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 < DP_MAX_PORTS; 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 < DP_MAX_PORTS)
 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;
 571         p = (in_port < DP_MAX_PORTS ? dp->ports[in_port]
 572              : in_port == OFPP_LOCAL ? dp->local_port
 573              : NULL);
 574         if (p)
 575                 skb->dev = p->dev;
 576          else
 577                 skb->dev = NULL;
 578 }
 579
 580 int
 581 dp_xmit_skb(struct sk_buff *skb)
 582 {
 583         int len = skb->len;
 584         if (packet_length(skb) > skb->dev->mtu) {
 585                 printk("dropped over-mtu packet: %d > %d\n",
 586                            packet_length(skb), skb->dev->mtu);
 587                 kfree_skb(skb);
 588                 return -E2BIG;
 589         }
 590
 591         dev_queue_xmit(skb);
 592
 593         return len;
 594 }
 595
 596 /* Takes ownership of 'skb' and transmits it to 'out_port' on 'dp'.
 597  */
 598 int dp_output_port(struct datapath *dp, struct sk_buff *skb, int out_port,
 599                    int ignore_no_fwd)
 600 {
 601         BUG_ON(!skb);
 602         switch (out_port){
 603         case OFPP_IN_PORT:
 604                 /* Send it out the port it came in on, which is already set in
 605                  * the skb. */
 606                 if (!skb->dev) {
 607                         if (net_ratelimit())
 608                                 printk("skb device not set forwarding to in_port\n");
 609                         kfree_skb(skb);
 610                         return -ESRCH;
 611                 }
 612                 return dp_xmit_skb(skb);
 613
 614         case OFPP_TABLE: {
 615                 int retval = run_flow_through_tables(dp->chain, skb,
 616                                                      skb->dev->br_port);
 617                 if (retval)
 618                         kfree_skb(skb);
 619                 return retval;
 620         }
 621
 622         case OFPP_FLOOD:
 623                 return output_all(dp, skb, 1);
 624
 625         case OFPP_ALL:
 626                 return output_all(dp, skb, 0);
 627
 628         case OFPP_CONTROLLER:
 629                 return dp_output_control(dp, skb, fwd_save_skb(skb), 0,
 630                                                   OFPR_ACTION);
 631
 632         case OFPP_LOCAL: {
 633                 struct net_device *dev = dp->netdev;
 634                 return dev ? dp_dev_recv(dev, skb) : -ESRCH;
 635         }
 636
 637         case 0 ... DP_MAX_PORTS - 1: {
 638                 struct net_bridge_port *p = dp->ports[out_port];
 639                 if (p == NULL)
 640                         goto bad_port;
 641                 if (p->dev == skb->dev) {
 642                         /* To send to the input port, must use OFPP_IN_PORT */
 643                         kfree_skb(skb);
 644                         if (net_ratelimit())
 645                                 printk("can't directly forward to input port\n");
 646                         return -EINVAL;
 647                 }
 648                 if (p->config & OFPPC_NO_FWD && !ignore_no_fwd) {
 649                         kfree_skb(skb);
 650                         return 0;
 651                 }
 652                 skb->dev = p->dev;
 653                 return dp_xmit_skb(skb);
 654         }
 655
 656         default:
 657                 goto bad_port;
 658         }
 659
 660 bad_port:
 661         kfree_skb(skb);
 662         if (net_ratelimit())
 663                 printk("can't forward to bad port %d\n", out_port);
 664         return -ENOENT;
 665 }
 666
 667 /* Takes ownership of 'skb' and transmits it to 'dp''s control path.  If
 668  * 'buffer_id' != -1, then only the first 64 bytes of 'skb' are sent;
 669  * otherwise, all of 'skb' is sent.  'reason' indicates why 'skb' is being
 670  * sent. 'max_len' sets the maximum number of bytes that the caller
 671  * wants to be sent; a value of 0 indicates the entire packet should be
 672  * sent. */
 673 int
 674 dp_output_control(struct datapath *dp, struct sk_buff *skb,
 675                            uint32_t buffer_id, size_t max_len, int reason)
 676 {
 677         /* FIXME?  Can we avoid creating a new skbuff in the case where we
 678          * forward the whole packet? */
 679         struct sk_buff *f_skb;
 680         struct ofp_packet_in *opi;
 681         struct net_bridge_port *p;
 682         size_t fwd_len, opi_len;
 683         int err;
 684
 685         fwd_len = skb->len;
 686         if ((buffer_id != (uint32_t) -1) && max_len)
 687                 fwd_len = min(fwd_len, max_len);
 688
 689         opi_len = offsetof(struct ofp_packet_in, data) + fwd_len;
 690         opi = alloc_openflow_skb(dp, opi_len, OFPT_PACKET_IN, NULL, &f_skb);
 691         if (!opi) {
 692                 err = -ENOMEM;
 693                 goto out;
 694         }
 695         opi->buffer_id      = htonl(buffer_id);
 696         opi->total_len      = htons(skb->len);
 697         p = skb->dev->br_port;
 698         opi->in_port        = htons(p ? p->port_no : OFPP_LOCAL);
 699         opi->reason         = reason;
 700         opi->pad            = 0;
 701         memcpy(opi->data, skb_mac_header(skb), fwd_len);
 702         err = send_openflow_skb(f_skb, NULL);
 703
 704 out:
 705         kfree_skb(skb);
 706         return err;
 707 }
 708
 709 static void fill_port_desc(struct net_bridge_port *p, struct ofp_phy_port *desc)
 710 {
 711         unsigned long flags;
 712         desc->port_no = htons(p->port_no);
 713         strncpy(desc->name, p->dev->name, OFP_MAX_PORT_NAME_LEN);
 714         desc->name[OFP_MAX_PORT_NAME_LEN-1] = '\0';
 715         memcpy(desc->hw_addr, p->dev->dev_addr, ETH_ALEN);
 716         desc->curr = 0;
 717         desc->supported = 0;
 718         desc->advertised = 0;
 719         desc->peer = 0;
 720
 721         spin_lock_irqsave(&p->lock, flags);
 722         desc->config = htonl(p->config);
 723         desc->state = htonl(p->state);
 724         spin_unlock_irqrestore(&p->lock, flags);
 725
 726 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,4,24)
 727         if (p->dev->ethtool_ops && p->dev->ethtool_ops->get_settings) {
 728                 struct ethtool_cmd ecmd = { .cmd = ETHTOOL_GSET };
 729
 730                 if (!p->dev->ethtool_ops->get_settings(p->dev, &ecmd)) {
 731                         /* Set the supported features */
 732                         if (ecmd.supported & SUPPORTED_10baseT_Half)
 733                                 desc->supported |= OFPPF_10MB_HD;
 734                         if (ecmd.supported & SUPPORTED_10baseT_Full)
 735                                 desc->supported |= OFPPF_10MB_FD;
 736                         if (ecmd.supported & SUPPORTED_100baseT_Half)
 737                                 desc->supported |= OFPPF_100MB_HD;
 738                         if (ecmd.supported & SUPPORTED_100baseT_Full)
 739                                 desc->supported |= OFPPF_100MB_FD;
 740                         if (ecmd.supported & SUPPORTED_1000baseT_Half)
 741                                 desc->supported |= OFPPF_1GB_HD;
 742                         if (ecmd.supported & SUPPORTED_1000baseT_Full)
 743                                 desc->supported |= OFPPF_1GB_FD;
 744                         if (ecmd.supported & SUPPORTED_10000baseT_Full)
 745                                 desc->supported |= OFPPF_10GB_FD;
 746                         if (ecmd.supported & SUPPORTED_TP)
 747                                 desc->supported |= OFPPF_COPPER;
 748                         if (ecmd.supported & SUPPORTED_FIBRE)
 749                                 desc->supported |= OFPPF_FIBER;
 750                         if (ecmd.supported & SUPPORTED_Autoneg)
 751                                 desc->supported |= OFPPF_AUTONEG;
 752 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,14)
 753                         if (ecmd.supported & SUPPORTED_Pause)
 754                                 desc->supported |= OFPPF_PAUSE;
 755                         if (ecmd.supported & SUPPORTED_Asym_Pause)
 756                                 desc->supported |= OFPPF_PAUSE_ASYM;
 757 #endif /* kernel >= 2.6.14 */
 758
 759                         /* Set the advertised features */
 760                         if (ecmd.advertising & ADVERTISED_10baseT_Half)
 761                                 desc->advertised |= OFPPF_10MB_HD;
 762                         if (ecmd.advertising & ADVERTISED_10baseT_Full)
 763                                 desc->advertised |= OFPPF_10MB_FD;
 764                         if (ecmd.advertising & ADVERTISED_100baseT_Half)
 765                                 desc->advertised |= OFPPF_100MB_HD;
 766                         if (ecmd.advertising & ADVERTISED_100baseT_Full)
 767                                 desc->advertised |= OFPPF_100MB_FD;
 768                         if (ecmd.advertising & ADVERTISED_1000baseT_Half)
 769                                 desc->advertised |= OFPPF_1GB_HD;
 770                         if (ecmd.advertising & ADVERTISED_1000baseT_Full)
 771                                 desc->advertised |= OFPPF_1GB_FD;
 772                         if (ecmd.advertising & ADVERTISED_10000baseT_Full)
 773                                 desc->advertised |= OFPPF_10GB_FD;
 774                         if (ecmd.advertising & ADVERTISED_TP)
 775                                 desc->advertised |= OFPPF_COPPER;
 776                         if (ecmd.advertising & ADVERTISED_FIBRE)
 777                                 desc->advertised |= OFPPF_FIBER;
 778                         if (ecmd.advertising & ADVERTISED_Autoneg)
 779                                 desc->advertised |= OFPPF_AUTONEG;
 780 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,14)
 781                         if (ecmd.advertising & ADVERTISED_Pause)
 782                                 desc->advertised |= OFPPF_PAUSE;
 783                         if (ecmd.advertising & ADVERTISED_Asym_Pause)
 784                                 desc->advertised |= OFPPF_PAUSE_ASYM;
 785 #endif /* kernel >= 2.6.14 */
 786
 787                         /* Set the current features */
 788                         if (ecmd.speed == SPEED_10)
 789                                 desc->curr = (ecmd.duplex) ? OFPPF_10MB_FD : OFPPF_10MB_HD;
 790                         else if (ecmd.speed == SPEED_100)
 791                                 desc->curr = (ecmd.duplex) ? OFPPF_100MB_FD : OFPPF_100MB_HD;
 792                         else if (ecmd.speed == SPEED_1000)
 793                                 desc->curr = (ecmd.duplex) ? OFPPF_1GB_FD : OFPPF_1GB_HD;
 794                         else if (ecmd.speed == SPEED_10000)
 795                                 desc->curr = OFPPF_10GB_FD;
 796
 797                         if (ecmd.port == PORT_TP)
 798                                 desc->curr |= OFPPF_COPPER;
 799                         else if (ecmd.port == PORT_FIBRE)
 800                                 desc->curr |= OFPPF_FIBER;
 801
 802                         if (ecmd.autoneg)
 803                                 desc->curr |= OFPPF_AUTONEG;
 804                 }
 805         }
 806 #endif
 807         desc->curr = htonl(desc->curr);
 808         desc->supported = htonl(desc->supported);
 809         desc->advertised = htonl(desc->advertised);
 810         desc->peer = htonl(desc->peer);
 811 }
 812
 813 static int
 814 fill_features_reply(struct datapath *dp, struct ofp_switch_features *ofr)
 815 {
 816         struct net_bridge_port *p;
 817         int port_count = 0;
 818
 819         ofr->datapath_id  = cpu_to_be64(dp->id);
 820
 821         ofr->n_buffers    = htonl(N_PKT_BUFFERS);
 822         ofr->n_tables     = dp->chain->n_tables;
 823         ofr->capabilities = htonl(OFP_SUPPORTED_CAPABILITIES);
 824         ofr->actions      = htonl(OFP_SUPPORTED_ACTIONS);
 825         memset(ofr->pad, 0, sizeof ofr->pad);
 826
 827         list_for_each_entry_rcu (p, &dp->port_list, node) {
 828                 fill_port_desc(p, &ofr->ports[port_count]);
 829                 port_count++;
 830         }
 831
 832         return port_count;
 833 }
 834
 835 int
 836 dp_send_features_reply(struct datapath *dp, const struct sender *sender)
 837 {
 838         struct sk_buff *skb;
 839         struct ofp_switch_features *ofr;
 840         size_t ofr_len, port_max_len;
 841         int port_count;
 842
 843         /* Overallocate. */
 844         port_max_len = sizeof(struct ofp_phy_port) * DP_MAX_PORTS;
 845         ofr = alloc_openflow_skb(dp, sizeof(*ofr) + port_max_len,
 846                                  OFPT_FEATURES_REPLY, sender, &skb);
 847         if (!ofr)
 848                 return -ENOMEM;
 849
 850         /* Fill. */
 851         port_count = fill_features_reply(dp, ofr);
 852
 853         /* Shrink to fit. */
 854         ofr_len = sizeof(*ofr) + (sizeof(struct ofp_phy_port) * port_count);
 855         resize_openflow_skb(skb, &ofr->header, ofr_len);
 856         return send_openflow_skb(skb, sender);
 857 }
 858
 859 int
 860 dp_send_config_reply(struct datapath *dp, const struct sender *sender)
 861 {
 862         struct sk_buff *skb;
 863         struct ofp_switch_config *osc;
 864
 865         osc = alloc_openflow_skb(dp, sizeof *osc, OFPT_GET_CONFIG_REPLY, sender,
 866                                  &skb);
 867         if (!osc)
 868                 return -ENOMEM;
 869
 870         osc->flags = htons(dp->flags);
 871         osc->miss_send_len = htons(dp->miss_send_len);
 872
 873         return send_openflow_skb(skb, sender);
 874 }
 875
 876 int
 877 dp_send_hello(struct datapath *dp, const struct sender *sender,
 878               const struct ofp_header *request)
 879 {
 880         if (request->version < OFP_VERSION) {
 881                 char err[64];
 882                 sprintf(err, "Only version 0x%02x supported", OFP_VERSION);
 883                 dp_send_error_msg(dp, sender, OFPET_HELLO_FAILED,
 884                                   OFPHFC_INCOMPATIBLE, err, strlen(err));
 885                 return -EINVAL;
 886         } else {
 887                 struct sk_buff *skb;
 888                 struct ofp_header *reply;
 889
 890                 reply = alloc_openflow_skb(dp, sizeof *reply,
 891                                            OFPT_HELLO, sender, &skb);
 892                 if (!reply)
 893                         return -ENOMEM;
 894
 895                 return send_openflow_skb(skb, sender);
 896         }
 897 }
 898
 899 /* Callback function for a workqueue to disable an interface */
 900 static void
 901 down_port_cb(struct work_struct *work)
 902 {
 903         struct net_bridge_port *p = container_of(work, struct net_bridge_port,
 904                         port_task);
 905
 906         rtnl_lock();
 907         if (dev_change_flags(p->dev, p->dev->flags & ~IFF_UP) < 0)
 908                 if (net_ratelimit())
 909                         printk("problem bringing up port %s\n", p->dev->name);
 910         rtnl_unlock();
 911         p->config |= OFPPC_PORT_DOWN;
 912 }
 913
 914 /* Callback function for a workqueue to enable an interface */
 915 static void
 916 up_port_cb(struct work_struct *work)
 917 {
 918         struct net_bridge_port *p = container_of(work, struct net_bridge_port,
 919                         port_task);
 920
 921         rtnl_lock();
 922         if (dev_change_flags(p->dev, p->dev->flags | IFF_UP) < 0)
 923                 if (net_ratelimit())
 924                         printk("problem bringing down port %s\n", p->dev->name);
 925         rtnl_unlock();
 926         p->config &= ~OFPPC_PORT_DOWN;
 927 }
 928
 929 int
 930 dp_update_port_flags(struct datapath *dp, const struct ofp_port_mod *opm)
 931 {
 932         unsigned long int flags;
 933         int port_no = ntohs(opm->port_no);
 934         struct net_bridge_port *p;
 935         p = (port_no < DP_MAX_PORTS ? dp->ports[port_no]
 936              : port_no == OFPP_LOCAL ? dp->local_port
 937              : NULL);
 938
 939         /* Make sure the port id hasn't changed since this was sent */
 940         if (!p || memcmp(opm->hw_addr, p->dev->dev_addr, ETH_ALEN))
 941                 return -1;
 942
 943         spin_lock_irqsave(&p->lock, flags);
 944         if (opm->mask) {
 945                 uint32_t config_mask = ntohl(opm->mask);
 946                 p->config &= ~config_mask;
 947                 p->config |= ntohl(opm->config) & config_mask;
 948         }
 949
 950         /* Modifying the status of an interface requires taking a lock
 951          * that cannot be done from here.  For this reason, we use a shared
 952          * workqueue, which will cause it to be executed from a safer
 953          * context. */
 954         if (opm->mask & htonl(OFPPC_PORT_DOWN)) {
 955                 if ((opm->config & htonl(OFPPC_PORT_DOWN))
 956                     && (p->config & OFPPC_PORT_DOWN) == 0) {
 957                         PREPARE_WORK(&p->port_task, down_port_cb);
 958                         schedule_work(&p->port_task);
 959                 } else if ((opm->config & htonl(OFPPC_PORT_DOWN)) == 0
 960                            && (p->config & OFPPC_PORT_DOWN)) {
 961                         PREPARE_WORK(&p->port_task, up_port_cb);
 962                         schedule_work(&p->port_task);
 963                 }
 964         }
 965         spin_unlock_irqrestore(&p->lock, flags);
 966
 967         return 0;
 968 }
 969
 970 /* Initialize the port status field of the bridge port. */
 971 static void
 972 init_port_status(struct net_bridge_port *p)
 973 {
 974         unsigned long int flags;
 975
 976         spin_lock_irqsave(&p->lock, flags);
 977
 978         if (p->dev->flags & IFF_UP)
 979                 p->config &= ~OFPPC_PORT_DOWN;
 980         else
 981                 p->config |= OFPPC_PORT_DOWN;
 982
 983         if (netif_carrier_ok(p->dev))
 984                 p->state &= ~OFPPS_LINK_DOWN;
 985         else
 986                 p->state |= OFPPS_LINK_DOWN;
 987
 988         spin_unlock_irqrestore(&p->lock, flags);
 989 }
 990
 991 int
 992 dp_send_port_status(struct net_bridge_port *p, uint8_t status)
 993 {
 994         struct sk_buff *skb;
 995         struct ofp_port_status *ops;
 996
 997         ops = alloc_openflow_skb(p->dp, sizeof *ops, OFPT_PORT_STATUS, NULL,
 998                                  &skb);
 999         if (!ops)
1000                 return -ENOMEM;
1001         ops->reason = status;
1002         memset(ops->pad, 0, sizeof ops->pad);
1003         fill_port_desc(p, &ops->desc);
1004
1005         return send_openflow_skb(skb, NULL);
1006 }
1007
1008 int
1009 dp_send_flow_expired(struct datapath *dp, struct sw_flow *flow,
1010                      enum ofp_flow_expired_reason reason)
1011 {
1012         struct sk_buff *skb;
1013         struct ofp_flow_expired *ofe;
1014
1015         if (!(dp->flags & OFPC_SEND_FLOW_EXP))
1016                 return 0;
1017
1018         ofe = alloc_openflow_skb(dp, sizeof *ofe, OFPT_FLOW_EXPIRED, 0, &skb);
1019         if (!ofe)
1020                 return -ENOMEM;
1021
1022         flow_fill_match(&ofe->match, &flow->key);
1023
1024         ofe->priority = htons(flow->priority);
1025         ofe->reason = reason;
1026         memset(ofe->pad, 0, sizeof ofe->pad);
1027
1028         ofe->duration     = htonl((jiffies - flow->init_time) / HZ);
1029         memset(ofe->pad2, 0, sizeof ofe->pad2);
1030         ofe->packet_count = cpu_to_be64(flow->packet_count);
1031         ofe->byte_count   = cpu_to_be64(flow->byte_count);
1032
1033         return send_openflow_skb(skb, NULL);
1034 }
1035 EXPORT_SYMBOL(dp_send_flow_expired);
1036
1037 int
1038 dp_send_error_msg(struct datapath *dp, const struct sender *sender,
1039                 uint16_t type, uint16_t code, const void *data, size_t len)
1040 {
1041         struct sk_buff *skb;
1042         struct ofp_error_msg *oem;
1043
1044
1045         oem = alloc_openflow_skb(dp, sizeof(*oem)+len, OFPT_ERROR,
1046                         sender, &skb);
1047         if (!oem)
1048                 return -ENOMEM;
1049
1050         oem->type = htons(type);
1051         oem->code = htons(code);
1052         memcpy(oem->data, data, len);
1053
1054         return send_openflow_skb(skb, sender);
1055 }
1056
1057 int
1058 dp_send_echo_reply(struct datapath *dp, const struct sender *sender,
1059                    const struct ofp_header *rq)
1060 {
1061         struct sk_buff *skb;
1062         struct ofp_header *reply;
1063
1064         reply = alloc_openflow_skb(dp, ntohs(rq->length), OFPT_ECHO_REPLY,
1065                                    sender, &skb);
1066         if (!reply)
1067                 return -ENOMEM;
1068
1069         memcpy(reply + 1, rq + 1, ntohs(rq->length) - sizeof *rq);
1070         return send_openflow_skb(skb, sender);
1071 }
1072
1073 /* Generic Netlink interface.
1074  *
1075  * See netlink(7) for an introduction to netlink.  See
1076  * http://linux-net.osdl.org/index.php/Netlink for more information and
1077  * pointers on how to work with netlink and Generic Netlink in the kernel and
1078  * in userspace. */
1079
1080 static struct genl_family dp_genl_family = {
1081         .id = GENL_ID_GENERATE,
1082         .hdrsize = 0,
1083         .name = DP_GENL_FAMILY_NAME,
1084         .version = 1,
1085         .maxattr = DP_GENL_A_MAX,
1086 };
1087
1088 /* Attribute policy: what each attribute may contain.  */
1089 static struct nla_policy dp_genl_policy[DP_GENL_A_MAX + 1] = {
1090         [DP_GENL_A_DP_IDX] = { .type = NLA_U32 },
1091         [DP_GENL_A_MC_GROUP] = { .type = NLA_U32 },
1092         [DP_GENL_A_PORTNAME] = { .type = NLA_STRING }
1093 };
1094
1095 static int dp_genl_add(struct sk_buff *skb, struct genl_info *info)
1096 {
1097         if (!info->attrs[DP_GENL_A_DP_IDX])
1098                 return -EINVAL;
1099
1100         return new_dp(nla_get_u32(info->attrs[DP_GENL_A_DP_IDX]));
1101 }
1102
1103 static struct genl_ops dp_genl_ops_add_dp = {
1104         .cmd = DP_GENL_C_ADD_DP,
1105         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1106         .policy = dp_genl_policy,
1107         .doit = dp_genl_add,
1108         .dumpit = NULL,
1109 };
1110
1111 struct datapath *dp_get(int dp_idx)
1112 {
1113         if (dp_idx < 0 || dp_idx > DP_MAX)
1114                 return NULL;
1115         return rcu_dereference(dps[dp_idx]);
1116 }
1117
1118 static int dp_genl_del(struct sk_buff *skb, struct genl_info *info)
1119 {
1120         struct datapath *dp;
1121         int err;
1122
1123         if (!info->attrs[DP_GENL_A_DP_IDX])
1124                 return -EINVAL;
1125
1126         dp = dp_get(nla_get_u32((info->attrs[DP_GENL_A_DP_IDX])));
1127         if (!dp)
1128                 err = -ENOENT;
1129         else {
1130                 del_dp(dp);
1131                 err = 0;
1132         }
1133         return err;
1134 }
1135
1136 static struct genl_ops dp_genl_ops_del_dp = {
1137         .cmd = DP_GENL_C_DEL_DP,
1138         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1139         .policy = dp_genl_policy,
1140         .doit = dp_genl_del,
1141         .dumpit = NULL,
1142 };
1143
1144 /* Queries a datapath for related information.  Currently the only relevant
1145  * information is the datapath's multicast group ID.  Really we want one
1146  * multicast group per datapath, but because of locking issues[*] we can't
1147  * easily get one.  Thus, every datapath will currently return the same
1148  * global multicast group ID, but in the future it would be nice to fix that.
1149  *
1150  * [*] dp_genl_add, to add a new datapath, is called under the genl_lock
1151  *       mutex, and genl_register_mc_group, called to acquire a new multicast
1152  *       group ID, also acquires genl_lock, thus deadlock.
1153  */
1154 static int dp_genl_query(struct sk_buff *skb, struct genl_info *info)
1155 {
1156         struct datapath *dp;
1157         struct sk_buff *ans_skb = NULL;
1158         int dp_idx;
1159         int err = -ENOMEM;
1160
1161         if (!info->attrs[DP_GENL_A_DP_IDX])
1162                 return -EINVAL;
1163
1164         rcu_read_lock();
1165         dp_idx = nla_get_u32((info->attrs[DP_GENL_A_DP_IDX]));
1166         dp = dp_get(dp_idx);
1167         if (!dp)
1168                 err = -ENOENT;
1169         else {
1170                 void *data;
1171                 ans_skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1172                 if (!ans_skb) {
1173                         err = -ENOMEM;
1174                         goto err;
1175                 }
1176                 data = genlmsg_put_reply(ans_skb, info, &dp_genl_family,
1177                                          0, DP_GENL_C_QUERY_DP);
1178                 if (data == NULL) {
1179                         err = -ENOMEM;
1180                         goto err;
1181                 }
1182                 NLA_PUT_U32(ans_skb, DP_GENL_A_DP_IDX, dp_idx);
1183                 NLA_PUT_U32(ans_skb, DP_GENL_A_MC_GROUP, mc_group.id);
1184
1185                 genlmsg_end(ans_skb, data);
1186                 err = genlmsg_reply(ans_skb, info);
1187                 if (!err)
1188                         ans_skb = NULL;
1189         }
1190 err:
1191 nla_put_failure:
1192         if (ans_skb)
1193                 kfree_skb(ans_skb);
1194         rcu_read_unlock();
1195         return err;
1196 }
1197
1198 static struct genl_ops dp_genl_ops_query_dp = {
1199         .cmd = DP_GENL_C_QUERY_DP,
1200         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1201         .policy = dp_genl_policy,
1202         .doit = dp_genl_query,
1203         .dumpit = NULL,
1204 };
1205
1206 static int dp_genl_add_del_port(struct sk_buff *skb, struct genl_info *info)
1207 {
1208         struct datapath *dp;
1209         struct net_device *port;
1210         int err;
1211
1212         if (!info->attrs[DP_GENL_A_DP_IDX] || !info->attrs[DP_GENL_A_PORTNAME])
1213                 return -EINVAL;
1214
1215         /* Get datapath. */
1216         dp = dp_get(nla_get_u32(info->attrs[DP_GENL_A_DP_IDX]));
1217         if (!dp) {
1218                 err = -ENOENT;
1219                 goto out;
1220         }
1221
1222         /* Get interface to add/remove. */
1223         port = dev_get_by_name(&init_net,
1224                         nla_data(info->attrs[DP_GENL_A_PORTNAME]));
1225         if (!port) {
1226                 err = -ENOENT;
1227                 goto out;
1228         }
1229
1230         /* Execute operation. */
1231         if (info->genlhdr->cmd == DP_GENL_C_ADD_PORT)
1232                 err = add_switch_port(dp, port);
1233         else {
1234                 if (port->br_port == NULL || port->br_port->dp != dp) {
1235                         err = -ENOENT;
1236                         goto out_put;
1237                 }
1238                 err = dp_del_switch_port(port->br_port);
1239         }
1240
1241 out_put:
1242         dev_put(port);
1243 out:
1244         return err;
1245 }
1246
1247 static struct genl_ops dp_genl_ops_add_port = {
1248         .cmd = DP_GENL_C_ADD_PORT,
1249         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1250         .policy = dp_genl_policy,
1251         .doit = dp_genl_add_del_port,
1252         .dumpit = NULL,
1253 };
1254
1255 static struct genl_ops dp_genl_ops_del_port = {
1256         .cmd = DP_GENL_C_DEL_PORT,
1257         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1258         .policy = dp_genl_policy,
1259         .doit = dp_genl_add_del_port,
1260         .dumpit = NULL,
1261 };
1262
1263 static int dp_genl_openflow(struct sk_buff *skb, struct genl_info *info)
1264 {
1265         struct nlattr *va = info->attrs[DP_GENL_A_OPENFLOW];
1266         struct datapath *dp;
1267         struct ofp_header *oh;
1268         struct sender sender;
1269         int err;
1270
1271         if (!info->attrs[DP_GENL_A_DP_IDX] || !va)
1272                 return -EINVAL;
1273
1274         dp = dp_get(nla_get_u32(info->attrs[DP_GENL_A_DP_IDX]));
1275         if (!dp)
1276                 return -ENOENT;
1277
1278         if (nla_len(va) < sizeof(struct ofp_header))
1279                 return -EINVAL;
1280         oh = nla_data(va);
1281
1282         sender.xid = oh->xid;
1283         sender.pid = info->snd_pid;
1284         sender.seq = info->snd_seq;
1285
1286         mutex_lock(&dp_mutex);
1287         err = fwd_control_input(dp->chain, &sender,
1288                                 nla_data(va), nla_len(va));
1289         mutex_unlock(&dp_mutex);
1290         return err;
1291 }
1292
1293 static struct nla_policy dp_genl_openflow_policy[DP_GENL_A_MAX + 1] = {
1294         [DP_GENL_A_DP_IDX] = { .type = NLA_U32 },
1295 };
1296
1297 static int desc_stats_dump(struct datapath *dp, void *state,
1298                             void *body, int *body_len)
1299 {
1300         struct ofp_desc_stats *ods = body;
1301         int n_bytes = sizeof *ods;
1302
1303         if (n_bytes > *body_len) {
1304                 return -ENOBUFS;
1305         }
1306         *body_len = n_bytes;
1307
1308         strncpy(ods->mfr_desc, mfr_desc, sizeof ods->mfr_desc);
1309         strncpy(ods->hw_desc, hw_desc, sizeof ods->hw_desc);
1310         strncpy(ods->sw_desc, sw_desc, sizeof ods->sw_desc);
1311         strncpy(ods->serial_num, serial_num, sizeof ods->serial_num);
1312
1313         return 0;
1314 }
1315
1316 struct flow_stats_state {
1317         int table_idx;
1318         struct sw_table_position position;
1319         const struct ofp_flow_stats_request *rq;
1320
1321         void *body;
1322         int bytes_used, bytes_allocated;
1323 };
1324
1325 static int flow_stats_init(struct datapath *dp, const void *body, int body_len,
1326                            void **state)
1327 {
1328         const struct ofp_flow_stats_request *fsr = body;
1329         struct flow_stats_state *s = kmalloc(sizeof *s, GFP_ATOMIC);
1330         if (!s)
1331                 return -ENOMEM;
1332         s->table_idx = fsr->table_id == 0xff ? 0 : fsr->table_id;
1333         memset(&s->position, 0, sizeof s->position);
1334         s->rq = fsr;
1335         *state = s;
1336         return 0;
1337 }
1338
1339 static int flow_stats_dump_callback(struct sw_flow *flow, void *private)
1340 {
1341         struct sw_flow_actions *sf_acts = rcu_dereference(flow->sf_acts);
1342         struct flow_stats_state *s = private;
1343         struct ofp_flow_stats *ofs;
1344         int length;
1345
1346         length = sizeof *ofs + sf_acts->actions_len;
1347         if (length + s->bytes_used > s->bytes_allocated)
1348                 return 1;
1349
1350         ofs = s->body + s->bytes_used;
1351         ofs->length          = htons(length);
1352         ofs->table_id        = s->table_idx;
1353         ofs->pad             = 0;
1354         ofs->match.wildcards = htonl(flow->key.wildcards);
1355         ofs->match.in_port   = flow->key.in_port;
1356         memcpy(ofs->match.dl_src, flow->key.dl_src, ETH_ALEN);
1357         memcpy(ofs->match.dl_dst, flow->key.dl_dst, ETH_ALEN);
1358         ofs->match.dl_vlan   = flow->key.dl_vlan;
1359         ofs->match.dl_type   = flow->key.dl_type;
1360         ofs->match.nw_src    = flow->key.nw_src;
1361         ofs->match.nw_dst    = flow->key.nw_dst;
1362         ofs->match.nw_proto  = flow->key.nw_proto;
1363         ofs->match.pad       = 0;
1364         ofs->match.tp_src    = flow->key.tp_src;
1365         ofs->match.tp_dst    = flow->key.tp_dst;
1366         ofs->duration        = htonl((jiffies - flow->init_time) / HZ);
1367         ofs->priority        = htons(flow->priority);
1368         ofs->idle_timeout    = htons(flow->idle_timeout);
1369         ofs->hard_timeout    = htons(flow->hard_timeout);
1370         memset(ofs->pad2, 0, sizeof ofs->pad2);
1371         ofs->packet_count    = cpu_to_be64(flow->packet_count);
1372         ofs->byte_count      = cpu_to_be64(flow->byte_count);
1373         memcpy(ofs->actions, sf_acts->actions, sf_acts->actions_len);
1374
1375         s->bytes_used += length;
1376         return 0;
1377 }
1378
1379 static int flow_stats_dump(struct datapath *dp, void *state,
1380                            void *body, int *body_len)
1381 {
1382         struct flow_stats_state *s = state;
1383         struct sw_flow_key match_key;
1384         int error = 0;
1385
1386         s->bytes_used = 0;
1387         s->bytes_allocated = *body_len;
1388         s->body = body;
1389
1390         flow_extract_match(&match_key, &s->rq->match);
1391         while (s->table_idx < dp->chain->n_tables
1392                && (s->rq->table_id == 0xff || s->rq->table_id == s->table_idx))
1393         {
1394                 struct sw_table *table = dp->chain->tables[s->table_idx];
1395
1396                 error = table->iterate(table, &match_key, &s->position,
1397                                        flow_stats_dump_callback, s);
1398                 if (error)
1399                         break;
1400
1401                 s->table_idx++;
1402                 memset(&s->position, 0, sizeof s->position);
1403         }
1404         *body_len = s->bytes_used;
1405
1406         /* If error is 0, we're done.
1407          * Otherwise, if some bytes were used, there are more flows to come.
1408          * Otherwise, we were not able to fit even a single flow in the body,
1409          * which indicates that we have a single flow with too many actions to
1410          * fit.  We won't ever make any progress at that rate, so give up. */
1411         return !error ? 0 : s->bytes_used ? 1 : -ENOMEM;
1412 }
1413
1414 static void flow_stats_done(void *state)
1415 {
1416         kfree(state);
1417 }
1418
1419 static int aggregate_stats_init(struct datapath *dp,
1420                                 const void *body, int body_len,
1421                                 void **state)
1422 {
1423         *state = (void *)body;
1424         return 0;
1425 }
1426
1427 static int aggregate_stats_dump_callback(struct sw_flow *flow, void *private)
1428 {
1429         struct ofp_aggregate_stats_reply *rpy = private;
1430         rpy->packet_count += flow->packet_count;
1431         rpy->byte_count += flow->byte_count;
1432         rpy->flow_count++;
1433         return 0;
1434 }
1435
1436 static int aggregate_stats_dump(struct datapath *dp, void *state,
1437                                 void *body, int *body_len)
1438 {
1439         struct ofp_aggregate_stats_request *rq = state;
1440         struct ofp_aggregate_stats_reply *rpy;
1441         struct sw_table_position position;
1442         struct sw_flow_key match_key;
1443         int table_idx;
1444
1445         if (*body_len < sizeof *rpy)
1446                 return -ENOBUFS;
1447         rpy = body;
1448         *body_len = sizeof *rpy;
1449
1450         memset(rpy, 0, sizeof *rpy);
1451
1452         flow_extract_match(&match_key, &rq->match);
1453         table_idx = rq->table_id == 0xff ? 0 : rq->table_id;
1454         memset(&position, 0, sizeof position);
1455         while (table_idx < dp->chain->n_tables
1456                && (rq->table_id == 0xff || rq->table_id == table_idx))
1457         {
1458                 struct sw_table *table = dp->chain->tables[table_idx];
1459                 int error;
1460
1461                 error = table->iterate(table, &match_key, &position,
1462                                        aggregate_stats_dump_callback, rpy);
1463                 if (error)
1464                         return error;
1465
1466                 table_idx++;
1467                 memset(&position, 0, sizeof position);
1468         }
1469
1470         rpy->packet_count = cpu_to_be64(rpy->packet_count);
1471         rpy->byte_count = cpu_to_be64(rpy->byte_count);
1472         rpy->flow_count = htonl(rpy->flow_count);
1473         return 0;
1474 }
1475
1476 static int table_stats_dump(struct datapath *dp, void *state,
1477                             void *body, int *body_len)
1478 {
1479         struct ofp_table_stats *ots;
1480         int n_bytes = dp->chain->n_tables * sizeof *ots;
1481         int i;
1482         if (n_bytes > *body_len)
1483                 return -ENOBUFS;
1484         *body_len = n_bytes;
1485         for (i = 0, ots = body; i < dp->chain->n_tables; i++, ots++) {
1486                 struct sw_table_stats stats;
1487                 dp->chain->tables[i]->stats(dp->chain->tables[i], &stats);
1488                 strncpy(ots->name, stats.name, sizeof ots->name);
1489                 ots->table_id = i;
1490                 ots->wildcards = htonl(stats.wildcards);
1491                 memset(ots->pad, 0, sizeof ots->pad);
1492                 ots->max_entries = htonl(stats.max_flows);
1493                 ots->active_count = htonl(stats.n_flows);
1494                 ots->lookup_count = cpu_to_be64(stats.n_lookup);
1495                 ots->matched_count = cpu_to_be64(stats.n_matched);
1496         }
1497         return 0;
1498 }
1499
1500 struct port_stats_state {
1501         int port;
1502 };
1503
1504 static int port_stats_init(struct datapath *dp, const void *body, int body_len,
1505                            void **state)
1506 {
1507         struct port_stats_state *s = kmalloc(sizeof *s, GFP_ATOMIC);
1508         if (!s)
1509                 return -ENOMEM;
1510         s->port = 0;
1511         *state = s;
1512         return 0;
1513 }
1514
1515 static int port_stats_dump(struct datapath *dp, void *state,
1516                            void *body, int *body_len)
1517 {
1518         struct port_stats_state *s = state;
1519         struct ofp_port_stats *ops;
1520         int n_ports, max_ports;
1521         int i;
1522
1523         max_ports = *body_len / sizeof *ops;
1524         if (!max_ports)
1525                 return -ENOMEM;
1526         ops = body;
1527
1528         n_ports = 0;
1529         for (i = s->port; i < DP_MAX_PORTS && n_ports < max_ports; i++) {
1530                 struct net_bridge_port *p = dp->ports[i];
1531                 struct net_device_stats *stats;
1532                 if (!p)
1533                         continue;
1534                 stats = p->dev->get_stats(p->dev);
1535                 ops->port_no = htons(p->port_no);
1536                 memset(ops->pad, 0, sizeof ops->pad);
1537                 ops->rx_packets   = cpu_to_be64(stats->rx_packets);
1538                 ops->tx_packets   = cpu_to_be64(stats->tx_packets);
1539                 ops->rx_bytes     = cpu_to_be64(stats->rx_bytes);
1540                 ops->tx_bytes     = cpu_to_be64(stats->tx_bytes);
1541                 ops->rx_dropped   = cpu_to_be64(stats->rx_dropped);
1542                 ops->tx_dropped   = cpu_to_be64(stats->tx_dropped);
1543                 ops->rx_errors    = cpu_to_be64(stats->rx_errors);
1544                 ops->tx_errors    = cpu_to_be64(stats->tx_errors);
1545                 ops->rx_frame_err = cpu_to_be64(stats->rx_frame_errors);
1546                 ops->rx_over_err  = cpu_to_be64(stats->rx_over_errors);
1547                 ops->rx_crc_err   = cpu_to_be64(stats->rx_crc_errors);
1548                 ops->collisions   = cpu_to_be64(stats->collisions);
1549                 n_ports++;
1550                 ops++;
1551         }
1552         s->port = i;
1553         *body_len = n_ports * sizeof *ops;
1554         return n_ports >= max_ports;
1555 }
1556
1557 static void port_stats_done(void *state)
1558 {
1559         kfree(state);
1560 }
1561
1562 struct stats_type {
1563         /* Minimum and maximum acceptable number of bytes in body member of
1564          * struct ofp_stats_request. */
1565         size_t min_body, max_body;
1566
1567         /* Prepares to dump some kind of statistics on 'dp'.  'body' and
1568          * 'body_len' are the 'body' member of the struct ofp_stats_request.
1569          * Returns zero if successful, otherwise a negative error code.
1570          * May initialize '*state' to state information.  May be null if no
1571          * initialization is required.*/
1572         int (*init)(struct datapath *dp, const void *body, int body_len,
1573                     void **state);
1574
1575         /* Dumps statistics for 'dp' into the '*body_len' bytes at 'body', and
1576          * modifies '*body_len' to reflect the number of bytes actually used.
1577          * ('body' will be transmitted as the 'body' member of struct
1578          * ofp_stats_reply.) */
1579         int (*dump)(struct datapath *dp, void *state,
1580                     void *body, int *body_len);
1581
1582         /* Cleans any state created by the init or dump functions.  May be null
1583          * if no cleanup is required. */
1584         void (*done)(void *state);
1585 };
1586
1587 static const struct stats_type stats[] = {
1588         [OFPST_DESC] = {
1589                 0,
1590                 0,
1591                 NULL,
1592                 desc_stats_dump,
1593                 NULL
1594         },
1595         [OFPST_FLOW] = {
1596                 sizeof(struct ofp_flow_stats_request),
1597                 sizeof(struct ofp_flow_stats_request),
1598                 flow_stats_init,
1599                 flow_stats_dump,
1600                 flow_stats_done
1601         },
1602         [OFPST_AGGREGATE] = {
1603                 sizeof(struct ofp_aggregate_stats_request),
1604                 sizeof(struct ofp_aggregate_stats_request),
1605                 aggregate_stats_init,
1606                 aggregate_stats_dump,
1607                 NULL
1608         },
1609         [OFPST_TABLE] = {
1610                 0,
1611                 0,
1612                 NULL,
1613                 table_stats_dump,
1614                 NULL
1615         },
1616         [OFPST_PORT] = {
1617                 0,
1618                 0,
1619                 port_stats_init,
1620                 port_stats_dump,
1621                 port_stats_done
1622         },
1623 };
1624
1625 static int
1626 dp_genl_openflow_dumpit(struct sk_buff *skb, struct netlink_callback *cb)
1627 {
1628         struct datapath *dp;
1629         struct sender sender;
1630         const struct stats_type *s;
1631         struct ofp_stats_reply *osr;
1632         int dp_idx;
1633         int max_openflow_len, body_len;
1634         void *body;
1635         int err;
1636
1637         /* Set up the cleanup function for this dump.  Linux 2.6.20 and later
1638          * support setting up cleanup functions via the .doneit member of
1639          * struct genl_ops.  This kluge supports earlier versions also. */
1640         cb->done = dp_genl_openflow_done;
1641
1642         sender.pid = NETLINK_CB(cb->skb).pid;
1643         sender.seq = cb->nlh->nlmsg_seq;
1644         if (!cb->args[0]) {
1645                 struct nlattr *attrs[DP_GENL_A_MAX + 1];
1646                 struct ofp_stats_request *rq;
1647                 struct nlattr *va;
1648                 size_t len, body_len;
1649                 int type;
1650
1651                 err = nlmsg_parse(cb->nlh, GENL_HDRLEN, attrs, DP_GENL_A_MAX,
1652                                   dp_genl_openflow_policy);
1653                 if (err < 0)
1654                         return err;
1655
1656                 if (!attrs[DP_GENL_A_DP_IDX])
1657                         return -EINVAL;
1658                 dp_idx = nla_get_u16(attrs[DP_GENL_A_DP_IDX]);
1659                 dp = dp_get(dp_idx);
1660                 if (!dp)
1661                         return -ENOENT;
1662
1663                 va = attrs[DP_GENL_A_OPENFLOW];
1664                 len = nla_len(va);
1665                 if (!va || len < sizeof *rq)
1666                         return -EINVAL;
1667
1668                 rq = nla_data(va);
1669                 sender.xid = rq->header.xid;
1670                 type = ntohs(rq->type);
1671                 if (rq->header.version != OFP_VERSION) {
1672                         dp_send_error_msg(dp, &sender, OFPET_BAD_REQUEST,
1673                                           OFPBRC_BAD_VERSION, rq, len);
1674                         return -EINVAL;
1675                 }
1676                 if (rq->header.type != OFPT_STATS_REQUEST
1677                     || ntohs(rq->header.length) != len)
1678                         return -EINVAL;
1679
1680                 if (type >= ARRAY_SIZE(stats) || !stats[type].dump) {
1681                         dp_send_error_msg(dp, &sender, OFPET_BAD_REQUEST,
1682                                           OFPBRC_BAD_STAT, rq, len);
1683                         return -EINVAL;
1684                 }
1685
1686                 s = &stats[type];
1687                 body_len = len - offsetof(struct ofp_stats_request, body);
1688                 if (body_len < s->min_body || body_len > s->max_body)
1689                         return -EINVAL;
1690
1691                 cb->args[0] = 1;
1692                 cb->args[1] = dp_idx;
1693                 cb->args[2] = type;
1694                 cb->args[3] = rq->header.xid;
1695                 if (s->init) {
1696                         void *state;
1697                         err = s->init(dp, rq->body, body_len, &state);
1698                         if (err)
1699                                 return err;
1700                         cb->args[4] = (long) state;
1701                 }
1702         } else if (cb->args[0] == 1) {
1703                 sender.xid = cb->args[3];
1704                 dp_idx = cb->args[1];
1705                 s = &stats[cb->args[2]];
1706
1707                 dp = dp_get(dp_idx);
1708                 if (!dp)
1709                         return -ENOENT;
1710         } else {
1711                 return 0;
1712         }
1713
1714         osr = put_openflow_headers(dp, skb, OFPT_STATS_REPLY, &sender,
1715                                    &max_openflow_len);
1716         if (IS_ERR(osr))
1717                 return PTR_ERR(osr);
1718         osr->type = htons(s - stats);
1719         osr->flags = 0;
1720         resize_openflow_skb(skb, &osr->header, max_openflow_len);
1721         body = osr->body;
1722         body_len = max_openflow_len - offsetof(struct ofp_stats_reply, body);
1723
1724         err = s->dump(dp, (void *) cb->args[4], body, &body_len);
1725         if (err >= 0) {
1726                 if (!err)
1727                         cb->args[0] = 2;
1728                 else
1729                         osr->flags = ntohs(OFPSF_REPLY_MORE);
1730                 resize_openflow_skb(skb, &osr->header,
1731                                     (offsetof(struct ofp_stats_reply, body)
1732                                      + body_len));
1733                 err = skb->len;
1734         }
1735
1736         return err;
1737 }
1738
1739 static int
1740 dp_genl_openflow_done(struct netlink_callback *cb)
1741 {
1742         if (cb->args[0]) {
1743                 const struct stats_type *s = &stats[cb->args[2]];
1744                 if (s->done)
1745                         s->done((void *) cb->args[4]);
1746         }
1747         return 0;
1748 }
1749
1750 static struct genl_ops dp_genl_ops_openflow = {
1751         .cmd = DP_GENL_C_OPENFLOW,
1752         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1753         .policy = dp_genl_openflow_policy,
1754         .doit = dp_genl_openflow,
1755         .dumpit = dp_genl_openflow_dumpit,
1756 };
1757
1758 static struct genl_ops *dp_genl_all_ops[] = {
1759         /* Keep this operation first.  Generic Netlink dispatching
1760          * looks up operations with linear search, so we want it at the
1761          * front. */
1762         &dp_genl_ops_openflow,
1763
1764         &dp_genl_ops_add_dp,
1765         &dp_genl_ops_del_dp,
1766         &dp_genl_ops_query_dp,
1767         &dp_genl_ops_add_port,
1768         &dp_genl_ops_del_port,
1769 };
1770
1771 static int dp_init_netlink(void)
1772 {
1773         int err;
1774         int i;
1775
1776         err = genl_register_family(&dp_genl_family);
1777         if (err)
1778                 return err;
1779
1780         for (i = 0; i < ARRAY_SIZE(dp_genl_all_ops); i++) {
1781                 err = genl_register_ops(&dp_genl_family, dp_genl_all_ops[i]);
1782                 if (err)
1783                         goto err_unregister;
1784         }
1785
1786         strcpy(mc_group.name, "openflow");
1787         err = genl_register_mc_group(&dp_genl_family, &mc_group);
1788         if (err < 0)
1789                 goto err_unregister;
1790
1791         return 0;
1792
1793 err_unregister:
1794         genl_unregister_family(&dp_genl_family);
1795                 return err;
1796 }
1797
1798 static void dp_uninit_netlink(void)
1799 {
1800         genl_unregister_family(&dp_genl_family);
1801 }
1802
1803 static int __init dp_init(void)
1804 {
1805         int err;
1806
1807         printk("OpenFlow "VERSION", built "__DATE__" "__TIME__", "
1808                "protocol 0x%02x\n", OFP_VERSION);
1809
1810         err = flow_init();
1811         if (err)
1812                 goto error;
1813
1814         err = register_netdevice_notifier(&dp_device_notifier);
1815         if (err)
1816                 goto error_flow_exit;
1817
1818         err = dp_init_netlink();
1819         if (err)
1820                 goto error_unreg_notifier;
1821
1822         /* Hook into callback used by the bridge to intercept packets.
1823          * Parasites we are. */
1824         if (br_handle_frame_hook)
1825                 printk("openflow: hijacking bridge hook\n");
1826         br_handle_frame_hook = dp_frame_hook;
1827
1828         return 0;
1829
1830 error_unreg_notifier:
1831         unregister_netdevice_notifier(&dp_device_notifier);
1832 error_flow_exit:
1833         flow_exit();
1834 error:
1835         printk(KERN_EMERG "openflow: failed to install!");
1836         return err;
1837 }
1838
1839 static void dp_cleanup(void)
1840 {
1841         fwd_exit();
1842         dp_uninit_netlink();
1843         unregister_netdevice_notifier(&dp_device_notifier);
1844         flow_exit();
1845         br_handle_frame_hook = NULL;
1846 }
1847
1848 module_init(dp_init);
1849 module_exit(dp_cleanup);
1850
1851 MODULE_DESCRIPTION("OpenFlow switching datapath");
1852 MODULE_AUTHOR("Copyright (c) 2007, 2008 The Board of Trustees of The Leland Stanford Junior University");
1853 MODULE_LICENSE("GPL");