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 int 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                 return 0;
 545         }
 546         return -ENOENT;
 547 }
 548
 549 static int xmit_skb(struct sk_buff *skb)
 550 {
 551         int len = skb->len;
 552         if (packet_length(skb) > skb->dev->mtu) {
 553                 printk("dropped over-mtu packet: %d > %d\n",
 554                            packet_length(skb), skb->dev->mtu);
 555                 kfree_skb(skb);
 556                 return -E2BIG;
 557         }
 558
 559         dev_queue_xmit(skb);
 560
 561         return len;
 562 }
 563
 564 /* Takes ownership of 'skb' and transmits it to 'out_port' on 'dp'.
 565  */
 566 int dp_output_port(struct datapath *dp, struct sk_buff *skb, int out_port,
 567                    int ignore_no_fwd)
 568 {
 569         BUG_ON(!skb);
 570         switch (out_port){
 571         case OFPP_IN_PORT:
 572                 /* Send it out the port it came in on, which is already set in
 573                  * the skb. */
 574                 if (!skb->dev) {
 575                         if (net_ratelimit())
 576                                 printk("skb device not set forwarding to in_port\n");
 577                         kfree(skb);
 578                         return -ESRCH;
 579                 }
 580                 return xmit_skb(skb);
 581
 582         case OFPP_TABLE: {
 583                 int retval = run_flow_through_tables(dp->chain, skb,
 584                                                      skb->dev->br_port);
 585                 if (retval)
 586                         kfree_skb(skb);
 587                 return retval;
 588         }
 589
 590         case OFPP_FLOOD:
 591                 return output_all(dp, skb, 1);
 592
 593         case OFPP_ALL:
 594                 return output_all(dp, skb, 0);
 595
 596         case OFPP_CONTROLLER:
 597                 return dp_output_control(dp, skb, fwd_save_skb(skb), 0,
 598                                                   OFPR_ACTION);
 599
 600         case OFPP_LOCAL: {
 601                 struct net_device *dev = dp->netdev;
 602                 return dev ? dp_dev_recv(dev, skb) : -ESRCH;
 603         }
 604
 605         case 0 ... OFPP_MAX-1: {
 606                 struct net_bridge_port *p = dp->ports[out_port];
 607                 if (p == NULL)
 608                         goto bad_port;
 609                 if (p->dev == skb->dev) {
 610                         /* To send to the input port, must use OFPP_IN_PORT */
 611                         kfree_skb(skb);
 612                         if (net_ratelimit())
 613                                 printk("can't directly forward to input port\n");
 614                         return -EINVAL;
 615                 }
 616                 if (p->config & OFPPC_NO_FWD && !ignore_no_fwd) {
 617                         kfree_skb(skb);
 618                         return 0;
 619                 }
 620                 skb->dev = p->dev;
 621                 return xmit_skb(skb);
 622         }
 623
 624         default:
 625                 goto bad_port;
 626         }
 627
 628 bad_port:
 629         kfree_skb(skb);
 630         if (net_ratelimit())
 631                 printk("can't forward to bad port %d\n", out_port);
 632         return -ENOENT;
 633 }
 634
 635 /* Takes ownership of 'skb' and transmits it to 'dp''s control path.  If
 636  * 'buffer_id' != -1, then only the first 64 bytes of 'skb' are sent;
 637  * otherwise, all of 'skb' is sent.  'reason' indicates why 'skb' is being
 638  * sent. 'max_len' sets the maximum number of bytes that the caller
 639  * wants to be sent; a value of 0 indicates the entire packet should be
 640  * sent. */
 641 int
 642 dp_output_control(struct datapath *dp, struct sk_buff *skb,
 643                            uint32_t buffer_id, size_t max_len, int reason)
 644 {
 645         /* FIXME?  Can we avoid creating a new skbuff in the case where we
 646          * forward the whole packet? */
 647         struct sk_buff *f_skb;
 648         struct ofp_packet_in *opi;
 649         struct net_bridge_port *p;
 650         size_t fwd_len, opi_len;
 651         int err;
 652
 653         fwd_len = skb->len;
 654         if ((buffer_id != (uint32_t) -1) && max_len)
 655                 fwd_len = min(fwd_len, max_len);
 656
 657         opi_len = offsetof(struct ofp_packet_in, data) + fwd_len;
 658         opi = alloc_openflow_skb(dp, opi_len, OFPT_PACKET_IN, NULL, &f_skb);
 659         if (!opi) {
 660                 err = -ENOMEM;
 661                 goto out;
 662         }
 663         opi->buffer_id      = htonl(buffer_id);
 664         opi->total_len      = htons(skb->len);
 665         p = skb->dev->br_port;
 666         opi->in_port        = htons(p ? p->port_no : OFPP_LOCAL);
 667         opi->reason         = reason;
 668         opi->pad            = 0;
 669         memcpy(opi->data, skb_mac_header(skb), fwd_len);
 670         err = send_openflow_skb(f_skb, NULL);
 671
 672 out:
 673         kfree_skb(skb);
 674         return err;
 675 }
 676
 677 static void fill_port_desc(struct net_bridge_port *p, struct ofp_phy_port *desc)
 678 {
 679         unsigned long flags;
 680         desc->port_no = htons(p->port_no);
 681         strncpy(desc->name, p->dev->name, OFP_MAX_PORT_NAME_LEN);
 682         desc->name[OFP_MAX_PORT_NAME_LEN-1] = '\0';
 683         memcpy(desc->hw_addr, p->dev->dev_addr, ETH_ALEN);
 684         desc->curr = 0;
 685         desc->supported = 0;
 686         desc->advertised = 0;
 687         desc->peer = 0;
 688
 689         spin_lock_irqsave(&p->lock, flags);
 690         desc->config = htonl(p->config);
 691         desc->state = htonl(p->state);
 692         spin_unlock_irqrestore(&p->lock, flags);
 693
 694 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,4,24)
 695         if (p->dev->ethtool_ops && p->dev->ethtool_ops->get_settings) {
 696                 struct ethtool_cmd ecmd = { .cmd = ETHTOOL_GSET };
 697
 698                 if (!p->dev->ethtool_ops->get_settings(p->dev, &ecmd)) {
 699                         /* Set the supported features */
 700                         if (ecmd.supported & SUPPORTED_10baseT_Half)
 701                                 desc->supported |= OFPPF_10MB_HD;
 702                         if (ecmd.supported & SUPPORTED_10baseT_Full)
 703                                 desc->supported |= OFPPF_10MB_FD;
 704                         if (ecmd.supported & SUPPORTED_100baseT_Half)
 705                                 desc->supported |= OFPPF_100MB_HD;
 706                         if (ecmd.supported & SUPPORTED_100baseT_Full)
 707                                 desc->supported |= OFPPF_100MB_FD;
 708                         if (ecmd.supported & SUPPORTED_1000baseT_Half)
 709                                 desc->supported |= OFPPF_1GB_HD;
 710                         if (ecmd.supported & SUPPORTED_1000baseT_Full)
 711                                 desc->supported |= OFPPF_1GB_FD;
 712                         if (ecmd.supported & SUPPORTED_10000baseT_Full)
 713                                 desc->supported |= OFPPF_10GB_FD;
 714                         if (ecmd.supported & SUPPORTED_TP)
 715                                 desc->supported |= OFPPF_COPPER;
 716                         if (ecmd.supported & SUPPORTED_FIBRE)
 717                                 desc->supported |= OFPPF_FIBER;
 718                         if (ecmd.supported & SUPPORTED_Autoneg)
 719                                 desc->supported |= OFPPF_AUTONEG;
 720 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,14)
 721                         if (ecmd.supported & SUPPORTED_Pause)
 722                                 desc->supported |= OFPPF_PAUSE;
 723                         if (ecmd.supported & SUPPORTED_Asym_Pause)
 724                                 desc->supported |= OFPPF_PAUSE_ASYM;
 725 #endif /* kernel >= 2.6.14 */
 726
 727                         /* Set the advertised features */
 728                         if (ecmd.advertising & ADVERTISED_10baseT_Half)
 729                                 desc->advertised |= OFPPF_10MB_HD;
 730                         if (ecmd.advertising & ADVERTISED_10baseT_Full)
 731                                 desc->advertised |= OFPPF_10MB_FD;
 732                         if (ecmd.advertising & ADVERTISED_100baseT_Half)
 733                                 desc->advertised |= OFPPF_100MB_HD;
 734                         if (ecmd.advertising & ADVERTISED_100baseT_Full)
 735                                 desc->advertised |= OFPPF_100MB_FD;
 736                         if (ecmd.advertising & ADVERTISED_1000baseT_Half)
 737                                 desc->advertised |= OFPPF_1GB_HD;
 738                         if (ecmd.advertising & ADVERTISED_1000baseT_Full)
 739                                 desc->advertised |= OFPPF_1GB_FD;
 740                         if (ecmd.advertising & ADVERTISED_10000baseT_Full)
 741                                 desc->advertised |= OFPPF_10GB_FD;
 742                         if (ecmd.advertising & ADVERTISED_TP)
 743                                 desc->advertised |= OFPPF_COPPER;
 744                         if (ecmd.advertising & ADVERTISED_FIBRE)
 745                                 desc->advertised |= OFPPF_FIBER;
 746                         if (ecmd.advertising & ADVERTISED_Autoneg)
 747                                 desc->advertised |= OFPPF_AUTONEG;
 748 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,14)
 749                         if (ecmd.advertising & ADVERTISED_Pause)
 750                                 desc->advertised |= OFPPF_PAUSE;
 751                         if (ecmd.advertising & ADVERTISED_Asym_Pause)
 752                                 desc->advertised |= OFPPF_PAUSE_ASYM;
 753 #endif /* kernel >= 2.6.14 */
 754
 755                         /* Set the current features */
 756                         if (ecmd.speed == SPEED_10)
 757                                 desc->curr = (ecmd.duplex) ? OFPPF_10MB_FD : OFPPF_10MB_HD;
 758                         else if (ecmd.speed == SPEED_100)
 759                                 desc->curr = (ecmd.duplex) ? OFPPF_100MB_FD : OFPPF_100MB_HD;
 760                         else if (ecmd.speed == SPEED_1000)
 761                                 desc->curr = (ecmd.duplex) ? OFPPF_1GB_FD : OFPPF_1GB_HD;
 762                         else if (ecmd.speed == SPEED_10000)
 763                                 desc->curr = OFPPF_10GB_FD;
 764
 765                         if (ecmd.port == PORT_TP)
 766                                 desc->curr |= OFPPF_COPPER;
 767                         else if (ecmd.port == PORT_FIBRE)
 768                                 desc->curr |= OFPPF_FIBER;
 769
 770                         if (ecmd.autoneg)
 771                                 desc->curr |= OFPPF_AUTONEG;
 772                 }
 773         }
 774 #endif
 775         desc->curr = htonl(desc->curr);
 776         desc->supported = htonl(desc->supported);
 777         desc->advertised = htonl(desc->advertised);
 778         desc->peer = htonl(desc->peer);
 779 }
 780
 781 static int
 782 fill_features_reply(struct datapath *dp, struct ofp_switch_features *ofr)
 783 {
 784         struct net_bridge_port *p;
 785         int port_count = 0;
 786
 787         ofr->datapath_id  = cpu_to_be64(dp->id);
 788
 789         ofr->n_buffers    = htonl(N_PKT_BUFFERS);
 790         ofr->n_tables     = dp->chain->n_tables;
 791         ofr->capabilities = htonl(OFP_SUPPORTED_CAPABILITIES);
 792         ofr->actions      = htonl(OFP_SUPPORTED_ACTIONS);
 793         memset(ofr->pad, 0, sizeof ofr->pad);
 794
 795         list_for_each_entry_rcu (p, &dp->port_list, node) {
 796                 fill_port_desc(p, &ofr->ports[port_count]);
 797                 port_count++;
 798         }
 799
 800         return port_count;
 801 }
 802
 803 int
 804 dp_send_features_reply(struct datapath *dp, const struct sender *sender)
 805 {
 806         struct sk_buff *skb;
 807         struct ofp_switch_features *ofr;
 808         size_t ofr_len, port_max_len;
 809         int port_count;
 810
 811         /* Overallocate. */
 812         port_max_len = sizeof(struct ofp_phy_port) * OFPP_MAX;
 813         ofr = alloc_openflow_skb(dp, sizeof(*ofr) + port_max_len,
 814                                  OFPT_FEATURES_REPLY, sender, &skb);
 815         if (!ofr)
 816                 return -ENOMEM;
 817
 818         /* Fill. */
 819         port_count = fill_features_reply(dp, ofr);
 820
 821         /* Shrink to fit. */
 822         ofr_len = sizeof(*ofr) + (sizeof(struct ofp_phy_port) * port_count);
 823         resize_openflow_skb(skb, &ofr->header, ofr_len);
 824         return send_openflow_skb(skb, sender);
 825 }
 826
 827 int
 828 dp_send_config_reply(struct datapath *dp, const struct sender *sender)
 829 {
 830         struct sk_buff *skb;
 831         struct ofp_switch_config *osc;
 832
 833         osc = alloc_openflow_skb(dp, sizeof *osc, OFPT_GET_CONFIG_REPLY, sender,
 834                                  &skb);
 835         if (!osc)
 836                 return -ENOMEM;
 837
 838         osc->flags = htons(dp->flags);
 839         osc->miss_send_len = htons(dp->miss_send_len);
 840
 841         return send_openflow_skb(skb, sender);
 842 }
 843
 844 int
 845 dp_send_hello(struct datapath *dp, const struct sender *sender,
 846               const struct ofp_header *request)
 847 {
 848         if (request->version < OFP_VERSION) {
 849                 char err[64];
 850                 sprintf(err, "Only version 0x%02x supported", OFP_VERSION);
 851                 dp_send_error_msg(dp, sender, OFPET_HELLO_FAILED,
 852                                   OFPHFC_INCOMPATIBLE, err, strlen(err));
 853                 return -EINVAL;
 854         } else {
 855                 struct sk_buff *skb;
 856                 struct ofp_header *reply;
 857
 858                 reply = alloc_openflow_skb(dp, sizeof *reply,
 859                                            OFPT_HELLO, sender, &skb);
 860                 if (!reply)
 861                         return -ENOMEM;
 862
 863                 return send_openflow_skb(skb, sender);
 864         }
 865 }
 866
 867 /* Callback function for a workqueue to disable an interface */
 868 static void
 869 down_port_cb(struct work_struct *work)
 870 {
 871         struct net_bridge_port *p = container_of(work, struct net_bridge_port,
 872                         port_task);
 873
 874         rtnl_lock();
 875         if (dev_change_flags(p->dev, p->dev->flags & ~IFF_UP) < 0)
 876                 if (net_ratelimit())
 877                         printk("problem bringing up port %s\n", p->dev->name);
 878         rtnl_unlock();
 879         p->config |= OFPPC_PORT_DOWN;
 880 }
 881
 882 /* Callback function for a workqueue to enable an interface */
 883 static void
 884 up_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 down port %s\n", p->dev->name);
 893         rtnl_unlock();
 894         p->config &= ~OFPPC_PORT_DOWN;
 895 }
 896
 897 int
 898 dp_update_port_flags(struct datapath *dp, const struct ofp_port_mod *opm)
 899 {
 900         unsigned long int flags;
 901         int port_no = ntohs(opm->port_no);
 902         struct net_bridge_port *p = (port_no < OFPP_MAX ? dp->ports[port_no]
 903                                      : port_no == OFPP_LOCAL ? dp->local_port
 904                                      : NULL);
 905
 906         /* Make sure the port id hasn't changed since this was sent */
 907         if (!p || memcmp(opm->hw_addr, p->dev->dev_addr, ETH_ALEN))
 908                 return -1;
 909
 910         spin_lock_irqsave(&p->lock, flags);
 911         if (opm->mask) {
 912                 uint32_t config_mask = ntohl(opm->mask);
 913                 p->config &= ~config_mask;
 914                 p->config |= ntohl(opm->config) & config_mask;
 915         }
 916
 917         /* Modifying the status of an interface requires taking a lock
 918          * that cannot be done from here.  For this reason, we use a shared
 919          * workqueue, which will cause it to be executed from a safer
 920          * context. */
 921         if (opm->mask & htonl(OFPPC_PORT_DOWN)) {
 922                 if ((opm->config & htonl(OFPPC_PORT_DOWN))
 923                     && (p->config & OFPPC_PORT_DOWN) == 0) {
 924                         PREPARE_WORK(&p->port_task, down_port_cb);
 925                         schedule_work(&p->port_task);
 926                 } else if ((opm->config & htonl(OFPPC_PORT_DOWN)) == 0
 927                            && (p->config & OFPPC_PORT_DOWN)) {
 928                         PREPARE_WORK(&p->port_task, up_port_cb);
 929                         schedule_work(&p->port_task);
 930                 }
 931         }
 932         spin_unlock_irqrestore(&p->lock, flags);
 933
 934         return 0;
 935 }
 936
 937 /* Initialize the port status field of the bridge port. */
 938 static void
 939 init_port_status(struct net_bridge_port *p)
 940 {
 941         unsigned long int flags;
 942
 943         spin_lock_irqsave(&p->lock, flags);
 944
 945         if (p->dev->flags & IFF_UP)
 946                 p->config &= ~OFPPC_PORT_DOWN;
 947         else
 948                 p->config |= OFPPC_PORT_DOWN;
 949
 950         if (netif_carrier_ok(p->dev))
 951                 p->state &= ~OFPPS_LINK_DOWN;
 952         else
 953                 p->state |= OFPPS_LINK_DOWN;
 954
 955         spin_unlock_irqrestore(&p->lock, flags);
 956 }
 957
 958 int
 959 dp_send_port_status(struct net_bridge_port *p, uint8_t status)
 960 {
 961         struct sk_buff *skb;
 962         struct ofp_port_status *ops;
 963
 964         ops = alloc_openflow_skb(p->dp, sizeof *ops, OFPT_PORT_STATUS, NULL,
 965                                  &skb);
 966         if (!ops)
 967                 return -ENOMEM;
 968         ops->reason = status;
 969         memset(ops->pad, 0, sizeof ops->pad);
 970         fill_port_desc(p, &ops->desc);
 971
 972         return send_openflow_skb(skb, NULL);
 973 }
 974
 975 int
 976 dp_send_flow_expired(struct datapath *dp, struct sw_flow *flow,
 977                      enum ofp_flow_expired_reason reason)
 978 {
 979         struct sk_buff *skb;
 980         struct ofp_flow_expired *ofe;
 981
 982         if (!(dp->flags & OFPC_SEND_FLOW_EXP))
 983                 return 0;
 984
 985         ofe = alloc_openflow_skb(dp, sizeof *ofe, OFPT_FLOW_EXPIRED, 0, &skb);
 986         if (!ofe)
 987                 return -ENOMEM;
 988
 989         flow_fill_match(&ofe->match, &flow->key);
 990
 991         ofe->priority = htons(flow->priority);
 992         ofe->reason = reason;
 993         memset(ofe->pad, 0, sizeof ofe->pad);
 994
 995         ofe->duration     = htonl((jiffies - flow->init_time) / HZ);
 996         memset(ofe->pad2, 0, sizeof ofe->pad2);
 997         ofe->packet_count = cpu_to_be64(flow->packet_count);
 998         ofe->byte_count   = cpu_to_be64(flow->byte_count);
 999
1000         return send_openflow_skb(skb, NULL);
1001 }
1002 EXPORT_SYMBOL(dp_send_flow_expired);
1003
1004 int
1005 dp_send_error_msg(struct datapath *dp, const struct sender *sender,
1006                 uint16_t type, uint16_t code, const void *data, size_t len)
1007 {
1008         struct sk_buff *skb;
1009         struct ofp_error_msg *oem;
1010
1011
1012         oem = alloc_openflow_skb(dp, sizeof(*oem)+len, OFPT_ERROR,
1013                         sender, &skb);
1014         if (!oem)
1015                 return -ENOMEM;
1016
1017         oem->type = htons(type);
1018         oem->code = htons(code);
1019         memcpy(oem->data, data, len);
1020
1021         return send_openflow_skb(skb, sender);
1022 }
1023
1024 int
1025 dp_send_echo_reply(struct datapath *dp, const struct sender *sender,
1026                    const struct ofp_header *rq)
1027 {
1028         struct sk_buff *skb;
1029         struct ofp_header *reply;
1030
1031         reply = alloc_openflow_skb(dp, ntohs(rq->length), OFPT_ECHO_REPLY,
1032                                    sender, &skb);
1033         if (!reply)
1034                 return -ENOMEM;
1035
1036         memcpy(reply + 1, rq + 1, ntohs(rq->length) - sizeof *rq);
1037         return send_openflow_skb(skb, sender);
1038 }
1039
1040 /* Generic Netlink interface.
1041  *
1042  * See netlink(7) for an introduction to netlink.  See
1043  * http://linux-net.osdl.org/index.php/Netlink for more information and
1044  * pointers on how to work with netlink and Generic Netlink in the kernel and
1045  * in userspace. */
1046
1047 static struct genl_family dp_genl_family = {
1048         .id = GENL_ID_GENERATE,
1049         .hdrsize = 0,
1050         .name = DP_GENL_FAMILY_NAME,
1051         .version = 1,
1052         .maxattr = DP_GENL_A_MAX,
1053 };
1054
1055 /* Attribute policy: what each attribute may contain.  */
1056 static struct nla_policy dp_genl_policy[DP_GENL_A_MAX + 1] = {
1057         [DP_GENL_A_DP_IDX] = { .type = NLA_U32 },
1058         [DP_GENL_A_MC_GROUP] = { .type = NLA_U32 },
1059         [DP_GENL_A_PORTNAME] = { .type = NLA_STRING }
1060 };
1061
1062 static int dp_genl_add(struct sk_buff *skb, struct genl_info *info)
1063 {
1064         if (!info->attrs[DP_GENL_A_DP_IDX])
1065                 return -EINVAL;
1066
1067         return new_dp(nla_get_u32(info->attrs[DP_GENL_A_DP_IDX]));
1068 }
1069
1070 static struct genl_ops dp_genl_ops_add_dp = {
1071         .cmd = DP_GENL_C_ADD_DP,
1072         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1073         .policy = dp_genl_policy,
1074         .doit = dp_genl_add,
1075         .dumpit = NULL,
1076 };
1077
1078 struct datapath *dp_get(int dp_idx)
1079 {
1080         if (dp_idx < 0 || dp_idx > DP_MAX)
1081                 return NULL;
1082         return rcu_dereference(dps[dp_idx]);
1083 }
1084
1085 static int dp_genl_del(struct sk_buff *skb, struct genl_info *info)
1086 {
1087         struct datapath *dp;
1088         int err;
1089
1090         if (!info->attrs[DP_GENL_A_DP_IDX])
1091                 return -EINVAL;
1092
1093         dp = dp_get(nla_get_u32((info->attrs[DP_GENL_A_DP_IDX])));
1094         if (!dp)
1095                 err = -ENOENT;
1096         else {
1097                 del_dp(dp);
1098                 err = 0;
1099         }
1100         return err;
1101 }
1102
1103 static struct genl_ops dp_genl_ops_del_dp = {
1104         .cmd = DP_GENL_C_DEL_DP,
1105         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1106         .policy = dp_genl_policy,
1107         .doit = dp_genl_del,
1108         .dumpit = NULL,
1109 };
1110
1111 /* Queries a datapath for related information.  Currently the only relevant
1112  * information is the datapath's multicast group ID.  Really we want one
1113  * multicast group per datapath, but because of locking issues[*] we can't
1114  * easily get one.  Thus, every datapath will currently return the same
1115  * global multicast group ID, but in the future it would be nice to fix that.
1116  *
1117  * [*] dp_genl_add, to add a new datapath, is called under the genl_lock
1118  *       mutex, and genl_register_mc_group, called to acquire a new multicast
1119  *       group ID, also acquires genl_lock, thus deadlock.
1120  */
1121 static int dp_genl_query(struct sk_buff *skb, struct genl_info *info)
1122 {
1123         struct datapath *dp;
1124         struct sk_buff *ans_skb = NULL;
1125         int dp_idx;
1126         int err = -ENOMEM;
1127
1128         if (!info->attrs[DP_GENL_A_DP_IDX])
1129                 return -EINVAL;
1130
1131         rcu_read_lock();
1132         dp_idx = nla_get_u32((info->attrs[DP_GENL_A_DP_IDX]));
1133         dp = dp_get(dp_idx);
1134         if (!dp)
1135                 err = -ENOENT;
1136         else {
1137                 void *data;
1138                 ans_skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1139                 if (!ans_skb) {
1140                         err = -ENOMEM;
1141                         goto err;
1142                 }
1143                 data = genlmsg_put_reply(ans_skb, info, &dp_genl_family,
1144                                          0, DP_GENL_C_QUERY_DP);
1145                 if (data == NULL) {
1146                         err = -ENOMEM;
1147                         goto err;
1148                 }
1149                 NLA_PUT_U32(ans_skb, DP_GENL_A_DP_IDX, dp_idx);
1150                 NLA_PUT_U32(ans_skb, DP_GENL_A_MC_GROUP, mc_group.id);
1151
1152                 genlmsg_end(ans_skb, data);
1153                 err = genlmsg_reply(ans_skb, info);
1154                 if (!err)
1155                         ans_skb = NULL;
1156         }
1157 err:
1158 nla_put_failure:
1159         if (ans_skb)
1160                 kfree_skb(ans_skb);
1161         rcu_read_unlock();
1162         return err;
1163 }
1164
1165 static struct genl_ops dp_genl_ops_query_dp = {
1166         .cmd = DP_GENL_C_QUERY_DP,
1167         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1168         .policy = dp_genl_policy,
1169         .doit = dp_genl_query,
1170         .dumpit = NULL,
1171 };
1172
1173 static int dp_genl_add_del_port(struct sk_buff *skb, struct genl_info *info)
1174 {
1175         struct datapath *dp;
1176         struct net_device *port;
1177         int err;
1178
1179         if (!info->attrs[DP_GENL_A_DP_IDX] || !info->attrs[DP_GENL_A_PORTNAME])
1180                 return -EINVAL;
1181
1182         /* Get datapath. */
1183         dp = dp_get(nla_get_u32(info->attrs[DP_GENL_A_DP_IDX]));
1184         if (!dp) {
1185                 err = -ENOENT;
1186                 goto out;
1187         }
1188
1189         /* Get interface to add/remove. */
1190         port = dev_get_by_name(&init_net,
1191                         nla_data(info->attrs[DP_GENL_A_PORTNAME]));
1192         if (!port) {
1193                 err = -ENOENT;
1194                 goto out;
1195         }
1196
1197         /* Execute operation. */
1198         if (info->genlhdr->cmd == DP_GENL_C_ADD_PORT)
1199                 err = add_switch_port(dp, port);
1200         else {
1201                 if (port->br_port == NULL || port->br_port->dp != dp) {
1202                         err = -ENOENT;
1203                         goto out_put;
1204                 }
1205                 err = dp_del_switch_port(port->br_port);
1206         }
1207
1208 out_put:
1209         dev_put(port);
1210 out:
1211         return err;
1212 }
1213
1214 static struct genl_ops dp_genl_ops_add_port = {
1215         .cmd = DP_GENL_C_ADD_PORT,
1216         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1217         .policy = dp_genl_policy,
1218         .doit = dp_genl_add_del_port,
1219         .dumpit = NULL,
1220 };
1221
1222 static struct genl_ops dp_genl_ops_del_port = {
1223         .cmd = DP_GENL_C_DEL_PORT,
1224         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1225         .policy = dp_genl_policy,
1226         .doit = dp_genl_add_del_port,
1227         .dumpit = NULL,
1228 };
1229
1230 static int dp_genl_openflow(struct sk_buff *skb, struct genl_info *info)
1231 {
1232         struct nlattr *va = info->attrs[DP_GENL_A_OPENFLOW];
1233         struct datapath *dp;
1234         struct ofp_header *oh;
1235         struct sender sender;
1236         int err;
1237
1238         if (!info->attrs[DP_GENL_A_DP_IDX] || !va)
1239                 return -EINVAL;
1240
1241         dp = dp_get(nla_get_u32(info->attrs[DP_GENL_A_DP_IDX]));
1242         if (!dp)
1243                 return -ENOENT;
1244
1245         if (nla_len(va) < sizeof(struct ofp_header))
1246                 return -EINVAL;
1247         oh = nla_data(va);
1248
1249         sender.xid = oh->xid;
1250         sender.pid = info->snd_pid;
1251         sender.seq = info->snd_seq;
1252
1253         mutex_lock(&dp_mutex);
1254         err = fwd_control_input(dp->chain, &sender,
1255                                 nla_data(va), nla_len(va));
1256         mutex_unlock(&dp_mutex);
1257         return err;
1258 }
1259
1260 static struct nla_policy dp_genl_openflow_policy[DP_GENL_A_MAX + 1] = {
1261         [DP_GENL_A_DP_IDX] = { .type = NLA_U32 },
1262 };
1263
1264 static int desc_stats_dump(struct datapath *dp, void *state,
1265                             void *body, int *body_len)
1266 {
1267         struct ofp_desc_stats *ods = body;
1268         int n_bytes = sizeof *ods;
1269
1270         if (n_bytes > *body_len) {
1271                 return -ENOBUFS;
1272         }
1273         *body_len = n_bytes;
1274
1275         strncpy(ods->mfr_desc, mfr_desc, sizeof ods->mfr_desc);
1276         strncpy(ods->hw_desc, hw_desc, sizeof ods->hw_desc);
1277         strncpy(ods->sw_desc, sw_desc, sizeof ods->sw_desc);
1278         strncpy(ods->serial_num, serial_num, sizeof ods->serial_num);
1279
1280         return 0;
1281 }
1282
1283 struct flow_stats_state {
1284         int table_idx;
1285         struct sw_table_position position;
1286         const struct ofp_flow_stats_request *rq;
1287
1288         void *body;
1289         int bytes_used, bytes_allocated;
1290 };
1291
1292 static int flow_stats_init(struct datapath *dp, const void *body, int body_len,
1293                            void **state)
1294 {
1295         const struct ofp_flow_stats_request *fsr = body;
1296         struct flow_stats_state *s = kmalloc(sizeof *s, GFP_ATOMIC);
1297         if (!s)
1298                 return -ENOMEM;
1299         s->table_idx = fsr->table_id == 0xff ? 0 : fsr->table_id;
1300         memset(&s->position, 0, sizeof s->position);
1301         s->rq = fsr;
1302         *state = s;
1303         return 0;
1304 }
1305
1306 static int flow_stats_dump_callback(struct sw_flow *flow, void *private)
1307 {
1308         struct sw_flow_actions *sf_acts = rcu_dereference(flow->sf_acts);
1309         struct flow_stats_state *s = private;
1310         struct ofp_flow_stats *ofs;
1311         int length;
1312
1313         length = sizeof *ofs + sf_acts->actions_len;
1314         if (length + s->bytes_used > s->bytes_allocated)
1315                 return 1;
1316
1317         ofs = s->body + s->bytes_used;
1318         ofs->length          = htons(length);
1319         ofs->table_id        = s->table_idx;
1320         ofs->pad             = 0;
1321         ofs->match.wildcards = htonl(flow->key.wildcards);
1322         ofs->match.in_port   = flow->key.in_port;
1323         memcpy(ofs->match.dl_src, flow->key.dl_src, ETH_ALEN);
1324         memcpy(ofs->match.dl_dst, flow->key.dl_dst, ETH_ALEN);
1325         ofs->match.dl_vlan   = flow->key.dl_vlan;
1326         ofs->match.dl_type   = flow->key.dl_type;
1327         ofs->match.nw_src    = flow->key.nw_src;
1328         ofs->match.nw_dst    = flow->key.nw_dst;
1329         ofs->match.nw_proto  = flow->key.nw_proto;
1330         ofs->match.pad       = 0;
1331         ofs->match.tp_src    = flow->key.tp_src;
1332         ofs->match.tp_dst    = flow->key.tp_dst;
1333         ofs->duration        = htonl((jiffies - flow->init_time) / HZ);
1334         ofs->priority        = htons(flow->priority);
1335         ofs->idle_timeout    = htons(flow->idle_timeout);
1336         ofs->hard_timeout    = htons(flow->hard_timeout);
1337         memset(ofs->pad2, 0, sizeof ofs->pad2);
1338         ofs->packet_count    = cpu_to_be64(flow->packet_count);
1339         ofs->byte_count      = cpu_to_be64(flow->byte_count);
1340         memcpy(ofs->actions, sf_acts->actions, sf_acts->actions_len);
1341
1342         s->bytes_used += length;
1343         return 0;
1344 }
1345
1346 static int flow_stats_dump(struct datapath *dp, void *state,
1347                            void *body, int *body_len)
1348 {
1349         struct flow_stats_state *s = state;
1350         struct sw_flow_key match_key;
1351         int error = 0;
1352
1353         s->bytes_used = 0;
1354         s->bytes_allocated = *body_len;
1355         s->body = body;
1356
1357         flow_extract_match(&match_key, &s->rq->match);
1358         while (s->table_idx < dp->chain->n_tables
1359                && (s->rq->table_id == 0xff || s->rq->table_id == s->table_idx))
1360         {
1361                 struct sw_table *table = dp->chain->tables[s->table_idx];
1362
1363                 error = table->iterate(table, &match_key, &s->position,
1364                                        flow_stats_dump_callback, s);
1365                 if (error)
1366                         break;
1367
1368                 s->table_idx++;
1369                 memset(&s->position, 0, sizeof s->position);
1370         }
1371         *body_len = s->bytes_used;
1372
1373         /* If error is 0, we're done.
1374          * Otherwise, if some bytes were used, there are more flows to come.
1375          * Otherwise, we were not able to fit even a single flow in the body,
1376          * which indicates that we have a single flow with too many actions to
1377          * fit.  We won't ever make any progress at that rate, so give up. */
1378         return !error ? 0 : s->bytes_used ? 1 : -ENOMEM;
1379 }
1380
1381 static void flow_stats_done(void *state)
1382 {
1383         kfree(state);
1384 }
1385
1386 static int aggregate_stats_init(struct datapath *dp,
1387                                 const void *body, int body_len,
1388                                 void **state)
1389 {
1390         *state = (void *)body;
1391         return 0;
1392 }
1393
1394 static int aggregate_stats_dump_callback(struct sw_flow *flow, void *private)
1395 {
1396         struct ofp_aggregate_stats_reply *rpy = private;
1397         rpy->packet_count += flow->packet_count;
1398         rpy->byte_count += flow->byte_count;
1399         rpy->flow_count++;
1400         return 0;
1401 }
1402
1403 static int aggregate_stats_dump(struct datapath *dp, void *state,
1404                                 void *body, int *body_len)
1405 {
1406         struct ofp_aggregate_stats_request *rq = state;
1407         struct ofp_aggregate_stats_reply *rpy;
1408         struct sw_table_position position;
1409         struct sw_flow_key match_key;
1410         int table_idx;
1411
1412         if (*body_len < sizeof *rpy)
1413                 return -ENOBUFS;
1414         rpy = body;
1415         *body_len = sizeof *rpy;
1416
1417         memset(rpy, 0, sizeof *rpy);
1418
1419         flow_extract_match(&match_key, &rq->match);
1420         table_idx = rq->table_id == 0xff ? 0 : rq->table_id;
1421         memset(&position, 0, sizeof position);
1422         while (table_idx < dp->chain->n_tables
1423                && (rq->table_id == 0xff || rq->table_id == table_idx))
1424         {
1425                 struct sw_table *table = dp->chain->tables[table_idx];
1426                 int error;
1427
1428                 error = table->iterate(table, &match_key, &position,
1429                                        aggregate_stats_dump_callback, rpy);
1430                 if (error)
1431                         return error;
1432
1433                 table_idx++;
1434                 memset(&position, 0, sizeof position);
1435         }
1436
1437         rpy->packet_count = cpu_to_be64(rpy->packet_count);
1438         rpy->byte_count = cpu_to_be64(rpy->byte_count);
1439         rpy->flow_count = htonl(rpy->flow_count);
1440         return 0;
1441 }
1442
1443 static int table_stats_dump(struct datapath *dp, void *state,
1444                             void *body, int *body_len)
1445 {
1446         struct ofp_table_stats *ots;
1447         int n_bytes = dp->chain->n_tables * sizeof *ots;
1448         int i;
1449         if (n_bytes > *body_len)
1450                 return -ENOBUFS;
1451         *body_len = n_bytes;
1452         for (i = 0, ots = body; i < dp->chain->n_tables; i++, ots++) {
1453                 struct sw_table_stats stats;
1454                 dp->chain->tables[i]->stats(dp->chain->tables[i], &stats);
1455                 strncpy(ots->name, stats.name, sizeof ots->name);
1456                 ots->table_id = i;
1457                 ots->wildcards = htonl(stats.wildcards);
1458                 memset(ots->pad, 0, sizeof ots->pad);
1459                 ots->max_entries = htonl(stats.max_flows);
1460                 ots->active_count = htonl(stats.n_flows);
1461                 ots->lookup_count = cpu_to_be64(stats.n_lookup);
1462                 ots->matched_count = cpu_to_be64(stats.n_matched);
1463         }
1464         return 0;
1465 }
1466
1467 struct port_stats_state {
1468         int port;
1469 };
1470
1471 static int port_stats_init(struct datapath *dp, const void *body, int body_len,
1472                            void **state)
1473 {
1474         struct port_stats_state *s = kmalloc(sizeof *s, GFP_ATOMIC);
1475         if (!s)
1476                 return -ENOMEM;
1477         s->port = 0;
1478         *state = s;
1479         return 0;
1480 }
1481
1482 static int port_stats_dump(struct datapath *dp, void *state,
1483                            void *body, int *body_len)
1484 {
1485         struct port_stats_state *s = state;
1486         struct ofp_port_stats *ops;
1487         int n_ports, max_ports;
1488         int i;
1489
1490         max_ports = *body_len / sizeof *ops;
1491         if (!max_ports)
1492                 return -ENOMEM;
1493         ops = body;
1494
1495         n_ports = 0;
1496         for (i = s->port; i < OFPP_MAX && n_ports < max_ports; i++) {
1497                 struct net_bridge_port *p = dp->ports[i];
1498                 struct net_device_stats *stats;
1499                 if (!p)
1500                         continue;
1501                 stats = p->dev->get_stats(p->dev);
1502                 ops->port_no = htons(p->port_no);
1503                 memset(ops->pad, 0, sizeof ops->pad);
1504                 ops->rx_packets   = cpu_to_be64(stats->rx_packets);
1505                 ops->tx_packets   = cpu_to_be64(stats->tx_packets);
1506                 ops->rx_bytes     = cpu_to_be64(stats->rx_bytes);
1507                 ops->tx_bytes     = cpu_to_be64(stats->tx_bytes);
1508                 ops->rx_dropped   = cpu_to_be64(stats->rx_dropped);
1509                 ops->tx_dropped   = cpu_to_be64(stats->tx_dropped);
1510                 ops->rx_errors    = cpu_to_be64(stats->rx_errors);
1511                 ops->tx_errors    = cpu_to_be64(stats->tx_errors);
1512                 ops->rx_frame_err = cpu_to_be64(stats->rx_frame_errors);
1513                 ops->rx_over_err  = cpu_to_be64(stats->rx_over_errors);
1514                 ops->rx_crc_err   = cpu_to_be64(stats->rx_crc_errors);
1515                 ops->collisions   = cpu_to_be64(stats->collisions);
1516                 n_ports++;
1517                 ops++;
1518         }
1519         s->port = i;
1520         *body_len = n_ports * sizeof *ops;
1521         return n_ports >= max_ports;
1522 }
1523
1524 static void port_stats_done(void *state)
1525 {
1526         kfree(state);
1527 }
1528
1529 struct stats_type {
1530         /* Minimum and maximum acceptable number of bytes in body member of
1531          * struct ofp_stats_request. */
1532         size_t min_body, max_body;
1533
1534         /* Prepares to dump some kind of statistics on 'dp'.  'body' and
1535          * 'body_len' are the 'body' member of the struct ofp_stats_request.
1536          * Returns zero if successful, otherwise a negative error code.
1537          * May initialize '*state' to state information.  May be null if no
1538          * initialization is required.*/
1539         int (*init)(struct datapath *dp, const void *body, int body_len,
1540                     void **state);
1541
1542         /* Dumps statistics for 'dp' into the '*body_len' bytes at 'body', and
1543          * modifies '*body_len' to reflect the number of bytes actually used.
1544          * ('body' will be transmitted as the 'body' member of struct
1545          * ofp_stats_reply.) */
1546         int (*dump)(struct datapath *dp, void *state,
1547                     void *body, int *body_len);
1548
1549         /* Cleans any state created by the init or dump functions.  May be null
1550          * if no cleanup is required. */
1551         void (*done)(void *state);
1552 };
1553
1554 static const struct stats_type stats[] = {
1555         [OFPST_DESC] = {
1556                 0,
1557                 0,
1558                 NULL,
1559                 desc_stats_dump,
1560                 NULL
1561         },
1562         [OFPST_FLOW] = {
1563                 sizeof(struct ofp_flow_stats_request),
1564                 sizeof(struct ofp_flow_stats_request),
1565                 flow_stats_init,
1566                 flow_stats_dump,
1567                 flow_stats_done
1568         },
1569         [OFPST_AGGREGATE] = {
1570                 sizeof(struct ofp_aggregate_stats_request),
1571                 sizeof(struct ofp_aggregate_stats_request),
1572                 aggregate_stats_init,
1573                 aggregate_stats_dump,
1574                 NULL
1575         },
1576         [OFPST_TABLE] = {
1577                 0,
1578                 0,
1579                 NULL,
1580                 table_stats_dump,
1581                 NULL
1582         },
1583         [OFPST_PORT] = {
1584                 0,
1585                 0,
1586                 port_stats_init,
1587                 port_stats_dump,
1588                 port_stats_done
1589         },
1590 };
1591
1592 static int
1593 dp_genl_openflow_dumpit(struct sk_buff *skb, struct netlink_callback *cb)
1594 {
1595         struct datapath *dp;
1596         struct sender sender;
1597         const struct stats_type *s;
1598         struct ofp_stats_reply *osr;
1599         int dp_idx;
1600         int max_openflow_len, body_len;
1601         void *body;
1602         int err;
1603
1604         /* Set up the cleanup function for this dump.  Linux 2.6.20 and later
1605          * support setting up cleanup functions via the .doneit member of
1606          * struct genl_ops.  This kluge supports earlier versions also. */
1607         cb->done = dp_genl_openflow_done;
1608
1609         sender.pid = NETLINK_CB(cb->skb).pid;
1610         sender.seq = cb->nlh->nlmsg_seq;
1611         if (!cb->args[0]) {
1612                 struct nlattr *attrs[DP_GENL_A_MAX + 1];
1613                 struct ofp_stats_request *rq;
1614                 struct nlattr *va;
1615                 size_t len, body_len;
1616                 int type;
1617
1618                 err = nlmsg_parse(cb->nlh, GENL_HDRLEN, attrs, DP_GENL_A_MAX,
1619                                   dp_genl_openflow_policy);
1620                 if (err < 0)
1621                         return err;
1622
1623                 if (!attrs[DP_GENL_A_DP_IDX])
1624                         return -EINVAL;
1625                 dp_idx = nla_get_u16(attrs[DP_GENL_A_DP_IDX]);
1626                 dp = dp_get(dp_idx);
1627                 if (!dp)
1628                         return -ENOENT;
1629
1630                 va = attrs[DP_GENL_A_OPENFLOW];
1631                 len = nla_len(va);
1632                 if (!va || len < sizeof *rq)
1633                         return -EINVAL;
1634
1635                 rq = nla_data(va);
1636                 sender.xid = rq->header.xid;
1637                 type = ntohs(rq->type);
1638                 if (rq->header.version != OFP_VERSION) {
1639                         dp_send_error_msg(dp, &sender, OFPET_BAD_REQUEST,
1640                                           OFPBRC_BAD_VERSION, rq, len);
1641                         return -EINVAL;
1642                 }
1643                 if (rq->header.type != OFPT_STATS_REQUEST
1644                     || ntohs(rq->header.length) != len)
1645                         return -EINVAL;
1646
1647                 if (type >= ARRAY_SIZE(stats) || !stats[type].dump) {
1648                         dp_send_error_msg(dp, &sender, OFPET_BAD_REQUEST,
1649                                           OFPBRC_BAD_STAT, rq, len);
1650                         return -EINVAL;
1651                 }
1652
1653                 s = &stats[type];
1654                 body_len = len - offsetof(struct ofp_stats_request, body);
1655                 if (body_len < s->min_body || body_len > s->max_body)
1656                         return -EINVAL;
1657
1658                 cb->args[0] = 1;
1659                 cb->args[1] = dp_idx;
1660                 cb->args[2] = type;
1661                 cb->args[3] = rq->header.xid;
1662                 if (s->init) {
1663                         void *state;
1664                         err = s->init(dp, rq->body, body_len, &state);
1665                         if (err)
1666                                 return err;
1667                         cb->args[4] = (long) state;
1668                 }
1669         } else if (cb->args[0] == 1) {
1670                 sender.xid = cb->args[3];
1671                 dp_idx = cb->args[1];
1672                 s = &stats[cb->args[2]];
1673
1674                 dp = dp_get(dp_idx);
1675                 if (!dp)
1676                         return -ENOENT;
1677         } else {
1678                 return 0;
1679         }
1680
1681         osr = put_openflow_headers(dp, skb, OFPT_STATS_REPLY, &sender,
1682                                    &max_openflow_len);
1683         if (IS_ERR(osr))
1684                 return PTR_ERR(osr);
1685         osr->type = htons(s - stats);
1686         osr->flags = 0;
1687         resize_openflow_skb(skb, &osr->header, max_openflow_len);
1688         body = osr->body;
1689         body_len = max_openflow_len - offsetof(struct ofp_stats_reply, body);
1690
1691         err = s->dump(dp, (void *) cb->args[4], body, &body_len);
1692         if (err >= 0) {
1693                 if (!err)
1694                         cb->args[0] = 2;
1695                 else
1696                         osr->flags = ntohs(OFPSF_REPLY_MORE);
1697                 resize_openflow_skb(skb, &osr->header,
1698                                     (offsetof(struct ofp_stats_reply, body)
1699                                      + body_len));
1700                 err = skb->len;
1701         }
1702
1703         return err;
1704 }
1705
1706 static int
1707 dp_genl_openflow_done(struct netlink_callback *cb)
1708 {
1709         if (cb->args[0]) {
1710                 const struct stats_type *s = &stats[cb->args[2]];
1711                 if (s->done)
1712                         s->done((void *) cb->args[4]);
1713         }
1714         return 0;
1715 }
1716
1717 static struct genl_ops dp_genl_ops_openflow = {
1718         .cmd = DP_GENL_C_OPENFLOW,
1719         .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1720         .policy = dp_genl_openflow_policy,
1721         .doit = dp_genl_openflow,
1722         .dumpit = dp_genl_openflow_dumpit,
1723 };
1724
1725 static struct genl_ops *dp_genl_all_ops[] = {
1726         /* Keep this operation first.  Generic Netlink dispatching
1727          * looks up operations with linear search, so we want it at the
1728          * front. */
1729         &dp_genl_ops_openflow,
1730
1731         &dp_genl_ops_add_dp,
1732         &dp_genl_ops_del_dp,
1733         &dp_genl_ops_query_dp,
1734         &dp_genl_ops_add_port,
1735         &dp_genl_ops_del_port,
1736 };
1737
1738 static int dp_init_netlink(void)
1739 {
1740         int err;
1741         int i;
1742
1743         err = genl_register_family(&dp_genl_family);
1744         if (err)
1745                 return err;
1746
1747         for (i = 0; i < ARRAY_SIZE(dp_genl_all_ops); i++) {
1748                 err = genl_register_ops(&dp_genl_family, dp_genl_all_ops[i]);
1749                 if (err)
1750                         goto err_unregister;
1751         }
1752
1753         strcpy(mc_group.name, "openflow");
1754         err = genl_register_mc_group(&dp_genl_family, &mc_group);
1755         if (err < 0)
1756                 goto err_unregister;
1757
1758         return 0;
1759
1760 err_unregister:
1761         genl_unregister_family(&dp_genl_family);
1762                 return err;
1763 }
1764
1765 static void dp_uninit_netlink(void)
1766 {
1767         genl_unregister_family(&dp_genl_family);
1768 }
1769
1770 static int __init dp_init(void)
1771 {
1772         int err;
1773
1774         printk("OpenFlow "VERSION", built "__DATE__" "__TIME__", "
1775                "protocol 0x%02x\n", OFP_VERSION);
1776
1777         err = flow_init();
1778         if (err)
1779                 goto error;
1780
1781         err = register_netdevice_notifier(&dp_device_notifier);
1782         if (err)
1783                 goto error_flow_exit;
1784
1785         err = dp_init_netlink();
1786         if (err)
1787                 goto error_unreg_notifier;
1788
1789         /* Hook into callback used by the bridge to intercept packets.
1790          * Parasites we are. */
1791         if (br_handle_frame_hook)
1792                 printk("openflow: hijacking bridge hook\n");
1793         br_handle_frame_hook = dp_frame_hook;
1794
1795         return 0;
1796
1797 error_unreg_notifier:
1798         unregister_netdevice_notifier(&dp_device_notifier);
1799 error_flow_exit:
1800         flow_exit();
1801 error:
1802         printk(KERN_EMERG "openflow: failed to install!");
1803         return err;
1804 }
1805
1806 static void dp_cleanup(void)
1807 {
1808         fwd_exit();
1809         dp_uninit_netlink();
1810         unregister_netdevice_notifier(&dp_device_notifier);
1811         flow_exit();
1812         br_handle_frame_hook = NULL;
1813 }
1814
1815 module_init(dp_init);
1816 module_exit(dp_cleanup);
1817
1818 MODULE_DESCRIPTION("OpenFlow switching datapath");
1819 MODULE_AUTHOR("Copyright (c) 2007, 2008 The Board of Trustees of The Leland Stanford Junior University");
1820 MODULE_LICENSE("GPL");