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/etherdevice.h>
  17 #include <linux/kernel.h>
  18 #include <linux/mutex.h>
  19 #include <linux/rtnetlink.h>
  20 #include <linux/timer.h>
  21 #include <linux/rcupdate.h>
  22 #include <linux/version.h>
  23 #include <linux/ethtool.h>
  24 #include <linux/random.h>
  25 #include <asm/system.h>
  26 #include <linux/netfilter_bridge.h>
  27 #include <linux/inetdevice.h>
  28 #include <linux/list.h>
  29
  30 #include "openflow-netlink.h"
  31 #include "datapath.h"
  32 #include "table.h"
  33 #include "chain.h"
  34 #include "forward.h"
  35 #include "flow.h"
  36 #include "datapath_t.h"
  37
  38 #include "compat.h"
  39
  40
  41 /* Number of seconds between runs of the flow expiration code. */
  42 #define EXPIRE_SECS 1
  43
  44 #define BRIDGE_PORT_NO_FLOOD    0x00000001
  45
  46 #define UINT32_MAX                        4294967295U
  47
  48 struct net_bridge_port {
  49         u16     port_no;
  50         u32 flags;
  51         struct datapath *dp;
  52         struct net_device *dev;
  53         struct list_head node; /* Element in datapath.ports. */
  54 };
  55
  56 static struct genl_family dp_genl_family;
  57 static struct genl_multicast_group mc_group;
  58
  59 int dp_dev_setup(struct net_device *dev);
  60
  61 /* It's hard to imagine wanting more than one datapath, but... */
  62 #define DP_MAX 32
  63
  64 /* datapaths.  Protected on the read side by rcu_read_lock, on the write side
  65  * by dp_mutex.
  66  *
  67  * It is safe to access the datapath and net_bridge_port structures with just
  68  * the dp_mutex, but to access the chain you need to take the rcu_read_lock
  69  * also (because dp_mutex doesn't prevent flows from being destroyed).
  70  */
  71 static struct datapath *dps[DP_MAX];
  72 static DEFINE_MUTEX(dp_mutex);
  73
  74 static void dp_timer_handler(unsigned long arg);
  75 static int send_port_status(struct net_bridge_port *p, uint8_t status);
  76
  77
  78 /* nla_unreserve - reduce amount of space reserved by nla_reserve
  79  * @skb: socket buffer from which to recover room
  80  * @nla: netlink attribute to adjust
  81  * @len: amount by which to reduce attribute payload
  82  *
  83  * Reduces amount of space reserved by a call to nla_reserve.
  84  *
  85  * No other attributes may be added between calling nla_reserve and this
  86  * function, since it will create a hole in the message.
  87  */
  88 void nla_unreserve(struct sk_buff *skb, struct nlattr *nla, int len)
  89 {
  90         skb->tail -= len;
  91         skb->len  -= len;
  92
  93         nla->nla_len -= len;
  94 }
  95
  96 /* Generates a unique datapath id.  It incorporates the datapath index
  97  * and a hardware address, if available.  If not, it generates a random
  98  * one.
  99  */
 100 static
 101 uint64_t gen_datapath_id(uint16_t dp_idx)
 102 {
 103         uint64_t id;
 104         int i;
 105         struct net_device *dev;
 106
 107         /* The top 16 bits are used to identify the datapath.  The lower 48 bits
 108          * use an interface address.  */
 109         id = (uint64_t)dp_idx << 48;
 110         if ((dev = dev_get_by_name(&init_net, "ctl0"))
 111                         || (dev = dev_get_by_name(&init_net, "eth0"))) {
 112                 for (i=0; i<ETH_ALEN; i++) {
 113                         id |= (uint64_t)dev->dev_addr[i] << (8*(ETH_ALEN-1 - i));
 114                 }
 115                 dev_put(dev);
 116         } else {
 117                 /* Randomly choose the lower 48 bits if we cannot find an
 118                  * address and mark the most significant bit to indicate that
 119                  * this was randomly generated. */
 120                 uint8_t rand[ETH_ALEN];
 121                 get_random_bytes(rand, ETH_ALEN);
 122                 id |= (uint64_t)1 << 63;
 123                 for (i=0; i<ETH_ALEN; i++) {
 124                         id |= (uint64_t)rand[i] << (8*(ETH_ALEN-1 - i));
 125                 }
 126         }
 127
 128         return id;
 129 }
 130
 131 /* Creates a new datapath numbered 'dp_idx'.  Returns 0 for success or a
 132  * negative error code.
 133  *
 134  * Not called with any locks. */
 135 static int new_dp(int dp_idx)
 136 {
 137         struct datapath *dp;
 138         int err;
 139
 140         if (dp_idx < 0 || dp_idx >= DP_MAX)
 141                 return -EINVAL;
 142
 143         if (!try_module_get(THIS_MODULE))
 144                 return -ENODEV;
 145
 146         mutex_lock(&dp_mutex);
 147         dp = rcu_dereference(dps[dp_idx]);
 148         if (dp != NULL) {
 149                 err = -EEXIST;
 150                 goto err_unlock;
 151         }
 152
 153         err = -ENOMEM;
 154         dp = kzalloc(sizeof *dp, GFP_KERNEL);
 155         if (dp == NULL)
 156                 goto err_unlock;
 157
 158         dp->dp_idx = dp_idx;
 159         dp->id = gen_datapath_id(dp_idx);
 160         dp->chain = chain_create(dp);
 161         if (dp->chain == NULL)
 162                 goto err_free_dp;
 163         INIT_LIST_HEAD(&dp->port_list);
 164
 165 #if 0
 166         /* Setup our "of" device */
 167         dp->dev.priv = dp;
 168         rtnl_lock();
 169         err = dp_dev_setup(&dp->dev);
 170         rtnl_unlock();
 171         if (err != 0)
 172                 printk("datapath: problem setting up 'of' device\n");
 173 #endif
 174
 175         dp->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
 176
 177         setup_timer(&dp->timer, dp_timer_handler, (unsigned long) dp);
 178         mod_timer(&dp->timer, round_jiffies(jiffies + (EXPIRE_SECS * HZ)));
 179
 180         rcu_assign_pointer(dps[dp_idx], dp);
 181         mutex_unlock(&dp_mutex);
 182
 183         return 0;
 184
 185 err_free_dp:
 186         kfree(dp);
 187 err_unlock:
 188         mutex_unlock(&dp_mutex);
 189         module_put(THIS_MODULE);
 190                 return err;
 191 }
 192
 193 /* Find and return a free port number under 'dp'.  Called under dp_mutex. */
 194 static int find_portno(struct datapath *dp)
 195 {
 196         int i;
 197         for (i = 0; i < OFPP_MAX; i++)
 198                 if (dp->ports[i] == NULL)
 199                         return i;
 200         return -EXFULL;
 201 }
 202
 203 static struct net_bridge_port *new_nbp(struct datapath *dp,
 204                                                                            struct net_device *dev)
 205 {
 206         struct net_bridge_port *p;
 207         int port_no;
 208
 209         port_no = find_portno(dp);
 210         if (port_no < 0)
 211                 return ERR_PTR(port_no);
 212
 213         p = kzalloc(sizeof(*p), GFP_KERNEL);
 214         if (p == NULL)
 215                 return ERR_PTR(-ENOMEM);
 216
 217         p->dp = dp;
 218         dev_hold(dev);
 219         p->dev = dev;
 220         p->port_no = port_no;
 221
 222         return p;
 223 }
 224
 225 /* Called with dp_mutex. */
 226 int add_switch_port(struct datapath *dp, struct net_device *dev)
 227 {
 228         struct net_bridge_port *p;
 229
 230         if (dev->flags & IFF_LOOPBACK || dev->type != ARPHRD_ETHER)
 231                 return -EINVAL;
 232
 233         if (dev->br_port != NULL)
 234                 return -EBUSY;
 235
 236         p = new_nbp(dp, dev);
 237         if (IS_ERR(p))
 238                 return PTR_ERR(p);
 239
 240         dev_hold(dev);
 241         rcu_assign_pointer(dev->br_port, p);
 242         rtnl_lock();
 243         dev_set_promiscuity(dev, 1);
 244         rtnl_unlock();
 245
 246         rcu_assign_pointer(dp->ports[p->port_no], p);
 247         list_add_rcu(&p->node, &dp->port_list);
 248
 249         /* Notify the ctlpath that this port has been added */
 250         send_port_status(p, OFPPR_ADD);
 251
 252         return 0;
 253 }
 254
 255 /* Delete 'p' from switch.
 256  * Called with dp_mutex. */
 257 static int del_switch_port(struct net_bridge_port *p)
 258 {
 259         /* First drop references to device. */
 260         rtnl_lock();
 261         dev_set_promiscuity(p->dev, -1);
 262         rtnl_unlock();
 263         list_del_rcu(&p->node);
 264         rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
 265         rcu_assign_pointer(p->dev->br_port, NULL);
 266
 267         /* Then wait until no one is still using it, and destroy it. */
 268         synchronize_rcu();
 269
 270         /* Notify the ctlpath that this port no longer exists */
 271         send_port_status(p, OFPPR_DELETE);
 272
 273         dev_put(p->dev);
 274         kfree(p);
 275
 276         return 0;
 277 }
 278
 279 /* Called with dp_mutex. */
 280 static void del_dp(struct datapath *dp)
 281 {
 282         struct net_bridge_port *p, *n;
 283
 284 #if 0
 285         /* Unregister the "of" device of this dp */
 286         rtnl_lock();
 287         unregister_netdevice(&dp->dev);
 288         rtnl_unlock();
 289 #endif
 290
 291         /* Drop references to DP. */
 292         list_for_each_entry_safe (p, n, &dp->port_list, node)
 293                 del_switch_port(p);
 294         del_timer_sync(&dp->timer);
 295         rcu_assign_pointer(dps[dp->dp_idx], NULL);
 296
 297         /* Wait until no longer in use, then destroy it. */
 298         synchronize_rcu();
 299         chain_destroy(dp->chain);
 300         kfree(dp);
 301         module_put(THIS_MODULE);
 302 }
 303
 304 static void dp_timer_handler(unsigned long arg)
 305 {
 306         struct datapath *dp = (struct datapath *) arg;
 307 #if 1
 308         chain_timeout(dp->chain);
 309 #else
 310         int count = chain_timeout(dp->chain);
 311         chain_print_stats(dp->chain);
 312         if (count)
 313                 printk("%d flows timed out\n", count);
 314 #endif
 315         mod_timer(&dp->timer, round_jiffies(jiffies + (EXPIRE_SECS * HZ)));
 316 }
 317
 318 /*
 319  * Used as br_handle_frame_hook.  (Cannot run bridge at the same time, even on
 320  * different set of devices!)  Returns 0 if *pskb should be processed further,
 321  * 1 if *pskb is handled. */
 322 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,22)
 323 /* Called with rcu_read_lock. */
 324 static struct sk_buff *dp_frame_hook(struct net_bridge_port *p,
 325                                          struct sk_buff *skb)
 326 {
 327         struct ethhdr *eh = eth_hdr(skb);
 328         struct sk_buff *skb_local = NULL;
 329
 330
 331         if (compare_ether_addr(eh->h_dest, skb->dev->dev_addr) == 0)
 332                 return skb;
 333
 334         if (is_broadcast_ether_addr(eh->h_dest)
 335                                 || is_multicast_ether_addr(eh->h_dest)
 336                                 || is_local_ether_addr(eh->h_dest))
 337                 skb_local = skb_clone(skb, GFP_ATOMIC);
 338
 339         /* Push the Ethernet header back on. */
 340         if (skb->protocol == htons(ETH_P_8021Q))
 341                 skb_push(skb, VLAN_ETH_HLEN);
 342         else
 343                 skb_push(skb, ETH_HLEN);
 344
 345         fwd_port_input(p->dp->chain, skb, p->port_no);
 346
 347         return skb_local;
 348 }
 349 #elif LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)
 350 static int dp_frame_hook(struct net_bridge_port *p, struct sk_buff **pskb)
 351 {
 352         /* Push the Ethernet header back on. */
 353         if ((*pskb)->protocol == htons(ETH_P_8021Q))
 354                 skb_push(*pskb, VLAN_ETH_HLEN);
 355         else
 356                 skb_push(*pskb, ETH_HLEN);
 357
 358         fwd_port_input(p->dp->chain, *pskb, p->port_no);
 359         return 1;
 360 }
 361 #else
 362 /* NB: This has only been tested on 2.4.35 */
 363
 364 /* Called without any locks (?) */
 365 static void dp_frame_hook(struct sk_buff *skb)
 366 {
 367         struct net_bridge_port *p = skb->dev->br_port;
 368
 369         /* Push the Ethernet header back on. */
 370         if (skb->protocol == htons(ETH_P_8021Q))
 371                 skb_push(skb, VLAN_ETH_HLEN);
 372         else
 373                 skb_push(skb, ETH_HLEN);
 374
 375         if (p) {
 376                 rcu_read_lock();
 377                 fwd_port_input(p->dp->chain, skb, p->port_no);
 378                 rcu_read_unlock();
 379         } else
 380                 kfree_skb(skb);
 381 }
 382 #endif
 383
 384 /* Forwarding output path.
 385  * Based on net/bridge/br_forward.c. */
 386
 387 /* Don't forward packets to originating port or with flooding disabled */
 388 static inline int should_deliver(const struct net_bridge_port *p,
 389                         const struct sk_buff *skb)
 390 {
 391         if ((skb->dev == p->dev) || (p->flags & BRIDGE_PORT_NO_FLOOD)) {
 392                 return 0;
 393         }
 394
 395         return 1;
 396 }
 397
 398 static inline unsigned packet_length(const struct sk_buff *skb)
 399 {
 400         int length = skb->len - ETH_HLEN;
 401         if (skb->protocol == htons(ETH_P_8021Q))
 402                 length -= VLAN_HLEN;
 403         return length;
 404 }
 405
 406 static int
 407 flood(struct datapath *dp, struct sk_buff *skb)
 408 {
 409         struct net_bridge_port *p;
 410         int prev_port;
 411
 412         prev_port = -1;
 413         list_for_each_entry_rcu (p, &dp->port_list, node) {
 414                 if (!should_deliver(p, skb))
 415                         continue;
 416                 if (prev_port != -1) {
 417                         struct sk_buff *clone = skb_clone(skb, GFP_ATOMIC);
 418                         if (!clone) {
 419                                 kfree_skb(skb);
 420                                 return -ENOMEM;
 421                         }
 422                         dp_output_port(dp, clone, prev_port);
 423                 }
 424                 prev_port = p->port_no;
 425         }
 426         if (prev_port != -1)
 427                 dp_output_port(dp, skb, prev_port);
 428         else
 429                 kfree_skb(skb);
 430
 431         return 0;
 432 }
 433
 434 /* Marks 'skb' as having originated from 'in_port' in 'dp'.
 435    FIXME: how are devices reference counted? */
 436 int dp_set_origin(struct datapath *dp, uint16_t in_port,
 437                            struct sk_buff *skb)
 438 {
 439         if (in_port < OFPP_MAX && dp->ports[in_port]) {
 440                 skb->dev = dp->ports[in_port]->dev;
 441                 return 0;
 442         }
 443         return -ENOENT;
 444 }
 445
 446 /* Takes ownership of 'skb' and transmits it to 'out_port' on 'dp'.
 447  */
 448 int dp_output_port(struct datapath *dp, struct sk_buff *skb, int out_port)
 449 {
 450         struct net_bridge_port *p;
 451         int len = skb->len;
 452
 453         BUG_ON(!skb);
 454         if (out_port == OFPP_FLOOD)
 455                 return flood(dp, skb);
 456         else if (out_port == OFPP_CONTROLLER)
 457                 return dp_output_control(dp, skb, fwd_save_skb(skb), 0,
 458                                                   OFPR_ACTION);
 459         else if (out_port >= OFPP_MAX)
 460                 goto bad_port;
 461
 462         p = dp->ports[out_port];
 463         if (p == NULL)
 464                 goto bad_port;
 465
 466         skb->dev = p->dev;
 467         if (packet_length(skb) > skb->dev->mtu) {
 468                 printk("dropped over-mtu packet: %d > %d\n",
 469                                         packet_length(skb), skb->dev->mtu);
 470                 kfree_skb(skb);
 471                 return -E2BIG;
 472         }
 473
 474         dev_queue_xmit(skb);
 475
 476         return len;
 477
 478 bad_port:
 479         kfree_skb(skb);
 480         if (net_ratelimit())
 481                 printk("can't forward to bad port %d\n", out_port);
 482         return -ENOENT;
 483 }
 484
 485 /* Takes ownership of 'skb' and transmits it to 'dp''s control path.  If
 486  * 'buffer_id' != -1, then only the first 64 bytes of 'skb' are sent;
 487  * otherwise, all of 'skb' is sent.  'reason' indicates why 'skb' is being
 488  * sent. 'max_len' sets the maximum number of bytes that the caller
 489  * wants to be sent; a value of 0 indicates the entire packet should be
 490  * sent. */
 491 int
 492 dp_output_control(struct datapath *dp, struct sk_buff *skb,
 493                            uint32_t buffer_id, size_t max_len, int reason)
 494 {
 495         /* FIXME? packet_rcv_spkt in net/packet/af_packet.c does some stuff
 496            that we should possibly be doing here too. */
 497         /* FIXME?  Can we avoid creating a new skbuff in the case where we
 498          * forward the whole packet? */
 499         struct sk_buff *f_skb;
 500         struct nlattr *attr;
 501         struct ofp_packet_in *opi;
 502         size_t opi_len;
 503         size_t len, fwd_len;
 504         void *data;
 505         int err = -ENOMEM;
 506
 507         fwd_len = skb->len;
 508         if ((buffer_id != (uint32_t) -1) && max_len)
 509                 fwd_len = min(fwd_len, max_len);
 510
 511         len = nla_total_size(offsetof(struct ofp_packet_in, data) + fwd_len)
 512                                 + nla_total_size(sizeof(uint32_t));
 513
 514         f_skb = genlmsg_new(len, GFP_ATOMIC);
 515         if (!f_skb)
 516                 goto error_free_skb;
 517
 518         data = genlmsg_put(f_skb, 0, 0, &dp_genl_family, 0,
 519                                 DP_GENL_C_OPENFLOW);
 520         if (data == NULL)
 521                 goto error_free_f_skb;
 522
 523         NLA_PUT_U32(f_skb, DP_GENL_A_DP_IDX, dp->dp_idx);
 524
 525         opi_len = offsetof(struct ofp_packet_in, data) + fwd_len;
 526         attr = nla_reserve(f_skb, DP_GENL_A_OPENFLOW, opi_len);
 527         if (!attr)
 528                 goto error_free_f_skb;
 529         opi = nla_data(attr);
 530         opi->header.version = OFP_VERSION;
 531         opi->header.type    = OFPT_PACKET_IN;
 532         opi->header.length  = htons(opi_len);
 533         opi->header.xid     = htonl(0);
 534
 535         opi->buffer_id      = htonl(buffer_id);
 536         opi->total_len      = htons(skb->len);
 537         opi->in_port        = htons(skb->dev->br_port->port_no);
 538         opi->reason         = reason;
 539         opi->pad            = 0;
 540         SKB_LINEAR_ASSERT(skb);
 541         memcpy(opi->data, skb_mac_header(skb), fwd_len);
 542
 543         err = genlmsg_end(f_skb, data);
 544         if (err < 0)
 545                 goto error_free_f_skb;
 546
 547         err = genlmsg_multicast(f_skb, 0, mc_group.id, GFP_ATOMIC);
 548         if (err && net_ratelimit())
 549                 printk(KERN_WARNING "dp_output_control: genlmsg_multicast failed: %d\n", err);
 550
 551         kfree_skb(skb);
 552
 553         return err;
 554
 555 nla_put_failure:
 556 error_free_f_skb:
 557         nlmsg_free(f_skb);
 558 error_free_skb:
 559         kfree_skb(skb);
 560         if (net_ratelimit())
 561                 printk(KERN_ERR "dp_output_control: failed to send: %d\n", err);
 562         return err;
 563 }
 564
 565 static void fill_port_desc(struct net_bridge_port *p, struct ofp_phy_port *desc)
 566 {
 567         desc->port_no = htons(p->port_no);
 568         strncpy(desc->name, p->dev->name, OFP_MAX_PORT_NAME_LEN);
 569         desc->name[OFP_MAX_PORT_NAME_LEN-1] = '\0';
 570         memcpy(desc->hw_addr, p->dev->dev_addr, ETH_ALEN);
 571         desc->flags = htonl(p->flags);
 572         desc->features = 0;
 573         desc->speed = 0;
 574
 575 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,4,24)
 576         if (p->dev->ethtool_ops && p->dev->ethtool_ops->get_settings) {
 577                 struct ethtool_cmd ecmd = { .cmd = ETHTOOL_GSET };
 578
 579                 if (!p->dev->ethtool_ops->get_settings(p->dev, &ecmd)) {
 580                         if (ecmd.supported & SUPPORTED_10baseT_Half)
 581                                 desc->features |= OFPPF_10MB_HD;
 582                         if (ecmd.supported & SUPPORTED_10baseT_Full)
 583                                 desc->features |= OFPPF_10MB_FD;
 584                         if (ecmd.supported & SUPPORTED_100baseT_Half)
 585                                 desc->features |= OFPPF_100MB_HD;
 586                         if (ecmd.supported & SUPPORTED_100baseT_Full)
 587                                 desc->features |= OFPPF_100MB_FD;
 588                         if (ecmd.supported & SUPPORTED_1000baseT_Half)
 589                                 desc->features |= OFPPF_1GB_HD;
 590                         if (ecmd.supported & SUPPORTED_1000baseT_Full)
 591                                 desc->features |= OFPPF_1GB_FD;
 592                         /* 10Gbps half-duplex doesn't exist... */
 593                         if (ecmd.supported & SUPPORTED_10000baseT_Full)
 594                                 desc->features |= OFPPF_10GB_FD;
 595
 596                         desc->features = htonl(desc->features);
 597                         desc->speed = htonl(ecmd.speed);
 598                 }
 599         }
 600 #endif
 601 }
 602
 603 static int
 604 fill_data_hello(struct datapath *dp, struct ofp_data_hello *odh)
 605 {
 606         struct net_bridge_port *p;
 607         int port_count = 0;
 608
 609         odh->header.version = OFP_VERSION;
 610         odh->header.type    = OFPT_DATA_HELLO;
 611         odh->header.xid     = htonl(0);
 612         odh->datapath_id    = cpu_to_be64(dp->id);
 613
 614         odh->n_exact        = htonl(2 * TABLE_HASH_MAX_FLOWS);
 615         odh->n_mac_only     = htonl(TABLE_MAC_MAX_FLOWS);
 616         odh->n_compression  = 0;                                           /* Not supported */
 617         odh->n_general      = htonl(TABLE_LINEAR_MAX_FLOWS);
 618         odh->buffer_mb      = htonl(UINT32_MAX);
 619         odh->n_buffers      = htonl(N_PKT_BUFFERS);
 620         odh->capabilities   = htonl(OFP_SUPPORTED_CAPABILITIES);
 621         odh->actions        = htonl(OFP_SUPPORTED_ACTIONS);
 622         odh->miss_send_len  = htons(dp->miss_send_len);
 623
 624         list_for_each_entry_rcu (p, &dp->port_list, node) {
 625                 fill_port_desc(p, &odh->ports[port_count]);
 626                 port_count++;
 627         }
 628
 629         return port_count;
 630 }
 631
 632 int
 633 dp_send_hello(struct datapath *dp)
 634 {
 635         struct sk_buff *skb;
 636         struct nlattr *attr;
 637         struct ofp_data_hello *odh;
 638         size_t odh_max_len, odh_len, port_max_len, len;
 639         void *data;
 640         int err = -ENOMEM;
 641         int port_count;
 642
 643
 644         /* Overallocate, since we can't reliably determine the number of
 645          * ports a priori. */
 646         port_max_len = sizeof(struct ofp_phy_port) * OFPP_MAX;
 647
 648         len = nla_total_size(sizeof(*odh) + port_max_len)
 649                                 + nla_total_size(sizeof(uint32_t));
 650
 651         skb = genlmsg_new(len, GFP_ATOMIC);
 652         if (!skb) {
 653                 if (net_ratelimit())
 654                         printk("dp_send_hello: genlmsg_new failed\n");
 655                 goto error;
 656         }
 657
 658         data = genlmsg_put(skb, 0, 0, &dp_genl_family, 0,
 659                            DP_GENL_C_OPENFLOW);
 660         if (data == NULL) {
 661                 if (net_ratelimit())
 662                         printk("dp_send_hello: genlmsg_put failed\n");
 663                 goto error;
 664         }
 665
 666         NLA_PUT_U32(skb, DP_GENL_A_DP_IDX, dp->dp_idx);
 667
 668         odh_max_len = sizeof(*odh) + port_max_len;
 669         attr = nla_reserve(skb, DP_GENL_A_OPENFLOW, odh_max_len);
 670         if (!attr) {
 671                 if (net_ratelimit())
 672                         printk("dp_send_hello: nla_reserve failed\n");
 673                 goto error;
 674         }
 675         odh = nla_data(attr);
 676         port_count = fill_data_hello(dp, odh);
 677
 678         /* Only now that we know how many ports we've added can we say
 679          * say something about the length. */
 680         odh_len = sizeof(*odh) + (sizeof(struct ofp_phy_port) * port_count);
 681         odh->header.length = htons(odh_len);
 682
 683         /* Take back the unused part that was reserved */
 684         nla_unreserve(skb, attr, (odh_max_len - odh_len));
 685
 686         err = genlmsg_end(skb, data);
 687         if (err < 0) {
 688                 if (net_ratelimit())
 689                         printk("dp_send_hello: genlmsg_end failed\n");
 690                 goto error;
 691         }
 692
 693         err = genlmsg_multicast(skb, 0, mc_group.id, GFP_ATOMIC);
 694         if (err && net_ratelimit())
 695                 printk(KERN_WARNING "dp_send_hello: genlmsg_multicast failed: %d\n", err);
 696
 697         return err;
 698
 699 nla_put_failure:
 700 error:
 701         kfree_skb(skb);
 702         if (net_ratelimit())
 703                 printk(KERN_ERR "dp_send_hello: failed to send: %d\n", err);
 704         return err;
 705 }
 706
 707 int
 708 dp_update_port_flags(struct datapath *dp, const struct ofp_phy_port *opp)
 709 {
 710         struct net_bridge_port *p;
 711
 712         p = dp->ports[htons(opp->port_no)];
 713
 714         /* Make sure the port id hasn't changed since this was sent */
 715         if (!p || memcmp(opp->hw_addr, p->dev->dev_addr, ETH_ALEN) != 0)
 716                 return -1;
 717
 718         p->flags = htonl(opp->flags);
 719
 720         return 0;
 721 }
 722
 723
 724 static int
 725 send_port_status(struct net_bridge_port *p, uint8_t status)
 726 {
 727         struct sk_buff *skb;
 728         struct nlattr *attr;
 729         struct ofp_port_status *ops;
 730         void *data;
 731         int err = -ENOMEM;
 732
 733
 734         skb = genlmsg_new(NLMSG_GOODSIZE, GFP_ATOMIC);
 735         if (!skb) {
 736                 if (net_ratelimit())
 737                         printk("send_port_status: genlmsg_new failed\n");
 738                 goto error;
 739         }
 740
 741         data = genlmsg_put(skb, 0, 0, &dp_genl_family, 0,
 742                            DP_GENL_C_OPENFLOW);
 743         if (data == NULL) {
 744                 if (net_ratelimit())
 745                         printk("send_port_status: genlmsg_put failed\n");
 746                 goto error;
 747         }
 748
 749         NLA_PUT_U32(skb, DP_GENL_A_DP_IDX, p->dp->dp_idx);
 750
 751         attr = nla_reserve(skb, DP_GENL_A_OPENFLOW, sizeof(*ops));
 752         if (!attr) {
 753                 if (net_ratelimit())
 754                         printk("send_port_status: nla_reserve failed\n");
 755                 goto error;
 756         }
 757
 758         ops = nla_data(attr);
 759         ops->header.version = OFP_VERSION;
 760         ops->header.type    = OFPT_PORT_STATUS;
 761         ops->header.length  = htons(sizeof(*ops));
 762         ops->header.xid     = htonl(0);
 763
 764         ops->reason         = status;
 765         fill_port_desc(p, &ops->desc);
 766
 767         err = genlmsg_end(skb, data);
 768         if (err < 0) {
 769                 if (net_ratelimit())
 770                         printk("send_port_status: genlmsg_end failed\n");
 771                 goto error;
 772         }
 773
 774         err = genlmsg_multicast(skb, 0, mc_group.id, GFP_ATOMIC);
 775         if (err && net_ratelimit())
 776                 printk(KERN_WARNING "send_port_status: genlmsg_multicast failed: %d\n", err);
 777
 778         return err;
 779
 780 nla_put_failure:
 781 error:
 782         kfree_skb(skb);
 783         if (net_ratelimit())
 784                 printk(KERN_ERR "send_port_status: failed to send: %d\n", err);
 785         return err;
 786 }
 787
 788 int
 789 dp_send_flow_expired(struct datapath *dp, struct sw_flow *flow)
 790 {
 791         struct sk_buff *skb;
 792         struct nlattr *attr;
 793         struct ofp_flow_expired *ofe;
 794         void *data;
 795         unsigned long duration_j;
 796         int err = -ENOMEM;
 797
 798
 799         skb = genlmsg_new(NLMSG_GOODSIZE, GFP_ATOMIC);
 800         if (!skb) {
 801                 if (net_ratelimit())
 802                         printk("dp_send_flow_expired: genlmsg_new failed\n");
 803                 goto error;
 804         }
 805
 806         data = genlmsg_put(skb, 0, 0, &dp_genl_family, 0,
 807                            DP_GENL_C_OPENFLOW);
 808         if (data == NULL) {
 809                 if (net_ratelimit())
 810                         printk("dp_send_flow_expired: genlmsg_put failed\n");
 811                 goto error;
 812         }
 813
 814         NLA_PUT_U32(skb, DP_GENL_A_DP_IDX, dp->dp_idx);
 815
 816         attr = nla_reserve(skb, DP_GENL_A_OPENFLOW, sizeof(*ofe));
 817         if (!attr) {
 818                 if (net_ratelimit())
 819                         printk("dp_send_flow_expired: nla_reserve failed\n");
 820                 goto error;
 821         }
 822
 823         ofe = nla_data(attr);
 824         ofe->header.version = OFP_VERSION;
 825         ofe->header.type    = OFPT_FLOW_EXPIRED;
 826         ofe->header.length  = htons(sizeof(*ofe));
 827         ofe->header.xid     = htonl(0);
 828
 829         flow_fill_match(&ofe->match, &flow->key);
 830         duration_j = (flow->timeout - HZ * flow->max_idle) - flow->init_time;
 831         ofe->duration   = htonl(duration_j / HZ);
 832         ofe->packet_count   = cpu_to_be64(flow->packet_count);
 833         ofe->byte_count     = cpu_to_be64(flow->byte_count);
 834
 835         err = genlmsg_end(skb, data);
 836         if (err < 0) {
 837                 if (net_ratelimit())
 838                         printk("dp_send_flow_expired: genlmsg_end failed\n");
 839                 goto error;
 840         }
 841
 842         err = genlmsg_multicast(skb, 0, mc_group.id, GFP_ATOMIC);
 843         if (err && net_ratelimit())
 844                 printk(KERN_WARNING "send_flow_expired: genlmsg_multicast failed: %d\n", err);
 845
 846         return err;
 847
 848 nla_put_failure:
 849 error:
 850         kfree_skb(skb);
 851         if (net_ratelimit())
 852                 printk(KERN_ERR "send_flow_expired: failed to send: %d\n", err);
 853         return err;
 854 }
 855
 856 /* Generic Netlink interface.
 857  *
 858  * See netlink(7) for an introduction to netlink.  See
 859  * http://linux-net.osdl.org/index.php/Netlink for more information and
 860  * pointers on how to work with netlink and Generic Netlink in the kernel and
 861  * in userspace. */
 862
 863 static struct genl_family dp_genl_family = {
 864         .id = GENL_ID_GENERATE,
 865         .hdrsize = 0,
 866         .name = DP_GENL_FAMILY_NAME,
 867         .version = 1,
 868         .maxattr = DP_GENL_A_MAX,
 869 };
 870
 871 /* Attribute policy: what each attribute may contain.  */
 872 static struct nla_policy dp_genl_policy[DP_GENL_A_MAX + 1] = {
 873         [DP_GENL_A_DP_IDX] = { .type = NLA_U32 },
 874         [DP_GENL_A_MC_GROUP] = { .type = NLA_U32 },
 875         [DP_GENL_A_PORTNAME] = { .type = NLA_STRING }
 876 };
 877
 878 static int dp_genl_add(struct sk_buff *skb, struct genl_info *info)
 879 {
 880         if (!info->attrs[DP_GENL_A_DP_IDX])
 881                 return -EINVAL;
 882
 883         return new_dp(nla_get_u32(info->attrs[DP_GENL_A_DP_IDX]));
 884 }
 885
 886 static struct genl_ops dp_genl_ops_add_dp = {
 887         .cmd = DP_GENL_C_ADD_DP,
 888         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 889         .policy = dp_genl_policy,
 890         .doit = dp_genl_add,
 891         .dumpit = NULL,
 892 };
 893
 894 struct datapath *dp_get(int dp_idx)
 895 {
 896         if (dp_idx < 0 || dp_idx > DP_MAX)
 897                 return NULL;
 898         return rcu_dereference(dps[dp_idx]);
 899 }
 900
 901 static int dp_genl_del(struct sk_buff *skb, struct genl_info *info)
 902 {
 903         struct datapath *dp;
 904         int err;
 905
 906         if (!info->attrs[DP_GENL_A_DP_IDX])
 907                 return -EINVAL;
 908
 909         mutex_lock(&dp_mutex);
 910         dp = dp_get(nla_get_u32((info->attrs[DP_GENL_A_DP_IDX])));
 911         if (!dp)
 912                 err = -ENOENT;
 913         else {
 914                 del_dp(dp);
 915                 err = 0;
 916         }
 917         mutex_unlock(&dp_mutex);
 918         return err;
 919 }
 920
 921 static struct genl_ops dp_genl_ops_del_dp = {
 922         .cmd = DP_GENL_C_DEL_DP,
 923         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 924         .policy = dp_genl_policy,
 925         .doit = dp_genl_del,
 926         .dumpit = NULL,
 927 };
 928
 929 /* Queries a datapath for related information.  Currently the only relevant
 930  * information is the datapath's multicast group ID.  Really we want one
 931  * multicast group per datapath, but because of locking issues[*] we can't
 932  * easily get one.  Thus, every datapath will currently return the same
 933  * global multicast group ID, but in the future it would be nice to fix that.
 934  *
 935  * [*] dp_genl_add, to add a new datapath, is called under the genl_lock
 936  *       mutex, and genl_register_mc_group, called to acquire a new multicast
 937  *       group ID, also acquires genl_lock, thus deadlock.
 938  */
 939 static int dp_genl_query(struct sk_buff *skb, struct genl_info *info)
 940 {
 941         struct datapath *dp;
 942         struct sk_buff *ans_skb = NULL;
 943         int dp_idx;
 944         int err = -ENOMEM;
 945
 946         if (!info->attrs[DP_GENL_A_DP_IDX])
 947                 return -EINVAL;
 948
 949         rcu_read_lock();
 950         dp_idx = nla_get_u32((info->attrs[DP_GENL_A_DP_IDX]));
 951         dp = dp_get(dp_idx);
 952         if (!dp)
 953                 err = -ENOENT;
 954         else {
 955                 void *data;
 956                 ans_skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
 957                 if (!ans_skb) {
 958                         err = -ENOMEM;
 959                         goto err;
 960                 }
 961                 data = genlmsg_put_reply(ans_skb, info, &dp_genl_family,
 962                                          0, DP_GENL_C_QUERY_DP);
 963                 if (data == NULL) {
 964                         err = -ENOMEM;
 965                         goto err;
 966                 }
 967                 NLA_PUT_U32(ans_skb, DP_GENL_A_DP_IDX, dp_idx);
 968                 NLA_PUT_U32(ans_skb, DP_GENL_A_MC_GROUP, mc_group.id);
 969
 970                 genlmsg_end(ans_skb, data);
 971                 err = genlmsg_reply(ans_skb, info);
 972                 if (!err)
 973                         ans_skb = NULL;
 974         }
 975 err:
 976 nla_put_failure:
 977         if (ans_skb)
 978                 kfree_skb(ans_skb);
 979         rcu_read_unlock();
 980         return err;
 981 }
 982
 983 /*
 984  * Fill flow entry for nl flow query.  Called with rcu_lock
 985  *
 986  */
 987 static
 988 int
 989 dp_fill_flow(struct ofp_flow_mod* ofm, struct swt_iterator* iter)
 990 {
 991         ofm->header.version  = OFP_VERSION;
 992         ofm->header.type     = OFPT_FLOW_MOD;
 993         ofm->header.length   = htons(sizeof(struct ofp_flow_mod)
 994                                 + sizeof(ofm->actions[0]));
 995         ofm->header.xid      = htonl(0);
 996
 997         ofm->match.wildcards = htons(iter->flow->key.wildcards);
 998         ofm->match.in_port   = iter->flow->key.in_port;
 999         ofm->match.dl_vlan   = iter->flow->key.dl_vlan;
1000         memcpy(ofm->match.dl_src, iter->flow->key.dl_src, ETH_ALEN);
1001         memcpy(ofm->match.dl_dst, iter->flow->key.dl_dst, ETH_ALEN);
1002         ofm->match.dl_type   = iter->flow->key.dl_type;
1003         ofm->match.nw_src    = iter->flow->key.nw_src;
1004         ofm->match.nw_dst    = iter->flow->key.nw_dst;
1005         ofm->match.nw_proto  = iter->flow->key.nw_proto;
1006         ofm->match.tp_src    = iter->flow->key.tp_src;
1007         ofm->match.tp_dst    = iter->flow->key.tp_dst;
1008         ofm->group_id        = iter->flow->group_id;
1009         ofm->max_idle        = iter->flow->max_idle;
1010         /* TODO support multiple actions  */
1011         ofm->actions[0]      = iter->flow->actions[0];
1012
1013         return 0;
1014 }
1015
1016 static int dp_genl_show(struct sk_buff *skb, struct genl_info *info)
1017 {
1018         struct datapath *dp;
1019         int err = -ENOMEM;
1020         struct sk_buff *ans_skb = NULL;
1021         void *data;
1022         struct nlattr *attr;
1023         struct ofp_data_hello *odh;
1024         size_t odh_max_len, odh_len, port_max_len, len;
1025         int port_count;
1026
1027         if (!info->attrs[DP_GENL_A_DP_IDX])
1028                 return -EINVAL;
1029
1030         mutex_lock(&dp_mutex);
1031         dp = dp_get(nla_get_u32((info->attrs[DP_GENL_A_DP_IDX])));
1032         if (!dp)
1033                 goto error;
1034
1035         /* Overallocate, since we can't reliably determine the number of
1036          * ports a priori. */
1037         port_max_len = sizeof(struct ofp_phy_port) * OFPP_MAX;
1038
1039         len = nla_total_size(sizeof(*odh) + port_max_len)
1040                         + nla_total_size(sizeof(uint32_t));
1041
1042         ans_skb = nlmsg_new(len, GFP_KERNEL);
1043         if (!ans_skb)
1044                 goto error;
1045
1046         data = genlmsg_put_reply(ans_skb, info, &dp_genl_family,
1047                                  0, DP_GENL_C_SHOW_DP);
1048         if (data == NULL)
1049                 goto error;
1050
1051         NLA_PUT_U32(ans_skb, DP_GENL_A_DP_IDX, dp->dp_idx);
1052
1053         odh_max_len = sizeof(*odh) + port_max_len;
1054         attr = nla_reserve(ans_skb, DP_GENL_A_DP_INFO, odh_max_len);
1055         if (!attr)
1056                 goto error;
1057         odh = nla_data(attr);
1058         port_count = fill_data_hello(dp, odh);
1059
1060         /* Only now that we know how many ports we've added can we say
1061          * say something about the length. */
1062         odh_len = sizeof(*odh) + (sizeof(struct ofp_phy_port) * port_count);
1063         odh->header.length = htons(odh_len);
1064
1065         /* Take back the unused part that was reserved */
1066         nla_unreserve(ans_skb, attr, (odh_max_len - odh_len));
1067
1068         genlmsg_end(ans_skb, data);
1069         err = genlmsg_reply(ans_skb, info);
1070         if (!err)
1071                 ans_skb = NULL;
1072
1073 error:
1074 nla_put_failure:
1075         if (ans_skb)
1076                 kfree_skb(ans_skb);
1077         mutex_unlock(&dp_mutex);
1078         return err;
1079 }
1080
1081 static struct genl_ops dp_genl_ops_show_dp = {
1082         .cmd = DP_GENL_C_SHOW_DP,
1083         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1084         .policy = dp_genl_policy,
1085         .doit = dp_genl_show,
1086         .dumpit = NULL,
1087 };
1088
1089 /* Convenience function */
1090 static
1091 void*
1092 dp_init_nl_flow_msg(uint32_t dp_idx, uint16_t table_idx,
1093                 struct genl_info *info, struct sk_buff* skb)
1094 {
1095         void* data;
1096
1097         data = genlmsg_put_reply(skb, info, &dp_genl_family, 0,
1098                                 DP_GENL_C_QUERY_FLOW);
1099         if (data == NULL)
1100                 return NULL;
1101         NLA_PUT_U32(skb, DP_GENL_A_DP_IDX,   dp_idx);
1102         NLA_PUT_U16(skb, DP_GENL_A_TABLEIDX, table_idx);
1103
1104         return data;
1105
1106 nla_put_failure:
1107         return NULL;
1108 }
1109
1110 /*  Iterate through the specified table and send all flow entries over
1111  *  netlink to userspace.  Each flow message has the following format:
1112  *
1113  *  32bit dpix
1114  *  16bit tabletype
1115  *  32bit number of flows
1116  *  openflow-flow-entries
1117  *
1118  *  The full table may require multiple messages.  A message with 0 flows
1119  *  signifies end-of message.
1120  */
1121
1122 static
1123 int
1124 dp_dump_table(struct datapath *dp, uint16_t table_idx, struct genl_info *info, struct ofp_flow_mod* matchme)
1125 {
1126         struct sk_buff  *skb = 0;
1127         struct sw_table *table = 0;
1128         struct swt_iterator iter;
1129         struct sw_flow_key in_flow;
1130         struct nlattr   *attr;
1131         int count = 0, sum_count = 0;
1132         void *data;
1133         uint8_t* ofm_ptr = 0;
1134         struct nlattr   *num_attr;
1135         int err = -ENOMEM;
1136
1137         table = dp->chain->tables[table_idx];
1138         if ( table == NULL ) {
1139                 dprintk("dp::dp_dump_table error, non-existant table at position %d\n", table_idx);
1140                 return -EINVAL;
1141         }
1142
1143         if (!table->iterator(table, &iter)) {
1144                 dprintk("dp::dp_dump_table couldn't initialize empty table iterator\n");
1145                 return -ENOMEM;
1146         }
1147
1148         while (iter.flow) {
1149
1150                 /* verify that we can fit all NL_FLOWS_PER_MESSAGE in a single
1151                  * sk_buf */
1152                 if( (sizeof(dp_genl_family) + sizeof(uint32_t) + sizeof(uint16_t) + sizeof(uint32_t) +
1153                                         (NL_FLOWS_PER_MESSAGE * sizeof(struct ofp_flow_mod))) > (8192 - 64)){
1154                         dprintk("dp::dp_dump_table NL_FLOWS_PER_MESSAGE may cause overrun in skbuf\n");
1155                         return -ENOMEM;
1156                 }
1157
1158                 skb = nlmsg_new(8192 - 64, GFP_ATOMIC);
1159                 if (skb == NULL) {
1160                         return -ENOMEM;
1161                 }
1162
1163                 data = dp_init_nl_flow_msg(dp->dp_idx, table_idx, info, skb);
1164                 if (data == NULL){
1165                         err= -ENOMEM;
1166                         goto error_free_skb;
1167                 }
1168
1169                 /* reserve space to put the number of flows for this message, to
1170                  * be filled after the loop*/
1171                 num_attr = nla_reserve(skb, DP_GENL_A_NUMFLOWS, sizeof(uint32_t));
1172                 if(!num_attr){
1173                         err = -ENOMEM;
1174                         goto error_free_skb;
1175                 }
1176
1177                 /* Only load NL_FLOWS_PER_MESSAGE flows at a time */
1178                 attr = nla_reserve(skb, DP_GENL_A_FLOW,
1179                                 (sizeof(struct ofp_flow_mod) + sizeof(struct ofp_action)) * NL_FLOWS_PER_MESSAGE);
1180                 if (!attr){
1181                         err = -ENOMEM;
1182                         goto error_free_skb;
1183                 }
1184
1185                 /* internal loop to fill NL_FLOWS_PER_MESSAGE flows */
1186                 ofm_ptr = nla_data(attr);
1187                 flow_extract_match(&in_flow, &matchme->match);
1188                 while (iter.flow && count < NL_FLOWS_PER_MESSAGE) {
1189                         if(flow_matches(&in_flow, &iter.flow->key)){
1190                                 if((err = dp_fill_flow((struct ofp_flow_mod*)ofm_ptr, &iter)))
1191                                         goto error_free_skb;
1192                                 count++;
1193                                 /* TODO support multiple actions  */
1194                                 ofm_ptr += sizeof(struct ofp_flow_mod) + sizeof(struct ofp_action);
1195                         }
1196                         table->iterator_next(&iter);
1197                 }
1198
1199                 *((uint32_t*)nla_data(num_attr)) = count;
1200                 genlmsg_end(skb, data);
1201
1202                 sum_count += count;
1203                 count = 0;
1204
1205                 err = genlmsg_unicast(skb, info->snd_pid);
1206                 skb = 0;
1207         }
1208
1209         /* send a sentinal message saying we're done */
1210         skb = nlmsg_new(NLMSG_GOODSIZE, GFP_ATOMIC);
1211         if (skb == NULL) {
1212                 return -ENOMEM;
1213         }
1214         data = dp_init_nl_flow_msg(dp->dp_idx, table_idx, info, skb);
1215         if (data == NULL){
1216                 err= -ENOMEM;
1217                 goto error_free_skb;
1218         }
1219
1220         NLA_PUT_U32(skb, DP_GENL_A_NUMFLOWS,   0);
1221         /* dummy flow so nl doesn't complain */
1222         attr = nla_reserve(skb, DP_GENL_A_FLOW, sizeof(struct ofp_flow_mod));
1223         if (!attr){
1224                 err = -ENOMEM;
1225                 goto error_free_skb;
1226         }
1227         genlmsg_end(skb, data);
1228         err = genlmsg_reply(skb, info); skb = 0;
1229
1230 nla_put_failure:
1231 error_free_skb:
1232         if(skb)
1233                 kfree_skb(skb);
1234         return err;
1235 }
1236
1237 /* Helper function to query_table which creates and sends a message packed with
1238  * table stats.  Message form is:
1239  *
1240  * u32 DP_IDX
1241  * u32 NUM_TABLES
1242  * OFP_TABLE (list of OFP_TABLES)
1243  *
1244  */
1245
1246 static
1247 int
1248 dp_dump_table_stats(struct datapath *dp, int dp_idx, struct genl_info *info)
1249 {
1250         struct sk_buff   *skb = 0;
1251         struct ofp_table *ot = 0;
1252         struct nlattr   *attr;
1253         struct sw_table_stats stats;
1254         void *data;
1255         int err = -ENOMEM;
1256         int i = 0;
1257         int nt = dp->chain->n_tables;
1258
1259         /* u32 IDX, u32 NUMTABLES, list-of-tables */
1260         skb = nlmsg_new(4 + 4 + (sizeof(struct ofp_table) * nt), GFP_ATOMIC);
1261         if (skb == NULL) {
1262                 return -ENOMEM;
1263         }
1264
1265         data = genlmsg_put_reply(skb, info, &dp_genl_family, 0,
1266                                 DP_GENL_C_QUERY_TABLE);
1267         if (data == NULL){
1268                 return -ENOMEM;
1269         }
1270
1271         NLA_PUT_U32(skb, DP_GENL_A_DP_IDX,      dp_idx);
1272         NLA_PUT_U32(skb, DP_GENL_A_NUMTABLES, nt);
1273
1274         /* ... we assume that all tables can fit in a single message.
1275          * Probably a reasonable assumption seeing that we only have
1276          * 3 atm */
1277         attr = nla_reserve(skb, DP_GENL_A_TABLE, (sizeof(struct ofp_table) * nt));
1278         if (!attr){
1279                 err = -ENOMEM;
1280                 goto error_free_skb;
1281         }
1282
1283         ot = nla_data(attr);
1284
1285         for (i = 0; i < nt; ++i) {
1286                 dp->chain->tables[i]->stats(dp->chain->tables[i], &stats);
1287                 ot->header.version = OFP_VERSION;
1288                 ot->header.type    = OFPT_TABLE;
1289                 ot->header.length  = htons(sizeof(struct ofp_table));
1290                 ot->header.xid     = htonl(0);
1291
1292                 strncpy(ot->name, stats.name, OFP_MAX_TABLE_NAME_LEN);
1293                 ot->table_id  = htons(i);
1294                 ot->n_flows   = htonl(stats.n_flows);
1295                 ot->max_flows = htonl(stats.max_flows);
1296                 ot++;
1297         }
1298
1299
1300         genlmsg_end(skb, data);
1301         err = genlmsg_reply(skb, info); skb = 0;
1302
1303 nla_put_failure:
1304 error_free_skb:
1305         if(skb)
1306                 kfree_skb(skb);
1307         return err;
1308 }
1309
1310 /*
1311  * Queries a datapath for flow-table statistics
1312  */
1313
1314
1315 static int dp_genl_table_query(struct sk_buff *skb, struct genl_info *info)
1316 {
1317         struct   datapath* dp;
1318         int       err = 0;
1319
1320         if (!info->attrs[DP_GENL_A_DP_IDX]) {
1321                 dprintk("dp::dp_genl_table_query received message with missing attributes\n");
1322                 return -EINVAL;
1323         }
1324
1325         rcu_read_lock();
1326         dp = dp_get(nla_get_u32(info->attrs[DP_GENL_A_DP_IDX]));
1327         if (!dp) {
1328                 err = -ENOENT;
1329                 goto err_out;
1330         }
1331
1332         err = dp_dump_table_stats(dp, nla_get_u32(info->attrs[DP_GENL_A_DP_IDX]), info);
1333
1334 err_out:
1335         rcu_read_unlock();
1336         return err;
1337 }
1338
1339 /*
1340  * Queries a datapath for flow-table entries.
1341  */
1342
1343 static int dp_genl_flow_query(struct sk_buff *skb, struct genl_info *info)
1344 {
1345         struct datapath* dp;
1346         struct ofp_flow_mod*  ofm;
1347         u16     table_idx;
1348         int     err = 0;
1349
1350         if (!info->attrs[DP_GENL_A_DP_IDX]
1351                                 || !info->attrs[DP_GENL_A_TABLEIDX]
1352                                 || !info->attrs[DP_GENL_A_FLOW]) {
1353                 dprintk("dp::dp_genl_flow_query received message with missing attributes\n");
1354                 return -EINVAL;
1355         }
1356
1357         rcu_read_lock();
1358         dp = dp_get(nla_get_u32(info->attrs[DP_GENL_A_DP_IDX]));
1359         if (!dp) {
1360                 err = -ENOENT;
1361                 goto err_out;
1362         }
1363
1364         table_idx = nla_get_u16(info->attrs[DP_GENL_A_TABLEIDX]);
1365
1366         if (dp->chain->n_tables <= table_idx){
1367                 printk("table index %d invalid (dp has %d tables)\n",
1368                                 table_idx, dp->chain->n_tables);
1369         err = -EINVAL;
1370                 goto err_out;
1371         }
1372
1373         ofm = nla_data(info->attrs[DP_GENL_A_FLOW]);
1374         err = dp_dump_table(dp, table_idx, info, ofm);
1375
1376 err_out:
1377         rcu_read_unlock();
1378         return err;
1379 }
1380
1381 static struct nla_policy dp_genl_flow_policy[DP_GENL_A_MAX + 1] = {
1382         [DP_GENL_A_DP_IDX]      = { .type = NLA_U32 },
1383         [DP_GENL_A_TABLEIDX] = { .type = NLA_U16 },
1384         [DP_GENL_A_NUMFLOWS]  = { .type = NLA_U32 },
1385 };
1386
1387 static struct genl_ops dp_genl_ops_query_flow = {
1388         .cmd    = DP_GENL_C_QUERY_FLOW,
1389         .flags  = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1390         .policy = dp_genl_flow_policy,
1391         .doit   = dp_genl_flow_query,
1392         .dumpit = NULL,
1393 };
1394
1395 static struct nla_policy dp_genl_table_policy[DP_GENL_A_MAX + 1] = {
1396         [DP_GENL_A_DP_IDX]      = { .type = NLA_U32 },
1397 };
1398
1399 static struct genl_ops dp_genl_ops_query_table = {
1400         .cmd    = DP_GENL_C_QUERY_TABLE,
1401         .flags  = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1402         .policy = dp_genl_table_policy,
1403         .doit   = dp_genl_table_query,
1404         .dumpit = NULL,
1405 };
1406
1407
1408 static struct genl_ops dp_genl_ops_query_dp = {
1409         .cmd = DP_GENL_C_QUERY_DP,
1410         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1411         .policy = dp_genl_policy,
1412         .doit = dp_genl_query,
1413         .dumpit = NULL,
1414 };
1415
1416 static int dp_genl_add_del_port(struct sk_buff *skb, struct genl_info *info)
1417 {
1418         struct datapath *dp;
1419         struct net_device *port;
1420         int err;
1421
1422         if (!info->attrs[DP_GENL_A_DP_IDX] || !info->attrs[DP_GENL_A_PORTNAME])
1423                 return -EINVAL;
1424
1425         /* Get datapath. */
1426         mutex_lock(&dp_mutex);
1427         dp = dp_get(nla_get_u32(info->attrs[DP_GENL_A_DP_IDX]));
1428         if (!dp) {
1429                 err = -ENOENT;
1430                 goto out;
1431         }
1432
1433         /* Get interface to add/remove. */
1434         port = dev_get_by_name(&init_net,
1435                         nla_data(info->attrs[DP_GENL_A_PORTNAME]));
1436         if (!port) {
1437                 err = -ENOENT;
1438                 goto out;
1439         }
1440
1441         /* Execute operation. */
1442         if (info->genlhdr->cmd == DP_GENL_C_ADD_PORT)
1443                 err = add_switch_port(dp, port);
1444         else {
1445                 if (port->br_port == NULL || port->br_port->dp != dp) {
1446                         err = -ENOENT;
1447                         goto out_put;
1448                 }
1449                 err = del_switch_port(port->br_port);
1450         }
1451
1452 out_put:
1453         dev_put(port);
1454 out:
1455         mutex_unlock(&dp_mutex);
1456         return err;
1457 }
1458
1459 static struct genl_ops dp_genl_ops_add_port = {
1460         .cmd = DP_GENL_C_ADD_PORT,
1461         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1462         .policy = dp_genl_policy,
1463         .doit = dp_genl_add_del_port,
1464         .dumpit = NULL,
1465 };
1466
1467 static struct genl_ops dp_genl_ops_del_port = {
1468         .cmd = DP_GENL_C_DEL_PORT,
1469         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1470         .policy = dp_genl_policy,
1471         .doit = dp_genl_add_del_port,
1472         .dumpit = NULL,
1473 };
1474
1475 static int dp_genl_openflow(struct sk_buff *skb, struct genl_info *info)
1476 {
1477         struct nlattr *va = info->attrs[DP_GENL_A_OPENFLOW];
1478         struct datapath *dp;
1479         int err;
1480
1481         if (!info->attrs[DP_GENL_A_DP_IDX] || !va)
1482                 return -EINVAL;
1483
1484         rcu_read_lock();
1485         dp = dp_get(nla_get_u32(info->attrs[DP_GENL_A_DP_IDX]));
1486         if (!dp) {
1487                 err = -ENOENT;
1488                 goto out;
1489         }
1490
1491         va = info->attrs[DP_GENL_A_OPENFLOW];
1492
1493         err = fwd_control_input(dp->chain, nla_data(va), nla_len(va));
1494
1495 out:
1496         rcu_read_unlock();
1497         return err;
1498 }
1499
1500 static struct nla_policy dp_genl_openflow_policy[DP_GENL_A_MAX + 1] = {
1501         [DP_GENL_A_DP_IDX] = { .type = NLA_U32 },
1502 };
1503
1504 static struct genl_ops dp_genl_ops_openflow = {
1505         .cmd = DP_GENL_C_OPENFLOW,
1506         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1507         .policy = dp_genl_openflow_policy,
1508         .doit = dp_genl_openflow,
1509         .dumpit = NULL,
1510 };
1511
1512 static struct nla_policy dp_genl_benchmark_policy[DP_GENL_A_MAX + 1] = {
1513         [DP_GENL_A_DP_IDX] = { .type = NLA_U32 },
1514         [DP_GENL_A_NPACKETS] = { .type = NLA_U32 },
1515         [DP_GENL_A_PSIZE] = { .type = NLA_U32 },
1516 };
1517
1518 static struct genl_ops dp_genl_ops_benchmark_nl = {
1519         .cmd = DP_GENL_C_BENCHMARK_NL,
1520         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1521         .policy = dp_genl_benchmark_policy,
1522         .doit = dp_genl_benchmark_nl,
1523         .dumpit = NULL,
1524 };
1525
1526 static struct genl_ops *dp_genl_all_ops[] = {
1527         /* Keep this operation first.  Generic Netlink dispatching
1528          * looks up operations with linear search, so we want it at the
1529          * front. */
1530         &dp_genl_ops_openflow,
1531
1532         &dp_genl_ops_query_flow,
1533         &dp_genl_ops_query_table,
1534         &dp_genl_ops_show_dp,
1535         &dp_genl_ops_add_dp,
1536         &dp_genl_ops_del_dp,
1537         &dp_genl_ops_query_dp,
1538         &dp_genl_ops_add_port,
1539         &dp_genl_ops_del_port,
1540         &dp_genl_ops_benchmark_nl,
1541 };
1542
1543 static int dp_init_netlink(void)
1544 {
1545         int err;
1546         int i;
1547
1548         err = genl_register_family(&dp_genl_family);
1549         if (err)
1550                 return err;
1551
1552         for (i = 0; i < ARRAY_SIZE(dp_genl_all_ops); i++) {
1553                 err = genl_register_ops(&dp_genl_family, dp_genl_all_ops[i]);
1554                 if (err)
1555                         goto err_unregister;
1556         }
1557
1558         strcpy(mc_group.name, "openflow");
1559         err = genl_register_mc_group(&dp_genl_family, &mc_group);
1560         if (err < 0)
1561                 goto err_unregister;
1562
1563         return 0;
1564
1565 err_unregister:
1566         genl_unregister_family(&dp_genl_family);
1567                 return err;
1568 }
1569
1570 static void dp_uninit_netlink(void)
1571 {
1572         genl_unregister_family(&dp_genl_family);
1573 }
1574
1575 #define DRV_NAME                "openflow"
1576 #define DRV_VERSION      VERSION
1577 #define DRV_DESCRIPTION "OpenFlow switching datapath implementation"
1578 #define DRV_COPYRIGHT   "Copyright (c) 2007, 2008 The Board of Trustees of The Leland Stanford Junior University"
1579
1580
1581 static int __init dp_init(void)
1582 {
1583         int err;
1584
1585         printk(KERN_INFO DRV_NAME ": " DRV_DESCRIPTION "\n");
1586         printk(KERN_INFO DRV_NAME ": " VERSION" built on "__DATE__" "__TIME__"\n");
1587         printk(KERN_INFO DRV_NAME ": " DRV_COPYRIGHT "\n");
1588
1589         err = flow_init();
1590         if (err)
1591                 goto error;
1592
1593         err = dp_init_netlink();
1594         if (err)
1595                 goto error_flow_exit;
1596
1597         /* Hook into callback used by the bridge to intercept packets.
1598          * Parasites we are. */
1599         if (br_handle_frame_hook)
1600                 printk("openflow: hijacking bridge hook\n");
1601         br_handle_frame_hook = dp_frame_hook;
1602
1603         return 0;
1604
1605 error_flow_exit:
1606         flow_exit();
1607 error:
1608         printk(KERN_EMERG "openflow: failed to install!");
1609         return err;
1610 }
1611
1612 static void dp_cleanup(void)
1613 {
1614         fwd_exit();
1615         dp_uninit_netlink();
1616         flow_exit();
1617         br_handle_frame_hook = NULL;
1618 }
1619
1620 module_init(dp_init);
1621 module_exit(dp_cleanup);
1622
1623 MODULE_DESCRIPTION(DRV_DESCRIPTION);
1624 MODULE_AUTHOR(DRV_COPYRIGHT);
1625 MODULE_LICENSE("GPL");