datapath/datapath.c

   1 /*
   2  * Distributed under the terms of the GNU GPL version 2.
   3  * Copyright (c) 2007, 2008 The Board of Trustees of The Leland
   4  * Stanford Junior University
   5  */
   6
   7 /* Functions for managing the dp interface/device. */
   8
   9 #include <linux/init.h>
  10 #include <linux/module.h>
  11 #include <linux/if_arp.h>
  12 #include <linux/if_bridge.h>
  13 #include <linux/if_vlan.h>
  14 #include <linux/in.h>
  15 #include <net/genetlink.h>
  16 #include <linux/ip.h>
  17 #include <linux/delay.h>
  18 #include <linux/etherdevice.h>
  19 #include <linux/kernel.h>
  20 #include <linux/kthread.h>
  21 #include <linux/mutex.h>
  22 #include <linux/rtnetlink.h>
  23 #include <linux/rcupdate.h>
  24 #include <linux/version.h>
  25 #include <linux/ethtool.h>
  26 #include <linux/random.h>
  27 #include <asm/system.h>
  28 #include <linux/netfilter_bridge.h>
  29 #include <linux/netfilter_ipv4.h>
  30 #include <linux/inetdevice.h>
  31 #include <linux/list.h>
  32 #include <linux/rculist.h>
  33 #include <linux/workqueue.h>
  34 #include <linux/dmi.h>
  35
  36 #include "openflow/nicira-ext.h"
  37 #include "openflow/openflow-netlink.h"
  38 #include "datapath.h"
  39 #include "nx_act_snat.h"
  40 #include "table.h"
  41 #include "chain.h"
  42 #include "dp_dev.h"
  43 #include "forward.h"
  44 #include "flow.h"
  45
  46 #include "compat.h"
  47
  48
  49 /* Strings to describe the manufacturer, hardware, and software.  This data
  50  * is queriable through the switch description stats message. */
  51 static char mfr_desc[DESC_STR_LEN] = "Nicira Networks, Inc.";
  52 static char hw_desc[DESC_STR_LEN] = "Reference Linux Kernel Module";
  53 static char sw_desc[DESC_STR_LEN] = VERSION BUILDNR;
  54 static char serial_num[SERIAL_NUM_LEN] = "None";
  55
  56 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)
  57 module_param_string(mfr_desc, mfr_desc, sizeof mfr_desc, 0444);
  58 module_param_string(hw_desc, hw_desc, sizeof hw_desc, 0444);
  59 module_param_string(sw_desc, sw_desc, sizeof sw_desc, 0444);
  60 module_param_string(serial_num, serial_num, sizeof serial_num, 0444);
  61 #else
  62 MODULE_PARM(mfr_desc, "s");
  63 MODULE_PARM(hw_desc, "s");
  64 MODULE_PARM(sw_desc, "s");
  65 MODULE_PARM(serial_num, "s");
  66 #endif
  67
  68
  69 /* Number of milliseconds between runs of the maintenance thread. */
  70 #define MAINT_SLEEP_MSECS 1000
  71
  72 #define UINT32_MAX                        4294967295U
  73 #define UINT16_MAX                        65535
  74 #define MAX(X, Y) ((X) > (Y) ? (X) : (Y))
  75
  76 static struct genl_family dp_genl_family;
  77 static struct genl_multicast_group mc_group;
  78
  79 /* It's hard to imagine wanting more than one datapath, but... */
  80 #define DP_MAX 32
  81
  82 /* Datapaths.  Protected on the read side by rcu_read_lock, on the write side
  83  * by dp_mutex.  dp_mutex is almost completely redundant with genl_mutex
  84  * maintained by the Generic Netlink code, but the timeout path needs mutual
  85  * exclusion too.
  86  *
  87  * It is safe to access the datapath and net_bridge_port structures with just
  88  * dp_mutex.
  89  */
  90 static struct datapath *dps[DP_MAX];
  91 DEFINE_MUTEX(dp_mutex);
  92 EXPORT_SYMBOL(dp_mutex);
  93
  94 static int dp_maint_func(void *data);
  95 static void init_port_status(struct net_bridge_port *p);
  96 static int dp_genl_openflow_done(struct netlink_callback *);
  97 static struct net_bridge_port *new_nbp(struct datapath *,
  98                                        struct net_device *, int port_no);
  99
 100 /* nla_shrink - reduce amount of space reserved by nla_reserve
 101  * @skb: socket buffer from which to recover room
 102  * @nla: netlink attribute to adjust
 103  * @len: new length of attribute payload
 104  *
 105  * Reduces amount of space reserved by a call to nla_reserve.
 106  *
 107  * No other attributes may be added between calling nla_reserve and this
 108  * function, since it will create a hole in the message.
 109  */
 110 void nla_shrink(struct sk_buff *skb, struct nlattr *nla, int len)
 111 {
 112         int delta = nla_total_size(len) - nla_total_size(nla_len(nla));
 113         BUG_ON(delta > 0);
 114         skb->tail += delta;
 115         skb->len  += delta;
 116         nla->nla_len = nla_attr_size(len);
 117 }
 118
 119 /* Puts a set of openflow headers for a message of the given 'type' into 'skb'.
 120  * If 'sender' is nonnull, then it is used as the message's destination.  'dp'
 121  * must specify the datapath to use.
 122  *
 123  * '*max_openflow_len' receives the maximum number of bytes that are available
 124  * for the embedded OpenFlow message.  The caller must call
 125  * resize_openflow_skb() to set the actual size of the message to this number
 126  * of bytes or less.
 127  *
 128  * Returns the openflow header if successful, otherwise (if 'skb' is too small)
 129  * an error code. */
 130 static void *
 131 put_openflow_headers(struct datapath *dp, struct sk_buff *skb, uint8_t type,
 132                      const struct sender *sender, int *max_openflow_len)
 133 {
 134         struct ofp_header *oh;
 135         struct nlattr *attr;
 136         int openflow_len;
 137
 138         /* Assemble the Generic Netlink wrapper. */
 139         if (!genlmsg_put(skb,
 140                          sender ? sender->pid : 0,
 141                          sender ? sender->seq : 0,
 142                          &dp_genl_family, 0, DP_GENL_C_OPENFLOW))
 143                 return ERR_PTR(-ENOBUFS);
 144         if (nla_put_u32(skb, DP_GENL_A_DP_IDX, dp->dp_idx) < 0)
 145                 return ERR_PTR(-ENOBUFS);
 146         openflow_len = (skb_tailroom(skb) - NLA_HDRLEN) & ~(NLA_ALIGNTO - 1);
 147         if (openflow_len < sizeof *oh)
 148                 return ERR_PTR(-ENOBUFS);
 149         *max_openflow_len = openflow_len;
 150         attr = nla_reserve(skb, DP_GENL_A_OPENFLOW, openflow_len);
 151         BUG_ON(!attr);
 152
 153         /* Fill in the header.  The caller is responsible for the length. */
 154         oh = nla_data(attr);
 155         oh->version = OFP_VERSION;
 156         oh->type = type;
 157         oh->xid = sender ? sender->xid : 0;
 158
 159         return oh;
 160 }
 161
 162 /* Resizes OpenFlow header 'oh', which must be at the tail end of 'skb', to new
 163  * length 'new_length' (in bytes), adjusting pointers and size values as
 164  * necessary. */
 165 static void
 166 resize_openflow_skb(struct sk_buff *skb,
 167                     struct ofp_header *oh, size_t new_length)
 168 {
 169         struct nlattr *attr = ((void *) oh) - NLA_HDRLEN;
 170         nla_shrink(skb, attr, new_length);
 171         oh->length = htons(new_length);
 172         nlmsg_end(skb, (struct nlmsghdr *) skb->data);
 173 }
 174
 175 /* Allocates a new skb to contain an OpenFlow message 'openflow_len' bytes in
 176  * length.  Returns a null pointer if memory is unavailable, otherwise returns
 177  * the OpenFlow header and stores a pointer to the skb in '*pskb'.
 178  *
 179  * 'type' is the OpenFlow message type.  If 'sender' is nonnull, then it is
 180  * used as the message's destination.  'dp' must specify the datapath to
 181  * use.  */
 182 static void *
 183 alloc_openflow_skb(struct datapath *dp, size_t openflow_len, uint8_t type,
 184                    const struct sender *sender, struct sk_buff **pskb)
 185 {
 186         struct ofp_header *oh;
 187         size_t genl_len;
 188         struct sk_buff *skb;
 189         int max_openflow_len;
 190
 191         if ((openflow_len + sizeof(struct ofp_header)) > UINT16_MAX) {
 192                 if (net_ratelimit())
 193                         printk("alloc_openflow_skb: openflow message too large: %zu\n",
 194                                         openflow_len);
 195                 return NULL;
 196         }
 197
 198         genl_len = nlmsg_total_size(GENL_HDRLEN + dp_genl_family.hdrsize);
 199         genl_len += nla_total_size(sizeof(uint32_t)); /* DP_GENL_A_DP_IDX */
 200         genl_len += nla_total_size(openflow_len);    /* DP_GENL_A_OPENFLOW */
 201         skb = *pskb = genlmsg_new(genl_len, GFP_ATOMIC);
 202         if (!skb) {
 203                 if (net_ratelimit())
 204                         printk("alloc_openflow_skb: genlmsg_new failed\n");
 205                 return NULL;
 206         }
 207
 208         oh = put_openflow_headers(dp, skb, type, sender, &max_openflow_len);
 209         BUG_ON(!oh || IS_ERR(oh));
 210         resize_openflow_skb(skb, oh, openflow_len);
 211
 212         return oh;
 213 }
 214
 215 /* Sends 'skb' to 'sender' if it is nonnull, otherwise multicasts 'skb' to all
 216  * listeners. */
 217 static int
 218 send_openflow_skb(struct sk_buff *skb, const struct sender *sender)
 219 {
 220         return (sender
 221                 ? genlmsg_unicast(skb, sender->pid)
 222                 : genlmsg_multicast(skb, 0, mc_group.id, GFP_ATOMIC));
 223 }
 224
 225 /* Retrieves the datapath id, which is the MAC address of the "of" device. */
 226 static
 227 uint64_t get_datapath_id(struct net_device *dev)
 228 {
 229         uint64_t id = 0;
 230         int i;
 231
 232         for (i=0; i<ETH_ALEN; i++)
 233                 id |= (uint64_t)dev->dev_addr[i] << (8*(ETH_ALEN-1 - i));
 234
 235         return id;
 236 }
 237
 238 /* Creates a new datapath numbered 'dp_idx'.  Returns 0 for success or a
 239  * negative error code. */
 240 static int new_dp(int dp_idx)
 241 {
 242         struct datapath *dp;
 243         int err;
 244
 245         if (dp_idx < 0 || dp_idx >= DP_MAX)
 246                 return -EINVAL;
 247
 248         if (!try_module_get(THIS_MODULE))
 249                 return -ENODEV;
 250
 251         /* Exit early if a datapath with that number already exists. */
 252         if (dps[dp_idx]) {
 253                 err = -EEXIST;
 254                 goto err_unlock;
 255         }
 256
 257         err = -ENOMEM;
 258         dp = kzalloc(sizeof *dp, GFP_KERNEL);
 259         if (dp == NULL)
 260                 goto err_unlock;
 261
 262         /* Setup our "of" device */
 263         err = dp_dev_setup(dp);
 264         if (err)
 265                 goto err_free_dp;
 266
 267         dp->dp_idx = dp_idx;
 268         dp->id = get_datapath_id(dp->netdev);
 269         dp->chain = chain_create(dp);
 270         if (dp->chain == NULL)
 271                 goto err_destroy_dp_dev;
 272         INIT_LIST_HEAD(&dp->port_list);
 273
 274         dp->local_port = new_nbp(dp, dp->netdev, OFPP_LOCAL);
 275         if (IS_ERR(dp->local_port)) {
 276                 err = PTR_ERR(dp->local_port);
 277                 goto err_destroy_local_port;
 278         }
 279
 280         dp->flags = 0;
 281         dp->miss_send_len = OFP_DEFAULT_MISS_SEND_LEN;
 282
 283         dp->dp_task = kthread_run(dp_maint_func, dp, "dp%d", dp_idx);
 284         if (IS_ERR(dp->dp_task))
 285                 goto err_destroy_chain;
 286
 287         dps[dp_idx] = dp;
 288
 289         return 0;
 290
 291 err_destroy_local_port:
 292         dp_del_switch_port(dp->local_port);
 293 err_destroy_chain:
 294         chain_destroy(dp->chain);
 295 err_destroy_dp_dev:
 296         dp_dev_destroy(dp);
 297 err_free_dp:
 298         kfree(dp);
 299 err_unlock:
 300         module_put(THIS_MODULE);
 301                 return err;
 302 }
 303
 304 /* Find and return a free port number under 'dp'. */
 305 static int find_portno(struct datapath *dp)
 306 {
 307         int i;
 308         for (i = 0; i < DP_MAX_PORTS; i++)
 309                 if (dp->ports[i] == NULL)
 310                         return i;
 311         return -EXFULL;
 312 }
 313
 314 static struct net_bridge_port *new_nbp(struct datapath *dp,
 315                                        struct net_device *dev, int port_no)
 316 {
 317         struct net_bridge_port *p;
 318
 319         if (dev->br_port != NULL)
 320                 return ERR_PTR(-EBUSY);
 321
 322         p = kzalloc(sizeof(*p), GFP_KERNEL);
 323         if (p == NULL)
 324                 return ERR_PTR(-ENOMEM);
 325
 326         rtnl_lock();
 327         dev_set_promiscuity(dev, 1);
 328         rtnl_unlock();
 329         dev_hold(dev);
 330         p->dp = dp;
 331         p->dev = dev;
 332         p->port_no = port_no;
 333         spin_lock_init(&p->lock);
 334         INIT_WORK(&p->port_task, NULL);
 335         if (port_no != OFPP_LOCAL)
 336                 rcu_assign_pointer(dev->br_port, p);
 337         if (port_no < DP_MAX_PORTS)
 338                 rcu_assign_pointer(dp->ports[port_no], p);
 339         list_add_rcu(&p->node, &dp->port_list);
 340
 341         return p;
 342 }
 343
 344 int add_switch_port(struct datapath *dp, struct net_device *dev)
 345 {
 346         struct net_bridge_port *p;
 347         int port_no;
 348
 349         if (dev->flags & IFF_LOOPBACK || dev->type != ARPHRD_ETHER
 350             || is_dp_dev(dev))
 351                 return -EINVAL;
 352
 353         port_no = find_portno(dp);
 354         if (port_no < 0)
 355                 return port_no;
 356
 357         p = new_nbp(dp, dev, port_no);
 358         if (IS_ERR(p))
 359                 return PTR_ERR(p);
 360
 361         init_port_status(p);
 362
 363         /* Notify the ctlpath that this port has been added */
 364         dp_send_port_status(p, OFPPR_ADD);
 365
 366         return 0;
 367 }
 368
 369 /* Delete 'p' from switch. */
 370 int dp_del_switch_port(struct net_bridge_port *p)
 371 {
 372 #ifdef SUPPORT_SNAT
 373         unsigned long flags;
 374 #endif
 375
 376         /* First drop references to device. */
 377         cancel_work_sync(&p->port_task);
 378         rtnl_lock();
 379         dev_set_promiscuity(p->dev, -1);
 380         rtnl_unlock();
 381         list_del_rcu(&p->node);
 382         if (p->port_no != OFPP_LOCAL)
 383                 rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
 384         rcu_assign_pointer(p->dev->br_port, NULL);
 385
 386         /* Then wait until no one is still using it, and destroy it. */
 387         synchronize_rcu();
 388
 389 #ifdef SUPPORT_SNAT
 390         /* Free any SNAT configuration on the port. */
 391         spin_lock_irqsave(&p->lock, flags);
 392         snat_free_conf(p);
 393         spin_unlock_irqrestore(&p->lock, flags);
 394 #endif
 395
 396         /* Notify the ctlpath that this port no longer exists */
 397         dp_send_port_status(p, OFPPR_DELETE);
 398
 399         dev_put(p->dev);
 400         kfree(p);
 401
 402         return 0;
 403 }
 404
 405 static void del_dp(struct datapath *dp)
 406 {
 407         struct net_bridge_port *p, *n;
 408
 409         kthread_stop(dp->dp_task);
 410
 411         /* Drop references to DP. */
 412         list_for_each_entry_safe (p, n, &dp->port_list, node)
 413                 dp_del_switch_port(p);
 414         rcu_assign_pointer(dps[dp->dp_idx], NULL);
 415
 416         /* Kill off local_port dev references from buffered packets that have
 417          * associated dst entries. */
 418         synchronize_rcu();
 419         fwd_discard_all();
 420
 421         /* Destroy dp->netdev.  (Must follow deleting switch ports since
 422          * dp->local_port has a reference to it.) */
 423         dp_dev_destroy(dp);
 424
 425         /* Wait until no longer in use, then destroy it. */
 426         synchronize_rcu();
 427         chain_destroy(dp->chain);
 428         kfree(dp);
 429         module_put(THIS_MODULE);
 430 }
 431
 432 static int dp_maint_func(void *data)
 433 {
 434         struct datapath *dp = (struct datapath *) data;
 435
 436         while (!kthread_should_stop()) {
 437 #ifdef SUPPORT_SNAT
 438                 struct net_bridge_port *p;
 439
 440                 /* Expire old SNAT entries */
 441                 rcu_read_lock();
 442                 list_for_each_entry_rcu (p, &dp->port_list, node)
 443                         snat_maint(p);
 444                 rcu_read_unlock();
 445 #endif
 446
 447                 /* Timeout old entries */
 448                 chain_timeout(dp->chain);
 449                 msleep_interruptible(MAINT_SLEEP_MSECS);
 450         }
 451
 452         return 0;
 453 }
 454
 455 static void
 456 do_port_input(struct net_bridge_port *p, struct sk_buff *skb)
 457 {
 458 #ifdef SUPPORT_SNAT
 459         /* Check if this packet needs early SNAT processing. */
 460         if (snat_pre_route(skb)) {
 461                 kfree_skb(skb);
 462                 return;
 463         }
 464 #endif
 465
 466         /* Push the Ethernet header back on. */
 467         skb_push(skb, ETH_HLEN);
 468         fwd_port_input(p->dp->chain, skb, p);
 469 }
 470
 471 /*
 472  * Used as br_handle_frame_hook.  (Cannot run bridge at the same time, even on
 473  * different set of devices!)
 474  */
 475 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,22)
 476 /* Called with rcu_read_lock. */
 477 static struct sk_buff *dp_frame_hook(struct net_bridge_port *p,
 478                                          struct sk_buff *skb)
 479 {
 480         do_port_input(p, skb);
 481         return NULL;
 482 }
 483 #elif LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)
 484 static int dp_frame_hook(struct net_bridge_port *p, struct sk_buff **pskb)
 485 {
 486         do_port_input(p, *pskb);
 487         return 1;
 488 }
 489 #else
 490 /* NB: This has only been tested on 2.4.35 */
 491 static void dp_frame_hook(struct sk_buff *skb)
 492 {
 493         struct net_bridge_port *p = skb->dev->br_port;
 494         if (p) {
 495                 rcu_read_lock();
 496                 do_port_input(p, skb);
 497                 rcu_read_unlock();
 498         } else
 499                 kfree_skb(skb);
 500 }
 501 #endif
 502
 503 /* Forwarding output path.
 504  * Based on net/bridge/br_forward.c. */
 505
 506 static inline unsigned packet_length(const struct sk_buff *skb)
 507 {
 508         int length = skb->len - ETH_HLEN;
 509         if (skb->protocol == htons(ETH_P_8021Q))
 510                 length -= VLAN_HLEN;
 511         return length;
 512 }
 513
 514 /* Send packets out all the ports except the originating one.  If the
 515  * "flood" argument is set, only send along the minimum spanning tree.
 516  */
 517 static int
 518 output_all(struct datapath *dp, struct sk_buff *skb, int flood)
 519 {
 520         u32 disable = flood ? OFPPC_NO_FLOOD : 0;
 521         struct net_bridge_port *p;
 522         int prev_port = -1;
 523
 524         list_for_each_entry_rcu (p, &dp->port_list, node) {
 525                 if (skb->dev == p->dev || p->config & disable)
 526                         continue;
 527                 if (prev_port != -1) {
 528                         struct sk_buff *clone = skb_clone(skb, GFP_ATOMIC);
 529                         if (!clone) {
 530                                 kfree_skb(skb);
 531                                 return -ENOMEM;
 532                         }
 533                         dp_output_port(dp, clone, prev_port, 0);
 534                 }
 535                 prev_port = p->port_no;
 536         }
 537         if (prev_port != -1)
 538                 dp_output_port(dp, skb, prev_port, 0);
 539         else
 540                 kfree_skb(skb);
 541
 542         return 0;
 543 }
 544
 545 /* Marks 'skb' as having originated from 'in_port' in 'dp'.
 546    FIXME: how are devices reference counted? */
 547 void dp_set_origin(struct datapath *dp, uint16_t in_port,
 548                            struct sk_buff *skb)
 549 {
 550         struct net_bridge_port *p;
 551         p = (in_port < DP_MAX_PORTS ? dp->ports[in_port]
 552              : in_port == OFPP_LOCAL ? dp->local_port
 553              : NULL);
 554         if (p)
 555                 skb->dev = p->dev;
 556          else
 557                 skb->dev = NULL;
 558 }
 559
 560 int
 561 dp_xmit_skb(struct sk_buff *skb)
 562 {
 563         int len = skb->len;
 564         if (packet_length(skb) > skb->dev->mtu) {
 565                 printk("dropped over-mtu packet: %d > %d\n",
 566                            packet_length(skb), skb->dev->mtu);
 567                 kfree_skb(skb);
 568                 return -E2BIG;
 569         }
 570
 571         dev_queue_xmit(skb);
 572
 573         return len;
 574 }
 575
 576 /* Takes ownership of 'skb' and transmits it to 'out_port' on 'dp'.
 577  */
 578 int dp_output_port(struct datapath *dp, struct sk_buff *skb, int out_port,
 579                    int ignore_no_fwd)
 580 {
 581         BUG_ON(!skb);
 582         switch (out_port){
 583         case OFPP_IN_PORT:
 584                 /* Send it out the port it came in on, which is already set in
 585                  * the skb. */
 586                 if (!skb->dev) {
 587                         if (net_ratelimit())
 588                                 printk("skb device not set forwarding to in_port\n");
 589                         kfree_skb(skb);
 590                         return -ESRCH;
 591                 }
 592                 return dp_xmit_skb(skb);
 593
 594         case OFPP_TABLE: {
 595                 int retval = run_flow_through_tables(dp->chain, skb,
 596                                                      skb->dev->br_port);
 597                 if (retval)
 598                         kfree_skb(skb);
 599                 return retval;
 600         }
 601
 602         case OFPP_FLOOD:
 603                 return output_all(dp, skb, 1);
 604
 605         case OFPP_ALL:
 606                 return output_all(dp, skb, 0);
 607
 608         case OFPP_CONTROLLER:
 609                 return dp_output_control(dp, skb, fwd_save_skb(skb), 0,
 610                                                   OFPR_ACTION);
 611
 612         case OFPP_LOCAL: {
 613                 struct net_device *dev = dp->netdev;
 614 #ifdef SUPPORT_SNAT
 615                 snat_local_in(skb);
 616 #endif
 617                 return dev ? dp_dev_recv(dev, skb) : -ESRCH;
 618         }
 619
 620         case 0 ... DP_MAX_PORTS - 1: {
 621                 struct net_bridge_port *p = dp->ports[out_port];
 622                 if (p == NULL)
 623                         goto bad_port;
 624                 if (p->dev == skb->dev) {
 625                         /* To send to the input port, must use OFPP_IN_PORT */
 626                         kfree_skb(skb);
 627                         if (net_ratelimit())
 628                                 printk("can't directly forward to input port\n");
 629                         return -EINVAL;
 630                 }
 631                 if (p->config & OFPPC_NO_FWD && !ignore_no_fwd) {
 632                         kfree_skb(skb);
 633                         return 0;
 634                 }
 635                 skb->dev = p->dev;
 636                 return dp_xmit_skb(skb);
 637         }
 638
 639         default:
 640                 goto bad_port;
 641         }
 642
 643 bad_port:
 644         kfree_skb(skb);
 645         if (net_ratelimit())
 646                 printk("can't forward to bad port %d\n", out_port);
 647         return -ENOENT;
 648 }
 649
 650 /* Takes ownership of 'skb' and transmits it to 'dp''s control path.  If
 651  * 'buffer_id' != -1, then only the first 64 bytes of 'skb' are sent;
 652  * otherwise, all of 'skb' is sent.  'reason' indicates why 'skb' is being
 653  * sent. 'max_len' sets the maximum number of bytes that the caller
 654  * wants to be sent; a value of 0 indicates the entire packet should be
 655  * sent. */
 656 int
 657 dp_output_control(struct datapath *dp, struct sk_buff *skb,
 658                            uint32_t buffer_id, size_t max_len, int reason)
 659 {
 660         /* FIXME?  Can we avoid creating a new skbuff in the case where we
 661          * forward the whole packet? */
 662         struct sk_buff *f_skb;
 663         struct ofp_packet_in *opi;
 664         struct net_bridge_port *p;
 665         size_t fwd_len, opi_len;
 666         int err;
 667
 668         fwd_len = skb->len;
 669         if ((buffer_id != (uint32_t) -1) && max_len)
 670                 fwd_len = min(fwd_len, max_len);
 671
 672         opi_len = offsetof(struct ofp_packet_in, data) + fwd_len;
 673         opi = alloc_openflow_skb(dp, opi_len, OFPT_PACKET_IN, NULL, &f_skb);
 674         if (!opi) {
 675                 err = -ENOMEM;
 676                 goto out;
 677         }
 678         opi->buffer_id      = htonl(buffer_id);
 679         opi->total_len      = htons(skb->len);
 680         p = skb->dev->br_port;
 681         opi->in_port        = htons(p ? p->port_no : OFPP_LOCAL);
 682         opi->reason         = reason;
 683         opi->pad            = 0;
 684         memcpy(opi->data, skb_mac_header(skb), fwd_len);
 685         err = send_openflow_skb(f_skb, NULL);
 686
 687 out:
 688         kfree_skb(skb);
 689         return err;
 690 }
 691
 692 static void fill_port_desc(struct net_bridge_port *p, struct ofp_phy_port *desc)
 693 {
 694         unsigned long flags;
 695         desc->port_no = htons(p->port_no);
 696         strncpy(desc->name, p->dev->name, OFP_MAX_PORT_NAME_LEN);
 697         desc->name[OFP_MAX_PORT_NAME_LEN-1] = '\0';
 698         memcpy(desc->hw_addr, p->dev->dev_addr, ETH_ALEN);
 699         desc->curr = 0;
 700         desc->supported = 0;
 701         desc->advertised = 0;
 702         desc->peer = 0;
 703
 704         spin_lock_irqsave(&p->lock, flags);
 705         desc->config = htonl(p->config);
 706         desc->state = htonl(p->state);
 707         spin_unlock_irqrestore(&p->lock, flags);
 708
 709 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,4,24)
 710         if (p->dev->ethtool_ops && p->dev->ethtool_ops->get_settings) {
 711                 struct ethtool_cmd ecmd = { .cmd = ETHTOOL_GSET };
 712
 713                 if (!p->dev->ethtool_ops->get_settings(p->dev, &ecmd)) {
 714                         /* Set the supported features */
 715                         if (ecmd.supported & SUPPORTED_10baseT_Half)
 716                                 desc->supported |= OFPPF_10MB_HD;
 717                         if (ecmd.supported & SUPPORTED_10baseT_Full)
 718                                 desc->supported |= OFPPF_10MB_FD;
 719                         if (ecmd.supported & SUPPORTED_100baseT_Half)
 720                                 desc->supported |= OFPPF_100MB_HD;
 721                         if (ecmd.supported & SUPPORTED_100baseT_Full)
 722                                 desc->supported |= OFPPF_100MB_FD;
 723                         if (ecmd.supported & SUPPORTED_1000baseT_Half)
 724                                 desc->supported |= OFPPF_1GB_HD;
 725                         if (ecmd.supported & SUPPORTED_1000baseT_Full)
 726                                 desc->supported |= OFPPF_1GB_FD;
 727                         if (ecmd.supported & SUPPORTED_10000baseT_Full)
 728                                 desc->supported |= OFPPF_10GB_FD;
 729                         if (ecmd.supported & SUPPORTED_TP)
 730                                 desc->supported |= OFPPF_COPPER;
 731                         if (ecmd.supported & SUPPORTED_FIBRE)
 732                                 desc->supported |= OFPPF_FIBER;
 733                         if (ecmd.supported & SUPPORTED_Autoneg)
 734                                 desc->supported |= OFPPF_AUTONEG;
 735 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,14)
 736                         if (ecmd.supported & SUPPORTED_Pause)
 737                                 desc->supported |= OFPPF_PAUSE;
 738                         if (ecmd.supported & SUPPORTED_Asym_Pause)
 739                                 desc->supported |= OFPPF_PAUSE_ASYM;
 740 #endif /* kernel >= 2.6.14 */
 741
 742                         /* Set the advertised features */
 743                         if (ecmd.advertising & ADVERTISED_10baseT_Half)
 744                                 desc->advertised |= OFPPF_10MB_HD;
 745                         if (ecmd.advertising & ADVERTISED_10baseT_Full)
 746                                 desc->advertised |= OFPPF_10MB_FD;
 747                         if (ecmd.advertising & ADVERTISED_100baseT_Half)
 748                                 desc->advertised |= OFPPF_100MB_HD;
 749                         if (ecmd.advertising & ADVERTISED_100baseT_Full)
 750                                 desc->advertised |= OFPPF_100MB_FD;
 751                         if (ecmd.advertising & ADVERTISED_1000baseT_Half)
 752                                 desc->advertised |= OFPPF_1GB_HD;
 753                         if (ecmd.advertising & ADVERTISED_1000baseT_Full)
 754                                 desc->advertised |= OFPPF_1GB_FD;
 755                         if (ecmd.advertising & ADVERTISED_10000baseT_Full)
 756                                 desc->advertised |= OFPPF_10GB_FD;
 757                         if (ecmd.advertising & ADVERTISED_TP)
 758                                 desc->advertised |= OFPPF_COPPER;
 759                         if (ecmd.advertising & ADVERTISED_FIBRE)
 760                                 desc->advertised |= OFPPF_FIBER;
 761                         if (ecmd.advertising & ADVERTISED_Autoneg)
 762                                 desc->advertised |= OFPPF_AUTONEG;
 763 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,14)
 764                         if (ecmd.advertising & ADVERTISED_Pause)
 765                                 desc->advertised |= OFPPF_PAUSE;
 766                         if (ecmd.advertising & ADVERTISED_Asym_Pause)
 767                                 desc->advertised |= OFPPF_PAUSE_ASYM;
 768 #endif /* kernel >= 2.6.14 */
 769
 770                         /* Set the current features */
 771                         if (ecmd.speed == SPEED_10)
 772                                 desc->curr = (ecmd.duplex) ? OFPPF_10MB_FD : OFPPF_10MB_HD;
 773                         else if (ecmd.speed == SPEED_100)
 774                                 desc->curr = (ecmd.duplex) ? OFPPF_100MB_FD : OFPPF_100MB_HD;
 775                         else if (ecmd.speed == SPEED_1000)
 776                                 desc->curr = (ecmd.duplex) ? OFPPF_1GB_FD : OFPPF_1GB_HD;
 777                         else if (ecmd.speed == SPEED_10000)
 778                                 desc->curr = OFPPF_10GB_FD;
 779
 780                         if (ecmd.port == PORT_TP)
 781                                 desc->curr |= OFPPF_COPPER;
 782                         else if (ecmd.port == PORT_FIBRE)
 783                                 desc->curr |= OFPPF_FIBER;
 784
 785                         if (ecmd.autoneg)
 786                                 desc->curr |= OFPPF_AUTONEG;
 787                 }
 788         }
 789 #endif
 790         desc->curr = htonl(desc->curr);
 791         desc->supported = htonl(desc->supported);
 792         desc->advertised = htonl(desc->advertised);
 793         desc->peer = htonl(desc->peer);
 794 }
 795
 796 static int
 797 fill_features_reply(struct datapath *dp, struct ofp_switch_features *ofr)
 798 {
 799         struct net_bridge_port *p;
 800         int port_count = 0;
 801
 802         ofr->datapath_id  = cpu_to_be64(dp->id);
 803
 804         ofr->n_buffers    = htonl(N_PKT_BUFFERS);
 805         ofr->n_tables     = dp->chain->n_tables;
 806         ofr->capabilities = htonl(OFP_SUPPORTED_CAPABILITIES);
 807         ofr->actions      = htonl(OFP_SUPPORTED_ACTIONS);
 808         memset(ofr->pad, 0, sizeof ofr->pad);
 809
 810         list_for_each_entry_rcu (p, &dp->port_list, node) {
 811                 fill_port_desc(p, &ofr->ports[port_count]);
 812                 port_count++;
 813         }
 814
 815         return port_count;
 816 }
 817
 818 int
 819 dp_send_features_reply(struct datapath *dp, const struct sender *sender)
 820 {
 821         struct sk_buff *skb;
 822         struct ofp_switch_features *ofr;
 823         size_t ofr_len, port_max_len;
 824         int port_count;
 825
 826         /* Overallocate. */
 827         port_max_len = sizeof(struct ofp_phy_port) * DP_MAX_PORTS;
 828         ofr = alloc_openflow_skb(dp, sizeof(*ofr) + port_max_len,
 829                                  OFPT_FEATURES_REPLY, sender, &skb);
 830         if (!ofr)
 831                 return -ENOMEM;
 832
 833         /* Fill. */
 834         port_count = fill_features_reply(dp, ofr);
 835
 836         /* Shrink to fit. */
 837         ofr_len = sizeof(*ofr) + (sizeof(struct ofp_phy_port) * port_count);
 838         resize_openflow_skb(skb, &ofr->header, ofr_len);
 839         return send_openflow_skb(skb, sender);
 840 }
 841
 842 int
 843 dp_send_config_reply(struct datapath *dp, const struct sender *sender)
 844 {
 845         struct sk_buff *skb;
 846         struct ofp_switch_config *osc;
 847
 848         osc = alloc_openflow_skb(dp, sizeof *osc, OFPT_GET_CONFIG_REPLY, sender,
 849                                  &skb);
 850         if (!osc)
 851                 return -ENOMEM;
 852
 853         osc->flags = htons(dp->flags);
 854         osc->miss_send_len = htons(dp->miss_send_len);
 855
 856         return send_openflow_skb(skb, sender);
 857 }
 858
 859 int
 860 dp_send_hello(struct datapath *dp, const struct sender *sender,
 861               const struct ofp_header *request)
 862 {
 863         if (request->version < OFP_VERSION) {
 864                 char err[64];
 865                 sprintf(err, "Only version 0x%02x supported", OFP_VERSION);
 866                 dp_send_error_msg(dp, sender, OFPET_HELLO_FAILED,
 867                                   OFPHFC_INCOMPATIBLE, err, strlen(err));
 868                 return -EINVAL;
 869         } else {
 870                 struct sk_buff *skb;
 871                 struct ofp_header *reply;
 872
 873                 reply = alloc_openflow_skb(dp, sizeof *reply,
 874                                            OFPT_HELLO, sender, &skb);
 875                 if (!reply)
 876                         return -ENOMEM;
 877
 878                 return send_openflow_skb(skb, sender);
 879         }
 880 }
 881
 882 /* Callback function for a workqueue to disable an interface */
 883 static void
 884 down_port_cb(struct work_struct *work)
 885 {
 886         struct net_bridge_port *p = container_of(work, struct net_bridge_port,
 887                         port_task);
 888
 889         rtnl_lock();
 890         if (dev_change_flags(p->dev, p->dev->flags & ~IFF_UP) < 0)
 891                 if (net_ratelimit())
 892                         printk("problem bringing up port %s\n", p->dev->name);
 893         rtnl_unlock();
 894         p->config |= OFPPC_PORT_DOWN;
 895 }
 896
 897 /* Callback function for a workqueue to enable an interface */
 898 static void
 899 up_port_cb(struct work_struct *work)
 900 {
 901         struct net_bridge_port *p = container_of(work, struct net_bridge_port,
 902                         port_task);
 903
 904         rtnl_lock();
 905         if (dev_change_flags(p->dev, p->dev->flags | IFF_UP) < 0)
 906                 if (net_ratelimit())
 907                         printk("problem bringing down port %s\n", p->dev->name);
 908         rtnl_unlock();
 909         p->config &= ~OFPPC_PORT_DOWN;
 910 }
 911
 912 int
 913 dp_update_port_flags(struct datapath *dp, const struct ofp_port_mod *opm)
 914 {
 915         unsigned long int flags;
 916         int port_no = ntohs(opm->port_no);
 917         struct net_bridge_port *p;
 918         p = (port_no < DP_MAX_PORTS ? dp->ports[port_no]
 919              : port_no == OFPP_LOCAL ? dp->local_port
 920              : NULL);
 921
 922         /* Make sure the port id hasn't changed since this was sent */
 923         if (!p || memcmp(opm->hw_addr, p->dev->dev_addr, ETH_ALEN))
 924                 return -1;
 925
 926         spin_lock_irqsave(&p->lock, flags);
 927         if (opm->mask) {
 928                 uint32_t config_mask = ntohl(opm->mask);
 929                 p->config &= ~config_mask;
 930                 p->config |= ntohl(opm->config) & config_mask;
 931         }
 932
 933         /* Modifying the status of an interface requires taking a lock
 934          * that cannot be done from here.  For this reason, we use a shared
 935          * workqueue, which will cause it to be executed from a safer
 936          * context. */
 937         if (opm->mask & htonl(OFPPC_PORT_DOWN)) {
 938                 if ((opm->config & htonl(OFPPC_PORT_DOWN))
 939                     && (p->config & OFPPC_PORT_DOWN) == 0) {
 940                         PREPARE_WORK(&p->port_task, down_port_cb);
 941                         schedule_work(&p->port_task);
 942                 } else if ((opm->config & htonl(OFPPC_PORT_DOWN)) == 0
 943                            && (p->config & OFPPC_PORT_DOWN)) {
 944                         PREPARE_WORK(&p->port_task, up_port_cb);
 945                         schedule_work(&p->port_task);
 946                 }
 947         }
 948         spin_unlock_irqrestore(&p->lock, flags);
 949
 950         return 0;
 951 }
 952
 953 /* Initialize the port status field of the bridge port. */
 954 static void
 955 init_port_status(struct net_bridge_port *p)
 956 {
 957         unsigned long int flags;
 958
 959         spin_lock_irqsave(&p->lock, flags);
 960
 961         if (p->dev->flags & IFF_UP)
 962                 p->config &= ~OFPPC_PORT_DOWN;
 963         else
 964                 p->config |= OFPPC_PORT_DOWN;
 965
 966         if (netif_carrier_ok(p->dev))
 967                 p->state &= ~OFPPS_LINK_DOWN;
 968         else
 969                 p->state |= OFPPS_LINK_DOWN;
 970
 971         spin_unlock_irqrestore(&p->lock, flags);
 972 }
 973
 974 int
 975 dp_send_port_status(struct net_bridge_port *p, uint8_t status)
 976 {
 977         struct sk_buff *skb;
 978         struct ofp_port_status *ops;
 979
 980         ops = alloc_openflow_skb(p->dp, sizeof *ops, OFPT_PORT_STATUS, NULL,
 981                                  &skb);
 982         if (!ops)
 983                 return -ENOMEM;
 984         ops->reason = status;
 985         memset(ops->pad, 0, sizeof ops->pad);
 986         fill_port_desc(p, &ops->desc);
 987
 988         return send_openflow_skb(skb, NULL);
 989 }
 990
 991 int
 992 dp_send_flow_expired(struct datapath *dp, struct sw_flow *flow,
 993                      enum ofp_flow_expired_reason reason)
 994 {
 995         struct sk_buff *skb;
 996         struct ofp_flow_expired *ofe;
 997
 998         if (!(dp->flags & OFPC_SEND_FLOW_EXP))
 999                 return 0;
1000
1001         ofe = alloc_openflow_skb(dp, sizeof *ofe, OFPT_FLOW_EXPIRED, 0, &skb);
1002         if (!ofe)
1003                 return -ENOMEM;
1004
1005         flow_fill_match(&ofe->match, &flow->key);
1006
1007         ofe->priority = htons(flow->priority);
1008         ofe->reason = reason;
1009         memset(ofe->pad, 0, sizeof ofe->pad);
1010
1011         ofe->duration     = htonl((jiffies - flow->init_time) / HZ);
1012         memset(ofe->pad2, 0, sizeof ofe->pad2);
1013         ofe->packet_count = cpu_to_be64(flow->packet_count);
1014         ofe->byte_count   = cpu_to_be64(flow->byte_count);
1015
1016         return send_openflow_skb(skb, NULL);
1017 }
1018 EXPORT_SYMBOL(dp_send_flow_expired);
1019
1020 int
1021 dp_send_error_msg(struct datapath *dp, const struct sender *sender,
1022                 uint16_t type, uint16_t code, const void *data, size_t len)
1023 {
1024         struct sk_buff *skb;
1025         struct ofp_error_msg *oem;
1026
1027
1028         oem = alloc_openflow_skb(dp, sizeof(*oem)+len, OFPT_ERROR,
1029                         sender, &skb);
1030         if (!oem)
1031                 return -ENOMEM;
1032
1033         oem->type = htons(type);
1034         oem->code = htons(code);
1035         memcpy(oem->data, data, len);
1036
1037         return send_openflow_skb(skb, sender);
1038 }
1039
1040 int
1041 dp_send_echo_reply(struct datapath *dp, const struct sender *sender,
1042                    const struct ofp_header *rq)
1043 {
1044         struct sk_buff *skb;
1045         struct ofp_header *reply;
1046
1047         reply = alloc_openflow_skb(dp, ntohs(rq->length), OFPT_ECHO_REPLY,
1048                                    sender, &skb);
1049         if (!reply)
1050                 return -ENOMEM;
1051
1052         memcpy(reply + 1, rq + 1, ntohs(rq->length) - sizeof *rq);
1053         return send_openflow_skb(skb, sender);
1054 }
1055
1056 /* Generic Netlink interface.
1057  *
1058  * See netlink(7) for an introduction to netlink.  See
1059  * http://linux-net.osdl.org/index.php/Netlink for more information and
1060  * pointers on how to work with netlink and Generic Netlink in the kernel and
1061  * in userspace. */
1062
1063 static struct genl_family dp_genl_family = {
1064         .id = GENL_ID_GENERATE,
1065         .hdrsize = 0,
1066         .name = DP_GENL_FAMILY_NAME,
1067         .version = 1,
1068         .maxattr = DP_GENL_A_MAX,
1069 };
1070
1071 /* Attribute policy: what each attribute may contain.  */
1072 static struct nla_policy dp_genl_policy[DP_GENL_A_MAX + 1] = {
1073         [DP_GENL_A_DP_IDX] = { .type = NLA_U32 },
1074         [DP_GENL_A_MC_GROUP] = { .type = NLA_U32 },
1075         [DP_GENL_A_PORTNAME] = { .type = NLA_STRING }
1076 };
1077
1078 static int dp_genl_add(struct sk_buff *skb, struct genl_info *info)
1079 {
1080         if (!info->attrs[DP_GENL_A_DP_IDX])
1081                 return -EINVAL;
1082
1083         return new_dp(nla_get_u32(info->attrs[DP_GENL_A_DP_IDX]));
1084 }
1085
1086 static struct genl_ops dp_genl_ops_add_dp = {
1087         .cmd = DP_GENL_C_ADD_DP,
1088         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1089         .policy = dp_genl_policy,
1090         .doit = dp_genl_add,
1091         .dumpit = NULL,
1092 };
1093
1094 struct datapath *dp_get(int dp_idx)
1095 {
1096         if (dp_idx < 0 || dp_idx > DP_MAX)
1097                 return NULL;
1098         return rcu_dereference(dps[dp_idx]);
1099 }
1100
1101 static int dp_genl_del(struct sk_buff *skb, struct genl_info *info)
1102 {
1103         struct datapath *dp;
1104         int err;
1105
1106         if (!info->attrs[DP_GENL_A_DP_IDX])
1107                 return -EINVAL;
1108
1109         dp = dp_get(nla_get_u32((info->attrs[DP_GENL_A_DP_IDX])));
1110         if (!dp)
1111                 err = -ENOENT;
1112         else {
1113                 del_dp(dp);
1114                 err = 0;
1115         }
1116         return err;
1117 }
1118
1119 static struct genl_ops dp_genl_ops_del_dp = {
1120         .cmd = DP_GENL_C_DEL_DP,
1121         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1122         .policy = dp_genl_policy,
1123         .doit = dp_genl_del,
1124         .dumpit = NULL,
1125 };
1126
1127 /* Queries a datapath for related information.  Currently the only relevant
1128  * information is the datapath's multicast group ID.  Really we want one
1129  * multicast group per datapath, but because of locking issues[*] we can't
1130  * easily get one.  Thus, every datapath will currently return the same
1131  * global multicast group ID, but in the future it would be nice to fix that.
1132  *
1133  * [*] dp_genl_add, to add a new datapath, is called under the genl_lock
1134  *       mutex, and genl_register_mc_group, called to acquire a new multicast
1135  *       group ID, also acquires genl_lock, thus deadlock.
1136  */
1137 static int dp_genl_query(struct sk_buff *skb, struct genl_info *info)
1138 {
1139         struct datapath *dp;
1140         struct sk_buff *ans_skb = NULL;
1141         int dp_idx;
1142         int err = -ENOMEM;
1143
1144         if (!info->attrs[DP_GENL_A_DP_IDX])
1145                 return -EINVAL;
1146
1147         rcu_read_lock();
1148         dp_idx = nla_get_u32((info->attrs[DP_GENL_A_DP_IDX]));
1149         dp = dp_get(dp_idx);
1150         if (!dp)
1151                 err = -ENOENT;
1152         else {
1153                 void *data;
1154                 ans_skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1155                 if (!ans_skb) {
1156                         err = -ENOMEM;
1157                         goto err;
1158                 }
1159                 data = genlmsg_put_reply(ans_skb, info, &dp_genl_family,
1160                                          0, DP_GENL_C_QUERY_DP);
1161                 if (data == NULL) {
1162                         err = -ENOMEM;
1163                         goto err;
1164                 }
1165                 NLA_PUT_U32(ans_skb, DP_GENL_A_DP_IDX, dp_idx);
1166                 NLA_PUT_U32(ans_skb, DP_GENL_A_MC_GROUP, mc_group.id);
1167
1168                 genlmsg_end(ans_skb, data);
1169                 err = genlmsg_reply(ans_skb, info);
1170                 if (!err)
1171                         ans_skb = NULL;
1172         }
1173 err:
1174 nla_put_failure:
1175         if (ans_skb)
1176                 kfree_skb(ans_skb);
1177         rcu_read_unlock();
1178         return err;
1179 }
1180
1181 static struct genl_ops dp_genl_ops_query_dp = {
1182         .cmd = DP_GENL_C_QUERY_DP,
1183         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1184         .policy = dp_genl_policy,
1185         .doit = dp_genl_query,
1186         .dumpit = NULL,
1187 };
1188
1189 static int dp_genl_add_del_port(struct sk_buff *skb, struct genl_info *info)
1190 {
1191         struct datapath *dp;
1192         struct net_device *port;
1193         int err;
1194
1195         if (!info->attrs[DP_GENL_A_DP_IDX] || !info->attrs[DP_GENL_A_PORTNAME])
1196                 return -EINVAL;
1197
1198         /* Get datapath. */
1199         dp = dp_get(nla_get_u32(info->attrs[DP_GENL_A_DP_IDX]));
1200         if (!dp) {
1201                 err = -ENOENT;
1202                 goto out;
1203         }
1204
1205         /* Get interface to add/remove. */
1206         port = dev_get_by_name(&init_net,
1207                         nla_data(info->attrs[DP_GENL_A_PORTNAME]));
1208         if (!port) {
1209                 err = -ENOENT;
1210                 goto out;
1211         }
1212
1213         /* Execute operation. */
1214         if (info->genlhdr->cmd == DP_GENL_C_ADD_PORT)
1215                 err = add_switch_port(dp, port);
1216         else {
1217                 if (port->br_port == NULL || port->br_port->dp != dp) {
1218                         err = -ENOENT;
1219                         goto out_put;
1220                 }
1221                 err = dp_del_switch_port(port->br_port);
1222         }
1223
1224 out_put:
1225         dev_put(port);
1226 out:
1227         return err;
1228 }
1229
1230 static struct genl_ops dp_genl_ops_add_port = {
1231         .cmd = DP_GENL_C_ADD_PORT,
1232         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1233         .policy = dp_genl_policy,
1234         .doit = dp_genl_add_del_port,
1235         .dumpit = NULL,
1236 };
1237
1238 static struct genl_ops dp_genl_ops_del_port = {
1239         .cmd = DP_GENL_C_DEL_PORT,
1240         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1241         .policy = dp_genl_policy,
1242         .doit = dp_genl_add_del_port,
1243         .dumpit = NULL,
1244 };
1245
1246 static int dp_genl_openflow(struct sk_buff *skb, struct genl_info *info)
1247 {
1248         struct nlattr *va = info->attrs[DP_GENL_A_OPENFLOW];
1249         struct datapath *dp;
1250         struct ofp_header *oh;
1251         struct sender sender;
1252         int err;
1253
1254         if (!info->attrs[DP_GENL_A_DP_IDX] || !va)
1255                 return -EINVAL;
1256
1257         dp = dp_get(nla_get_u32(info->attrs[DP_GENL_A_DP_IDX]));
1258         if (!dp)
1259                 return -ENOENT;
1260
1261         if (nla_len(va) < sizeof(struct ofp_header))
1262                 return -EINVAL;
1263         oh = nla_data(va);
1264
1265         sender.xid = oh->xid;
1266         sender.pid = info->snd_pid;
1267         sender.seq = info->snd_seq;
1268
1269         mutex_lock(&dp_mutex);
1270         err = fwd_control_input(dp->chain, &sender,
1271                                 nla_data(va), nla_len(va));
1272         mutex_unlock(&dp_mutex);
1273         return err;
1274 }
1275
1276 static struct nla_policy dp_genl_openflow_policy[DP_GENL_A_MAX + 1] = {
1277         [DP_GENL_A_DP_IDX] = { .type = NLA_U32 },
1278 };
1279
1280 static int desc_stats_dump(struct datapath *dp, void *state,
1281                             void *body, int *body_len)
1282 {
1283         struct ofp_desc_stats *ods = body;
1284         int n_bytes = sizeof *ods;
1285
1286         if (n_bytes > *body_len) {
1287                 return -ENOBUFS;
1288         }
1289         *body_len = n_bytes;
1290
1291         strncpy(ods->mfr_desc, mfr_desc, sizeof ods->mfr_desc);
1292         strncpy(ods->hw_desc, hw_desc, sizeof ods->hw_desc);
1293         strncpy(ods->sw_desc, sw_desc, sizeof ods->sw_desc);
1294         strncpy(ods->serial_num, serial_num, sizeof ods->serial_num);
1295
1296         return 0;
1297 }
1298
1299 struct flow_stats_state {
1300         int table_idx;
1301         struct sw_table_position position;
1302         const struct ofp_flow_stats_request *rq;
1303
1304         void *body;
1305         int bytes_used, bytes_allocated;
1306 };
1307
1308 static int flow_stats_init(struct datapath *dp, const void *body, int body_len,
1309                            void **state)
1310 {
1311         const struct ofp_flow_stats_request *fsr = body;
1312         struct flow_stats_state *s = kmalloc(sizeof *s, GFP_ATOMIC);
1313         if (!s)
1314                 return -ENOMEM;
1315         s->table_idx = fsr->table_id == 0xff ? 0 : fsr->table_id;
1316         memset(&s->position, 0, sizeof s->position);
1317         s->rq = fsr;
1318         *state = s;
1319         return 0;
1320 }
1321
1322 static int flow_stats_dump_callback(struct sw_flow *flow, void *private)
1323 {
1324         struct sw_flow_actions *sf_acts = rcu_dereference(flow->sf_acts);
1325         struct flow_stats_state *s = private;
1326         struct ofp_flow_stats *ofs;
1327         int length;
1328
1329         length = sizeof *ofs + sf_acts->actions_len;
1330         if (length + s->bytes_used > s->bytes_allocated)
1331                 return 1;
1332
1333         ofs = s->body + s->bytes_used;
1334         ofs->length          = htons(length);
1335         ofs->table_id        = s->table_idx;
1336         ofs->pad             = 0;
1337         ofs->match.wildcards = htonl(flow->key.wildcards);
1338         ofs->match.in_port   = flow->key.in_port;
1339         memcpy(ofs->match.dl_src, flow->key.dl_src, ETH_ALEN);
1340         memcpy(ofs->match.dl_dst, flow->key.dl_dst, ETH_ALEN);
1341         ofs->match.dl_vlan   = flow->key.dl_vlan;
1342         ofs->match.dl_type   = flow->key.dl_type;
1343         ofs->match.nw_src    = flow->key.nw_src;
1344         ofs->match.nw_dst    = flow->key.nw_dst;
1345         ofs->match.nw_proto  = flow->key.nw_proto;
1346         ofs->match.pad       = 0;
1347         ofs->match.tp_src    = flow->key.tp_src;
1348         ofs->match.tp_dst    = flow->key.tp_dst;
1349         ofs->duration        = htonl((jiffies - flow->init_time) / HZ);
1350         ofs->priority        = htons(flow->priority);
1351         ofs->idle_timeout    = htons(flow->idle_timeout);
1352         ofs->hard_timeout    = htons(flow->hard_timeout);
1353         memset(ofs->pad2, 0, sizeof ofs->pad2);
1354         ofs->packet_count    = cpu_to_be64(flow->packet_count);
1355         ofs->byte_count      = cpu_to_be64(flow->byte_count);
1356         memcpy(ofs->actions, sf_acts->actions, sf_acts->actions_len);
1357
1358         s->bytes_used += length;
1359         return 0;
1360 }
1361
1362 static int flow_stats_dump(struct datapath *dp, void *state,
1363                            void *body, int *body_len)
1364 {
1365         struct flow_stats_state *s = state;
1366         struct sw_flow_key match_key;
1367         int error = 0;
1368
1369         s->bytes_used = 0;
1370         s->bytes_allocated = *body_len;
1371         s->body = body;
1372
1373         flow_extract_match(&match_key, &s->rq->match);
1374         while (s->table_idx < dp->chain->n_tables
1375                && (s->rq->table_id == 0xff || s->rq->table_id == s->table_idx))
1376         {
1377                 struct sw_table *table = dp->chain->tables[s->table_idx];
1378
1379                 error = table->iterate(table, &match_key, &s->position,
1380                                        flow_stats_dump_callback, s);
1381                 if (error)
1382                         break;
1383
1384                 s->table_idx++;
1385                 memset(&s->position, 0, sizeof s->position);
1386         }
1387         *body_len = s->bytes_used;
1388
1389         /* If error is 0, we're done.
1390          * Otherwise, if some bytes were used, there are more flows to come.
1391          * Otherwise, we were not able to fit even a single flow in the body,
1392          * which indicates that we have a single flow with too many actions to
1393          * fit.  We won't ever make any progress at that rate, so give up. */
1394         return !error ? 0 : s->bytes_used ? 1 : -ENOMEM;
1395 }
1396
1397 static void flow_stats_done(void *state)
1398 {
1399         kfree(state);
1400 }
1401
1402 static int aggregate_stats_init(struct datapath *dp,
1403                                 const void *body, int body_len,
1404                                 void **state)
1405 {
1406         *state = (void *)body;
1407         return 0;
1408 }
1409
1410 static int aggregate_stats_dump_callback(struct sw_flow *flow, void *private)
1411 {
1412         struct ofp_aggregate_stats_reply *rpy = private;
1413         rpy->packet_count += flow->packet_count;
1414         rpy->byte_count += flow->byte_count;
1415         rpy->flow_count++;
1416         return 0;
1417 }
1418
1419 static int aggregate_stats_dump(struct datapath *dp, void *state,
1420                                 void *body, int *body_len)
1421 {
1422         struct ofp_aggregate_stats_request *rq = state;
1423         struct ofp_aggregate_stats_reply *rpy;
1424         struct sw_table_position position;
1425         struct sw_flow_key match_key;
1426         int table_idx;
1427
1428         if (*body_len < sizeof *rpy)
1429                 return -ENOBUFS;
1430         rpy = body;
1431         *body_len = sizeof *rpy;
1432
1433         memset(rpy, 0, sizeof *rpy);
1434
1435         flow_extract_match(&match_key, &rq->match);
1436         table_idx = rq->table_id == 0xff ? 0 : rq->table_id;
1437         memset(&position, 0, sizeof position);
1438         while (table_idx < dp->chain->n_tables
1439                && (rq->table_id == 0xff || rq->table_id == table_idx))
1440         {
1441                 struct sw_table *table = dp->chain->tables[table_idx];
1442                 int error;
1443
1444                 error = table->iterate(table, &match_key, &position,
1445                                        aggregate_stats_dump_callback, rpy);
1446                 if (error)
1447                         return error;
1448
1449                 table_idx++;
1450                 memset(&position, 0, sizeof position);
1451         }
1452
1453         rpy->packet_count = cpu_to_be64(rpy->packet_count);
1454         rpy->byte_count = cpu_to_be64(rpy->byte_count);
1455         rpy->flow_count = htonl(rpy->flow_count);
1456         return 0;
1457 }
1458
1459 static int table_stats_dump(struct datapath *dp, void *state,
1460                             void *body, int *body_len)
1461 {
1462         struct ofp_table_stats *ots;
1463         int n_bytes = dp->chain->n_tables * sizeof *ots;
1464         int i;
1465         if (n_bytes > *body_len)
1466                 return -ENOBUFS;
1467         *body_len = n_bytes;
1468         for (i = 0, ots = body; i < dp->chain->n_tables; i++, ots++) {
1469                 struct sw_table_stats stats;
1470                 dp->chain->tables[i]->stats(dp->chain->tables[i], &stats);
1471                 strncpy(ots->name, stats.name, sizeof ots->name);
1472                 ots->table_id = i;
1473                 ots->wildcards = htonl(stats.wildcards);
1474                 memset(ots->pad, 0, sizeof ots->pad);
1475                 ots->max_entries = htonl(stats.max_flows);
1476                 ots->active_count = htonl(stats.n_flows);
1477                 ots->lookup_count = cpu_to_be64(stats.n_lookup);
1478                 ots->matched_count = cpu_to_be64(stats.n_matched);
1479         }
1480         return 0;
1481 }
1482
1483 struct port_stats_state {
1484         int port;
1485 };
1486
1487 static int port_stats_init(struct datapath *dp, const void *body, int body_len,
1488                            void **state)
1489 {
1490         struct port_stats_state *s = kmalloc(sizeof *s, GFP_ATOMIC);
1491         if (!s)
1492                 return -ENOMEM;
1493         s->port = 0;
1494         *state = s;
1495         return 0;
1496 }
1497
1498 static int port_stats_dump(struct datapath *dp, void *state,
1499                            void *body, int *body_len)
1500 {
1501         struct port_stats_state *s = state;
1502         struct ofp_port_stats *ops;
1503         int n_ports, max_ports;
1504         int i;
1505
1506         max_ports = *body_len / sizeof *ops;
1507         if (!max_ports)
1508                 return -ENOMEM;
1509         ops = body;
1510
1511         n_ports = 0;
1512         for (i = s->port; i < DP_MAX_PORTS && n_ports < max_ports; i++) {
1513                 struct net_bridge_port *p = dp->ports[i];
1514                 struct net_device_stats *stats;
1515                 if (!p)
1516                         continue;
1517                 stats = p->dev->get_stats(p->dev);
1518                 ops->port_no = htons(p->port_no);
1519                 memset(ops->pad, 0, sizeof ops->pad);
1520                 ops->rx_packets   = cpu_to_be64(stats->rx_packets);
1521                 ops->tx_packets   = cpu_to_be64(stats->tx_packets);
1522                 ops->rx_bytes     = cpu_to_be64(stats->rx_bytes);
1523                 ops->tx_bytes     = cpu_to_be64(stats->tx_bytes);
1524                 ops->rx_dropped   = cpu_to_be64(stats->rx_dropped);
1525                 ops->tx_dropped   = cpu_to_be64(stats->tx_dropped);
1526                 ops->rx_errors    = cpu_to_be64(stats->rx_errors);
1527                 ops->tx_errors    = cpu_to_be64(stats->tx_errors);
1528                 ops->rx_frame_err = cpu_to_be64(stats->rx_frame_errors);
1529                 ops->rx_over_err  = cpu_to_be64(stats->rx_over_errors);
1530                 ops->rx_crc_err   = cpu_to_be64(stats->rx_crc_errors);
1531                 ops->collisions   = cpu_to_be64(stats->collisions);
1532                 n_ports++;
1533                 ops++;
1534         }
1535         s->port = i;
1536         *body_len = n_ports * sizeof *ops;
1537         return n_ports >= max_ports;
1538 }
1539
1540 static void port_stats_done(void *state)
1541 {
1542         kfree(state);
1543 }
1544
1545 struct stats_type {
1546         /* Minimum and maximum acceptable number of bytes in body member of
1547          * struct ofp_stats_request. */
1548         size_t min_body, max_body;
1549
1550         /* Prepares to dump some kind of statistics on 'dp'.  'body' and
1551          * 'body_len' are the 'body' member of the struct ofp_stats_request.
1552          * Returns zero if successful, otherwise a negative error code.
1553          * May initialize '*state' to state information.  May be null if no
1554          * initialization is required.*/
1555         int (*init)(struct datapath *dp, const void *body, int body_len,
1556                     void **state);
1557
1558         /* Dumps statistics for 'dp' into the '*body_len' bytes at 'body', and
1559          * modifies '*body_len' to reflect the number of bytes actually used.
1560          * ('body' will be transmitted as the 'body' member of struct
1561          * ofp_stats_reply.) */
1562         int (*dump)(struct datapath *dp, void *state,
1563                     void *body, int *body_len);
1564
1565         /* Cleans any state created by the init or dump functions.  May be null
1566          * if no cleanup is required. */
1567         void (*done)(void *state);
1568 };
1569
1570 static const struct stats_type stats[] = {
1571         [OFPST_DESC] = {
1572                 0,
1573                 0,
1574                 NULL,
1575                 desc_stats_dump,
1576                 NULL
1577         },
1578         [OFPST_FLOW] = {
1579                 sizeof(struct ofp_flow_stats_request),
1580                 sizeof(struct ofp_flow_stats_request),
1581                 flow_stats_init,
1582                 flow_stats_dump,
1583                 flow_stats_done
1584         },
1585         [OFPST_AGGREGATE] = {
1586                 sizeof(struct ofp_aggregate_stats_request),
1587                 sizeof(struct ofp_aggregate_stats_request),
1588                 aggregate_stats_init,
1589                 aggregate_stats_dump,
1590                 NULL
1591         },
1592         [OFPST_TABLE] = {
1593                 0,
1594                 0,
1595                 NULL,
1596                 table_stats_dump,
1597                 NULL
1598         },
1599         [OFPST_PORT] = {
1600                 0,
1601                 0,
1602                 port_stats_init,
1603                 port_stats_dump,
1604                 port_stats_done
1605         },
1606 };
1607
1608 static int
1609 dp_genl_openflow_dumpit(struct sk_buff *skb, struct netlink_callback *cb)
1610 {
1611         struct datapath *dp;
1612         struct sender sender;
1613         const struct stats_type *s;
1614         struct ofp_stats_reply *osr;
1615         int dp_idx;
1616         int max_openflow_len, body_len;
1617         void *body;
1618         int err;
1619
1620         /* Set up the cleanup function for this dump.  Linux 2.6.20 and later
1621          * support setting up cleanup functions via the .doneit member of
1622          * struct genl_ops.  This kluge supports earlier versions also. */
1623         cb->done = dp_genl_openflow_done;
1624
1625         sender.pid = NETLINK_CB(cb->skb).pid;
1626         sender.seq = cb->nlh->nlmsg_seq;
1627         if (!cb->args[0]) {
1628                 struct nlattr *attrs[DP_GENL_A_MAX + 1];
1629                 struct ofp_stats_request *rq;
1630                 struct nlattr *va;
1631                 size_t len, body_len;
1632                 int type;
1633
1634                 err = nlmsg_parse(cb->nlh, GENL_HDRLEN, attrs, DP_GENL_A_MAX,
1635                                   dp_genl_openflow_policy);
1636                 if (err < 0)
1637                         return err;
1638
1639                 if (!attrs[DP_GENL_A_DP_IDX])
1640                         return -EINVAL;
1641                 dp_idx = nla_get_u16(attrs[DP_GENL_A_DP_IDX]);
1642                 dp = dp_get(dp_idx);
1643                 if (!dp)
1644                         return -ENOENT;
1645
1646                 va = attrs[DP_GENL_A_OPENFLOW];
1647                 len = nla_len(va);
1648                 if (!va || len < sizeof *rq)
1649                         return -EINVAL;
1650
1651                 rq = nla_data(va);
1652                 sender.xid = rq->header.xid;
1653                 type = ntohs(rq->type);
1654                 if (rq->header.version != OFP_VERSION) {
1655                         dp_send_error_msg(dp, &sender, OFPET_BAD_REQUEST,
1656                                           OFPBRC_BAD_VERSION, rq, len);
1657                         return -EINVAL;
1658                 }
1659                 if (rq->header.type != OFPT_STATS_REQUEST
1660                     || ntohs(rq->header.length) != len)
1661                         return -EINVAL;
1662
1663                 if (type >= ARRAY_SIZE(stats) || !stats[type].dump) {
1664                         dp_send_error_msg(dp, &sender, OFPET_BAD_REQUEST,
1665                                           OFPBRC_BAD_STAT, rq, len);
1666                         return -EINVAL;
1667                 }
1668
1669                 s = &stats[type];
1670                 body_len = len - offsetof(struct ofp_stats_request, body);
1671                 if (body_len < s->min_body || body_len > s->max_body)
1672                         return -EINVAL;
1673
1674                 cb->args[0] = 1;
1675                 cb->args[1] = dp_idx;
1676                 cb->args[2] = type;
1677                 cb->args[3] = rq->header.xid;
1678                 if (s->init) {
1679                         void *state;
1680                         err = s->init(dp, rq->body, body_len, &state);
1681                         if (err)
1682                                 return err;
1683                         cb->args[4] = (long) state;
1684                 }
1685         } else if (cb->args[0] == 1) {
1686                 sender.xid = cb->args[3];
1687                 dp_idx = cb->args[1];
1688                 s = &stats[cb->args[2]];
1689
1690                 dp = dp_get(dp_idx);
1691                 if (!dp)
1692                         return -ENOENT;
1693         } else {
1694                 return 0;
1695         }
1696
1697         osr = put_openflow_headers(dp, skb, OFPT_STATS_REPLY, &sender,
1698                                    &max_openflow_len);
1699         if (IS_ERR(osr))
1700                 return PTR_ERR(osr);
1701         osr->type = htons(s - stats);
1702         osr->flags = 0;
1703         resize_openflow_skb(skb, &osr->header, max_openflow_len);
1704         body = osr->body;
1705         body_len = max_openflow_len - offsetof(struct ofp_stats_reply, body);
1706
1707         err = s->dump(dp, (void *) cb->args[4], body, &body_len);
1708         if (err >= 0) {
1709                 if (!err)
1710                         cb->args[0] = 2;
1711                 else
1712                         osr->flags = ntohs(OFPSF_REPLY_MORE);
1713                 resize_openflow_skb(skb, &osr->header,
1714                                     (offsetof(struct ofp_stats_reply, body)
1715                                      + body_len));
1716                 err = skb->len;
1717         }
1718
1719         return err;
1720 }
1721
1722 static int
1723 dp_genl_openflow_done(struct netlink_callback *cb)
1724 {
1725         if (cb->args[0]) {
1726                 const struct stats_type *s = &stats[cb->args[2]];
1727                 if (s->done)
1728                         s->done((void *) cb->args[4]);
1729         }
1730         return 0;
1731 }
1732
1733 static struct genl_ops dp_genl_ops_openflow = {
1734         .cmd = DP_GENL_C_OPENFLOW,
1735         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1736         .policy = dp_genl_openflow_policy,
1737         .doit = dp_genl_openflow,
1738         .dumpit = dp_genl_openflow_dumpit,
1739 };
1740
1741 static struct genl_ops *dp_genl_all_ops[] = {
1742         /* Keep this operation first.  Generic Netlink dispatching
1743          * looks up operations with linear search, so we want it at the
1744          * front. */
1745         &dp_genl_ops_openflow,
1746
1747         &dp_genl_ops_add_dp,
1748         &dp_genl_ops_del_dp,
1749         &dp_genl_ops_query_dp,
1750         &dp_genl_ops_add_port,
1751         &dp_genl_ops_del_port,
1752 };
1753
1754 static int dp_init_netlink(void)
1755 {
1756         int err;
1757         int i;
1758
1759         err = genl_register_family(&dp_genl_family);
1760         if (err)
1761                 return err;
1762
1763         for (i = 0; i < ARRAY_SIZE(dp_genl_all_ops); i++) {
1764                 err = genl_register_ops(&dp_genl_family, dp_genl_all_ops[i]);
1765                 if (err)
1766                         goto err_unregister;
1767         }
1768
1769         strcpy(mc_group.name, "openflow");
1770         err = genl_register_mc_group(&dp_genl_family, &mc_group);
1771         if (err < 0)
1772                 goto err_unregister;
1773
1774         return 0;
1775
1776 err_unregister:
1777         genl_unregister_family(&dp_genl_family);
1778                 return err;
1779 }
1780
1781 static void dp_uninit_netlink(void)
1782 {
1783         genl_unregister_family(&dp_genl_family);
1784 }
1785
1786 /* Set the description strings if appropriate values are available from
1787  * the DMI. */
1788 static void set_desc(void)
1789 {
1790         const char *uuid = dmi_get_system_info(DMI_PRODUCT_UUID);
1791         const char *uptr = uuid + 24;
1792
1793         if (!uuid || *uuid == '\0' || strlen(uuid) != 36)
1794                 return;
1795
1796         /* We are only interested version 1 UUIDs, since the last six bytes
1797          * are an IEEE 802 MAC address. */
1798         if (uuid[14] != '1')
1799                 return;
1800
1801         /* Only set if the UUID is from Nicira. */
1802         if (strncmp(uptr, NICIRA_OUI_STR, strlen(NICIRA_OUI_STR)))
1803                 return;
1804
1805         strlcpy(mfr_desc, dmi_get_system_info(DMI_SYS_VENDOR), sizeof(mfr_desc));
1806         snprintf(hw_desc, sizeof(hw_desc), "%s %s",
1807                         dmi_get_system_info(DMI_PRODUCT_NAME),
1808                         dmi_get_system_info(DMI_PRODUCT_VERSION));
1809         strlcpy(serial_num, dmi_get_system_info(DMI_PRODUCT_SERIAL),
1810                         sizeof(serial_num));
1811 }
1812
1813 static int __init dp_init(void)
1814 {
1815         int err;
1816
1817         printk("OpenFlow %s, built "__DATE__" "__TIME__", "
1818                "protocol 0x%02x\n", VERSION BUILDNR, OFP_VERSION);
1819
1820         err = flow_init();
1821         if (err)
1822                 goto error;
1823
1824         err = register_netdevice_notifier(&dp_device_notifier);
1825         if (err)
1826                 goto error_flow_exit;
1827
1828         err = dp_init_netlink();
1829         if (err)
1830                 goto error_unreg_notifier;
1831
1832         /* Check if better descriptions of the switch are available than the
1833          * defaults. */
1834         set_desc();
1835
1836         /* Hook into callback used by the bridge to intercept packets.
1837          * Parasites we are. */
1838         if (br_handle_frame_hook)
1839                 printk("openflow: hijacking bridge hook\n");
1840         br_handle_frame_hook = dp_frame_hook;
1841
1842         return 0;
1843
1844 error_unreg_notifier:
1845         unregister_netdevice_notifier(&dp_device_notifier);
1846 error_flow_exit:
1847         flow_exit();
1848 error:
1849         printk(KERN_EMERG "openflow: failed to install!");
1850         return err;
1851 }
1852
1853 static void dp_cleanup(void)
1854 {
1855         fwd_exit();
1856         dp_uninit_netlink();
1857         unregister_netdevice_notifier(&dp_device_notifier);
1858         flow_exit();
1859         br_handle_frame_hook = NULL;
1860 }
1861
1862 module_init(dp_init);
1863 module_exit(dp_cleanup);
1864
1865 MODULE_DESCRIPTION("OpenFlow switching datapath");
1866 MODULE_AUTHOR("Copyright (c) 2007, 2008 The Board of Trustees of The Leland Stanford Junior University");
1867 MODULE_LICENSE("GPL");