datapath/datapath.c

   1 /*
   2  * Distributed under the terms of the GNU GPL version 2.
   3  * Copyright (c) 2007, 2008 The Board of Trustees of The Leland
   4  * Stanford Junior University
   5  */
   6
   7 /* Functions for managing the dp interface/device. */
   8
   9 #include <linux/init.h>
  10 #include <linux/module.h>
  11 #include <linux/if_arp.h>
  12 #include <linux/if_bridge.h>
  13 #include <linux/if_vlan.h>
  14 #include <linux/in.h>
  15 #include <net/genetlink.h>
  16 #include <linux/ip.h>
  17 #include <linux/delay.h>
  18 #include <linux/etherdevice.h>
  19 #include <linux/kernel.h>
  20 #include <linux/kthread.h>
  21 #include <linux/mutex.h>
  22 #include <linux/rtnetlink.h>
  23 #include <linux/rcupdate.h>
  24 #include <linux/version.h>
  25 #include <linux/ethtool.h>
  26 #include <linux/random.h>
  27 #include <asm/system.h>
  28 #include <linux/netfilter_bridge.h>
  29 #include <linux/inetdevice.h>
  30 #include <linux/list.h>
  31 #include <linux/rculist.h>
  32 #include <linux/workqueue.h>
  33
  34 #include "openflow-netlink.h"
  35 #include "datapath.h"
  36 #include "table.h"
  37 #include "chain.h"
  38 #include "dp_dev.h"
  39 #include "forward.h"
  40 #include "flow.h"
  41
  42 #include "compat.h"
  43
  44
  45 /* Strings to describe the manufacturer, hardware, and software.  This data
  46  * is queriable through the switch description stats message. */
  47 static char mfr_desc[DESC_STR_LEN] = "Nicira Networks";
  48 static char hw_desc[DESC_STR_LEN] = "Reference Linux Kernel Module";
  49 static char sw_desc[DESC_STR_LEN] = VERSION;
  50 static char serial_num[SERIAL_NUM_LEN] = "None";
  51
  52 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)
  53 module_param_string(mfr_desc, mfr_desc, sizeof mfr_desc, 0444);
  54 module_param_string(hw_desc, hw_desc, sizeof hw_desc, 0444);
  55 module_param_string(sw_desc, sw_desc, sizeof sw_desc, 0444);
  56 module_param_string(serial_num, serial_num, sizeof serial_num, 0444);
  57 #else
  58 MODULE_PARM(mfr_desc, "s");
  59 MODULE_PARM(hw_desc, "s");
  60 MODULE_PARM(sw_desc, "s");
  61 MODULE_PARM(serial_num, "s");
  62 #endif
  63
  64
  65 /* Number of milliseconds between runs of the maintenance thread. */
  66 #define MAINT_SLEEP_MSECS 1000
  67
  68 #define UINT32_MAX                        4294967295U
  69 #define UINT16_MAX                        65535
  70 #define MAX(X, Y) ((X) > (Y) ? (X) : (Y))
  71
  72 static struct genl_family dp_genl_family;
  73 static struct genl_multicast_group mc_group;
  74
  75 /* It's hard to imagine wanting more than one datapath, but... */
  76 #define DP_MAX 32
  77
  78 /* Datapaths.  Protected on the read side by rcu_read_lock, on the write side
  79  * by dp_mutex.  dp_mutex is almost completely redundant with genl_mutex
  80  * maintained by the Generic Netlink code, but the timeout path needs mutual
  81  * exclusion too.
  82  *
  83  * It is safe to access the datapath and net_bridge_port structures with just
  84  * dp_mutex.
  85  */
  86 static struct datapath *dps[DP_MAX];
  87 DEFINE_MUTEX(dp_mutex);
  88 EXPORT_SYMBOL(dp_mutex);
  89
  90 static int dp_maint_func(void *data);
  91 static void init_port_status(struct net_bridge_port *p);
  92 static int dp_genl_openflow_done(struct netlink_callback *);
  93 static struct net_bridge_port *new_nbp(struct datapath *,
  94                                        struct net_device *, int port_no);
  95
  96 /* nla_shrink - reduce amount of space reserved by nla_reserve
  97  * @skb: socket buffer from which to recover room
  98  * @nla: netlink attribute to adjust
  99  * @len: new length of attribute payload
 100  *
 101  * Reduces amount of space reserved by a call to nla_reserve.
 102  *
 103  * No other attributes may be added between calling nla_reserve and this
 104  * function, since it will create a hole in the message.
 105  */
 106 void nla_shrink(struct sk_buff *skb, struct nlattr *nla, int len)
 107 {
 108         int delta = nla_total_size(len) - nla_total_size(nla_len(nla));
 109         BUG_ON(delta > 0);
 110         skb->tail += delta;
 111         skb->len  += delta;
 112         nla->nla_len = nla_attr_size(len);
 113 }
 114
 115 /* Puts a set of openflow headers for a message of the given 'type' into 'skb'.
 116  * If 'sender' is nonnull, then it is used as the message's destination.  'dp'
 117  * must specify the datapath to use.
 118  *
 119  * '*max_openflow_len' receives the maximum number of bytes that are available
 120  * for the embedded OpenFlow message.  The caller must call
 121  * resize_openflow_skb() to set the actual size of the message to this number
 122  * of bytes or less.
 123  *
 124  * Returns the openflow header if successful, otherwise (if 'skb' is too small)
 125  * an error code. */
 126 static void *
 127 put_openflow_headers(struct datapath *dp, struct sk_buff *skb, uint8_t type,
 128                      const struct sender *sender, int *max_openflow_len)
 129 {
 130         struct ofp_header *oh;
 131         struct nlattr *attr;
 132         int openflow_len;
 133
 134         /* Assemble the Generic Netlink wrapper. */
 135         if (!genlmsg_put(skb,
 136                          sender ? sender->pid : 0,
 137                          sender ? sender->seq : 0,
 138                          &dp_genl_family, 0, DP_GENL_C_OPENFLOW))
 139                 return ERR_PTR(-ENOBUFS);
 140         if (nla_put_u32(skb, DP_GENL_A_DP_IDX, dp->dp_idx) < 0)
 141                 return ERR_PTR(-ENOBUFS);
 142         openflow_len = (skb_tailroom(skb) - NLA_HDRLEN) & ~(NLA_ALIGNTO - 1);
 143         if (openflow_len < sizeof *oh)
 144                 return ERR_PTR(-ENOBUFS);
 145         *max_openflow_len = openflow_len;
 146         attr = nla_reserve(skb, DP_GENL_A_OPENFLOW, openflow_len);
 147         BUG_ON(!attr);
 148
 149         /* Fill in the header.  The caller is responsible for the length. */
 150         oh = nla_data(attr);
 151         oh->version = OFP_VERSION;
 152         oh->type = type;
 153         oh->xid = sender ? sender->xid : 0;
 154
 155         return oh;
 156 }
 157
 158 /* Resizes OpenFlow header 'oh', which must be at the tail end of 'skb', to new
 159  * length 'new_length' (in bytes), adjusting pointers and size values as
 160  * necessary. */
 161 static void
 162 resize_openflow_skb(struct sk_buff *skb,
 163                     struct ofp_header *oh, size_t new_length)
 164 {
 165         struct nlattr *attr = ((void *) oh) - NLA_HDRLEN;
 166         nla_shrink(skb, attr, new_length);
 167         oh->length = htons(new_length);
 168         nlmsg_end(skb, (struct nlmsghdr *) skb->data);
 169 }
 170
 171 /* Allocates a new skb to contain an OpenFlow message 'openflow_len' bytes in
 172  * length.  Returns a null pointer if memory is unavailable, otherwise returns
 173  * the OpenFlow header and stores a pointer to the skb in '*pskb'.
 174  *
 175  * 'type' is the OpenFlow message type.  If 'sender' is nonnull, then it is
 176  * used as the message's destination.  'dp' must specify the datapath to
 177  * use.  */
 178 static void *
 179 alloc_openflow_skb(struct datapath *dp, size_t openflow_len, uint8_t type,
 180                    const struct sender *sender, struct sk_buff **pskb)
 181 {
 182         struct ofp_header *oh;
 183         size_t genl_len;
 184         struct sk_buff *skb;
 185         int max_openflow_len;
 186
 187         if ((openflow_len + sizeof(struct ofp_header)) > UINT16_MAX) {
 188                 if (net_ratelimit())
 189                         printk("alloc_openflow_skb: openflow message too large: %zu\n",
 190                                         openflow_len);
 191                 return NULL;
 192         }
 193
 194         genl_len = nlmsg_total_size(GENL_HDRLEN + dp_genl_family.hdrsize);
 195         genl_len += nla_total_size(sizeof(uint32_t)); /* DP_GENL_A_DP_IDX */
 196         genl_len += nla_total_size(openflow_len);    /* DP_GENL_A_OPENFLOW */
 197         skb = *pskb = genlmsg_new(genl_len, GFP_ATOMIC);
 198         if (!skb) {
 199                 if (net_ratelimit())
 200                         printk("alloc_openflow_skb: genlmsg_new failed\n");
 201                 return NULL;
 202         }
 203
 204         oh = put_openflow_headers(dp, skb, type, sender, &max_openflow_len);
 205         BUG_ON(!oh || IS_ERR(oh));
 206         resize_openflow_skb(skb, oh, openflow_len);
 207
 208         return oh;
 209 }
 210
 211 /* Sends 'skb' to 'sender' if it is nonnull, otherwise multicasts 'skb' to all
 212  * listeners. */
 213 static int
 214 send_openflow_skb(struct sk_buff *skb, const struct sender *sender)
 215 {
 216         return (sender
 217                 ? genlmsg_unicast(skb, sender->pid)
 218                 : genlmsg_multicast(skb, 0, mc_group.id, GFP_ATOMIC));
 219 }
 220
 221 /* Generates a unique datapath id.  It incorporates the datapath index
 222  * and a hardware address, if available.  If not, it generates a random
 223  * one.
 224  */
 225 static
 226 uint64_t gen_datapath_id(uint16_t dp_idx)
 227 {
 228         uint64_t id;
 229         int i;
 230         struct net_device *dev;
 231
 232         /* The top 16 bits are used to identify the datapath.  The lower 48 bits
 233          * use an interface address.  */
 234         id = (uint64_t)dp_idx << 48;
 235         if ((dev = dev_get_by_name(&init_net, "ctl0"))
 236                         || (dev = dev_get_by_name(&init_net, "eth0"))) {
 237                 for (i=0; i<ETH_ALEN; i++) {
 238                         id |= (uint64_t)dev->dev_addr[i] << (8*(ETH_ALEN-1 - i));
 239                 }
 240                 dev_put(dev);
 241         } else {
 242                 /* Randomly choose the lower 48 bits if we cannot find an
 243                  * address and mark the most significant bit to indicate that
 244                  * this was randomly generated. */
 245                 uint8_t rand[ETH_ALEN];
 246                 get_random_bytes(rand, ETH_ALEN);
 247                 id |= (uint64_t)1 << 63;
 248                 for (i=0; i<ETH_ALEN; i++) {
 249                         id |= (uint64_t)rand[i] << (8*(ETH_ALEN-1 - i));
 250                 }
 251         }
 252
 253         return id;
 254 }
 255
 256 /* Creates a new datapath numbered 'dp_idx'.  Returns 0 for success or a
 257  * negative error code. */
 258 static int new_dp(int dp_idx)
 259 {
 260         struct datapath *dp;
 261         int err;
 262
 263         if (dp_idx < 0 || dp_idx >= DP_MAX)
 264                 return -EINVAL;
 265
 266         if (!try_module_get(THIS_MODULE))
 267                 return -ENODEV;
 268
 269         /* Exit early if a datapath with that number already exists. */
 270         if (dps[dp_idx]) {
 271                 err = -EEXIST;
 272                 goto err_unlock;
 273         }
 274
 275         err = -ENOMEM;
 276         dp = kzalloc(sizeof *dp, GFP_KERNEL);
 277         if (dp == NULL)
 278                 goto err_unlock;
 279
 280         /* Setup our "of" device */
 281         err = dp_dev_setup(dp);
 282         if (err)
 283                 goto err_free_dp;
 284
 285         dp->dp_idx = dp_idx;
 286         dp->id = gen_datapath_id(dp_idx);
 287         dp->chain = chain_create(dp);
 288         if (dp->chain == NULL)
 289                 goto err_destroy_dp_dev;
 290         INIT_LIST_HEAD(&dp->port_list);
 291
 292         dp->local_port = new_nbp(dp, dp->netdev, OFPP_LOCAL);
 293         if (IS_ERR(dp->local_port)) {
 294                 err = PTR_ERR(dp->local_port);
 295                 goto err_destroy_local_port;
 296         }
 297
 298         dp->flags = 0;
 299         dp->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
 300
 301         dp->dp_task = kthread_run(dp_maint_func, dp, "dp%d", dp_idx);
 302         if (IS_ERR(dp->dp_task))
 303                 goto err_destroy_chain;
 304
 305         dps[dp_idx] = dp;
 306
 307         return 0;
 308
 309 err_destroy_local_port:
 310         dp_del_switch_port(dp->local_port);
 311 err_destroy_chain:
 312         chain_destroy(dp->chain);
 313 err_destroy_dp_dev:
 314         dp_dev_destroy(dp);
 315 err_free_dp:
 316         kfree(dp);
 317 err_unlock:
 318         module_put(THIS_MODULE);
 319                 return err;
 320 }
 321
 322 /* Find and return a free port number under 'dp'. */
 323 static int find_portno(struct datapath *dp)
 324 {
 325         int i;
 326         for (i = 0; i < OFPP_MAX; i++)
 327                 if (dp->ports[i] == NULL)
 328                         return i;
 329         return -EXFULL;
 330 }
 331
 332 static struct net_bridge_port *new_nbp(struct datapath *dp,
 333                                        struct net_device *dev, int port_no)
 334 {
 335         struct net_bridge_port *p;
 336
 337         if (dev->br_port != NULL)
 338                 return ERR_PTR(-EBUSY);
 339
 340         p = kzalloc(sizeof(*p), GFP_KERNEL);
 341         if (p == NULL)
 342                 return ERR_PTR(-ENOMEM);
 343
 344         rtnl_lock();
 345         dev_set_promiscuity(dev, 1);
 346         rtnl_unlock();
 347         dev_hold(dev);
 348         p->dp = dp;
 349         p->dev = dev;
 350         p->port_no = port_no;
 351         spin_lock_init(&p->lock);
 352         INIT_WORK(&p->port_task, NULL);
 353         if (port_no != OFPP_LOCAL)
 354                 rcu_assign_pointer(dev->br_port, p);
 355         if (port_no < OFPP_MAX)
 356                 rcu_assign_pointer(dp->ports[port_no], p);
 357         list_add_rcu(&p->node, &dp->port_list);
 358
 359         return p;
 360 }
 361
 362 int add_switch_port(struct datapath *dp, struct net_device *dev)
 363 {
 364         struct net_bridge_port *p;
 365         int port_no;
 366
 367         if (dev->flags & IFF_LOOPBACK || dev->type != ARPHRD_ETHER
 368             || is_dp_dev(dev))
 369                 return -EINVAL;
 370
 371         port_no = find_portno(dp);
 372         if (port_no < 0)
 373                 return port_no;
 374
 375         p = new_nbp(dp, dev, port_no);
 376         if (IS_ERR(p))
 377                 return PTR_ERR(p);
 378
 379         init_port_status(p);
 380
 381         /* Notify the ctlpath that this port has been added */
 382         dp_send_port_status(p, OFPPR_ADD);
 383
 384         return 0;
 385 }
 386
 387 /* Delete 'p' from switch. */
 388 int dp_del_switch_port(struct net_bridge_port *p)
 389 {
 390         /* First drop references to device. */
 391         cancel_work_sync(&p->port_task);
 392         rtnl_lock();
 393         dev_set_promiscuity(p->dev, -1);
 394         rtnl_unlock();
 395         list_del_rcu(&p->node);
 396         if (p->port_no != OFPP_LOCAL)
 397                 rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
 398         rcu_assign_pointer(p->dev->br_port, NULL);
 399
 400         /* Then wait until no one is still using it, and destroy it. */
 401         synchronize_rcu();
 402
 403         /* Notify the ctlpath that this port no longer exists */
 404         dp_send_port_status(p, OFPPR_DELETE);
 405
 406         dev_put(p->dev);
 407         kfree(p);
 408
 409         return 0;
 410 }
 411
 412 static void del_dp(struct datapath *dp)
 413 {
 414         struct net_bridge_port *p, *n;
 415
 416         kthread_stop(dp->dp_task);
 417
 418         /* Drop references to DP. */
 419         list_for_each_entry_safe (p, n, &dp->port_list, node)
 420                 dp_del_switch_port(p);
 421         rcu_assign_pointer(dps[dp->dp_idx], NULL);
 422
 423         /* Kill off local_port dev references from buffered packets that have
 424          * associated dst entries. */
 425         synchronize_rcu();
 426         fwd_discard_all();
 427
 428         /* Destroy dp->netdev.  (Must follow deleting switch ports since
 429          * dp->local_port has a reference to it.) */
 430         dp_dev_destroy(dp);
 431
 432         /* Wait until no longer in use, then destroy it. */
 433         synchronize_rcu();
 434         chain_destroy(dp->chain);
 435         kfree(dp);
 436         module_put(THIS_MODULE);
 437 }
 438
 439 static int dp_maint_func(void *data)
 440 {
 441         struct datapath *dp = (struct datapath *) data;
 442
 443         while (!kthread_should_stop()) {
 444                 /* Timeout old entries */
 445                 chain_timeout(dp->chain);
 446                 msleep_interruptible(MAINT_SLEEP_MSECS);
 447         }
 448
 449         return 0;
 450 }
 451
 452 static void
 453 do_port_input(struct net_bridge_port *p, struct sk_buff *skb)
 454 {
 455         /* Push the Ethernet header back on. */
 456         skb_push(skb, ETH_HLEN);
 457         fwd_port_input(p->dp->chain, skb, p);
 458 }
 459
 460 /*
 461  * Used as br_handle_frame_hook.  (Cannot run bridge at the same time, even on
 462  * different set of devices!)
 463  */
 464 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,22)
 465 /* Called with rcu_read_lock. */
 466 static struct sk_buff *dp_frame_hook(struct net_bridge_port *p,
 467                                          struct sk_buff *skb)
 468 {
 469         do_port_input(p, skb);
 470         return NULL;
 471 }
 472 #elif LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)
 473 static int dp_frame_hook(struct net_bridge_port *p, struct sk_buff **pskb)
 474 {
 475         do_port_input(p, *pskb);
 476         return 1;
 477 }
 478 #else
 479 /* NB: This has only been tested on 2.4.35 */
 480 static void dp_frame_hook(struct sk_buff *skb)
 481 {
 482         struct net_bridge_port *p = skb->dev->br_port;
 483         if (p) {
 484                 rcu_read_lock();
 485                 do_port_input(p, skb);
 486                 rcu_read_unlock();
 487         } else
 488                 kfree_skb(skb);
 489 }
 490 #endif
 491
 492 /* Forwarding output path.
 493  * Based on net/bridge/br_forward.c. */
 494
 495 static inline unsigned packet_length(const struct sk_buff *skb)
 496 {
 497         int length = skb->len - ETH_HLEN;
 498         if (skb->protocol == htons(ETH_P_8021Q))
 499                 length -= VLAN_HLEN;
 500         return length;
 501 }
 502
 503 /* Send packets out all the ports except the originating one.  If the
 504  * "flood" argument is set, only send along the minimum spanning tree.
 505  */
 506 static int
 507 output_all(struct datapath *dp, struct sk_buff *skb, int flood)
 508 {
 509         u32 disable = flood ? OFPPC_NO_FLOOD : 0;
 510         struct net_bridge_port *p;
 511         int prev_port = -1;
 512
 513         list_for_each_entry_rcu (p, &dp->port_list, node) {
 514                 if (skb->dev == p->dev || p->config & disable)
 515                         continue;
 516                 if (prev_port != -1) {
 517                         struct sk_buff *clone = skb_clone(skb, GFP_ATOMIC);
 518                         if (!clone) {
 519                                 kfree_skb(skb);
 520                                 return -ENOMEM;
 521                         }
 522                         dp_output_port(dp, clone, prev_port, 0);
 523                 }
 524                 prev_port = p->port_no;
 525         }
 526         if (prev_port != -1)
 527                 dp_output_port(dp, skb, prev_port, 0);
 528         else
 529                 kfree_skb(skb);
 530
 531         return 0;
 532 }
 533
 534 /* Marks 'skb' as having originated from 'in_port' in 'dp'.
 535    FIXME: how are devices reference counted? */
 536 void dp_set_origin(struct datapath *dp, uint16_t in_port,
 537                            struct sk_buff *skb)
 538 {
 539         struct net_bridge_port *p = (in_port < OFPP_MAX ? dp->ports[in_port]
 540                                      : in_port == OFPP_LOCAL ? dp->local_port
 541                                      : NULL);
 542         if (p)
 543                 skb->dev = p->dev;
 544          else
 545                 skb->dev = NULL;
 546 }
 547
 548 static int xmit_skb(struct sk_buff *skb)
 549 {
 550         int len = skb->len;
 551         if (packet_length(skb) > skb->dev->mtu) {
 552                 printk("dropped over-mtu packet: %d > %d\n",
 553                            packet_length(skb), skb->dev->mtu);
 554                 kfree_skb(skb);
 555                 return -E2BIG;
 556         }
 557
 558         dev_queue_xmit(skb);
 559
 560         return len;
 561 }
 562
 563 /* Takes ownership of 'skb' and transmits it to 'out_port' on 'dp'.
 564  */
 565 int dp_output_port(struct datapath *dp, struct sk_buff *skb, int out_port,
 566                    int ignore_no_fwd)
 567 {
 568         BUG_ON(!skb);
 569         switch (out_port){
 570         case OFPP_IN_PORT:
 571                 /* Send it out the port it came in on, which is already set in
 572                  * the skb. */
 573                 if (!skb->dev) {
 574                         if (net_ratelimit())
 575                                 printk("skb device not set forwarding to in_port\n");
 576                         kfree_skb(skb);
 577                         return -ESRCH;
 578                 }
 579                 return xmit_skb(skb);
 580
 581         case OFPP_TABLE: {
 582                 int retval = run_flow_through_tables(dp->chain, skb,
 583                                                      skb->dev->br_port);
 584                 if (retval)
 585                         kfree_skb(skb);
 586                 return retval;
 587         }
 588
 589         case OFPP_FLOOD:
 590                 return output_all(dp, skb, 1);
 591
 592         case OFPP_ALL:
 593                 return output_all(dp, skb, 0);
 594
 595         case OFPP_CONTROLLER:
 596                 return dp_output_control(dp, skb, fwd_save_skb(skb), 0,
 597                                                   OFPR_ACTION);
 598
 599         case OFPP_LOCAL: {
 600                 struct net_device *dev = dp->netdev;
 601                 return dev ? dp_dev_recv(dev, skb) : -ESRCH;
 602         }
 603
 604         case 0 ... OFPP_MAX-1: {
 605                 struct net_bridge_port *p = dp->ports[out_port];
 606                 if (p == NULL)
 607                         goto bad_port;
 608                 if (p->dev == skb->dev) {
 609                         /* To send to the input port, must use OFPP_IN_PORT */
 610                         kfree_skb(skb);
 611                         if (net_ratelimit())
 612                                 printk("can't directly forward to input port\n");
 613                         return -EINVAL;
 614                 }
 615                 if (p->config & OFPPC_NO_FWD && !ignore_no_fwd) {
 616                         kfree_skb(skb);
 617                         return 0;
 618                 }
 619                 skb->dev = p->dev;
 620                 return xmit_skb(skb);
 621         }
 622
 623         default:
 624                 goto bad_port;
 625         }
 626
 627 bad_port:
 628         kfree_skb(skb);
 629         if (net_ratelimit())
 630                 printk("can't forward to bad port %d\n", out_port);
 631         return -ENOENT;
 632 }
 633
 634 /* Takes ownership of 'skb' and transmits it to 'dp''s control path.  If
 635  * 'buffer_id' != -1, then only the first 64 bytes of 'skb' are sent;
 636  * otherwise, all of 'skb' is sent.  'reason' indicates why 'skb' is being
 637  * sent. 'max_len' sets the maximum number of bytes that the caller
 638  * wants to be sent; a value of 0 indicates the entire packet should be
 639  * sent. */
 640 int
 641 dp_output_control(struct datapath *dp, struct sk_buff *skb,
 642                            uint32_t buffer_id, size_t max_len, int reason)
 643 {
 644         /* FIXME?  Can we avoid creating a new skbuff in the case where we
 645          * forward the whole packet? */
 646         struct sk_buff *f_skb;
 647         struct ofp_packet_in *opi;
 648         struct net_bridge_port *p;
 649         size_t fwd_len, opi_len;
 650         int err;
 651
 652         fwd_len = skb->len;
 653         if ((buffer_id != (uint32_t) -1) && max_len)
 654                 fwd_len = min(fwd_len, max_len);
 655
 656         opi_len = offsetof(struct ofp_packet_in, data) + fwd_len;
 657         opi = alloc_openflow_skb(dp, opi_len, OFPT_PACKET_IN, NULL, &f_skb);
 658         if (!opi) {
 659                 err = -ENOMEM;
 660                 goto out;
 661         }
 662         opi->buffer_id      = htonl(buffer_id);
 663         opi->total_len      = htons(skb->len);
 664         p = skb->dev->br_port;
 665         opi->in_port        = htons(p ? p->port_no : OFPP_LOCAL);
 666         opi->reason         = reason;
 667         opi->pad            = 0;
 668         memcpy(opi->data, skb_mac_header(skb), fwd_len);
 669         err = send_openflow_skb(f_skb, NULL);
 670
 671 out:
 672         kfree_skb(skb);
 673         return err;
 674 }
 675
 676 static void fill_port_desc(struct net_bridge_port *p, struct ofp_phy_port *desc)
 677 {
 678         unsigned long flags;
 679         desc->port_no = htons(p->port_no);
 680         strncpy(desc->name, p->dev->name, OFP_MAX_PORT_NAME_LEN);
 681         desc->name[OFP_MAX_PORT_NAME_LEN-1] = '\0';
 682         memcpy(desc->hw_addr, p->dev->dev_addr, ETH_ALEN);
 683         desc->curr = 0;
 684         desc->supported = 0;
 685         desc->advertised = 0;
 686         desc->peer = 0;
 687
 688         spin_lock_irqsave(&p->lock, flags);
 689         desc->config = htonl(p->config);
 690         desc->state = htonl(p->state);
 691         spin_unlock_irqrestore(&p->lock, flags);
 692
 693 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,4,24)
 694         if (p->dev->ethtool_ops && p->dev->ethtool_ops->get_settings) {
 695                 struct ethtool_cmd ecmd = { .cmd = ETHTOOL_GSET };
 696
 697                 if (!p->dev->ethtool_ops->get_settings(p->dev, &ecmd)) {
 698                         /* Set the supported features */
 699                         if (ecmd.supported & SUPPORTED_10baseT_Half)
 700                                 desc->supported |= OFPPF_10MB_HD;
 701                         if (ecmd.supported & SUPPORTED_10baseT_Full)
 702                                 desc->supported |= OFPPF_10MB_FD;
 703                         if (ecmd.supported & SUPPORTED_100baseT_Half)
 704                                 desc->supported |= OFPPF_100MB_HD;
 705                         if (ecmd.supported & SUPPORTED_100baseT_Full)
 706                                 desc->supported |= OFPPF_100MB_FD;
 707                         if (ecmd.supported & SUPPORTED_1000baseT_Half)
 708                                 desc->supported |= OFPPF_1GB_HD;
 709                         if (ecmd.supported & SUPPORTED_1000baseT_Full)
 710                                 desc->supported |= OFPPF_1GB_FD;
 711                         if (ecmd.supported & SUPPORTED_10000baseT_Full)
 712                                 desc->supported |= OFPPF_10GB_FD;
 713                         if (ecmd.supported & SUPPORTED_TP)
 714                                 desc->supported |= OFPPF_COPPER;
 715                         if (ecmd.supported & SUPPORTED_FIBRE)
 716                                 desc->supported |= OFPPF_FIBER;
 717                         if (ecmd.supported & SUPPORTED_Autoneg)
 718                                 desc->supported |= OFPPF_AUTONEG;
 719 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,14)
 720                         if (ecmd.supported & SUPPORTED_Pause)
 721                                 desc->supported |= OFPPF_PAUSE;
 722                         if (ecmd.supported & SUPPORTED_Asym_Pause)
 723                                 desc->supported |= OFPPF_PAUSE_ASYM;
 724 #endif /* kernel >= 2.6.14 */
 725
 726                         /* Set the advertised features */
 727                         if (ecmd.advertising & ADVERTISED_10baseT_Half)
 728                                 desc->advertised |= OFPPF_10MB_HD;
 729                         if (ecmd.advertising & ADVERTISED_10baseT_Full)
 730                                 desc->advertised |= OFPPF_10MB_FD;
 731                         if (ecmd.advertising & ADVERTISED_100baseT_Half)
 732                                 desc->advertised |= OFPPF_100MB_HD;
 733                         if (ecmd.advertising & ADVERTISED_100baseT_Full)
 734                                 desc->advertised |= OFPPF_100MB_FD;
 735                         if (ecmd.advertising & ADVERTISED_1000baseT_Half)
 736                                 desc->advertised |= OFPPF_1GB_HD;
 737                         if (ecmd.advertising & ADVERTISED_1000baseT_Full)
 738                                 desc->advertised |= OFPPF_1GB_FD;
 739                         if (ecmd.advertising & ADVERTISED_10000baseT_Full)
 740                                 desc->advertised |= OFPPF_10GB_FD;
 741                         if (ecmd.advertising & ADVERTISED_TP)
 742                                 desc->advertised |= OFPPF_COPPER;
 743                         if (ecmd.advertising & ADVERTISED_FIBRE)
 744                                 desc->advertised |= OFPPF_FIBER;
 745                         if (ecmd.advertising & ADVERTISED_Autoneg)
 746                                 desc->advertised |= OFPPF_AUTONEG;
 747 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,14)
 748                         if (ecmd.advertising & ADVERTISED_Pause)
 749                                 desc->advertised |= OFPPF_PAUSE;
 750                         if (ecmd.advertising & ADVERTISED_Asym_Pause)
 751                                 desc->advertised |= OFPPF_PAUSE_ASYM;
 752 #endif /* kernel >= 2.6.14 */
 753
 754                         /* Set the current features */
 755                         if (ecmd.speed == SPEED_10)
 756                                 desc->curr = (ecmd.duplex) ? OFPPF_10MB_FD : OFPPF_10MB_HD;
 757                         else if (ecmd.speed == SPEED_100)
 758                                 desc->curr = (ecmd.duplex) ? OFPPF_100MB_FD : OFPPF_100MB_HD;
 759                         else if (ecmd.speed == SPEED_1000)
 760                                 desc->curr = (ecmd.duplex) ? OFPPF_1GB_FD : OFPPF_1GB_HD;
 761                         else if (ecmd.speed == SPEED_10000)
 762                                 desc->curr = OFPPF_10GB_FD;
 763
 764                         if (ecmd.port == PORT_TP)
 765                                 desc->curr |= OFPPF_COPPER;
 766                         else if (ecmd.port == PORT_FIBRE)
 767                                 desc->curr |= OFPPF_FIBER;
 768
 769                         if (ecmd.autoneg)
 770                                 desc->curr |= OFPPF_AUTONEG;
 771                 }
 772         }
 773 #endif
 774         desc->curr = htonl(desc->curr);
 775         desc->supported = htonl(desc->supported);
 776         desc->advertised = htonl(desc->advertised);
 777         desc->peer = htonl(desc->peer);
 778 }
 779
 780 static int
 781 fill_features_reply(struct datapath *dp, struct ofp_switch_features *ofr)
 782 {
 783         struct net_bridge_port *p;
 784         int port_count = 0;
 785
 786         ofr->datapath_id  = cpu_to_be64(dp->id);
 787
 788         ofr->n_buffers    = htonl(N_PKT_BUFFERS);
 789         ofr->n_tables     = dp->chain->n_tables;
 790         ofr->capabilities = htonl(OFP_SUPPORTED_CAPABILITIES);
 791         ofr->actions      = htonl(OFP_SUPPORTED_ACTIONS);
 792         memset(ofr->pad, 0, sizeof ofr->pad);
 793
 794         list_for_each_entry_rcu (p, &dp->port_list, node) {
 795                 fill_port_desc(p, &ofr->ports[port_count]);
 796                 port_count++;
 797         }
 798
 799         return port_count;
 800 }
 801
 802 int
 803 dp_send_features_reply(struct datapath *dp, const struct sender *sender)
 804 {
 805         struct sk_buff *skb;
 806         struct ofp_switch_features *ofr;
 807         size_t ofr_len, port_max_len;
 808         int port_count;
 809
 810         /* Overallocate. */
 811         port_max_len = sizeof(struct ofp_phy_port) * OFPP_MAX;
 812         ofr = alloc_openflow_skb(dp, sizeof(*ofr) + port_max_len,
 813                                  OFPT_FEATURES_REPLY, sender, &skb);
 814         if (!ofr)
 815                 return -ENOMEM;
 816
 817         /* Fill. */
 818         port_count = fill_features_reply(dp, ofr);
 819
 820         /* Shrink to fit. */
 821         ofr_len = sizeof(*ofr) + (sizeof(struct ofp_phy_port) * port_count);
 822         resize_openflow_skb(skb, &ofr->header, ofr_len);
 823         return send_openflow_skb(skb, sender);
 824 }
 825
 826 int
 827 dp_send_config_reply(struct datapath *dp, const struct sender *sender)
 828 {
 829         struct sk_buff *skb;
 830         struct ofp_switch_config *osc;
 831
 832         osc = alloc_openflow_skb(dp, sizeof *osc, OFPT_GET_CONFIG_REPLY, sender,
 833                                  &skb);
 834         if (!osc)
 835                 return -ENOMEM;
 836
 837         osc->flags = htons(dp->flags);
 838         osc->miss_send_len = htons(dp->miss_send_len);
 839
 840         return send_openflow_skb(skb, sender);
 841 }
 842
 843 int
 844 dp_send_hello(struct datapath *dp, const struct sender *sender,
 845               const struct ofp_header *request)
 846 {
 847         if (request->version < OFP_VERSION) {
 848                 char err[64];
 849                 sprintf(err, "Only version 0x%02x supported", OFP_VERSION);
 850                 dp_send_error_msg(dp, sender, OFPET_HELLO_FAILED,
 851                                   OFPHFC_INCOMPATIBLE, err, strlen(err));
 852                 return -EINVAL;
 853         } else {
 854                 struct sk_buff *skb;
 855                 struct ofp_header *reply;
 856
 857                 reply = alloc_openflow_skb(dp, sizeof *reply,
 858                                            OFPT_HELLO, sender, &skb);
 859                 if (!reply)
 860                         return -ENOMEM;
 861
 862                 return send_openflow_skb(skb, sender);
 863         }
 864 }
 865
 866 /* Callback function for a workqueue to disable an interface */
 867 static void
 868 down_port_cb(struct work_struct *work)
 869 {
 870         struct net_bridge_port *p = container_of(work, struct net_bridge_port,
 871                         port_task);
 872
 873         rtnl_lock();
 874         if (dev_change_flags(p->dev, p->dev->flags & ~IFF_UP) < 0)
 875                 if (net_ratelimit())
 876                         printk("problem bringing up port %s\n", p->dev->name);
 877         rtnl_unlock();
 878         p->config |= OFPPC_PORT_DOWN;
 879 }
 880
 881 /* Callback function for a workqueue to enable an interface */
 882 static void
 883 up_port_cb(struct work_struct *work)
 884 {
 885         struct net_bridge_port *p = container_of(work, struct net_bridge_port,
 886                         port_task);
 887
 888         rtnl_lock();
 889         if (dev_change_flags(p->dev, p->dev->flags | IFF_UP) < 0)
 890                 if (net_ratelimit())
 891                         printk("problem bringing down port %s\n", p->dev->name);
 892         rtnl_unlock();
 893         p->config &= ~OFPPC_PORT_DOWN;
 894 }
 895
 896 int
 897 dp_update_port_flags(struct datapath *dp, const struct ofp_port_mod *opm)
 898 {
 899         unsigned long int flags;
 900         int port_no = ntohs(opm->port_no);
 901         struct net_bridge_port *p = (port_no < OFPP_MAX ? dp->ports[port_no]
 902                                      : port_no == OFPP_LOCAL ? dp->local_port
 903                                      : NULL);
 904
 905         /* Make sure the port id hasn't changed since this was sent */
 906         if (!p || memcmp(opm->hw_addr, p->dev->dev_addr, ETH_ALEN))
 907                 return -1;
 908
 909         spin_lock_irqsave(&p->lock, flags);
 910         if (opm->mask) {
 911                 uint32_t config_mask = ntohl(opm->mask);
 912                 p->config &= ~config_mask;
 913                 p->config |= ntohl(opm->config) & config_mask;
 914         }
 915
 916         /* Modifying the status of an interface requires taking a lock
 917          * that cannot be done from here.  For this reason, we use a shared
 918          * workqueue, which will cause it to be executed from a safer
 919          * context. */
 920         if (opm->mask & htonl(OFPPC_PORT_DOWN)) {
 921                 if ((opm->config & htonl(OFPPC_PORT_DOWN))
 922                     && (p->config & OFPPC_PORT_DOWN) == 0) {
 923                         PREPARE_WORK(&p->port_task, down_port_cb);
 924                         schedule_work(&p->port_task);
 925                 } else if ((opm->config & htonl(OFPPC_PORT_DOWN)) == 0
 926                            && (p->config & OFPPC_PORT_DOWN)) {
 927                         PREPARE_WORK(&p->port_task, up_port_cb);
 928                         schedule_work(&p->port_task);
 929                 }
 930         }
 931         spin_unlock_irqrestore(&p->lock, flags);
 932
 933         return 0;
 934 }
 935
 936 /* Initialize the port status field of the bridge port. */
 937 static void
 938 init_port_status(struct net_bridge_port *p)
 939 {
 940         unsigned long int flags;
 941
 942         spin_lock_irqsave(&p->lock, flags);
 943
 944         if (p->dev->flags & IFF_UP)
 945                 p->config &= ~OFPPC_PORT_DOWN;
 946         else
 947                 p->config |= OFPPC_PORT_DOWN;
 948
 949         if (netif_carrier_ok(p->dev))
 950                 p->state &= ~OFPPS_LINK_DOWN;
 951         else
 952                 p->state |= OFPPS_LINK_DOWN;
 953
 954         spin_unlock_irqrestore(&p->lock, flags);
 955 }
 956
 957 int
 958 dp_send_port_status(struct net_bridge_port *p, uint8_t status)
 959 {
 960         struct sk_buff *skb;
 961         struct ofp_port_status *ops;
 962
 963         ops = alloc_openflow_skb(p->dp, sizeof *ops, OFPT_PORT_STATUS, NULL,
 964                                  &skb);
 965         if (!ops)
 966                 return -ENOMEM;
 967         ops->reason = status;
 968         memset(ops->pad, 0, sizeof ops->pad);
 969         fill_port_desc(p, &ops->desc);
 970
 971         return send_openflow_skb(skb, NULL);
 972 }
 973
 974 int
 975 dp_send_flow_expired(struct datapath *dp, struct sw_flow *flow,
 976                      enum ofp_flow_expired_reason reason)
 977 {
 978         struct sk_buff *skb;
 979         struct ofp_flow_expired *ofe;
 980
 981         if (!(dp->flags & OFPC_SEND_FLOW_EXP))
 982                 return 0;
 983
 984         ofe = alloc_openflow_skb(dp, sizeof *ofe, OFPT_FLOW_EXPIRED, 0, &skb);
 985         if (!ofe)
 986                 return -ENOMEM;
 987
 988         flow_fill_match(&ofe->match, &flow->key);
 989
 990         ofe->priority = htons(flow->priority);
 991         ofe->reason = reason;
 992         memset(ofe->pad, 0, sizeof ofe->pad);
 993
 994         ofe->duration     = htonl((jiffies - flow->init_time) / HZ);
 995         memset(ofe->pad2, 0, sizeof ofe->pad2);
 996         ofe->packet_count = cpu_to_be64(flow->packet_count);
 997         ofe->byte_count   = cpu_to_be64(flow->byte_count);
 998
 999         return send_openflow_skb(skb, NULL);
1000 }
1001 EXPORT_SYMBOL(dp_send_flow_expired);
1002
1003 int
1004 dp_send_error_msg(struct datapath *dp, const struct sender *sender,
1005                 uint16_t type, uint16_t code, const void *data, size_t len)
1006 {
1007         struct sk_buff *skb;
1008         struct ofp_error_msg *oem;
1009
1010
1011         oem = alloc_openflow_skb(dp, sizeof(*oem)+len, OFPT_ERROR,
1012                         sender, &skb);
1013         if (!oem)
1014                 return -ENOMEM;
1015
1016         oem->type = htons(type);
1017         oem->code = htons(code);
1018         memcpy(oem->data, data, len);
1019
1020         return send_openflow_skb(skb, sender);
1021 }
1022
1023 int
1024 dp_send_echo_reply(struct datapath *dp, const struct sender *sender,
1025                    const struct ofp_header *rq)
1026 {
1027         struct sk_buff *skb;
1028         struct ofp_header *reply;
1029
1030         reply = alloc_openflow_skb(dp, ntohs(rq->length), OFPT_ECHO_REPLY,
1031                                    sender, &skb);
1032         if (!reply)
1033                 return -ENOMEM;
1034
1035         memcpy(reply + 1, rq + 1, ntohs(rq->length) - sizeof *rq);
1036         return send_openflow_skb(skb, sender);
1037 }
1038
1039 /* Generic Netlink interface.
1040  *
1041  * See netlink(7) for an introduction to netlink.  See
1042  * http://linux-net.osdl.org/index.php/Netlink for more information and
1043  * pointers on how to work with netlink and Generic Netlink in the kernel and
1044  * in userspace. */
1045
1046 static struct genl_family dp_genl_family = {
1047         .id = GENL_ID_GENERATE,
1048         .hdrsize = 0,
1049         .name = DP_GENL_FAMILY_NAME,
1050         .version = 1,
1051         .maxattr = DP_GENL_A_MAX,
1052 };
1053
1054 /* Attribute policy: what each attribute may contain.  */
1055 static struct nla_policy dp_genl_policy[DP_GENL_A_MAX + 1] = {
1056         [DP_GENL_A_DP_IDX] = { .type = NLA_U32 },
1057         [DP_GENL_A_MC_GROUP] = { .type = NLA_U32 },
1058         [DP_GENL_A_PORTNAME] = { .type = NLA_STRING }
1059 };
1060
1061 static int dp_genl_add(struct sk_buff *skb, struct genl_info *info)
1062 {
1063         if (!info->attrs[DP_GENL_A_DP_IDX])
1064                 return -EINVAL;
1065
1066         return new_dp(nla_get_u32(info->attrs[DP_GENL_A_DP_IDX]));
1067 }
1068
1069 static struct genl_ops dp_genl_ops_add_dp = {
1070         .cmd = DP_GENL_C_ADD_DP,
1071         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1072         .policy = dp_genl_policy,
1073         .doit = dp_genl_add,
1074         .dumpit = NULL,
1075 };
1076
1077 struct datapath *dp_get(int dp_idx)
1078 {
1079         if (dp_idx < 0 || dp_idx > DP_MAX)
1080                 return NULL;
1081         return rcu_dereference(dps[dp_idx]);
1082 }
1083
1084 static int dp_genl_del(struct sk_buff *skb, struct genl_info *info)
1085 {
1086         struct datapath *dp;
1087         int err;
1088
1089         if (!info->attrs[DP_GENL_A_DP_IDX])
1090                 return -EINVAL;
1091
1092         dp = dp_get(nla_get_u32((info->attrs[DP_GENL_A_DP_IDX])));
1093         if (!dp)
1094                 err = -ENOENT;
1095         else {
1096                 del_dp(dp);
1097                 err = 0;
1098         }
1099         return err;
1100 }
1101
1102 static struct genl_ops dp_genl_ops_del_dp = {
1103         .cmd = DP_GENL_C_DEL_DP,
1104         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1105         .policy = dp_genl_policy,
1106         .doit = dp_genl_del,
1107         .dumpit = NULL,
1108 };
1109
1110 /* Queries a datapath for related information.  Currently the only relevant
1111  * information is the datapath's multicast group ID.  Really we want one
1112  * multicast group per datapath, but because of locking issues[*] we can't
1113  * easily get one.  Thus, every datapath will currently return the same
1114  * global multicast group ID, but in the future it would be nice to fix that.
1115  *
1116  * [*] dp_genl_add, to add a new datapath, is called under the genl_lock
1117  *       mutex, and genl_register_mc_group, called to acquire a new multicast
1118  *       group ID, also acquires genl_lock, thus deadlock.
1119  */
1120 static int dp_genl_query(struct sk_buff *skb, struct genl_info *info)
1121 {
1122         struct datapath *dp;
1123         struct sk_buff *ans_skb = NULL;
1124         int dp_idx;
1125         int err = -ENOMEM;
1126
1127         if (!info->attrs[DP_GENL_A_DP_IDX])
1128                 return -EINVAL;
1129
1130         rcu_read_lock();
1131         dp_idx = nla_get_u32((info->attrs[DP_GENL_A_DP_IDX]));
1132         dp = dp_get(dp_idx);
1133         if (!dp)
1134                 err = -ENOENT;
1135         else {
1136                 void *data;
1137                 ans_skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1138                 if (!ans_skb) {
1139                         err = -ENOMEM;
1140                         goto err;
1141                 }
1142                 data = genlmsg_put_reply(ans_skb, info, &dp_genl_family,
1143                                          0, DP_GENL_C_QUERY_DP);
1144                 if (data == NULL) {
1145                         err = -ENOMEM;
1146                         goto err;
1147                 }
1148                 NLA_PUT_U32(ans_skb, DP_GENL_A_DP_IDX, dp_idx);
1149                 NLA_PUT_U32(ans_skb, DP_GENL_A_MC_GROUP, mc_group.id);
1150
1151                 genlmsg_end(ans_skb, data);
1152                 err = genlmsg_reply(ans_skb, info);
1153                 if (!err)
1154                         ans_skb = NULL;
1155         }
1156 err:
1157 nla_put_failure:
1158         if (ans_skb)
1159                 kfree_skb(ans_skb);
1160         rcu_read_unlock();
1161         return err;
1162 }
1163
1164 static struct genl_ops dp_genl_ops_query_dp = {
1165         .cmd = DP_GENL_C_QUERY_DP,
1166         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1167         .policy = dp_genl_policy,
1168         .doit = dp_genl_query,
1169         .dumpit = NULL,
1170 };
1171
1172 static int dp_genl_add_del_port(struct sk_buff *skb, struct genl_info *info)
1173 {
1174         struct datapath *dp;
1175         struct net_device *port;
1176         int err;
1177
1178         if (!info->attrs[DP_GENL_A_DP_IDX] || !info->attrs[DP_GENL_A_PORTNAME])
1179                 return -EINVAL;
1180
1181         /* Get datapath. */
1182         dp = dp_get(nla_get_u32(info->attrs[DP_GENL_A_DP_IDX]));
1183         if (!dp) {
1184                 err = -ENOENT;
1185                 goto out;
1186         }
1187
1188         /* Get interface to add/remove. */
1189         port = dev_get_by_name(&init_net,
1190                         nla_data(info->attrs[DP_GENL_A_PORTNAME]));
1191         if (!port) {
1192                 err = -ENOENT;
1193                 goto out;
1194         }
1195
1196         /* Execute operation. */
1197         if (info->genlhdr->cmd == DP_GENL_C_ADD_PORT)
1198                 err = add_switch_port(dp, port);
1199         else {
1200                 if (port->br_port == NULL || port->br_port->dp != dp) {
1201                         err = -ENOENT;
1202                         goto out_put;
1203                 }
1204                 err = dp_del_switch_port(port->br_port);
1205         }
1206
1207 out_put:
1208         dev_put(port);
1209 out:
1210         return err;
1211 }
1212
1213 static struct genl_ops dp_genl_ops_add_port = {
1214         .cmd = DP_GENL_C_ADD_PORT,
1215         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1216         .policy = dp_genl_policy,
1217         .doit = dp_genl_add_del_port,
1218         .dumpit = NULL,
1219 };
1220
1221 static struct genl_ops dp_genl_ops_del_port = {
1222         .cmd = DP_GENL_C_DEL_PORT,
1223         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1224         .policy = dp_genl_policy,
1225         .doit = dp_genl_add_del_port,
1226         .dumpit = NULL,
1227 };
1228
1229 static int dp_genl_openflow(struct sk_buff *skb, struct genl_info *info)
1230 {
1231         struct nlattr *va = info->attrs[DP_GENL_A_OPENFLOW];
1232         struct datapath *dp;
1233         struct ofp_header *oh;
1234         struct sender sender;
1235         int err;
1236
1237         if (!info->attrs[DP_GENL_A_DP_IDX] || !va)
1238                 return -EINVAL;
1239
1240         dp = dp_get(nla_get_u32(info->attrs[DP_GENL_A_DP_IDX]));
1241         if (!dp)
1242                 return -ENOENT;
1243
1244         if (nla_len(va) < sizeof(struct ofp_header))
1245                 return -EINVAL;
1246         oh = nla_data(va);
1247
1248         sender.xid = oh->xid;
1249         sender.pid = info->snd_pid;
1250         sender.seq = info->snd_seq;
1251
1252         mutex_lock(&dp_mutex);
1253         err = fwd_control_input(dp->chain, &sender,
1254                                 nla_data(va), nla_len(va));
1255         mutex_unlock(&dp_mutex);
1256         return err;
1257 }
1258
1259 static struct nla_policy dp_genl_openflow_policy[DP_GENL_A_MAX + 1] = {
1260         [DP_GENL_A_DP_IDX] = { .type = NLA_U32 },
1261 };
1262
1263 static int desc_stats_dump(struct datapath *dp, void *state,
1264                             void *body, int *body_len)
1265 {
1266         struct ofp_desc_stats *ods = body;
1267         int n_bytes = sizeof *ods;
1268
1269         if (n_bytes > *body_len) {
1270                 return -ENOBUFS;
1271         }
1272         *body_len = n_bytes;
1273
1274         strncpy(ods->mfr_desc, mfr_desc, sizeof ods->mfr_desc);
1275         strncpy(ods->hw_desc, hw_desc, sizeof ods->hw_desc);
1276         strncpy(ods->sw_desc, sw_desc, sizeof ods->sw_desc);
1277         strncpy(ods->serial_num, serial_num, sizeof ods->serial_num);
1278
1279         return 0;
1280 }
1281
1282 struct flow_stats_state {
1283         int table_idx;
1284         struct sw_table_position position;
1285         const struct ofp_flow_stats_request *rq;
1286
1287         void *body;
1288         int bytes_used, bytes_allocated;
1289 };
1290
1291 static int flow_stats_init(struct datapath *dp, const void *body, int body_len,
1292                            void **state)
1293 {
1294         const struct ofp_flow_stats_request *fsr = body;
1295         struct flow_stats_state *s = kmalloc(sizeof *s, GFP_ATOMIC);
1296         if (!s)
1297                 return -ENOMEM;
1298         s->table_idx = fsr->table_id == 0xff ? 0 : fsr->table_id;
1299         memset(&s->position, 0, sizeof s->position);
1300         s->rq = fsr;
1301         *state = s;
1302         return 0;
1303 }
1304
1305 static int flow_stats_dump_callback(struct sw_flow *flow, void *private)
1306 {
1307         struct sw_flow_actions *sf_acts = rcu_dereference(flow->sf_acts);
1308         struct flow_stats_state *s = private;
1309         struct ofp_flow_stats *ofs;
1310         int length;
1311
1312         length = sizeof *ofs + sf_acts->actions_len;
1313         if (length + s->bytes_used > s->bytes_allocated)
1314                 return 1;
1315
1316         ofs = s->body + s->bytes_used;
1317         ofs->length          = htons(length);
1318         ofs->table_id        = s->table_idx;
1319         ofs->pad             = 0;
1320         ofs->match.wildcards = htonl(flow->key.wildcards);
1321         ofs->match.in_port   = flow->key.in_port;
1322         memcpy(ofs->match.dl_src, flow->key.dl_src, ETH_ALEN);
1323         memcpy(ofs->match.dl_dst, flow->key.dl_dst, ETH_ALEN);
1324         ofs->match.dl_vlan   = flow->key.dl_vlan;
1325         ofs->match.dl_type   = flow->key.dl_type;
1326         ofs->match.nw_src    = flow->key.nw_src;
1327         ofs->match.nw_dst    = flow->key.nw_dst;
1328         ofs->match.nw_proto  = flow->key.nw_proto;
1329         ofs->match.pad       = 0;
1330         ofs->match.tp_src    = flow->key.tp_src;
1331         ofs->match.tp_dst    = flow->key.tp_dst;
1332         ofs->duration        = htonl((jiffies - flow->init_time) / HZ);
1333         ofs->priority        = htons(flow->priority);
1334         ofs->idle_timeout    = htons(flow->idle_timeout);
1335         ofs->hard_timeout    = htons(flow->hard_timeout);
1336         memset(ofs->pad2, 0, sizeof ofs->pad2);
1337         ofs->packet_count    = cpu_to_be64(flow->packet_count);
1338         ofs->byte_count      = cpu_to_be64(flow->byte_count);
1339         memcpy(ofs->actions, sf_acts->actions, sf_acts->actions_len);
1340
1341         s->bytes_used += length;
1342         return 0;
1343 }
1344
1345 static int flow_stats_dump(struct datapath *dp, void *state,
1346                            void *body, int *body_len)
1347 {
1348         struct flow_stats_state *s = state;
1349         struct sw_flow_key match_key;
1350         int error = 0;
1351
1352         s->bytes_used = 0;
1353         s->bytes_allocated = *body_len;
1354         s->body = body;
1355
1356         flow_extract_match(&match_key, &s->rq->match);
1357         while (s->table_idx < dp->chain->n_tables
1358                && (s->rq->table_id == 0xff || s->rq->table_id == s->table_idx))
1359         {
1360                 struct sw_table *table = dp->chain->tables[s->table_idx];
1361
1362                 error = table->iterate(table, &match_key, &s->position,
1363                                        flow_stats_dump_callback, s);
1364                 if (error)
1365                         break;
1366
1367                 s->table_idx++;
1368                 memset(&s->position, 0, sizeof s->position);
1369         }
1370         *body_len = s->bytes_used;
1371
1372         /* If error is 0, we're done.
1373          * Otherwise, if some bytes were used, there are more flows to come.
1374          * Otherwise, we were not able to fit even a single flow in the body,
1375          * which indicates that we have a single flow with too many actions to
1376          * fit.  We won't ever make any progress at that rate, so give up. */
1377         return !error ? 0 : s->bytes_used ? 1 : -ENOMEM;
1378 }
1379
1380 static void flow_stats_done(void *state)
1381 {
1382         kfree(state);
1383 }
1384
1385 static int aggregate_stats_init(struct datapath *dp,
1386                                 const void *body, int body_len,
1387                                 void **state)
1388 {
1389         *state = (void *)body;
1390         return 0;
1391 }
1392
1393 static int aggregate_stats_dump_callback(struct sw_flow *flow, void *private)
1394 {
1395         struct ofp_aggregate_stats_reply *rpy = private;
1396         rpy->packet_count += flow->packet_count;
1397         rpy->byte_count += flow->byte_count;
1398         rpy->flow_count++;
1399         return 0;
1400 }
1401
1402 static int aggregate_stats_dump(struct datapath *dp, void *state,
1403                                 void *body, int *body_len)
1404 {
1405         struct ofp_aggregate_stats_request *rq = state;
1406         struct ofp_aggregate_stats_reply *rpy;
1407         struct sw_table_position position;
1408         struct sw_flow_key match_key;
1409         int table_idx;
1410
1411         if (*body_len < sizeof *rpy)
1412                 return -ENOBUFS;
1413         rpy = body;
1414         *body_len = sizeof *rpy;
1415
1416         memset(rpy, 0, sizeof *rpy);
1417
1418         flow_extract_match(&match_key, &rq->match);
1419         table_idx = rq->table_id == 0xff ? 0 : rq->table_id;
1420         memset(&position, 0, sizeof position);
1421         while (table_idx < dp->chain->n_tables
1422                && (rq->table_id == 0xff || rq->table_id == table_idx))
1423         {
1424                 struct sw_table *table = dp->chain->tables[table_idx];
1425                 int error;
1426
1427                 error = table->iterate(table, &match_key, &position,
1428                                        aggregate_stats_dump_callback, rpy);
1429                 if (error)
1430                         return error;
1431
1432                 table_idx++;
1433                 memset(&position, 0, sizeof position);
1434         }
1435
1436         rpy->packet_count = cpu_to_be64(rpy->packet_count);
1437         rpy->byte_count = cpu_to_be64(rpy->byte_count);
1438         rpy->flow_count = htonl(rpy->flow_count);
1439         return 0;
1440 }
1441
1442 static int table_stats_dump(struct datapath *dp, void *state,
1443                             void *body, int *body_len)
1444 {
1445         struct ofp_table_stats *ots;
1446         int n_bytes = dp->chain->n_tables * sizeof *ots;
1447         int i;
1448         if (n_bytes > *body_len)
1449                 return -ENOBUFS;
1450         *body_len = n_bytes;
1451         for (i = 0, ots = body; i < dp->chain->n_tables; i++, ots++) {
1452                 struct sw_table_stats stats;
1453                 dp->chain->tables[i]->stats(dp->chain->tables[i], &stats);
1454                 strncpy(ots->name, stats.name, sizeof ots->name);
1455                 ots->table_id = i;
1456                 ots->wildcards = htonl(stats.wildcards);
1457                 memset(ots->pad, 0, sizeof ots->pad);
1458                 ots->max_entries = htonl(stats.max_flows);
1459                 ots->active_count = htonl(stats.n_flows);
1460                 ots->lookup_count = cpu_to_be64(stats.n_lookup);
1461                 ots->matched_count = cpu_to_be64(stats.n_matched);
1462         }
1463         return 0;
1464 }
1465
1466 struct port_stats_state {
1467         int port;
1468 };
1469
1470 static int port_stats_init(struct datapath *dp, const void *body, int body_len,
1471                            void **state)
1472 {
1473         struct port_stats_state *s = kmalloc(sizeof *s, GFP_ATOMIC);
1474         if (!s)
1475                 return -ENOMEM;
1476         s->port = 0;
1477         *state = s;
1478         return 0;
1479 }
1480
1481 static int port_stats_dump(struct datapath *dp, void *state,
1482                            void *body, int *body_len)
1483 {
1484         struct port_stats_state *s = state;
1485         struct ofp_port_stats *ops;
1486         int n_ports, max_ports;
1487         int i;
1488
1489         max_ports = *body_len / sizeof *ops;
1490         if (!max_ports)
1491                 return -ENOMEM;
1492         ops = body;
1493
1494         n_ports = 0;
1495         for (i = s->port; i < OFPP_MAX && n_ports < max_ports; i++) {
1496                 struct net_bridge_port *p = dp->ports[i];
1497                 struct net_device_stats *stats;
1498                 if (!p)
1499                         continue;
1500                 stats = p->dev->get_stats(p->dev);
1501                 ops->port_no = htons(p->port_no);
1502                 memset(ops->pad, 0, sizeof ops->pad);
1503                 ops->rx_packets   = cpu_to_be64(stats->rx_packets);
1504                 ops->tx_packets   = cpu_to_be64(stats->tx_packets);
1505                 ops->rx_bytes     = cpu_to_be64(stats->rx_bytes);
1506                 ops->tx_bytes     = cpu_to_be64(stats->tx_bytes);
1507                 ops->rx_dropped   = cpu_to_be64(stats->rx_dropped);
1508                 ops->tx_dropped   = cpu_to_be64(stats->tx_dropped);
1509                 ops->rx_errors    = cpu_to_be64(stats->rx_errors);
1510                 ops->tx_errors    = cpu_to_be64(stats->tx_errors);
1511                 ops->rx_frame_err = cpu_to_be64(stats->rx_frame_errors);
1512                 ops->rx_over_err  = cpu_to_be64(stats->rx_over_errors);
1513                 ops->rx_crc_err   = cpu_to_be64(stats->rx_crc_errors);
1514                 ops->collisions   = cpu_to_be64(stats->collisions);
1515                 n_ports++;
1516                 ops++;
1517         }
1518         s->port = i;
1519         *body_len = n_ports * sizeof *ops;
1520         return n_ports >= max_ports;
1521 }
1522
1523 static void port_stats_done(void *state)
1524 {
1525         kfree(state);
1526 }
1527
1528 struct stats_type {
1529         /* Minimum and maximum acceptable number of bytes in body member of
1530          * struct ofp_stats_request. */
1531         size_t min_body, max_body;
1532
1533         /* Prepares to dump some kind of statistics on 'dp'.  'body' and
1534          * 'body_len' are the 'body' member of the struct ofp_stats_request.
1535          * Returns zero if successful, otherwise a negative error code.
1536          * May initialize '*state' to state information.  May be null if no
1537          * initialization is required.*/
1538         int (*init)(struct datapath *dp, const void *body, int body_len,
1539                     void **state);
1540
1541         /* Dumps statistics for 'dp' into the '*body_len' bytes at 'body', and
1542          * modifies '*body_len' to reflect the number of bytes actually used.
1543          * ('body' will be transmitted as the 'body' member of struct
1544          * ofp_stats_reply.) */
1545         int (*dump)(struct datapath *dp, void *state,
1546                     void *body, int *body_len);
1547
1548         /* Cleans any state created by the init or dump functions.  May be null
1549          * if no cleanup is required. */
1550         void (*done)(void *state);
1551 };
1552
1553 static const struct stats_type stats[] = {
1554         [OFPST_DESC] = {
1555                 0,
1556                 0,
1557                 NULL,
1558                 desc_stats_dump,
1559                 NULL
1560         },
1561         [OFPST_FLOW] = {
1562                 sizeof(struct ofp_flow_stats_request),
1563                 sizeof(struct ofp_flow_stats_request),
1564                 flow_stats_init,
1565                 flow_stats_dump,
1566                 flow_stats_done
1567         },
1568         [OFPST_AGGREGATE] = {
1569                 sizeof(struct ofp_aggregate_stats_request),
1570                 sizeof(struct ofp_aggregate_stats_request),
1571                 aggregate_stats_init,
1572                 aggregate_stats_dump,
1573                 NULL
1574         },
1575         [OFPST_TABLE] = {
1576                 0,
1577                 0,
1578                 NULL,
1579                 table_stats_dump,
1580                 NULL
1581         },
1582         [OFPST_PORT] = {
1583                 0,
1584                 0,
1585                 port_stats_init,
1586                 port_stats_dump,
1587                 port_stats_done
1588         },
1589 };
1590
1591 static int
1592 dp_genl_openflow_dumpit(struct sk_buff *skb, struct netlink_callback *cb)
1593 {
1594         struct datapath *dp;
1595         struct sender sender;
1596         const struct stats_type *s;
1597         struct ofp_stats_reply *osr;
1598         int dp_idx;
1599         int max_openflow_len, body_len;
1600         void *body;
1601         int err;
1602
1603         /* Set up the cleanup function for this dump.  Linux 2.6.20 and later
1604          * support setting up cleanup functions via the .doneit member of
1605          * struct genl_ops.  This kluge supports earlier versions also. */
1606         cb->done = dp_genl_openflow_done;
1607
1608         sender.pid = NETLINK_CB(cb->skb).pid;
1609         sender.seq = cb->nlh->nlmsg_seq;
1610         if (!cb->args[0]) {
1611                 struct nlattr *attrs[DP_GENL_A_MAX + 1];
1612                 struct ofp_stats_request *rq;
1613                 struct nlattr *va;
1614                 size_t len, body_len;
1615                 int type;
1616
1617                 err = nlmsg_parse(cb->nlh, GENL_HDRLEN, attrs, DP_GENL_A_MAX,
1618                                   dp_genl_openflow_policy);
1619                 if (err < 0)
1620                         return err;
1621
1622                 if (!attrs[DP_GENL_A_DP_IDX])
1623                         return -EINVAL;
1624                 dp_idx = nla_get_u16(attrs[DP_GENL_A_DP_IDX]);
1625                 dp = dp_get(dp_idx);
1626                 if (!dp)
1627                         return -ENOENT;
1628
1629                 va = attrs[DP_GENL_A_OPENFLOW];
1630                 len = nla_len(va);
1631                 if (!va || len < sizeof *rq)
1632                         return -EINVAL;
1633
1634                 rq = nla_data(va);
1635                 sender.xid = rq->header.xid;
1636                 type = ntohs(rq->type);
1637                 if (rq->header.version != OFP_VERSION) {
1638                         dp_send_error_msg(dp, &sender, OFPET_BAD_REQUEST,
1639                                           OFPBRC_BAD_VERSION, rq, len);
1640                         return -EINVAL;
1641                 }
1642                 if (rq->header.type != OFPT_STATS_REQUEST
1643                     || ntohs(rq->header.length) != len)
1644                         return -EINVAL;
1645
1646                 if (type >= ARRAY_SIZE(stats) || !stats[type].dump) {
1647                         dp_send_error_msg(dp, &sender, OFPET_BAD_REQUEST,
1648                                           OFPBRC_BAD_STAT, rq, len);
1649                         return -EINVAL;
1650                 }
1651
1652                 s = &stats[type];
1653                 body_len = len - offsetof(struct ofp_stats_request, body);
1654                 if (body_len < s->min_body || body_len > s->max_body)
1655                         return -EINVAL;
1656
1657                 cb->args[0] = 1;
1658                 cb->args[1] = dp_idx;
1659                 cb->args[2] = type;
1660                 cb->args[3] = rq->header.xid;
1661                 if (s->init) {
1662                         void *state;
1663                         err = s->init(dp, rq->body, body_len, &state);
1664                         if (err)
1665                                 return err;
1666                         cb->args[4] = (long) state;
1667                 }
1668         } else if (cb->args[0] == 1) {
1669                 sender.xid = cb->args[3];
1670                 dp_idx = cb->args[1];
1671                 s = &stats[cb->args[2]];
1672
1673                 dp = dp_get(dp_idx);
1674                 if (!dp)
1675                         return -ENOENT;
1676         } else {
1677                 return 0;
1678         }
1679
1680         osr = put_openflow_headers(dp, skb, OFPT_STATS_REPLY, &sender,
1681                                    &max_openflow_len);
1682         if (IS_ERR(osr))
1683                 return PTR_ERR(osr);
1684         osr->type = htons(s - stats);
1685         osr->flags = 0;
1686         resize_openflow_skb(skb, &osr->header, max_openflow_len);
1687         body = osr->body;
1688         body_len = max_openflow_len - offsetof(struct ofp_stats_reply, body);
1689
1690         err = s->dump(dp, (void *) cb->args[4], body, &body_len);
1691         if (err >= 0) {
1692                 if (!err)
1693                         cb->args[0] = 2;
1694                 else
1695                         osr->flags = ntohs(OFPSF_REPLY_MORE);
1696                 resize_openflow_skb(skb, &osr->header,
1697                                     (offsetof(struct ofp_stats_reply, body)
1698                                      + body_len));
1699                 err = skb->len;
1700         }
1701
1702         return err;
1703 }
1704
1705 static int
1706 dp_genl_openflow_done(struct netlink_callback *cb)
1707 {
1708         if (cb->args[0]) {
1709                 const struct stats_type *s = &stats[cb->args[2]];
1710                 if (s->done)
1711                         s->done((void *) cb->args[4]);
1712         }
1713         return 0;
1714 }
1715
1716 static struct genl_ops dp_genl_ops_openflow = {
1717         .cmd = DP_GENL_C_OPENFLOW,
1718         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1719         .policy = dp_genl_openflow_policy,
1720         .doit = dp_genl_openflow,
1721         .dumpit = dp_genl_openflow_dumpit,
1722 };
1723
1724 static struct genl_ops *dp_genl_all_ops[] = {
1725         /* Keep this operation first.  Generic Netlink dispatching
1726          * looks up operations with linear search, so we want it at the
1727          * front. */
1728         &dp_genl_ops_openflow,
1729
1730         &dp_genl_ops_add_dp,
1731         &dp_genl_ops_del_dp,
1732         &dp_genl_ops_query_dp,
1733         &dp_genl_ops_add_port,
1734         &dp_genl_ops_del_port,
1735 };
1736
1737 static int dp_init_netlink(void)
1738 {
1739         int err;
1740         int i;
1741
1742         err = genl_register_family(&dp_genl_family);
1743         if (err)
1744                 return err;
1745
1746         for (i = 0; i < ARRAY_SIZE(dp_genl_all_ops); i++) {
1747                 err = genl_register_ops(&dp_genl_family, dp_genl_all_ops[i]);
1748                 if (err)
1749                         goto err_unregister;
1750         }
1751
1752         strcpy(mc_group.name, "openflow");
1753         err = genl_register_mc_group(&dp_genl_family, &mc_group);
1754         if (err < 0)
1755                 goto err_unregister;
1756
1757         return 0;
1758
1759 err_unregister:
1760         genl_unregister_family(&dp_genl_family);
1761                 return err;
1762 }
1763
1764 static void dp_uninit_netlink(void)
1765 {
1766         genl_unregister_family(&dp_genl_family);
1767 }
1768
1769 static int __init dp_init(void)
1770 {
1771         int err;
1772
1773         printk("OpenFlow "VERSION", built "__DATE__" "__TIME__", "
1774                "protocol 0x%02x\n", OFP_VERSION);
1775
1776         err = flow_init();
1777         if (err)
1778                 goto error;
1779
1780         err = register_netdevice_notifier(&dp_device_notifier);
1781         if (err)
1782                 goto error_flow_exit;
1783
1784         err = dp_init_netlink();
1785         if (err)
1786                 goto error_unreg_notifier;
1787
1788         /* Hook into callback used by the bridge to intercept packets.
1789          * Parasites we are. */
1790         if (br_handle_frame_hook)
1791                 printk("openflow: hijacking bridge hook\n");
1792         br_handle_frame_hook = dp_frame_hook;
1793
1794         return 0;
1795
1796 error_unreg_notifier:
1797         unregister_netdevice_notifier(&dp_device_notifier);
1798 error_flow_exit:
1799         flow_exit();
1800 error:
1801         printk(KERN_EMERG "openflow: failed to install!");
1802         return err;
1803 }
1804
1805 static void dp_cleanup(void)
1806 {
1807         fwd_exit();
1808         dp_uninit_netlink();
1809         unregister_netdevice_notifier(&dp_device_notifier);
1810         flow_exit();
1811         br_handle_frame_hook = NULL;
1812 }
1813
1814 module_init(dp_init);
1815 module_exit(dp_cleanup);
1816
1817 MODULE_DESCRIPTION("OpenFlow switching datapath");
1818 MODULE_AUTHOR("Copyright (c) 2007, 2008 The Board of Trustees of The Leland Stanford Junior University");
1819 MODULE_LICENSE("GPL");