         if (ecmd.port == PORT_TP)
 797                                 desc->curr |= OFPPF_COPPER;
 798                         else if (ecmd.port == PORT_FIBRE)
 799                                 desc->curr |= OFPPF_FIBER;
 800
 801                         if (ecmd.autoneg)
 802                                 desc->curr |= OFPPF_AUTONEG;
 803                 }
 804         }
 805 #endif
 806         desc->curr = htonl(desc->curr);
 807         desc->supported = htonl(desc->supported);
 808         desc->advertised = htonl(desc->advertised);
 809         desc->peer = htonl(desc->peer);
 810 }
 811
 812 static int
 813 fill_features_reply(struct datapath *dp, struct ofp_switch_features *ofr)
 814 {
 815         struct net_bridge_port *p;
 816         int port_count = 0;
 817
 818         ofr->datapath_id  = cpu_to_be64(dp->id);
 819
 820         ofr->n_buffers    = htonl(N_PKT_BUFFERS);
 821         ofr->n_tables     = dp->chain->n_tables;
 822         ofr->capabilities = htonl(OFP_SUPPORTED_CAPABILITIES);
 823         ofr->actions      = htonl(OFP_SUPPORTED_ACTIONS);
 824         memset(ofr->pad, 0, sizeof ofr->pad);
 825
 826         list_for_each_entry_rcu (p, &dp->port_list, node) {
 827                 fill_port_desc(p, &ofr->ports[port_count]);
 828                 port_count++;
 829         }
 830
 831         return port_count;
 832 }
 833
 834 int
 835 dp_send_features_reply(struct datapath *dp, const struct sender *sender)
 836 {
 837         struct sk_buff *skb;
 838         struct ofp_switch_features *ofr;
 839         size_t ofr_len, port_max_len;
 840         int port_count;
 841
 842         /* Overallocate. */
 843         port_max_len = sizeof(struct ofp_phy_port) * OFPP_MAX;
 844         ofr = alloc_openflow_skb(dp, sizeof(*ofr) + port_max_len,
 845                                  OFPT_FEATURES_REPLY, sender, &skb);
 846         if (!ofr)
 847                 return -ENOMEM;
 848
 849         /* Fill. */
 850         port_count = fill_features_reply(dp, ofr);
 851
 852         /* Shrink to fit. */
 853         ofr_len = sizeof(*ofr) + (sizeof(struct ofp_phy_port) * port_count);
 854         resize_openflow_skb(skb, &ofr->header, ofr_len);
 855         return send_openflow_skb(skb, sender);
 856 }
 857
 858 int
 859 dp_send_config_reply(struct datapath *dp, const struct sender *sender)
 860 {
 861         struct sk_buff *skb;
 862         struct ofp_switch_config *osc;
 863
 864         osc = alloc_openflow_skb(dp, sizeof *osc, OFPT_GET_CONFIG_REPLY, sender,
 865                                  &skb);
 866         if (!osc)
 867                 return -ENOMEM;
 868
 869         osc->flags = htons(dp->flags);
 870         osc->miss_send_len = htons(dp->miss_send_len);
 871
 872         return send_openflow_skb(skb, sender);
 873 }
 874
 875 int
 876 dp_send_hello(struct datapath *dp, const struct sender *sender,
 877               const struct ofp_header *request)
 878 {
 879         if (request->version < OFP_VERSION) {
 880                 char err[64];
 881                 sprintf(err, "Only version 0x%02x supported", OFP_VERSION);
 882                 dp_send_error_msg(dp, sender, OFPET_HELLO_FAILED,
 883                                   OFPHFC_INCOMPATIBLE, err, strlen(err));
 884                 return -EINVAL;
 885         } else {
 886                 struct sk_buff *skb;
 887                 struct ofp_header *reply;
 888
 889                 reply = alloc_openflow_skb(dp, sizeof *reply,
 890                                            OFPT_HELLO, sender, &skb);
 891                 if (!reply)
 892                         return -ENOMEM;
 893
 894                 return send_openflow_skb(skb, sender);
 895         }
 896 }
 897
 898 /* Callback function for a workqueue to disable an interface */
 899 static void
 900 down_port_cb(struct work_struct *work)
 901 {
 902         struct net_bridge_port *p = container_of(work, struct net_bridge_port,
 903                         port_task);
 904
 905         rtnl_lock();
 906         if (dev_change_flags(p->dev, p->dev->flags & ~IFF_UP) < 0)
 907                 if (net_ratelimit())
 908                         printk("problem bringing up port %s\n", p->dev->name);
 909         rtnl_unlock();
 910         p->config |= OFPPC_PORT_DOWN;
 911 }
 912
 913 /* Callback function for a workqueue to enable an interface */
 914 static void
 915 up_port_cb(struct work_struct *work)
 916 {
 917         struct net_bridge_port *p = container_of(work, struct net_bridge_port,
 918                         port_task);
 919
 920         rtnl_lock();
 921         if (dev_change_flags(p->dev, p->dev->flags | IFF_UP) < 0)
 922                 if (net_ratelimit())
 923                         printk("problem bringing down port %s\n", p->dev->name);
 924         rtnl_unlock();
 925         p->config &= ~OFPPC_PORT_DOWN;
 926 }
 927
 928 int
 929 dp_update_port_flags(struct datapath *dp, const struct ofp_port_mod *opm)
 930 {
 931         unsigned long int flags;
 932         int port_no = ntohs(opm->port_no);
 933         struct net_bridge_port *p = (port_no < OFPP_MAX ? dp->ports[port_no]
 934                                      : port_no == OFPP_LOCAL ? dp->local_port
 935                                      : NULL);
 936
 937         /* Make sure the port id hasn't changed since this was sent */
 938         if (!p || memcmp(opm->hw_addr, p->dev->dev_addr, ETH_ALEN))
 939                 return -1;
 940
 941         spin_lock_irqsave(&p->lock, flags);
 942         if (opm->mask) {
 943                 uint32_t config_mask = ntohl(opm->mask);
 944                 p->config &= ~config_mask;
 945                 p->config |= ntohl(opm->config) & config_mask;
 946         }
 947
 948         /* Modifying the status of an interface requires taking a lock
 949          * that cannot be done from here.  For this reason, we use a shared
 950          * workqueue, which will cause it to be executed from a safer
 951          * context. */
 952         if (opm->mask & htonl(OFPPC_PORT_DOWN)) {
 953                 if ((opm->config & htonl(OFPPC_PORT_DOWN))
 954                     && (p->config & OFPPC_PORT_DOWN) == 0) {
 955                         PREPARE_WORK(&p->port_task, down_port_cb);
 956                         schedule_work(&p->port_task);
 957                 } else if ((opm->config & htonl(OFPPC_PORT_DOWN)) == 0
 958                            && (p->config & OFPPC_PORT_DOWN)) {
 959                         PREPARE_WORK(&p->port_task, up_port_cb);
 960                         schedule_work(&p->port_task);
 961                 }
 962         }
 963         spin_unlock_irqrestore(&p->lock, flags);
 964
 965         return 0;
 966 }
 967
 968 /* Initialize the port status field of the bridge port. */
 969 static void
 970 init_port_status(struct net_bridge_port *p)
 971 {
 972         unsigned long int flags;
 973
 974         spin_lock_irqsave(&p->lock, flags);
 975
 976         if (p->dev->flags & IFF_UP)
 977                 p->config &= ~OFPPC_PORT_DOWN;
 978         else
 979                 p->config |= OFPPC_PORT_DOWN;
 980
 981         if (netif_carrier_ok(p->dev))
 982                 p->state &= ~OFPPS_LINK_DOWN;
 983         else
 984                 p->state |= OFPPS_LINK_DOWN;
 985
 986         spin_unlock_irqrestore(&p->lock, flags);
 987 }
 988
 989 int
 990 dp_send_port_status(struct net_bridge_port *p, uint8_t status)
 991 {
 992         struct sk_buff *skb;
 993         struct ofp_port_status *ops;
 994
 995         ops = alloc_openflow_skb(p->dp, sizeof *ops, OFPT_PORT_STATUS, NULL,
 996                                  &skb);
 997         if (!ops)
 998                 return -ENOMEM;
 999         ops->reason = status;
1000         memset(ops->pad, 0, sizeof ops->pad);
1001         fill_port_desc(p, &ops->desc);
1002
1003         return send_openflow_skb(skb, NULL);
1004 }
1005
1006 int
1007 dp_send_flow_expired(struct datapath *dp, struct sw_flow *flow,
1008                      enum ofp_flow_expired_reason reason)
1009 {
1010         struct sk_buff *skb;
1011         struct ofp_flow_expired *ofe;
1012
1013         if (!(dp->flags & OFPC_SEND_FLOW_EXP))
1014                 return 0;
1015
1016         ofe = alloc_openflow_skb(dp, sizeof *ofe, OFPT_FLOW_EXPIRED, 0, &skb);
1017         if (!ofe)
1018                 return -ENOMEM;
1019
1020         flow_fill_match(&ofe->match, &flow->key);
1021
1022         ofe->priority = htons(flow->priority);
1023         ofe->reason = reason;
1024         memset(ofe->pad, 0, sizeof ofe->pad);
1025
1026         ofe->duration     = htonl((jiffies - flow->init_time) / HZ);
1027         memset(ofe->pad2, 0, sizeof ofe->pad2);
1028         ofe->packet_count = cpu_to_be64(flow->packet_count);
1029         ofe->byte_count   = cpu_to_be64(flow->byte_count);
1030
1031         return send_openflow_skb(skb, NULL);
1032 }
1033 EXPORT_SYMBOL(dp_send_flow_expired);
1034
1035 int
1036 dp_send_error_msg(struct datapath *dp, const struct sender *sender,
1037                 uint16_t type, uint16_t code, const void *data, size_t len)
1038 {
1039         struct sk_buff *skb;
1040         struct ofp_error_msg *oem;
1041
1042
1043         oem = alloc_openflow_skb(dp, sizeof(*oem)+len, OFPT_ERROR,
1044                         sender, &skb);
1045         if (!oem)
1046                 return -ENOMEM;
1047
1048         oem->type = htons(type);
1049         oem->code = htons(code);
1050         memcpy(oem->data, data, len);
1051
1052         return send_openflow_skb(skb, sender);
1053 }
1054
1055 int
1056 dp_send_echo_reply(struct datapath *dp, const struct sender *sender,
1057                    const struct ofp_header *rq)
1058 {
1059         struct sk_buff *skb;
1060         struct ofp_header *reply;
1061
1062         reply = alloc_openflow_skb(dp, ntohs(rq->length), OFPT_ECHO_REPLY,
1063                                    sender, &skb);
1064         if (!reply)
1065                 return -ENOMEM;
1066
1067         memcpy(reply + 1, rq + 1, ntohs(rq->length) - sizeof *rq);
1068         return send_openflow_skb(skb, sender);
1069 }
1070
1071 /* Generic Netlink interface.
1072  *
1073  * See netlink(7) for an introduction to netlink.  See
1074  * http://linux-net.osdl.org/index.php/Netlink for more information and
1075  * pointers on how to work with netlink and Generic Netlink in the kernel and
1076  * in userspace. */
1077
1078 static struct genl_family dp_genl_family = {
1079         .id = GENL_ID_GENERATE,
1080         .hdrsize = 0,
1081         .name = DP_GENL_FAMILY_NAME,
1082         .version = 1,
1083         .maxattr = DP_GENL_A_MAX,
1084 };
1085
1086 /* Attribute policy: what each attribute may contain.  */
1087 static struct nla_policy dp_genl_policy[DP_GENL_A_MAX + 1] = {
1088         [DP_GENL_A_DP_IDX] = { .type = NLA_U32 },
1089         [DP_GENL_A_MC_GROUP] = { .type = NLA_U32 },
1090         [DP_GENL_A_PORTNAME] = { .type = NLA_STRING }
1091 };
1092
1093 static int dp_genl_add(struct sk_buff *skb, struct genl_info *info)
1094 {
1095         if (!info->attrs[DP_GENL_A_DP_IDX])
1096                 return -EINVAL;
1097
1098         return new_dp(nla_get_u32(info->attrs[DP_GENL_A_DP_IDX]));
1099 }
1100
1101 static struct genl_ops dp_genl_ops_add_dp = {
1102         .cmd = DP_GENL_C_ADD_DP,
1103         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1104         .policy = dp_genl_policy,
1105         .doit = dp_genl_add,
1106         .dumpit = NULL,
1107 };
1108
1109 struct datapath *dp_get(int dp_idx)
1110 {
1111         if (dp_idx < 0 || dp_idx > DP_MAX)
1112                 return NULL;
1113         return rcu_dereference(dps[dp_idx]);
1114 }
1115
1116 static int dp_genl_del(struct sk_buff *skb, struct genl_info *info)
1117 {
1118         struct datapath *dp;
1119         int err;
1120
1121         if (!info->attrs[DP_GENL_A_DP_IDX])
1122                 return -EINVAL;
1123
1124         dp = dp_get(nla_get_u32((info->attrs[DP_GENL_A_DP_IDX])));
1125         if (!dp)
1126                 err = -ENOENT;
1127         else {
1128                 del_dp(dp);
1129                 err = 0;
1130         }
1131         return err;
1132 }
1133
1134 static struct genl_ops dp_genl_ops_del_dp = {
1135         .cmd = DP_GENL_C_DEL_DP,
1136         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1137         .policy = dp_genl_policy,
1138         .doit = dp_genl_del,
1139         .dumpit = NULL,
1140 };
1141
1142 /* Queries a datapath for related information.  Currently the only relevant
1143  * information is the datapath's multicast group ID.  Really we want one
1144  * multicast group per datapath, but because of locking issues[*] we can't
1145  * easily get one.  Thus, every datapath will currently return the same
1146  * global multicast group ID, but in the future it would be nice to fix that.
1147  *
1148  * [*] dp_genl_add, to add a new datapath, is called under the genl_lock
1149  *       mutex, and genl_register_mc_group, called to acquire a new multicast
1150  *       group ID, also acquires genl_lock, thus deadlock.
1151  */
1152 static int dp_genl_query(struct sk_buff *skb, struct genl_info *info)
1153 {
1154         struct datapath *dp;
1155         struct sk_buff *ans_skb = NULL;
1156         int dp_idx;
1157         int err = -ENOMEM;
1158
1159         if (!info->attrs[DP_GENL_A_DP_IDX])
1160                 return -EINVAL;
1161
1162         rcu_read_lock();
1163         dp_idx = nla_get_u32((info->attrs[DP_GENL_A_DP_IDX]));
1164         dp = dp_get(dp_idx);
1165         if (!dp)
1166                 err = -ENOENT;
1167         else {
1168                 void *data;
1169                 ans_skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1170                 if (!ans_skb) {
1171                         err = -ENOMEM;
1172                         goto err;
1173                 }
1174                 data = genlmsg_put_reply(ans_skb, info, &dp_genl_family,
1175                                          0, DP_GENL_C_QUERY_DP);
1176                 if (data == NULL) {
1177                         err = -ENOMEM;
1178                         goto err;
1179                 }
1180                 NLA_PUT_U32(ans_skb, DP_GENL_A_DP_IDX, dp_idx);
1181                 NLA_PUT_U32(ans_skb, DP_GENL_A_MC_GROUP, mc_group.id);
1182
1183                 genlmsg_end(ans_skb, data);
1184                 err = genlmsg_reply(ans_skb, info);
1185                 if (!err)
1186                         ans_skb = NULL;
1187         }
1188 err:
1189 nla_put_failure:
1190         if (ans_skb)
1191                 kfree_skb(ans_skb);
1192         rcu_read_unlock();
1193         return err;
1194 }
1195
1196 static struct genl_ops dp_genl_ops_query_dp = {
1197         .cmd = DP_GENL_C_QUERY_DP,
1198         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1199         .policy = dp_genl_policy,
1200         .doit = dp_genl_query,
1201         .dumpit = NULL,
1202 };
1203
1204 static int dp_genl_add_del_port(struct sk_buff *skb, struct genl_info *info)
1205 {
1206         struct datapath *dp;
1207         struct net_device *port;
1208         int err;
1209
1210         if (!info->attrs[DP_GENL_A_DP_IDX] || !info->attrs[DP_GENL_A_PORTNAME])
1211                 return -EINVAL;
1212
1213         /* Get datapath. */
1214         dp = dp_get(nla_get_u32(info->attrs[DP_GENL_A_DP_IDX]));
1215         if (!dp) {
1216                 err = -ENOENT;
1217                 goto out;
1218         }
1219
1220         /* Get interface to add/remove. */
1221         port = dev_get_by_name(&init_net,
1222                         nla_data(info->attrs[DP_GENL_A_PORTNAME]));
1223         if (!port) {
1224                 err = -ENOENT;
1225                 goto out;
1226         }
1227
1228         /* Execute operation. */
1229         if (info->genlhdr->cmd == DP_GENL_C_ADD_PORT)
1230                 err = add_switch_port(dp, port);
1231         else {
1232                 if (port->br_port == NULL || port->br_port->dp != dp) {
1233                         err = -ENOENT;
1234                         goto out_put;
1235                 }
1236                 err = dp_del_switch_port(port->br_port);
1237         }
1238
1239 out_put:
1240         dev_put(port);
1241 out:
1242         return err;
1243 }
1244
1245 static struct genl_ops dp_genl_ops_add_port = {
1246         .cmd = DP_GENL_C_ADD_PORT,
1247         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1248         .policy = dp_genl_policy,
1249         .doit = dp_genl_add_del_port,
1250         .dumpit = NULL,
1251 };
1252
1253 static struct genl_ops dp_genl_ops_del_port = {
1254         .cmd = DP_GENL_C_DEL_PORT,
1255         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1256         .policy = dp_genl_policy,
1257         .doit = dp_genl_add_del_port,
1258         .dumpit = NULL,
1259 };
1260
1261 static int dp_genl_openflow(struct sk_buff *skb, struct genl_info *info)
1262 {
1263         struct nlattr *va = info->attrs[DP_GENL_A_OPENFLOW];
1264         struct datapath *dp;
1265         struct ofp_header *oh;
1266         struct sender sender;
1267         int err;
1268
1269         if (!info->attrs[DP_GENL_A_DP_IDX] || !va)
1270                 return -EINVAL;
1271
1272         dp = dp_get(nla_get_u32(info->attrs[DP_GENL_A_DP_IDX]));
1273         if (!dp)
1274                 return -ENOENT;
1275
1276         if (nla_len(va) < sizeof(struct ofp_header))
1277                 return -EINVAL;
1278         oh = nla_data(va);
1279
1280         sender.xid = oh->xid;
1281         sender.pid = info->snd_pid;
1282         sender.seq = info->snd_seq;
1283
1284         mutex_lock(&dp_mutex);
1285         err = fwd_control_input(dp->chain, &sender,
1286                                 nla_data(va), nla_len(va));
1287         mutex_unlock(&dp_mutex);
1288         return err;
1289 }
1290
1291 static struct nla_policy dp_genl_openflow_policy[DP_GENL_A_MAX + 1] = {
1292         [DP_GENL_A_DP_IDX] = { .type = NLA_U32 },
1293 };
1294
1295 static int desc_stats_dump(struct datapath *dp, void *state,
1296                             void *body, int *body_len)
1297 {
1298         struct ofp_desc_stats *ods = body;
1299         int n_bytes = sizeof *ods;
1300
1301         if (n_bytes > *body_len) {
1302                 return -ENOBUFS;
1303         }
1304         *body_len = n_bytes;
1305
1306         strncpy(ods->mfr_desc, mfr_desc, sizeof ods->mfr_desc);
1307         strncpy(ods->hw_desc, hw_desc, sizeof ods->hw_desc);
1308         strncpy(ods->sw_desc, sw_desc, sizeof ods->sw_desc);
1309         strncpy(ods->serial_num, serial_num, sizeof ods->serial_num);
1310
1311         return 0;
1312 }
1313
1314 struct flow_stats_state {
1315         int table_idx;
1316         struct sw_table_position position;
1317         const struct ofp_flow_stats_request *rq;
1318
1319         void *body;
1320         int bytes_used, bytes_allocated;
1321 };
1322
1323 static int flow_stats_init(struct datapath *dp, const void *body, int body_len,
1324                            void **state)
1325 {
1326         const struct ofp_flow_stats_request *fsr = body;
1327         struct flow_stats_state *s = kmalloc(sizeof *s, GFP_ATOMIC);
1328         if (!s)
1329                 return -ENOMEM;
1330         s->table_idx = fsr->table_id == 0xff ? 0 : fsr->table_id;
1331         memset(&s->position, 0, sizeof s->position);
1332         s->rq = fsr;
1333         *state = s;
1334         return 0;
1335 }
1336
1337 static int flow_stats_dump_callback(struct sw_flow *flow, void *private)
1338 {
1339         struct sw_flow_actions *sf_acts = rcu_dereference(flow->sf_acts);
1340         struct flow_stats_state *s = private;
1341         struct ofp_flow_stats *ofs;
1342         int length;
1343
1344         length = sizeof *ofs + sf_acts->actions_len;
1345         if (length + s->bytes_used > s->bytes_allocated)
1346                 return 1;
1347
1348         ofs = s->body + s->bytes_used;
1349         ofs->length          = htons(length);
1350         ofs->table_id        = s->table_idx;
1351         ofs->pad             = 0;
1352         ofs->match.wildcards = htonl(flow->key.wildcards);
1353         ofs->match.in_port   = flow->key.in_port;
1354         memcpy(ofs->match.dl_src, flow->key.dl_src, ETH_ALEN);
1355         memcpy(ofs->match.dl_dst, flow->key.dl_dst, ETH_ALEN);
1356         ofs->match.dl_vlan   = flow->key.dl_vlan;
1357         ofs->match.dl_type   = flow->key.dl_type;
1358         ofs->match.nw_src    = flow->key.nw_src;
1359         ofs->match.nw_dst    = flow->key.nw_dst;
1360         ofs->match.nw_proto  = flow->key.nw_proto;
1361         ofs->match.pad       = 0;
1362         ofs->match.tp_src    = flow->key.tp_src;
1363         ofs->match.tp_dst    = flow->key.tp_dst;
1364         ofs->duration        = htonl((jiffies - flow->init_time) / HZ);
1365         ofs->priority        = htons(flow->priority);
1366         ofs->idle_timeout    = htons(flow->idle_timeout);
1367         ofs->hard_timeout    = htons(flow->hard_timeout);
1368         memset(ofs->pad2, 0, sizeof ofs->pad2);
1369         ofs->packet_count    = cpu_to_be64(flow->packet_count);
1370         ofs->byte_count      = cpu_to_be64(flow->byte_count);
1371         memcpy(ofs->actions, sf_acts->actions, sf_acts->actions_len);
1372
1373         s->bytes_used += length;
1374         return 0;
1375 }
1376
1377 static int flow_stats_dump(struct datapath *dp, void *state,
1378                            void *body, int *body_len)
1379 {
1380         struct flow_stats_state *s = state;
1381         struct sw_flow_key match_key;
1382         int error = 0;
1383
1384         s->bytes_used = 0;
1385         s->bytes_allocated = *body_len;
1386         s->body = body;
1387
1388         flow_extract_match(&match_key, &s->rq->match);
1389         while (s->table_idx < dp->chain->n_tables
1390                && (s->rq->table_id == 0xff || s->rq->table_id == s->table_idx))
1391         {
1392                 struct sw_table *table = dp->chain->tables[s->table_idx];
1393
1394                 error = table->iterate(table, &match_key, &s->position,
1395                                        flow_stats_dump_callback, s);
1396                 if (error)
1397                         break;
1398
1399                 s->table_idx++;
1400                 memset(&s->position, 0, sizeof s->position);
1401         }
1402         *body_len = s->bytes_used;
1403
1404         /* If error is 0, we're done.
1405          * Otherwise, if some bytes were used, there are more flows to come.
1406          * Otherwise, we were not able to fit even a single flow in the body,
1407          * which indicates that we have a single flow with too many actions to
1408          * fit.  We won't ever make any progress at that rate, so give up. */
1409         return !error ? 0 : s->bytes_used ? 1 : -ENOMEM;
1410 }
1411
1412 static void flow_stats_done(void *state)
1413 {
1414         kfree(state);
1415 }
1416
1417 static int aggregate_stats_init(struct datapath *dp,
1418                                 const void *body, int body_len,
1419                                 void **state)
1420 {
1421         *state = (void *)body;
1422         return 0;
1423 }
1424
1425 static int aggregate_stats_dump_callback(struct sw_flow *flow, void *private)
1426 {
1427         struct ofp_aggregate_stats_reply *rpy = private;
1428         rpy->packet_count += flow->packet_count;
1429         rpy->byte_count += flow->byte_count;
1430         rpy->flow_count++;
1431         return 0;
1432 }
1433
1434 static int aggregate_stats_dump(struct datapath *dp, void *state,
1435                                 void *body, int *body_len)
1436 {
1437         struct ofp_aggregate_stats_request *rq = state;
1438         struct ofp_aggregate_stats_reply *rpy;
1439         struct sw_table_position position;
1440         struct sw_flow_key match_key;
1441         int table_idx;
1442
1443         if (*body_len < sizeof *rpy)
1444                 return -ENOBUFS;
1445         rpy = body;
1446         *body_len = sizeof *rpy;
1447
1448         memset(rpy, 0, sizeof *rpy);
1449
1450         flow_extract_match(&match_key, &rq->match);
1451         table_idx = rq->table_id == 0xff ? 0 : rq->table_id;
1452         memset(&position, 0, sizeof position);
1453         while (table_idx < dp->chain->n_tables
1454                && (rq->table_id == 0xff || rq->table_id == table_idx))
1455         {
1456                 struct sw_table *table = dp->chain->tables[table_idx];
1457                 int error;
1458
1459                 error = table->iterate(table, &match_key, &position,
1460                                        aggregate_stats_dump_callback, rpy);
1461                 if (error)
1462                         return error;
1463
1464                 table_idx++;
1465                 memset(&position, 0, sizeof position);
1466         }
1467
1468         rpy->packet_count = cpu_to_be64(rpy->packet_count);
1469         rpy->byte_count = cpu_to_be64(rpy->byte_count);
1470         rpy->flow_count = htonl(rpy->flow_count);
1471         return 0;
1472 }
1473
1474 static int table_stats_dump(struct datapath *dp, void *state,
1475                             void *body, int *body_len)
1476 {
1477         struct ofp_table_stats *ots;
1478         int n_bytes = dp->chain->n_tables * sizeof *ots;
1479         int i;
1480         if (n_bytes > *body_len)
1481                 return -ENOBUFS;
1482         *body_len = n_bytes;
1483         for (i = 0, ots = body; i < dp->chain->n_tables; i++, ots++) {
1484                 struct sw_table_stats stats;
1485                 dp->chain->tables[i]->stats(dp->chain->tables[i], &stats);
1486                 strncpy(ots->name, stats.name, sizeof ots->name);
1487                 ots->table_id = i;
1488                 ots->wildcards = htonl(stats.wildcards);
1489                 memset(ots->pad, 0, sizeof ots->pad);
1490                 ots->max_entries = htonl(stats.max_flows);
1491                 ots->active_count = htonl(stats.n_flows);
1492                 ots->lookup_count = cpu_to_be64(stats.n_lookup);
1493                 ots->matched_count = cpu_to_be64(stats.n_matched);
1494         }
1495         return 0;
1496 }
1497
1498 struct port_stats_state {
1499         int port;
1500 };
1501
1502 static int port_stats_init(struct datapath *dp, const void *body, int body_len,
1503                            void **state)
1504 {
1505         struct port_stats_state *s = kmalloc(sizeof *s, GFP_ATOMIC);
1506         if (!s)
1507                 return -ENOMEM;
1508         s->port = 0;
1509         *state = s;
1510         return 0;
1511 }
1512
1513 static int port_stats_dump(struct datapath *dp, void *state,
1514                            void *body, int *body_len)
1515 {
1516         struct port_stats_state *s = state;
1517         struct ofp_port_stats *ops;
1518         int n_ports, max_ports;
1519         int i;
1520
1521         max_ports = *body_len / sizeof *ops;
1522         if (!max_ports)
1523                 return -ENOMEM;
1524         ops = body;
1525
1526         n_ports = 0;
1527         for (i = s->port; i < OFPP_MAX && n_ports < max_ports; i++) {
1528                 struct net_bridge_port *p = dp->ports[i];
1529                 struct net_device_stats *stats;
1530                 if (!p)
1531                         continue;
1532                 stats = p->dev->get_stats(p->dev);
1533                 ops->port_no = htons(p->port_no);
1534                 memset(ops->pad, 0, sizeof ops->pad);
1535                 ops->rx_packets   = cpu_to_be64(stats->rx_packets);
1536                 ops->tx_packets   = cpu_to_be64(stats->tx_packets);
1537                 ops->rx_bytes     = cpu_to_be64(stats->rx_bytes);
1538                 ops->tx_bytes     = cpu_to_be64(stats->tx_bytes);
1539                 ops->rx_dropped   = cpu_to_be64(stats->rx_dropped);
1540                 ops->tx_dropped   = cpu_to_be64(stats->tx_dropped);
1541                 ops->rx_errors    = cpu_to_be64(stats->rx_errors);
1542                 ops->tx_errors    = cpu_to_be64(stats->tx_errors);
1543                 ops->rx_frame_err = cpu_to_be64(stats->rx_frame_errors);
1544                 ops->rx_over_err  = cpu_to_be64(stats->rx_over_errors);
1545                 ops->rx_crc_err   = cpu_to_be64(stats->rx_crc_errors);
1546                 ops->collisions   = cpu_to_be64(stats->collisions);
1547                 n_ports++;
1548                 ops++;
1549         }
1550         s->port = i;
1551         *body_len = n_ports * sizeof *ops;
1552         return n_ports >= max_ports;
1553 }
1554
1555 static void port_stats_done(void *state)
1556 {
1557         kfree(state);
1558 }
1559
1560 struct stats_type {
1561         /* Minimum and maximum acceptable number of bytes in body member of
1562          * struct ofp_stats_request. */
1563         size_t min_body, max_body;
1564
1565         /* Prepares to dump some kind of statistics on 'dp'.  'body' and
1566          * 'body_len' are the 'body' member of the struct ofp_stats_request.
1567          * Returns zero if successful, otherwise a negative error code.
1568          * May initialize '*state' to state information.  May be null if no
1569          * initialization is required.*/
1570         int (*init)(struct datapath *dp, const void *body, int body_len,
1571                     void **state);
1572
1573         /* Dumps statistics for 'dp' into the '*body_len' bytes at 'body', and
1574          * modifies '*body_len' to reflect the number of bytes actually used.
1575          * ('body' will be transmitted as the 'body' member of struct
1576          * ofp_stats_reply.) */
1577         int (*dump)(struct datapath *dp, void *state,
1578                     void *body, int *body_len);
1579
1580         /* Cleans any state created by the init or dump functions.  May be null
1581          * if no cleanup is required. */
1582         void (*done)(void *state);
1583 };
1584
1585 static const struct stats_type stats[] = {
1586         [OFPST_DESC] = {
1587                 0,
1588                 0,
1589                 NULL,
1590                 desc_stats_dump,
1591                 NULL
1592         },
1593         [OFPST_FLOW] = {
1594                 sizeof(struct ofp_flow_stats_request),
1595                 sizeof(struct ofp_flow_stats_request),
1596                 flow_stats_init,
1597                 flow_stats_dump,
1598                 flow_stats_done
1599         },
1600         [OFPST_AGGREGATE] = {
1601                 sizeof(struct ofp_aggregate_stats_request),
1602                 sizeof(struct ofp_aggregate_stats_request),
1603                 aggregate_stats_init,
1604                 aggregate_stats_dump,
1605                 NULL
1606         },
1607         [OFPST_TABLE] = {
1608                 0,
1609                 0,
1610                 NULL,
1611                 table_stats_dump,
1612                 NULL
1613         },
1614         [OFPST_PORT] = {
1615                 0,
1616                 0,
1617                 port_stats_init,
1618                 port_stats_dump,
1619                 port_stats_done
1620         },
1621 };
1622
1623 static int
1624 dp_genl_openflow_dumpit(struct sk_buff *skb, struct netlink_callback *cb)
1625 {
1626         struct datapath *dp;
1627         struct sender sender;
1628         const struct stats_type *s;
1629         struct ofp_stats_reply *osr;
1630         int dp_idx;
1631         int max_openflow_len, body_len;
1632         void *body;
1633         int err;
1634
1635         /* Set up the cleanup function for this dump.  Linux 2.6.20 and later
1636          * support setting up cleanup functions via the .doneit member of
1637          * struct genl_ops.  This kluge supports earlier versions also. */
1638         cb->done = dp_genl_openflow_done;
1639
1640         sender.pid = NETLINK_CB(cb->skb).pid;
1641         sender.seq = cb->nlh->nlmsg_seq;
1642         if (!cb->args[0]) {
1643                 struct nlattr *attrs[DP_GENL_A_MAX + 1];
1644                 struct ofp_stats_request *rq;
1645                 struct nlattr *va;
1646                 size_t len, body_len;
1647                 int type;
1648
1649                 err = nlmsg_parse(cb->nlh, GENL_HDRLEN, attrs, DP_GENL_A_MAX,
1650                                   dp_genl_openflow_policy);
1651                 if (err < 0)
1652                         return err;
1653
1654                 if (!attrs[DP_GENL_A_DP_IDX])
1655                         return -EINVAL;
1656                 dp_idx = nla_get_u16(attrs[DP_GENL_A_DP_IDX]);
1657                 dp = dp_get(dp_idx);
1658                 if (!dp)
1659                         return -ENOENT;
1660
1661                 va = attrs[DP_GENL_A_OPENFLOW];
1662                 len = nla_len(va);
1663                 if (!va || len < sizeof *rq)
1664                         return -EINVAL;
1665
1666                 rq = nla_data(va);
1667                 sender.xid = rq->header.xid;
1668                 type = ntohs(rq->type);
1669                 if (rq->header.version != OFP_VERSION) {
1670                         dp_send_error_msg(dp, &sender, OFPET_BAD_REQUEST,
1671                                           OFPBRC_BAD_VERSION, rq, len);
1672                         return -EINVAL;
1673                 }
1674                 if (rq->header.type != OFPT_STATS_REQUEST
1675                     || ntohs(rq->header.length) != len)
1676                         return -EINVAL;
1677
1678                 if (type >= ARRAY_SIZE(stats) || !stats[type].dump) {
1679                         dp_send_error_msg(dp, &sender, OFPET_BAD_REQUEST,
1680                                           OFPBRC_BAD_STAT, rq, len);
1681                         return -EINVAL;
1682                 }
1683
1684                 s = &stats[type];
1685                 body_len = len - offsetof(struct ofp_stats_request, body);
1686                 if (body_len < s->min_body || body_len > s->max_body)
1687                         return -EINVAL;
1688
1689                 cb->args[0] = 1;
1690                 cb->args[1] = dp_idx;
1691                 cb->args[2] = type;
1692                 cb->args[3] = rq->header.xid;
1693                 if (s->init) {
1694                         void *state;
1695                         err = s->init(dp, rq->body, body_len, &state);
1696                         if (err)
1697                                 return err;
1698                         cb->args[4] = (long) state;
1699                 }
1700         } else if (cb->args[0] == 1) {
1701                 sender.xid = cb->args[3];
1702                 dp_idx = cb->args[1];
1703                 s = &stats[cb->args[2]];
1704
1705                 dp = dp_get(dp_idx);
1706                 if (!dp)
1707                         return -ENOENT;
1708         } else {
1709                 return 0;
1710         }
1711
1712         osr = put_openflow_headers(dp, skb, OFPT_STATS_REPLY, &sender,
1713                                    &max_openflow_len);
1714         if (IS_ERR(osr))
1715                 return PTR_ERR(osr);
1716         osr->type = htons(s - stats);
1717         osr->flags = 0;
1718         resize_openflow_skb(skb, &osr->header, max_openflow_len);
1719         body = osr->body;
1720         body_len = max_openflow_len - offsetof(struct ofp_stats_reply, body);
1721
1722         err = s->dump(dp, (void *) cb->args[4], body, &body_len);
1723         if (err >= 0) {
1724                 if (!err)
1725                         cb->args[0] = 2;
1726                 else
1727                         osr->flags = ntohs(OFPSF_REPLY_MORE);
1728                 resize_openflow_skb(skb, &osr->header,
1729                                     (offsetof(struct ofp_stats_reply, body)
1730                                      + body_len));
1731                 err = skb->len;
1732         }
1733
1734         return err;
1735 }
1736
1737 static int
1738 dp_genl_openflow_done(struct netlink_callback *cb)
1739 {
1740         if (cb->args[0]) {
1741                 const struct stats_type *s = &stats[cb->args[2]];
1742                 if (s->done)
1743                         s->done((void *) cb->args[4]);
1744         }
1745         return 0;
1746 }
1747
1748 static struct genl_ops dp_genl_ops_openflow = {
1749         .cmd = DP_GENL_C_OPENFLOW,
1750         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1751         .policy = dp_genl_openflow_policy,
1752         .doit = dp_genl_openflow,
1753         .dumpit = dp_genl_openflow_dumpit,
1754 };
1755
1756 static struct genl_ops *dp_genl_all_ops[] = {
1757         /* Keep this operation first.  Generic Netlink dispatching
1758          * looks up operations with linear search, so we want it at the
1759          * front. */
1760         &dp_genl_ops_openflow,
1761
1762         &dp_genl_ops_add_dp,
1763         &dp_genl_ops_del_dp,
1764         &dp_genl_ops_query_dp,
1765         &dp_genl_ops_add_port,
1766         &dp_genl_ops_del_port,
1767 };
1768
1769 static int dp_init_netlink(void)
1770 {
1771         int err;
1772         int i;
1773
1774         err = genl_register_family(&dp_genl_family);
1775         if (err)
1776                 return err;
1777
1778         for (i = 0; i < ARRAY_SIZE(dp_genl_all_ops); i++) {
1779                 err = genl_register_ops(&dp_genl_family, dp_genl_all_ops[i]);
1780                 if (err)
1781                         goto err_unregister;
1782         }
1783
1784         strcpy(mc_group.name, "openflow");
1785         err = genl_register_mc_group(&dp_genl_family, &mc_group);
1786         if (err < 0)
1787                 goto err_unregister;
1788
1789         return 0;
1790
1791 err_unregister:
1792         genl_unregister_family(&dp_genl_family);
1793                 return err;
1794 }
1795
1796 static void dp_uninit_netlink(void)
1797 {
1798         genl_unregister_family(&dp_genl_family);
1799 }
1800
1801 static int __init dp_init(void)
1802 {
1803         int err;
1804
1805         printk("OpenFlow "VERSION", built "__DATE__" "__TIME__", "
1806                "protocol 0x%02x\n", OFP_VERSION);
1807
1808         err = flow_init();
1809         if (err)
1810                 goto error;
1811
1812         err = register_netdevice_notifier(&dp_device_notifier);
1813         if (err)
1814                 goto error_flow_exit;
1815
1816         err = dp_init_netlink();
1817         if (err)
1818                 goto error_unreg_notifier;
1819
1820         /* Hook into callback used by the bridge to intercept packets.
1821          * Parasites we are. */
1822         if (br_handle_frame_hook)
1823                 printk("openflow: hijacking bridge hook\n");
1824         br_handle_frame_hook = dp_frame_hook;
1825
1826         return 0;
1827
1828 error_unreg_notifier:
1829         unregister_netdevice_notifier(&dp_device_notifier);
1830 error_flow_exit:
1831         flow_exit();
1832 error:
1833         printk(KERN_EMERG "openflow: failed to install!");
1834         return err;
1835 }
1836
1837 static void dp_cleanup(void)
1838 {
1839         fwd_exit();
1840         dp_uninit_netlink();
1841         unregister_netdevice_notifier(&dp_device_notifier);
1842         flow_exit();
1843         br_handle_frame_hook = NULL;
1844 }
1845
1846 module_init(dp_init);
1847 module_exit(dp_cleanup);
1848
1849 MODULE_DESCRIPTION("OpenFlow switching datapath");
1850 MODULE_AUTHOR("Copyright (c) 2007, 2008 The Board of Trustees of The Leland Stanford Junior University");
1851 MODULE_LICENSE("GPL");