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/module.h>
  10 #include <linux/if_arp.h>
  11 #include <linux/if_bridge.h>
  12 #include <linux/if_vlan.h>
  13 #include <linux/in.h>
  14 #include <net/genetlink.h>
  15 #include <linux/ip.h>
  16 #include <linux/delay.h>
  17 #include <linux/etherdevice.h>
  18 #include <linux/kernel.h>
  19 #include <linux/kthread.h>
  20 #include <linux/mutex.h>
  21 #include <linux/rtnetlink.h>
  22 #include <linux/rcupdate.h>
  23 #include <linux/version.h>
  24 #include <linux/ethtool.h>
  25 #include <linux/random.h>
  26 #include <asm/system.h>
  27 #include <linux/netfilter_bridge.h>
  28 #include <linux/inetdevice.h>
  29 #include <linux/list.h>
  30
  31 #include "openflow-netlink.h"
  32 #include "datapath.h"
  33 #include "table.h"
  34 #include "chain.h"
  35 #include "forward.h"
  36 #include "flow.h"
  37 #include "datapath_t.h"
  38
  39 #include "compat.h"
  40
  41
  42 /* Number of milliseconds between runs of the maintenance thread. */
  43 #define MAINT_SLEEP_MSECS 1000
  44
  45 #define BRIDGE_PORT_NO_FLOOD    0x00000001
  46
  47 #define UINT32_MAX                        4294967295U
  48 #define MAX(X, Y) ((X) > (Y) ? (X) : (Y))
  49
  50 struct net_bridge_port {
  51         u16     port_no;
  52         u32 flags;
  53         struct datapath *dp;
  54         struct net_device *dev;
  55         struct list_head node; /* Element in datapath.ports. */
  56 };
  57
  58 static struct genl_family dp_genl_family;
  59 static struct genl_multicast_group mc_group;
  60
  61 int dp_dev_setup(struct net_device *dev);
  62
  63 /* It's hard to imagine wanting more than one datapath, but... */
  64 #define DP_MAX 32
  65
  66 /* datapaths.  Protected on the read side by rcu_read_lock, on the write side
  67  * by dp_mutex.
  68  *
  69  * It is safe to access the datapath and net_bridge_port structures with just
  70  * the dp_mutex, but to access the chain you need to take the rcu_read_lock
  71  * also (because dp_mutex doesn't prevent flows from being destroyed).
  72  */
  73 static struct datapath *dps[DP_MAX];
  74 static DEFINE_MUTEX(dp_mutex);
  75
  76 static int dp_maint_func(void *data);
  77 static int send_port_status(struct net_bridge_port *p, uint8_t status);
  78
  79
  80 /* nla_unreserve - reduce amount of space reserved by nla_reserve
  81  * @skb: socket buffer from which to recover room
  82  * @nla: netlink attribute to adjust
  83  * @len: amount by which to reduce attribute payload
  84  *
  85  * Reduces amount of space reserved by a call to nla_reserve.
  86  *
  87  * No other attributes may be added between calling nla_reserve and this
  88  * function, since it will create a hole in the message.
  89  */
  90 void nla_unreserve(struct sk_buff *skb, struct nlattr *nla, int len)
  91 {
  92         skb->tail -= len;
  93         skb->len  -= len;
  94
  95         nla->nla_len -= len;
  96 }
  97
  98 static void *
  99 alloc_openflow_skb(struct datapath *dp, size_t openflow_len, uint8_t type,
 100                    const struct sender *sender, struct sk_buff **pskb)
 101 {
 102         size_t genl_len;
 103         struct sk_buff *skb;
 104         struct nlattr *attr;
 105         struct ofp_header *oh;
 106
 107         genl_len = nla_total_size(sizeof(uint32_t)); /* DP_GENL_A_DP_IDX */
 108         genl_len += nla_total_size(openflow_len);    /* DP_GENL_A_OPENFLOW */
 109         skb = *pskb = genlmsg_new(genl_len, GFP_ATOMIC);
 110         if (!skb) {
 111                 if (net_ratelimit())
 112                         printk("alloc_openflow_skb: genlmsg_new failed\n");
 113                 return NULL;
 114         }
 115
 116         /* Assemble the Generic Netlink wrapper. */
 117         if (!genlmsg_put(skb,
 118                          sender ? sender->pid : 0,
 119                          sender ? sender->seq : 0,
 120                          &dp_genl_family, 0, DP_GENL_C_OPENFLOW))
 121                 BUG();
 122         if (nla_put_u32(skb, DP_GENL_A_DP_IDX, dp->dp_idx) < 0)
 123                 BUG();
 124         attr = nla_reserve(skb, DP_GENL_A_OPENFLOW, openflow_len);
 125         BUG_ON(!attr);
 126         nlmsg_end(skb, (struct nlmsghdr *) skb->data);
 127
 128         /* Fill in the header. */
 129         oh = nla_data(attr);
 130         oh->version = OFP_VERSION;
 131         oh->type = type;
 132         oh->length = htons(openflow_len);
 133         oh->xid = sender ? sender->xid : 0;
 134
 135         return oh;
 136 }
 137
 138 static void
 139 resize_openflow_skb(struct sk_buff *skb,
 140                     struct ofp_header *oh, size_t new_length)
 141 {
 142         struct nlattr *attr;
 143
 144         BUG_ON(new_length > ntohs(oh->length));
 145         attr = ((void *) oh) - NLA_HDRLEN;
 146         nla_unreserve(skb, attr, ntohs(oh->length) - new_length);
 147         oh->length = htons(new_length);
 148         nlmsg_end(skb, (struct nlmsghdr *) skb->data);
 149 }
 150
 151 static int
 152 send_openflow_skb(struct sk_buff *skb, const struct sender *sender)
 153 {
 154         int err = (sender
 155                    ? genlmsg_unicast(skb, sender->pid)
 156                    : genlmsg_multicast(skb, 0, mc_group.id, GFP_ATOMIC));
 157         if (err && net_ratelimit())
 158                 printk(KERN_WARNING "send_openflow_skb: send failed: %d\n",
 159                        err);
 160         return err;
 161 }
 162
 163 /* Generates a unique datapath id.  It incorporates the datapath index
 164  * and a hardware address, if available.  If not, it generates a random
 165  * one.
 166  */
 167 static
 168 uint64_t gen_datapath_id(uint16_t dp_idx)
 169 {
 170         uint64_t id;
 171         int i;
 172         struct net_device *dev;
 173
 174         /* The top 16 bits are used to identify the datapath.  The lower 48 bits
 175          * use an interface address.  */
 176         id = (uint64_t)dp_idx << 48;
 177         if ((dev = dev_get_by_name(&init_net, "ctl0"))
 178                         || (dev = dev_get_by_name(&init_net, "eth0"))) {
 179                 for (i=0; i<ETH_ALEN; i++) {
 180                         id |= (uint64_t)dev->dev_addr[i] << (8*(ETH_ALEN-1 - i));
 181                 }
 182                 dev_put(dev);
 183         } else {
 184                 /* Randomly choose the lower 48 bits if we cannot find an
 185                  * address and mark the most significant bit to indicate that
 186                  * this was randomly generated. */
 187                 uint8_t rand[ETH_ALEN];
 188                 get_random_bytes(rand, ETH_ALEN);
 189                 id |= (uint64_t)1 << 63;
 190                 for (i=0; i<ETH_ALEN; i++) {
 191                         id |= (uint64_t)rand[i] << (8*(ETH_ALEN-1 - i));
 192                 }
 193         }
 194
 195         return id;
 196 }
 197
 198 /* Creates a new datapath numbered 'dp_idx'.  Returns 0 for success or a
 199  * negative error code.
 200  *
 201  * Not called with any locks. */
 202 static int new_dp(int dp_idx)
 203 {
 204         struct datapath *dp;
 205         int err;
 206
 207         if (dp_idx < 0 || dp_idx >= DP_MAX)
 208                 return -EINVAL;
 209
 210         if (!try_module_get(THIS_MODULE))
 211                 return -ENODEV;
 212
 213         mutex_lock(&dp_mutex);
 214         dp = rcu_dereference(dps[dp_idx]);
 215         if (dp != NULL) {
 216                 err = -EEXIST;
 217                 goto err_unlock;
 218         }
 219
 220         err = -ENOMEM;
 221         dp = kzalloc(sizeof *dp, GFP_KERNEL);
 222         if (dp == NULL)
 223                 goto err_unlock;
 224
 225         dp->dp_idx = dp_idx;
 226         dp->id = gen_datapath_id(dp_idx);
 227         dp->chain = chain_create(dp);
 228         if (dp->chain == NULL)
 229                 goto err_free_dp;
 230         INIT_LIST_HEAD(&dp->port_list);
 231
 232 #if 0
 233         /* Setup our "of" device */
 234         dp->dev.priv = dp;
 235         rtnl_lock();
 236         err = dp_dev_setup(&dp->dev);
 237         rtnl_unlock();
 238         if (err != 0)
 239                 printk("datapath: problem setting up 'of' device\n");
 240 #endif
 241
 242         dp->config.flags = 0;
 243         dp->config.miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
 244
 245         dp->dp_task = kthread_run(dp_maint_func, dp, "dp%d", dp_idx);
 246         if (IS_ERR(dp->dp_task))
 247                 goto err_free_dp;
 248
 249         rcu_assign_pointer(dps[dp_idx], dp);
 250         mutex_unlock(&dp_mutex);
 251
 252         return 0;
 253
 254 err_free_dp:
 255         kfree(dp);
 256 err_unlock:
 257         mutex_unlock(&dp_mutex);
 258         module_put(THIS_MODULE);
 259                 return err;
 260 }
 261
 262 /* Find and return a free port number under 'dp'.  Called under dp_mutex. */
 263 static int find_portno(struct datapath *dp)
 264 {
 265         int i;
 266         for (i = 0; i < OFPP_MAX; i++)
 267                 if (dp->ports[i] == NULL)
 268                         return i;
 269         return -EXFULL;
 270 }
 271
 272 static struct net_bridge_port *new_nbp(struct datapath *dp,
 273                                                                            struct net_device *dev)
 274 {
 275         struct net_bridge_port *p;
 276         int port_no;
 277
 278         port_no = find_portno(dp);
 279         if (port_no < 0)
 280                 return ERR_PTR(port_no);
 281
 282         p = kzalloc(sizeof(*p), GFP_KERNEL);
 283         if (p == NULL)
 284                 return ERR_PTR(-ENOMEM);
 285
 286         p->dp = dp;
 287         dev_hold(dev);
 288         p->dev = dev;
 289         p->port_no = port_no;
 290
 291         return p;
 292 }
 293
 294 /* Called with dp_mutex. */
 295 int add_switch_port(struct datapath *dp, struct net_device *dev)
 296 {
 297         struct net_bridge_port *p;
 298
 299         if (dev->flags & IFF_LOOPBACK || dev->type != ARPHRD_ETHER)
 300                 return -EINVAL;
 301
 302         if (dev->br_port != NULL)
 303                 return -EBUSY;
 304
 305         p = new_nbp(dp, dev);
 306         if (IS_ERR(p))
 307                 return PTR_ERR(p);
 308
 309         dev_hold(dev);
 310         rcu_assign_pointer(dev->br_port, p);
 311         rtnl_lock();
 312         dev_set_promiscuity(dev, 1);
 313         rtnl_unlock();
 314
 315         rcu_assign_pointer(dp->ports[p->port_no], p);
 316         list_add_rcu(&p->node, &dp->port_list);
 317
 318         /* Notify the ctlpath that this port has been added */
 319         send_port_status(p, OFPPR_ADD);
 320
 321         return 0;
 322 }
 323
 324 /* Delete 'p' from switch.
 325  * Called with dp_mutex. */
 326 static int del_switch_port(struct net_bridge_port *p)
 327 {
 328         /* First drop references to device. */
 329         rtnl_lock();
 330         dev_set_promiscuity(p->dev, -1);
 331         rtnl_unlock();
 332         list_del_rcu(&p->node);
 333         rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
 334         rcu_assign_pointer(p->dev->br_port, NULL);
 335
 336         /* Then wait until no one is still using it, and destroy it. */
 337         synchronize_rcu();
 338
 339         /* Notify the ctlpath that this port no longer exists */
 340         send_port_status(p, OFPPR_DELETE);
 341
 342         dev_put(p->dev);
 343         kfree(p);
 344
 345         return 0;
 346 }
 347
 348 /* Called with dp_mutex. */
 349 static void del_dp(struct datapath *dp)
 350 {
 351         struct net_bridge_port *p, *n;
 352
 353 #if 0
 354         /* Unregister the "of" device of this dp */
 355         rtnl_lock();
 356         unregister_netdevice(&dp->dev);
 357         rtnl_unlock();
 358 #endif
 359
 360         kthread_stop(dp->dp_task);
 361
 362         /* Drop references to DP. */
 363         list_for_each_entry_safe (p, n, &dp->port_list, node)
 364                 del_switch_port(p);
 365         rcu_assign_pointer(dps[dp->dp_idx], NULL);
 366
 367         /* Wait until no longer in use, then destroy it. */
 368         synchronize_rcu();
 369         chain_destroy(dp->chain);
 370         kfree(dp);
 371         module_put(THIS_MODULE);
 372 }
 373
 374 static int dp_maint_func(void *data)
 375 {
 376         struct datapath *dp = (struct datapath *) data;
 377
 378         while (!kthread_should_stop()) {
 379 #if 1
 380                 chain_timeout(dp->chain);
 381 #else
 382                 int count = chain_timeout(dp->chain);
 383                 chain_print_stats(dp->chain);
 384                 if (count)
 385                         printk("%d flows timed out\n", count);
 386 #endif
 387                 msleep_interruptible(MAINT_SLEEP_MSECS);
 388         }
 389
 390         return 0;
 391 }
 392
 393 /*
 394  * Used as br_handle_frame_hook.  (Cannot run bridge at the same time, even on
 395  * different set of devices!)  Returns 0 if *pskb should be processed further,
 396  * 1 if *pskb is handled. */
 397 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,22)
 398 /* Called with rcu_read_lock. */
 399 static struct sk_buff *dp_frame_hook(struct net_bridge_port *p,
 400                                          struct sk_buff *skb)
 401 {
 402         struct ethhdr *eh = eth_hdr(skb);
 403         struct sk_buff *skb_local = NULL;
 404
 405
 406         if (compare_ether_addr(eh->h_dest, skb->dev->dev_addr) == 0)
 407                 return skb;
 408
 409         if (is_broadcast_ether_addr(eh->h_dest)
 410                                 || is_multicast_ether_addr(eh->h_dest)
 411                                 || is_local_ether_addr(eh->h_dest))
 412                 skb_local = skb_clone(skb, GFP_ATOMIC);
 413
 414         /* Push the Ethernet header back on. */
 415         if (skb->protocol == htons(ETH_P_8021Q))
 416                 skb_push(skb, VLAN_ETH_HLEN);
 417         else
 418                 skb_push(skb, ETH_HLEN);
 419
 420         fwd_port_input(p->dp->chain, skb, p->port_no);
 421
 422         return skb_local;
 423 }
 424 #elif LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)
 425 static int dp_frame_hook(struct net_bridge_port *p, struct sk_buff **pskb)
 426 {
 427         /* Push the Ethernet header back on. */
 428         if ((*pskb)->protocol == htons(ETH_P_8021Q))
 429                 skb_push(*pskb, VLAN_ETH_HLEN);
 430         else
 431                 skb_push(*pskb, ETH_HLEN);
 432
 433         fwd_port_input(p->dp->chain, *pskb, p->port_no);
 434         return 1;
 435 }
 436 #else
 437 /* NB: This has only been tested on 2.4.35 */
 438
 439 /* Called without any locks (?) */
 440 static void dp_frame_hook(struct sk_buff *skb)
 441 {
 442         struct net_bridge_port *p = skb->dev->br_port;
 443
 444         /* Push the Ethernet header back on. */
 445         if (skb->protocol == htons(ETH_P_8021Q))
 446                 skb_push(skb, VLAN_ETH_HLEN);
 447         else
 448                 skb_push(skb, ETH_HLEN);
 449
 450         if (p) {
 451                 rcu_read_lock();
 452                 fwd_port_input(p->dp->chain, skb, p->port_no);
 453                 rcu_read_unlock();
 454         } else
 455                 kfree_skb(skb);
 456 }
 457 #endif
 458
 459 /* Forwarding output path.
 460  * Based on net/bridge/br_forward.c. */
 461
 462 /* Don't forward packets to originating port or with flooding disabled */
 463 static inline int should_deliver(const struct net_bridge_port *p,
 464                         const struct sk_buff *skb)
 465 {
 466         if ((skb->dev == p->dev) || (p->flags & BRIDGE_PORT_NO_FLOOD)) {
 467                 return 0;
 468         }
 469
 470         return 1;
 471 }
 472
 473 static inline unsigned packet_length(const struct sk_buff *skb)
 474 {
 475         int length = skb->len - ETH_HLEN;
 476         if (skb->protocol == htons(ETH_P_8021Q))
 477                 length -= VLAN_HLEN;
 478         return length;
 479 }
 480
 481 static int
 482 flood(struct datapath *dp, struct sk_buff *skb)
 483 {
 484         struct net_bridge_port *p;
 485         int prev_port;
 486
 487         prev_port = -1;
 488         list_for_each_entry_rcu (p, &dp->port_list, node) {
 489                 if (!should_deliver(p, skb))
 490                         continue;
 491                 if (prev_port != -1) {
 492                         struct sk_buff *clone = skb_clone(skb, GFP_ATOMIC);
 493                         if (!clone) {
 494                                 kfree_skb(skb);
 495                                 return -ENOMEM;
 496                         }
 497                         dp_output_port(dp, clone, prev_port);
 498                 }
 499                 prev_port = p->port_no;
 500         }
 501         if (prev_port != -1)
 502                 dp_output_port(dp, skb, prev_port);
 503         else
 504                 kfree_skb(skb);
 505
 506         return 0;
 507 }
 508
 509 /* Marks 'skb' as having originated from 'in_port' in 'dp'.
 510    FIXME: how are devices reference counted? */
 511 int dp_set_origin(struct datapath *dp, uint16_t in_port,
 512                            struct sk_buff *skb)
 513 {
 514         if (in_port < OFPP_MAX && dp->ports[in_port]) {
 515                 skb->dev = dp->ports[in_port]->dev;
 516                 return 0;
 517         }
 518         return -ENOENT;
 519 }
 520
 521 /* Takes ownership of 'skb' and transmits it to 'out_port' on 'dp'.
 522  */
 523 int dp_output_port(struct datapath *dp, struct sk_buff *skb, int out_port)
 524 {
 525         struct net_bridge_port *p;
 526         int len = skb->len;
 527
 528         BUG_ON(!skb);
 529         if (out_port == OFPP_FLOOD)
 530                 return flood(dp, skb);
 531         else if (out_port == OFPP_CONTROLLER)
 532                 return dp_output_control(dp, skb, fwd_save_skb(skb), 0,
 533                                                   OFPR_ACTION);
 534         else if (out_port >= OFPP_MAX)
 535                 goto bad_port;
 536
 537         p = dp->ports[out_port];
 538         if (p == NULL)
 539                 goto bad_port;
 540
 541         skb->dev = p->dev;
 542         if (packet_length(skb) > skb->dev->mtu) {
 543                 printk("dropped over-mtu packet: %d > %d\n",
 544                                         packet_length(skb), skb->dev->mtu);
 545                 kfree_skb(skb);
 546                 return -E2BIG;
 547         }
 548
 549         dev_queue_xmit(skb);
 550
 551         return len;
 552
 553 bad_port:
 554         kfree_skb(skb);
 555         if (net_ratelimit())
 556                 printk("can't forward to bad port %d\n", out_port);
 557         return -ENOENT;
 558 }
 559
 560 /* Takes ownership of 'skb' and transmits it to 'dp''s control path.  If
 561  * 'buffer_id' != -1, then only the first 64 bytes of 'skb' are sent;
 562  * otherwise, all of 'skb' is sent.  'reason' indicates why 'skb' is being
 563  * sent. 'max_len' sets the maximum number of bytes that the caller
 564  * wants to be sent; a value of 0 indicates the entire packet should be
 565  * sent. */
 566 int
 567 dp_output_control(struct datapath *dp, struct sk_buff *skb,
 568                            uint32_t buffer_id, size_t max_len, int reason)
 569 {
 570         /* FIXME?  Can we avoid creating a new skbuff in the case where we
 571          * forward the whole packet? */
 572         struct sk_buff *f_skb;
 573         struct ofp_packet_in *opi;
 574         size_t fwd_len, opi_len;
 575         int err;
 576
 577         fwd_len = skb->len;
 578         if ((buffer_id != (uint32_t) -1) && max_len)
 579                 fwd_len = min(fwd_len, max_len);
 580
 581         opi_len = offsetof(struct ofp_packet_in, data) + fwd_len;
 582         opi = alloc_openflow_skb(dp, opi_len, OFPT_PACKET_IN, NULL, &f_skb);
 583         opi->buffer_id      = htonl(buffer_id);
 584         opi->total_len      = htons(skb->len);
 585         opi->in_port        = htons(skb->dev->br_port->port_no);
 586         opi->reason         = reason;
 587         opi->pad            = 0;
 588         memcpy(opi->data, skb_mac_header(skb), fwd_len);
 589         err = send_openflow_skb(f_skb, NULL);
 590
 591         kfree_skb(skb);
 592
 593         return err;
 594 }
 595
 596 static void fill_port_desc(struct net_bridge_port *p, struct ofp_phy_port *desc)
 597 {
 598         desc->port_no = htons(p->port_no);
 599         strncpy(desc->name, p->dev->name, OFP_MAX_PORT_NAME_LEN);
 600         desc->name[OFP_MAX_PORT_NAME_LEN-1] = '\0';
 601         memcpy(desc->hw_addr, p->dev->dev_addr, ETH_ALEN);
 602         desc->flags = htonl(p->flags);
 603         desc->features = 0;
 604         desc->speed = 0;
 605
 606 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,4,24)
 607         if (p->dev->ethtool_ops && p->dev->ethtool_ops->get_settings) {
 608                 struct ethtool_cmd ecmd = { .cmd = ETHTOOL_GSET };
 609
 610                 if (!p->dev->ethtool_ops->get_settings(p->dev, &ecmd)) {
 611                         if (ecmd.supported & SUPPORTED_10baseT_Half)
 612                                 desc->features |= OFPPF_10MB_HD;
 613                         if (ecmd.supported & SUPPORTED_10baseT_Full)
 614                                 desc->features |= OFPPF_10MB_FD;
 615                         if (ecmd.supported & SUPPORTED_100baseT_Half)
 616                                 desc->features |= OFPPF_100MB_HD;
 617                         if (ecmd.supported & SUPPORTED_100baseT_Full)
 618                                 desc->features |= OFPPF_100MB_FD;
 619                         if (ecmd.supported & SUPPORTED_1000baseT_Half)
 620                                 desc->features |= OFPPF_1GB_HD;
 621                         if (ecmd.supported & SUPPORTED_1000baseT_Full)
 622                                 desc->features |= OFPPF_1GB_FD;
 623                         /* 10Gbps half-duplex doesn't exist... */
 624                         if (ecmd.supported & SUPPORTED_10000baseT_Full)
 625                                 desc->features |= OFPPF_10GB_FD;
 626
 627                         desc->features = htonl(desc->features);
 628                         desc->speed = htonl(ecmd.speed);
 629                 }
 630         }
 631 #endif
 632 }
 633
 634 static int
 635 fill_features_reply(struct datapath *dp, struct ofp_switch_features *ofr)
 636 {
 637         struct net_bridge_port *p;
 638         int port_count = 0;
 639
 640         ofr->datapath_id    = cpu_to_be64(dp->id);
 641
 642         ofr->n_exact        = htonl(2 * TABLE_HASH_MAX_FLOWS);
 643         ofr->n_mac_only     = htonl(TABLE_MAC_MAX_FLOWS);
 644         ofr->n_compression  = 0;                                           /* Not supported */
 645         ofr->n_general      = htonl(TABLE_LINEAR_MAX_FLOWS);
 646         ofr->buffer_mb      = htonl(UINT32_MAX);
 647         ofr->n_buffers      = htonl(N_PKT_BUFFERS);
 648         ofr->capabilities   = htonl(OFP_SUPPORTED_CAPABILITIES);
 649         ofr->actions        = htonl(OFP_SUPPORTED_ACTIONS);
 650
 651         list_for_each_entry_rcu (p, &dp->port_list, node) {
 652                 fill_port_desc(p, &ofr->ports[port_count]);
 653                 port_count++;
 654         }
 655
 656         return port_count;
 657 }
 658
 659 int
 660 dp_send_features_reply(struct datapath *dp, const struct sender *sender)
 661 {
 662         struct sk_buff *skb;
 663         struct ofp_switch_features *ofr;
 664         size_t ofr_len, port_max_len;
 665         int port_count;
 666
 667         /* Overallocate. */
 668         port_max_len = sizeof(struct ofp_phy_port) * OFPP_MAX;
 669         ofr = alloc_openflow_skb(dp, sizeof(*ofr) + port_max_len,
 670                                  OFPT_FEATURES_REPLY, sender, &skb);
 671         if (!ofr)
 672                 return -ENOMEM;
 673
 674         /* Fill. */
 675         port_count = fill_features_reply(dp, ofr);
 676
 677         /* Shrink to fit. */
 678         ofr_len = sizeof(*ofr) + (sizeof(struct ofp_phy_port) * port_count);
 679         resize_openflow_skb(skb, &ofr->header, ofr_len);
 680         return send_openflow_skb(skb, sender);
 681 }
 682
 683 int
 684 dp_send_config_reply(struct datapath *dp, const struct sender *sender)
 685 {
 686         struct sk_buff *skb;
 687         struct ofp_switch_config *osc;
 688
 689         osc = alloc_openflow_skb(dp, sizeof *osc, OFPT_PORT_STATUS, sender,
 690                                  &skb);
 691         if (!osc)
 692                 return -ENOMEM;
 693         memcpy(((char *)osc) + sizeof osc->header,
 694                ((char *)&dp->config) + sizeof dp->config.header,
 695                sizeof dp->config - sizeof dp->config.header);
 696         return send_openflow_skb(skb, sender);
 697 }
 698
 699 int
 700 dp_update_port_flags(struct datapath *dp, const struct ofp_phy_port *opp)
 701 {
 702         struct net_bridge_port *p;
 703
 704         p = dp->ports[htons(opp->port_no)];
 705
 706         /* Make sure the port id hasn't changed since this was sent */
 707         if (!p || memcmp(opp->hw_addr, p->dev->dev_addr, ETH_ALEN) != 0)
 708                 return -1;
 709
 710         p->flags = htonl(opp->flags);
 711
 712         return 0;
 713 }
 714
 715
 716 static int
 717 send_port_status(struct net_bridge_port *p, uint8_t status)
 718 {
 719         struct sk_buff *skb;
 720         struct ofp_port_status *ops;
 721
 722         ops = alloc_openflow_skb(p->dp, sizeof *ops, OFPT_PORT_STATUS, NULL,
 723                                  &skb);
 724         if (!ops)
 725                 return -ENOMEM;
 726         ops->reason = status;
 727         fill_port_desc(p, &ops->desc);
 728
 729         return send_openflow_skb(skb, NULL);
 730 }
 731
 732 int
 733 dp_send_flow_expired(struct datapath *dp, struct sw_flow *flow)
 734 {
 735         struct sk_buff *skb;
 736         struct ofp_flow_expired *ofe;
 737         unsigned long duration_j;
 738
 739         ofe = alloc_openflow_skb(dp, sizeof *ofe, OFPT_FLOW_EXPIRED, 0, &skb);
 740         if (!ofe)
 741                 return -ENOMEM;
 742
 743         flow_fill_match(&ofe->match, &flow->key);
 744         duration_j = (flow->timeout - HZ * flow->max_idle) - flow->init_time;
 745         ofe->duration   = htonl(duration_j / HZ);
 746         ofe->packet_count   = cpu_to_be64(flow->packet_count);
 747         ofe->byte_count     = cpu_to_be64(flow->byte_count);
 748         return send_openflow_skb(skb, NULL);
 749 }
 750
 751 static void
 752 fill_flow_stats(struct ofp_flow_stats *ofs, struct sw_flow *flow,
 753                 int table_idx)
 754 {
 755         int duration;
 756
 757         ofs->match.wildcards = htons(flow->key.wildcards);
 758         ofs->match.in_port   = flow->key.in_port;
 759         memcpy(ofs->match.dl_src, flow->key.dl_src, ETH_ALEN);
 760         memcpy(ofs->match.dl_dst, flow->key.dl_dst, ETH_ALEN);
 761         ofs->match.dl_vlan   = flow->key.dl_vlan;
 762         ofs->match.dl_type   = flow->key.dl_type;
 763         ofs->match.nw_src    = flow->key.nw_src;
 764         ofs->match.nw_dst    = flow->key.nw_dst;
 765         ofs->match.nw_proto  = flow->key.nw_proto;
 766         memset(ofs->match.pad, 0, sizeof ofs->match.pad);
 767         ofs->match.tp_src    = flow->key.tp_src;
 768         ofs->match.tp_dst    = flow->key.tp_dst;
 769         duration = (jiffies - flow->init_time) / HZ;
 770         ofs->duration        = htons(min(65535, duration));
 771         ofs->table_id        = htons(table_idx);
 772         ofs->packet_count    = cpu_to_be64(flow->packet_count);
 773         ofs->byte_count      = cpu_to_be64(flow->byte_count);
 774 }
 775
 776 int
 777 dp_send_flow_stats(struct datapath *dp, const struct sender *sender,
 778                    const struct ofp_match *match)
 779 {
 780         struct sk_buff *skb;
 781         struct ofp_flow_stat_reply *fsr;
 782         size_t header_size, fudge, flow_size;
 783         struct sw_flow_key match_key;
 784         int table_idx, n_flows, max_flows;
 785
 786         header_size = offsetof(struct ofp_flow_stat_reply, flows);
 787         fudge = 128;
 788         flow_size = sizeof fsr->flows[0];
 789         max_flows = (NLMSG_GOODSIZE - header_size - fudge) / flow_size;
 790         fsr = alloc_openflow_skb(dp, header_size + max_flows * flow_size,
 791                                  OFPT_FLOW_STAT_REPLY, sender, &skb);
 792         if (!fsr)
 793                 return -ENOMEM;
 794
 795         n_flows = 0;
 796         flow_extract_match(&match_key, match);
 797         for (table_idx = 0; table_idx < dp->chain->n_tables; table_idx++) {
 798                 struct sw_table *table = dp->chain->tables[table_idx];
 799                 struct swt_iterator iter;
 800
 801                 if (n_flows >= max_flows) {
 802                         break;
 803                 }
 804
 805                 if (!table->iterator(table, &iter)) {
 806                         if (net_ratelimit())
 807                                 printk("iterator failed for table %d\n",
 808                                        table_idx);
 809                         continue;
 810                 }
 811
 812                 for (; iter.flow; table->iterator_next(&iter)) {
 813                         if (flow_matches(&match_key, &iter.flow->key)) {
 814                                 fill_flow_stats(&fsr->flows[n_flows],
 815                                                 iter.flow, table_idx);
 816                                 if (++n_flows >= max_flows) {
 817                                         break;
 818                                 }
 819                         }
 820                 }
 821                 table->iterator_destroy(&iter);
 822         }
 823         resize_openflow_skb(skb, &fsr->header,
 824                             header_size + flow_size * n_flows);
 825         return send_openflow_skb(skb, sender);
 826 }
 827
 828 static int
 829 fill_port_stat_reply(struct datapath *dp, struct ofp_port_stat_reply *psr)
 830 {
 831         struct net_bridge_port *p;
 832         int port_count = 0;
 833
 834         list_for_each_entry_rcu (p, &dp->port_list, node) {
 835                 struct ofp_port_stats *ps = &psr->ports[port_count++];
 836                 struct net_device_stats *stats = p->dev->get_stats(p->dev);
 837                 ps->port_no = htons(p->port_no);
 838                 memset(ps->pad, 0, sizeof ps->pad);
 839                 ps->rx_count = cpu_to_be64(stats->rx_packets);
 840                 ps->tx_count = cpu_to_be64(stats->tx_packets);
 841                 ps->drop_count = cpu_to_be64(stats->rx_dropped
 842                                              + stats->tx_dropped);
 843         }
 844
 845         return port_count;
 846 }
 847
 848 int
 849 dp_send_port_stats(struct datapath *dp, const struct sender *sender)
 850 {
 851         struct sk_buff *skb;
 852         struct ofp_port_stat_reply *psr;
 853         size_t psr_len, port_max_len;
 854         int port_count;
 855
 856         /* Overallocate. */
 857         port_max_len = sizeof(struct ofp_port_stats) * OFPP_MAX;
 858         psr = alloc_openflow_skb(dp, sizeof *psr + port_max_len,
 859                                  OFPT_PORT_STAT_REPLY, sender, &skb);
 860         if (!psr)
 861                 return -ENOMEM;
 862
 863         /* Fill. */
 864         port_count = fill_port_stat_reply(dp, psr);
 865
 866         /* Shrink to fit. */
 867         psr_len = sizeof *psr + sizeof(struct ofp_port_stats) * port_count;
 868         resize_openflow_skb(skb, &psr->header, psr_len);
 869         return send_openflow_skb(skb, sender);
 870 }
 871
 872 int
 873 dp_send_table_stats(struct datapath *dp, const struct sender *sender)
 874 {
 875         struct sk_buff *skb;
 876         struct ofp_table_stat_reply *tsr;
 877         int i, n_tables;
 878
 879         n_tables = dp->chain->n_tables;
 880         tsr = alloc_openflow_skb(dp, (offsetof(struct ofp_table_stat_reply,
 881                                                tables)
 882                                       + sizeof tsr->tables[0] * n_tables),
 883                                  OFPT_TABLE_STAT_REPLY, sender, &skb);
 884         if (!tsr)
 885                 return -ENOMEM;
 886         for (i = 0; i < n_tables; i++) {
 887                 struct ofp_table_stats *ots = &tsr->tables[i];
 888                 struct sw_table_stats stats;
 889                 dp->chain->tables[i]->stats(dp->chain->tables[i], &stats);
 890                 strncpy(ots->name, stats.name, sizeof ots->name);
 891                 ots->table_id = htons(i);
 892                 ots->pad[0] = ots->pad[1] = 0;
 893                 ots->max_entries = htonl(stats.max_flows);
 894                 ots->active_count = htonl(stats.n_flows);
 895                 ots->matched_count = cpu_to_be64(0); /* FIXME */
 896         }
 897         return send_openflow_skb(skb, sender);
 898 }
 899
 900 /* Generic Netlink interface.
 901  *
 902  * See netlink(7) for an introduction to netlink.  See
 903  * http://linux-net.osdl.org/index.php/Netlink for more information and
 904  * pointers on how to work with netlink and Generic Netlink in the kernel and
 905  * in userspace. */
 906
 907 static struct genl_family dp_genl_family = {
 908         .id = GENL_ID_GENERATE,
 909         .hdrsize = 0,
 910         .name = DP_GENL_FAMILY_NAME,
 911         .version = 1,
 912         .maxattr = DP_GENL_A_MAX,
 913 };
 914
 915 /* Attribute policy: what each attribute may contain.  */
 916 static struct nla_policy dp_genl_policy[DP_GENL_A_MAX + 1] = {
 917         [DP_GENL_A_DP_IDX] = { .type = NLA_U32 },
 918         [DP_GENL_A_MC_GROUP] = { .type = NLA_U32 },
 919         [DP_GENL_A_PORTNAME] = { .type = NLA_STRING }
 920 };
 921
 922 static int dp_genl_add(struct sk_buff *skb, struct genl_info *info)
 923 {
 924         if (!info->attrs[DP_GENL_A_DP_IDX])
 925                 return -EINVAL;
 926
 927         return new_dp(nla_get_u32(info->attrs[DP_GENL_A_DP_IDX]));
 928 }
 929
 930 static struct genl_ops dp_genl_ops_add_dp = {
 931         .cmd = DP_GENL_C_ADD_DP,
 932         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 933         .policy = dp_genl_policy,
 934         .doit = dp_genl_add,
 935         .dumpit = NULL,
 936 };
 937
 938 struct datapath *dp_get(int dp_idx)
 939 {
 940         if (dp_idx < 0 || dp_idx > DP_MAX)
 941                 return NULL;
 942         return rcu_dereference(dps[dp_idx]);
 943 }
 944
 945 static int dp_genl_del(struct sk_buff *skb, struct genl_info *info)
 946 {
 947         struct datapath *dp;
 948         int err;
 949
 950         if (!info->attrs[DP_GENL_A_DP_IDX])
 951                 return -EINVAL;
 952
 953         mutex_lock(&dp_mutex);
 954         dp = dp_get(nla_get_u32((info->attrs[DP_GENL_A_DP_IDX])));
 955         if (!dp)
 956                 err = -ENOENT;
 957         else {
 958                 del_dp(dp);
 959                 err = 0;
 960         }
 961         mutex_unlock(&dp_mutex);
 962         return err;
 963 }
 964
 965 static struct genl_ops dp_genl_ops_del_dp = {
 966         .cmd = DP_GENL_C_DEL_DP,
 967         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 968         .policy = dp_genl_policy,
 969         .doit = dp_genl_del,
 970         .dumpit = NULL,
 971 };
 972
 973 /* Queries a datapath for related information.  Currently the only relevant
 974  * information is the datapath's multicast group ID.  Really we want one
 975  * multicast group per datapath, but because of locking issues[*] we can't
 976  * easily get one.  Thus, every datapath will currently return the same
 977  * global multicast group ID, but in the future it would be nice to fix that.
 978  *
 979  * [*] dp_genl_add, to add a new datapath, is called under the genl_lock
 980  *       mutex, and genl_register_mc_group, called to acquire a new multicast
 981  *       group ID, also acquires genl_lock, thus deadlock.
 982  */
 983 static int dp_genl_query(struct sk_buff *skb, struct genl_info *info)
 984 {
 985         struct datapath *dp;
 986         struct sk_buff *ans_skb = NULL;
 987         int dp_idx;
 988         int err = -ENOMEM;
 989
 990         if (!info->attrs[DP_GENL_A_DP_IDX])
 991                 return -EINVAL;
 992
 993         rcu_read_lock();
 994         dp_idx = nla_get_u32((info->attrs[DP_GENL_A_DP_IDX]));
 995         dp = dp_get(dp_idx);
 996         if (!dp)
 997                 err = -ENOENT;
 998         else {
 999                 void *data;
1000                 ans_skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1001                 if (!ans_skb) {
1002                         err = -ENOMEM;
1003                         goto err;
1004                 }
1005                 data = genlmsg_put_reply(ans_skb, info, &dp_genl_family,
1006                                          0, DP_GENL_C_QUERY_DP);
1007                 if (data == NULL) {
1008                         err = -ENOMEM;
1009                         goto err;
1010                 }
1011                 NLA_PUT_U32(ans_skb, DP_GENL_A_DP_IDX, dp_idx);
1012                 NLA_PUT_U32(ans_skb, DP_GENL_A_MC_GROUP, mc_group.id);
1013
1014                 genlmsg_end(ans_skb, data);
1015                 err = genlmsg_reply(ans_skb, info);
1016                 if (!err)
1017                         ans_skb = NULL;
1018         }
1019 err:
1020 nla_put_failure:
1021         if (ans_skb)
1022                 kfree_skb(ans_skb);
1023         rcu_read_unlock();
1024         return err;
1025 }
1026
1027 static struct genl_ops dp_genl_ops_query_dp = {
1028         .cmd = DP_GENL_C_QUERY_DP,
1029         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1030         .policy = dp_genl_policy,
1031         .doit = dp_genl_query,
1032         .dumpit = NULL,
1033 };
1034
1035 static int dp_genl_add_del_port(struct sk_buff *skb, struct genl_info *info)
1036 {
1037         struct datapath *dp;
1038         struct net_device *port;
1039         int err;
1040
1041         if (!info->attrs[DP_GENL_A_DP_IDX] || !info->attrs[DP_GENL_A_PORTNAME])
1042                 return -EINVAL;
1043
1044         /* Get datapath. */
1045         mutex_lock(&dp_mutex);
1046         dp = dp_get(nla_get_u32(info->attrs[DP_GENL_A_DP_IDX]));
1047         if (!dp) {
1048                 err = -ENOENT;
1049                 goto out;
1050         }
1051
1052         /* Get interface to add/remove. */
1053         port = dev_get_by_name(&init_net,
1054                         nla_data(info->attrs[DP_GENL_A_PORTNAME]));
1055         if (!port) {
1056                 err = -ENOENT;
1057                 goto out;
1058         }
1059
1060         /* Execute operation. */
1061         if (info->genlhdr->cmd == DP_GENL_C_ADD_PORT)
1062                 err = add_switch_port(dp, port);
1063         else {
1064                 if (port->br_port == NULL || port->br_port->dp != dp) {
1065                         err = -ENOENT;
1066                         goto out_put;
1067                 }
1068                 err = del_switch_port(port->br_port);
1069         }
1070
1071 out_put:
1072         dev_put(port);
1073 out:
1074         mutex_unlock(&dp_mutex);
1075         return err;
1076 }
1077
1078 static struct genl_ops dp_genl_ops_add_port = {
1079         .cmd = DP_GENL_C_ADD_PORT,
1080         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1081         .policy = dp_genl_policy,
1082         .doit = dp_genl_add_del_port,
1083         .dumpit = NULL,
1084 };
1085
1086 static struct genl_ops dp_genl_ops_del_port = {
1087         .cmd = DP_GENL_C_DEL_PORT,
1088         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1089         .policy = dp_genl_policy,
1090         .doit = dp_genl_add_del_port,
1091         .dumpit = NULL,
1092 };
1093
1094 static int dp_genl_openflow(struct sk_buff *skb, struct genl_info *info)
1095 {
1096         struct nlattr *va = info->attrs[DP_GENL_A_OPENFLOW];
1097         struct datapath *dp;
1098         struct ofp_header *oh;
1099         struct sender sender;
1100         int err;
1101
1102         if (!info->attrs[DP_GENL_A_DP_IDX] || !va)
1103                 return -EINVAL;
1104
1105         rcu_read_lock();
1106         dp = dp_get(nla_get_u32(info->attrs[DP_GENL_A_DP_IDX]));
1107         if (!dp) {
1108                 err = -ENOENT;
1109                 goto out;
1110         }
1111
1112         va = info->attrs[DP_GENL_A_OPENFLOW];
1113         if (nla_len(va) < sizeof(struct ofp_header)) {
1114                 err = -EINVAL;
1115                 goto out;
1116         }
1117         oh = nla_data(va);
1118
1119         sender.xid = oh->xid;
1120         sender.pid = info->snd_pid;
1121         sender.seq = info->snd_seq;
1122         err = fwd_control_input(dp->chain, &sender, nla_data(va), nla_len(va));
1123
1124 out:
1125         rcu_read_unlock();
1126         return err;
1127 }
1128
1129 static struct nla_policy dp_genl_openflow_policy[DP_GENL_A_MAX + 1] = {
1130         [DP_GENL_A_DP_IDX] = { .type = NLA_U32 },
1131 };
1132
1133 static struct genl_ops dp_genl_ops_openflow = {
1134         .cmd = DP_GENL_C_OPENFLOW,
1135         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1136         .policy = dp_genl_openflow_policy,
1137         .doit = dp_genl_openflow,
1138         .dumpit = NULL,
1139 };
1140
1141 static struct nla_policy dp_genl_benchmark_policy[DP_GENL_A_MAX + 1] = {
1142         [DP_GENL_A_DP_IDX] = { .type = NLA_U32 },
1143         [DP_GENL_A_NPACKETS] = { .type = NLA_U32 },
1144         [DP_GENL_A_PSIZE] = { .type = NLA_U32 },
1145 };
1146
1147 static struct genl_ops dp_genl_ops_benchmark_nl = {
1148         .cmd = DP_GENL_C_BENCHMARK_NL,
1149         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1150         .policy = dp_genl_benchmark_policy,
1151         .doit = dp_genl_benchmark_nl,
1152         .dumpit = NULL,
1153 };
1154
1155 static struct genl_ops *dp_genl_all_ops[] = {
1156         /* Keep this operation first.  Generic Netlink dispatching
1157          * looks up operations with linear search, so we want it at the
1158          * front. */
1159         &dp_genl_ops_openflow,
1160
1161         &dp_genl_ops_add_dp,
1162         &dp_genl_ops_del_dp,
1163         &dp_genl_ops_query_dp,
1164         &dp_genl_ops_add_port,
1165         &dp_genl_ops_del_port,
1166         &dp_genl_ops_benchmark_nl,
1167 };
1168
1169 static int dp_init_netlink(void)
1170 {
1171         int err;
1172         int i;
1173
1174         err = genl_register_family(&dp_genl_family);
1175         if (err)
1176                 return err;
1177
1178         for (i = 0; i < ARRAY_SIZE(dp_genl_all_ops); i++) {
1179                 err = genl_register_ops(&dp_genl_family, dp_genl_all_ops[i]);
1180                 if (err)
1181                         goto err_unregister;
1182         }
1183
1184         strcpy(mc_group.name, "openflow");
1185         err = genl_register_mc_group(&dp_genl_family, &mc_group);
1186         if (err < 0)
1187                 goto err_unregister;
1188
1189         return 0;
1190
1191 err_unregister:
1192         genl_unregister_family(&dp_genl_family);
1193                 return err;
1194 }
1195
1196 static void dp_uninit_netlink(void)
1197 {
1198         genl_unregister_family(&dp_genl_family);
1199 }
1200
1201 #define DRV_NAME                "openflow"
1202 #define DRV_VERSION      VERSION
1203 #define DRV_DESCRIPTION "OpenFlow switching datapath implementation"
1204 #define DRV_COPYRIGHT   "Copyright (c) 2007, 2008 The Board of Trustees of The Leland Stanford Junior University"
1205
1206
1207 static int __init dp_init(void)
1208 {
1209         int err;
1210
1211         printk(KERN_INFO DRV_NAME ": " DRV_DESCRIPTION "\n");
1212         printk(KERN_INFO DRV_NAME ": " VERSION" built on "__DATE__" "__TIME__"\n");
1213         printk(KERN_INFO DRV_NAME ": " DRV_COPYRIGHT "\n");
1214
1215         err = flow_init();
1216         if (err)
1217                 goto error;
1218
1219         err = dp_init_netlink();
1220         if (err)
1221                 goto error_flow_exit;
1222
1223         /* Hook into callback used by the bridge to intercept packets.
1224          * Parasites we are. */
1225         if (br_handle_frame_hook)
1226                 printk("openflow: hijacking bridge hook\n");
1227         br_handle_frame_hook = dp_frame_hook;
1228
1229         return 0;
1230
1231 error_flow_exit:
1232         flow_exit();
1233 error:
1234         printk(KERN_EMERG "openflow: failed to install!");
1235         return err;
1236 }
1237
1238 static void dp_cleanup(void)
1239 {
1240         fwd_exit();
1241         dp_uninit_netlink();
1242         flow_exit();
1243         br_handle_frame_hook = NULL;
1244 }
1245
1246 module_init(dp_init);
1247 module_exit(dp_cleanup);
1248
1249 MODULE_DESCRIPTION(DRV_DESCRIPTION);
1250 MODULE_AUTHOR(DRV_COPYRIGHT);
1251 MODULE_LICENSE("GPL");