datapath/datapath.c

   1 /*
   2  * Copyright (c) 2007, 2008, 2009, 2010, 2011 Nicira Networks.
   3  * Distributed under the terms of the GNU GPL version 2.
   4  *
   5  * Significant portions of this file may be copied from parts of the Linux
   6  * kernel, by Linus Torvalds and others.
   7  */
   8
   9 /* Functions for managing the dp interface/device. */
  10
  11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  12
  13 #include <linux/init.h>
  14 #include <linux/module.h>
  15 #include <linux/if_arp.h>
  16 #include <linux/if_vlan.h>
  17 #include <linux/in.h>
  18 #include <linux/ip.h>
  19 #include <linux/jhash.h>
  20 #include <linux/delay.h>
  21 #include <linux/time.h>
  22 #include <linux/etherdevice.h>
  23 #include <linux/genetlink.h>
  24 #include <linux/kernel.h>
  25 #include <linux/kthread.h>
  26 #include <linux/mutex.h>
  27 #include <linux/percpu.h>
  28 #include <linux/rcupdate.h>
  29 #include <linux/tcp.h>
  30 #include <linux/udp.h>
  31 #include <linux/version.h>
  32 #include <linux/ethtool.h>
  33 #include <linux/wait.h>
  34 #include <asm/system.h>
  35 #include <asm/div64.h>
  36 #include <asm/bug.h>
  37 #include <linux/highmem.h>
  38 #include <linux/netfilter_bridge.h>
  39 #include <linux/netfilter_ipv4.h>
  40 #include <linux/inetdevice.h>
  41 #include <linux/list.h>
  42 #include <linux/rculist.h>
  43 #include <linux/dmi.h>
  44 #include <net/inet_ecn.h>
  45 #include <net/genetlink.h>
  46
  47 #include "openvswitch/datapath-protocol.h"
  48 #include "checksum.h"
  49 #include "datapath.h"
  50 #include "actions.h"
  51 #include "flow.h"
  52 #include "table.h"
  53 #include "vlan.h"
  54 #include "vport-internal_dev.h"
  55
  56 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,18) || \
  57     LINUX_VERSION_CODE >= KERNEL_VERSION(3,1,0)
  58 #error Kernels before 2.6.18 or after 3.0 are not supported by this version of Open vSwitch.
  59 #endif
  60
  61 int (*dp_ioctl_hook)(struct net_device *dev, struct ifreq *rq, int cmd);
  62 EXPORT_SYMBOL(dp_ioctl_hook);
  63
  64 /**
  65  * DOC: Locking:
  66  *
  67  * Writes to device state (add/remove datapath, port, set operations on vports,
  68  * etc.) are protected by RTNL.
  69  *
  70  * Writes to other state (flow table modifications, set miscellaneous datapath
  71  * parameters such as drop frags, etc.) are protected by genl_mutex.  The RTNL
  72  * lock nests inside genl_mutex.
  73  *
  74  * Reads are protected by RCU.
  75  *
  76  * There are a few special cases (mostly stats) that have their own
  77  * synchronization but they nest under all of above and don't interact with
  78  * each other.
  79  */
  80
  81 /* Global list of datapaths to enable dumping them all out.
  82  * Protected by genl_mutex.
  83  */
  84 static LIST_HEAD(dps);
  85
  86 static struct vport *new_vport(const struct vport_parms *);
  87 static int queue_userspace_packets(struct datapath *, struct sk_buff *,
  88                                  const struct dp_upcall_info *);
  89
  90 /* Must be called with rcu_read_lock, genl_mutex, or RTNL lock. */
  91 struct datapath *get_dp(int dp_ifindex)
  92 {
  93         struct datapath *dp = NULL;
  94         struct net_device *dev;
  95
  96         rcu_read_lock();
  97         dev = dev_get_by_index_rcu(&init_net, dp_ifindex);
  98         if (dev) {
  99                 struct vport *vport = internal_dev_get_vport(dev);
 100                 if (vport)
 101                         dp = vport->dp;
 102         }
 103         rcu_read_unlock();
 104
 105         return dp;
 106 }
 107 EXPORT_SYMBOL_GPL(get_dp);
 108
 109 /* Must be called with genl_mutex. */
 110 static struct tbl *get_table_protected(struct datapath *dp)
 111 {
 112         return rcu_dereference_protected(dp->table, lockdep_genl_is_held());
 113 }
 114
 115 /* Must be called with rcu_read_lock or RTNL lock. */
 116 static struct vport *get_vport_protected(struct datapath *dp, u16 port_no)
 117 {
 118         return rcu_dereference_rtnl(dp->ports[port_no]);
 119 }
 120
 121 /* Must be called with rcu_read_lock or RTNL lock. */
 122 const char *dp_name(const struct datapath *dp)
 123 {
 124         return vport_get_name(rcu_dereference_rtnl(dp->ports[ODPP_LOCAL]));
 125 }
 126
 127 static int get_dpifindex(struct datapath *dp)
 128 {
 129         struct vport *local;
 130         int ifindex;
 131
 132         rcu_read_lock();
 133
 134         local = get_vport_protected(dp, ODPP_LOCAL);
 135         if (local)
 136                 ifindex = vport_get_ifindex(local);
 137         else
 138                 ifindex = 0;
 139
 140         rcu_read_unlock();
 141
 142         return ifindex;
 143 }
 144
 145 static inline size_t br_nlmsg_size(void)
 146 {
 147         return NLMSG_ALIGN(sizeof(struct ifinfomsg))
 148                + nla_total_size(IFNAMSIZ) /* IFLA_IFNAME */
 149                + nla_total_size(MAX_ADDR_LEN) /* IFLA_ADDRESS */
 150                + nla_total_size(4) /* IFLA_MASTER */
 151                + nla_total_size(4) /* IFLA_MTU */
 152                + nla_total_size(4) /* IFLA_LINK */
 153                + nla_total_size(1); /* IFLA_OPERSTATE */
 154 }
 155
 156 /* Caller must hold RTNL lock. */
 157 static int dp_fill_ifinfo(struct sk_buff *skb,
 158                           const struct vport *port,
 159                           int event, unsigned int flags)
 160 {
 161         struct datapath *dp = port->dp;
 162         int ifindex = vport_get_ifindex(port);
 163         int iflink = vport_get_iflink(port);
 164         struct ifinfomsg *hdr;
 165         struct nlmsghdr *nlh;
 166
 167         if (ifindex < 0)
 168                 return ifindex;
 169
 170         if (iflink < 0)
 171                 return iflink;
 172
 173         nlh = nlmsg_put(skb, 0, 0, event, sizeof(*hdr), flags);
 174         if (nlh == NULL)
 175                 return -EMSGSIZE;
 176
 177         hdr = nlmsg_data(nlh);
 178         hdr->ifi_family = AF_BRIDGE;
 179         hdr->__ifi_pad = 0;
 180         hdr->ifi_type = ARPHRD_ETHER;
 181         hdr->ifi_index = ifindex;
 182         hdr->ifi_flags = vport_get_flags(port);
 183         hdr->ifi_change = 0;
 184
 185         NLA_PUT_STRING(skb, IFLA_IFNAME, vport_get_name(port));
 186         NLA_PUT_U32(skb, IFLA_MASTER, get_dpifindex(dp));
 187         NLA_PUT_U32(skb, IFLA_MTU, vport_get_mtu(port));
 188 #ifdef IFLA_OPERSTATE
 189         NLA_PUT_U8(skb, IFLA_OPERSTATE,
 190                    vport_is_running(port)
 191                         ? vport_get_operstate(port)
 192                         : IF_OPER_DOWN);
 193 #endif
 194
 195         NLA_PUT(skb, IFLA_ADDRESS, ETH_ALEN, vport_get_addr(port));
 196
 197         if (ifindex != iflink)
 198                 NLA_PUT_U32(skb, IFLA_LINK,iflink);
 199
 200         return nlmsg_end(skb, nlh);
 201
 202 nla_put_failure:
 203         nlmsg_cancel(skb, nlh);
 204         return -EMSGSIZE;
 205 }
 206
 207 /* Caller must hold RTNL lock. */
 208 static void dp_ifinfo_notify(int event, struct vport *port)
 209 {
 210         struct sk_buff *skb;
 211         int err = -ENOBUFS;
 212
 213         skb = nlmsg_new(br_nlmsg_size(), GFP_KERNEL);
 214         if (skb == NULL)
 215                 goto errout;
 216
 217         err = dp_fill_ifinfo(skb, port, event, 0);
 218         if (err < 0) {
 219                 /* -EMSGSIZE implies BUG in br_nlmsg_size() */
 220                 WARN_ON(err == -EMSGSIZE);
 221                 kfree_skb(skb);
 222                 goto errout;
 223         }
 224         rtnl_notify(skb, &init_net, 0, RTNLGRP_LINK, NULL, GFP_KERNEL);
 225         return;
 226 errout:
 227         if (err < 0)
 228                 rtnl_set_sk_err(&init_net, RTNLGRP_LINK, err);
 229 }
 230
 231 static void release_dp(struct kobject *kobj)
 232 {
 233         struct datapath *dp = container_of(kobj, struct datapath, ifobj);
 234         kfree(dp);
 235 }
 236
 237 static struct kobj_type dp_ktype = {
 238         .release = release_dp
 239 };
 240
 241 static void destroy_dp_rcu(struct rcu_head *rcu)
 242 {
 243         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 244
 245         tbl_destroy((struct tbl __force *)dp->table, flow_free_tbl);
 246         free_percpu(dp->stats_percpu);
 247         kobject_put(&dp->ifobj);
 248 }
 249
 250 /* Called with RTNL lock and genl_lock. */
 251 static struct vport *new_vport(const struct vport_parms *parms)
 252 {
 253         struct vport *vport;
 254
 255         vport = vport_add(parms);
 256         if (!IS_ERR(vport)) {
 257                 struct datapath *dp = parms->dp;
 258
 259                 rcu_assign_pointer(dp->ports[parms->port_no], vport);
 260                 list_add(&vport->node, &dp->port_list);
 261
 262                 dp_ifinfo_notify(RTM_NEWLINK, vport);
 263         }
 264
 265         return vport;
 266 }
 267
 268 /* Called with RTNL lock. */
 269 int dp_detach_port(struct vport *p)
 270 {
 271         ASSERT_RTNL();
 272
 273         if (p->port_no != ODPP_LOCAL)
 274                 dp_sysfs_del_if(p);
 275         dp_ifinfo_notify(RTM_DELLINK, p);
 276
 277         /* First drop references to device. */
 278         list_del(&p->node);
 279         rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
 280
 281         /* Then destroy it. */
 282         return vport_del(p);
 283 }
 284
 285 /* Must be called with rcu_read_lock. */
 286 void dp_process_received_packet(struct vport *p, struct sk_buff *skb)
 287 {
 288         struct datapath *dp = p->dp;
 289         struct dp_stats_percpu *stats;
 290         int stats_counter_off;
 291         int error;
 292
 293         OVS_CB(skb)->vport = p;
 294
 295         if (!OVS_CB(skb)->flow) {
 296                 struct sw_flow_key key;
 297                 struct tbl_node *flow_node;
 298                 int key_len;
 299                 bool is_frag;
 300
 301                 /* Extract flow from 'skb' into 'key'. */
 302                 error = flow_extract(skb, p->port_no, &key, &key_len, &is_frag);
 303                 if (unlikely(error)) {
 304                         kfree_skb(skb);
 305                         return;
 306                 }
 307
 308                 if (is_frag && dp->drop_frags) {
 309                         consume_skb(skb);
 310                         stats_counter_off = offsetof(struct dp_stats_percpu, n_frags);
 311                         goto out;
 312                 }
 313
 314                 /* Look up flow. */
 315                 flow_node = tbl_lookup(rcu_dereference(dp->table), &key, key_len,
 316                                        flow_hash(&key, key_len), flow_cmp);
 317                 if (unlikely(!flow_node)) {
 318                         struct dp_upcall_info upcall;
 319
 320                         upcall.cmd = ODP_PACKET_CMD_MISS;
 321                         upcall.key = &key;
 322                         upcall.userdata = 0;
 323                         upcall.sample_pool = 0;
 324                         upcall.actions = NULL;
 325                         upcall.actions_len = 0;
 326                         dp_upcall(dp, skb, &upcall);
 327                         stats_counter_off = offsetof(struct dp_stats_percpu, n_missed);
 328                         goto out;
 329                 }
 330
 331                 OVS_CB(skb)->flow = flow_cast(flow_node);
 332         }
 333
 334         stats_counter_off = offsetof(struct dp_stats_percpu, n_hit);
 335         flow_used(OVS_CB(skb)->flow, skb);
 336         execute_actions(dp, skb);
 337
 338 out:
 339         /* Update datapath statistics. */
 340         local_bh_disable();
 341         stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
 342
 343         write_seqcount_begin(&stats->seqlock);
 344         (*(u64 *)((u8 *)stats + stats_counter_off))++;
 345         write_seqcount_end(&stats->seqlock);
 346
 347         local_bh_enable();
 348 }
 349
 350 static void copy_and_csum_skb(struct sk_buff *skb, void *to)
 351 {
 352         u16 csum_start, csum_offset;
 353         __wsum csum;
 354
 355         get_skb_csum_pointers(skb, &csum_start, &csum_offset);
 356         csum_start -= skb_headroom(skb);
 357
 358         skb_copy_bits(skb, 0, to, csum_start);
 359
 360         csum = skb_copy_and_csum_bits(skb, csum_start, to + csum_start,
 361                                       skb->len - csum_start, 0);
 362         *(__sum16 *)(to + csum_start + csum_offset) = csum_fold(csum);
 363 }
 364
 365 static struct genl_family dp_packet_genl_family = {
 366         .id = GENL_ID_GENERATE,
 367         .hdrsize = sizeof(struct odp_header),
 368         .name = ODP_PACKET_FAMILY,
 369         .version = 1,
 370         .maxattr = ODP_PACKET_ATTR_MAX
 371 };
 372
 373 /* Generic Netlink multicast groups for upcalls.
 374  *
 375  * We really want three unique multicast groups per datapath, but we can't even
 376  * get one, because genl_register_mc_group() takes genl_lock, which is also
 377  * held during Generic Netlink message processing, so trying to acquire
 378  * multicast groups during ODP_DP_NEW processing deadlocks.  Instead, we
 379  * preallocate a few groups and use them round-robin for datapaths.  Collision
 380  * isn't fatal--multicast listeners should check that the family is the one
 381  * that they want and discard others--but it wastes time and memory to receive
 382  * unwanted messages.
 383  */
 384 #define PACKET_N_MC_GROUPS 16
 385 static struct genl_multicast_group packet_mc_groups[PACKET_N_MC_GROUPS];
 386
 387 static u32 packet_mc_group(int dp_ifindex, u8 cmd)
 388 {
 389         u32 idx;
 390         BUILD_BUG_ON_NOT_POWER_OF_2(PACKET_N_MC_GROUPS);
 391
 392         idx = jhash_2words(dp_ifindex, cmd, 0) & (PACKET_N_MC_GROUPS - 1);
 393         return packet_mc_groups[idx].id;
 394 }
 395
 396 static int packet_register_mc_groups(void)
 397 {
 398         int i;
 399
 400         for (i = 0; i < PACKET_N_MC_GROUPS; i++) {
 401                 struct genl_multicast_group *group = &packet_mc_groups[i];
 402                 int error;
 403
 404                 sprintf(group->name, "packet%d", i);
 405                 error = genl_register_mc_group(&dp_packet_genl_family, group);
 406                 if (error)
 407                         return error;
 408         }
 409         return 0;
 410 }
 411
 412 int dp_upcall(struct datapath *dp, struct sk_buff *skb, const struct dp_upcall_info *upcall_info)
 413 {
 414         struct dp_stats_percpu *stats;
 415         int err;
 416
 417         WARN_ON_ONCE(skb_shared(skb));
 418
 419         forward_ip_summed(skb, true);
 420
 421         /* Break apart GSO packets into their component pieces.  Otherwise
 422          * userspace may try to stuff a 64kB packet into a 1500-byte MTU. */
 423         if (skb_is_gso(skb)) {
 424                 struct sk_buff *nskb = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
 425
 426                 if (IS_ERR(nskb)) {
 427                         kfree_skb(skb);
 428                         err = PTR_ERR(nskb);
 429                         goto err;
 430                 }
 431                 consume_skb(skb);
 432                 skb = nskb;
 433         }
 434
 435         err = queue_userspace_packets(dp, skb, upcall_info);
 436         if (err)
 437                 goto err;
 438
 439         return 0;
 440
 441 err:
 442         local_bh_disable();
 443         stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
 444
 445         write_seqcount_begin(&stats->seqlock);
 446         stats->n_lost++;
 447         write_seqcount_end(&stats->seqlock);
 448
 449         local_bh_enable();
 450
 451         return err;
 452 }
 453
 454 /* Send each packet in the 'skb' list to userspace for 'dp' as directed by
 455  * 'upcall_info'.  There will be only one packet unless we broke up a GSO
 456  * packet.
 457  */
 458 static int queue_userspace_packets(struct datapath *dp, struct sk_buff *skb,
 459                                  const struct dp_upcall_info *upcall_info)
 460 {
 461         int dp_ifindex;
 462         u32 group;
 463         struct sk_buff *nskb;
 464         int err;
 465
 466         dp_ifindex = get_dpifindex(dp);
 467         if (!dp_ifindex) {
 468                 err = -ENODEV;
 469                 nskb = skb->next;
 470                 goto err_kfree_skbs;
 471         }
 472
 473         group = packet_mc_group(dp_ifindex, upcall_info->cmd);
 474
 475         do {
 476                 struct odp_header *upcall;
 477                 struct sk_buff *user_skb; /* to be queued to userspace */
 478                 struct nlattr *nla;
 479                 unsigned int len;
 480
 481                 nskb = skb->next;
 482                 skb->next = NULL;
 483
 484                 err = vlan_deaccel_tag(skb);
 485                 if (unlikely(err))
 486                         goto err_kfree_skbs;
 487
 488                 if (nla_attr_size(skb->len) > USHRT_MAX) {
 489                         err = -EFBIG;
 490                         goto err_kfree_skbs;
 491                 }
 492
 493                 len = sizeof(struct odp_header);
 494                 len += nla_total_size(skb->len);
 495                 len += nla_total_size(FLOW_BUFSIZE);
 496                 if (upcall_info->userdata)
 497                         len += nla_total_size(8);
 498                 if (upcall_info->sample_pool)
 499                         len += nla_total_size(4);
 500                 if (upcall_info->actions_len)
 501                         len += nla_total_size(upcall_info->actions_len);
 502
 503                 user_skb = genlmsg_new(len, GFP_ATOMIC);
 504                 if (!user_skb) {
 505                         netlink_set_err(INIT_NET_GENL_SOCK, 0, group, -ENOBUFS);
 506                         err = -ENOMEM;
 507                         goto err_kfree_skbs;
 508                 }
 509
 510                 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family, 0, upcall_info->cmd);
 511                 upcall->dp_ifindex = dp_ifindex;
 512
 513                 nla = nla_nest_start(user_skb, ODP_PACKET_ATTR_KEY);
 514                 flow_to_nlattrs(upcall_info->key, user_skb);
 515                 nla_nest_end(user_skb, nla);
 516
 517                 if (upcall_info->userdata)
 518                         nla_put_u64(user_skb, ODP_PACKET_ATTR_USERDATA, upcall_info->userdata);
 519                 if (upcall_info->sample_pool)
 520                         nla_put_u32(user_skb, ODP_PACKET_ATTR_SAMPLE_POOL, upcall_info->sample_pool);
 521                 if (upcall_info->actions_len) {
 522                         const struct nlattr *actions = upcall_info->actions;
 523                         u32 actions_len = upcall_info->actions_len;
 524
 525                         nla = nla_nest_start(user_skb, ODP_PACKET_ATTR_ACTIONS);
 526                         memcpy(__skb_put(user_skb, actions_len), actions, actions_len);
 527                         nla_nest_end(user_skb, nla);
 528                 }
 529
 530                 nla = __nla_reserve(user_skb, ODP_PACKET_ATTR_PACKET, skb->len);
 531                 if (skb->ip_summed == CHECKSUM_PARTIAL)
 532                         copy_and_csum_skb(skb, nla_data(nla));
 533                 else
 534                         skb_copy_bits(skb, 0, nla_data(nla), skb->len);
 535
 536                 err = genlmsg_multicast(user_skb, 0, group, GFP_ATOMIC);
 537                 if (err)
 538                         goto err_kfree_skbs;
 539
 540                 consume_skb(skb);
 541                 skb = nskb;
 542         } while (skb);
 543         return 0;
 544
 545 err_kfree_skbs:
 546         kfree_skb(skb);
 547         while ((skb = nskb) != NULL) {
 548                 nskb = skb->next;
 549                 kfree_skb(skb);
 550         }
 551         return err;
 552 }
 553
 554 /* Called with genl_mutex. */
 555 static int flush_flows(int dp_ifindex)
 556 {
 557         struct tbl *old_table;
 558         struct tbl *new_table;
 559         struct datapath *dp;
 560
 561         dp = get_dp(dp_ifindex);
 562         if (!dp)
 563                 return -ENODEV;
 564
 565         old_table = get_table_protected(dp);
 566         new_table = tbl_create(TBL_MIN_BUCKETS);
 567         if (!new_table)
 568                 return -ENOMEM;
 569
 570         rcu_assign_pointer(dp->table, new_table);
 571
 572         tbl_deferred_destroy(old_table, flow_free_tbl);
 573
 574         return 0;
 575 }
 576
 577 static int validate_actions(const struct nlattr *attr)
 578 {
 579         const struct nlattr *a;
 580         int rem;
 581
 582         nla_for_each_nested(a, attr, rem) {
 583                 static const u32 action_lens[ODP_ACTION_ATTR_MAX + 1] = {
 584                         [ODP_ACTION_ATTR_OUTPUT] = 4,
 585                         [ODP_ACTION_ATTR_USERSPACE] = 8,
 586                         [ODP_ACTION_ATTR_SET_DL_TCI] = 2,
 587                         [ODP_ACTION_ATTR_STRIP_VLAN] = 0,
 588                         [ODP_ACTION_ATTR_SET_DL_SRC] = ETH_ALEN,
 589                         [ODP_ACTION_ATTR_SET_DL_DST] = ETH_ALEN,
 590                         [ODP_ACTION_ATTR_SET_NW_SRC] = 4,
 591                         [ODP_ACTION_ATTR_SET_NW_DST] = 4,
 592                         [ODP_ACTION_ATTR_SET_NW_TOS] = 1,
 593                         [ODP_ACTION_ATTR_SET_TP_SRC] = 2,
 594                         [ODP_ACTION_ATTR_SET_TP_DST] = 2,
 595                         [ODP_ACTION_ATTR_SET_TUNNEL] = 8,
 596                         [ODP_ACTION_ATTR_SET_PRIORITY] = 4,
 597                         [ODP_ACTION_ATTR_POP_PRIORITY] = 0,
 598                 };
 599                 int type = nla_type(a);
 600
 601                 if (type > ODP_ACTION_ATTR_MAX || nla_len(a) != action_lens[type])
 602                         return -EINVAL;
 603
 604                 switch (type) {
 605                 case ODP_ACTION_ATTR_UNSPEC:
 606                         return -EINVAL;
 607
 608                 case ODP_ACTION_ATTR_USERSPACE:
 609                 case ODP_ACTION_ATTR_STRIP_VLAN:
 610                 case ODP_ACTION_ATTR_SET_DL_SRC:
 611                 case ODP_ACTION_ATTR_SET_DL_DST:
 612                 case ODP_ACTION_ATTR_SET_NW_SRC:
 613                 case ODP_ACTION_ATTR_SET_NW_DST:
 614                 case ODP_ACTION_ATTR_SET_TP_SRC:
 615                 case ODP_ACTION_ATTR_SET_TP_DST:
 616                 case ODP_ACTION_ATTR_SET_TUNNEL:
 617                 case ODP_ACTION_ATTR_SET_PRIORITY:
 618                 case ODP_ACTION_ATTR_POP_PRIORITY:
 619                         /* No validation needed. */
 620                         break;
 621
 622                 case ODP_ACTION_ATTR_OUTPUT:
 623                         if (nla_get_u32(a) >= DP_MAX_PORTS)
 624                                 return -EINVAL;
 625                         break;
 626
 627                 case ODP_ACTION_ATTR_SET_DL_TCI:
 628                         if (nla_get_be16(a) & htons(VLAN_CFI_MASK))
 629                                 return -EINVAL;
 630                         break;
 631
 632                 case ODP_ACTION_ATTR_SET_NW_TOS:
 633                         if (nla_get_u8(a) & INET_ECN_MASK)
 634                                 return -EINVAL;
 635                         break;
 636
 637                 default:
 638                         return -EOPNOTSUPP;
 639                 }
 640         }
 641
 642         if (rem > 0)
 643                 return -EINVAL;
 644
 645         return 0;
 646 }
 647 static void clear_stats(struct sw_flow *flow)
 648 {
 649         flow->used = 0;
 650         flow->tcp_flags = 0;
 651         flow->packet_count = 0;
 652         flow->byte_count = 0;
 653 }
 654
 655 /* Called with genl_mutex. */
 656 static int expand_table(struct datapath *dp)
 657 {
 658         struct tbl *old_table = get_table_protected(dp);
 659         struct tbl *new_table;
 660
 661         new_table = tbl_expand(old_table);
 662         if (IS_ERR(new_table))
 663                 return PTR_ERR(new_table);
 664
 665         rcu_assign_pointer(dp->table, new_table);
 666         tbl_deferred_destroy(old_table, NULL);
 667
 668         return 0;
 669 }
 670
 671 static int odp_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 672 {
 673         struct odp_header *odp_header = info->userhdr;
 674         struct nlattr **a = info->attrs;
 675         struct sw_flow_actions *acts;
 676         struct sk_buff *packet;
 677         struct sw_flow *flow;
 678         struct datapath *dp;
 679         struct ethhdr *eth;
 680         bool is_frag;
 681         int len;
 682         int err;
 683         int key_len;
 684
 685         err = -EINVAL;
 686         if (!a[ODP_PACKET_ATTR_PACKET] || !a[ODP_PACKET_ATTR_KEY] ||
 687             !a[ODP_PACKET_ATTR_ACTIONS] ||
 688             nla_len(a[ODP_PACKET_ATTR_PACKET]) < ETH_HLEN)
 689                 goto err;
 690
 691         err = validate_actions(a[ODP_PACKET_ATTR_ACTIONS]);
 692         if (err)
 693                 goto err;
 694
 695         len = nla_len(a[ODP_PACKET_ATTR_PACKET]);
 696         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 697         err = -ENOMEM;
 698         if (!packet)
 699                 goto err;
 700         skb_reserve(packet, NET_IP_ALIGN);
 701
 702         memcpy(__skb_put(packet, len), nla_data(a[ODP_PACKET_ATTR_PACKET]), len);
 703
 704         skb_reset_mac_header(packet);
 705         eth = eth_hdr(packet);
 706
 707         /* Normally, setting the skb 'protocol' field would be handled by a
 708          * call to eth_type_trans(), but it assumes there's a sending
 709          * device, which we may not have. */
 710         if (ntohs(eth->h_proto) >= 1536)
 711                 packet->protocol = eth->h_proto;
 712         else
 713                 packet->protocol = htons(ETH_P_802_2);
 714
 715         /* Build an sw_flow for sending this packet. */
 716         flow = flow_alloc();
 717         err = PTR_ERR(flow);
 718         if (IS_ERR(flow))
 719                 goto err_kfree_skb;
 720
 721         err = flow_extract(packet, -1, &flow->key, &key_len, &is_frag);
 722         if (err)
 723                 goto err_flow_put;
 724
 725         err = flow_metadata_from_nlattrs(&flow->key.eth.in_port,
 726                                          &flow->key.eth.tun_id,
 727                                          a[ODP_PACKET_ATTR_KEY]);
 728         if (err)
 729                 goto err_flow_put;
 730
 731         flow->tbl_node.hash = flow_hash(&flow->key, key_len);
 732
 733         acts = flow_actions_alloc(a[ODP_PACKET_ATTR_ACTIONS]);
 734         err = PTR_ERR(acts);
 735         if (IS_ERR(acts))
 736                 goto err_flow_put;
 737         rcu_assign_pointer(flow->sf_acts, acts);
 738
 739         OVS_CB(packet)->flow = flow;
 740
 741         rcu_read_lock();
 742         dp = get_dp(odp_header->dp_ifindex);
 743         err = -ENODEV;
 744         if (!dp)
 745                 goto err_unlock;
 746
 747         if (flow->key.eth.in_port < DP_MAX_PORTS)
 748                 OVS_CB(packet)->vport = get_vport_protected(dp,
 749                                                         flow->key.eth.in_port);
 750
 751         err = execute_actions(dp, packet);
 752         rcu_read_unlock();
 753
 754         flow_put(flow);
 755         return err;
 756
 757 err_unlock:
 758         rcu_read_unlock();
 759 err_flow_put:
 760         flow_put(flow);
 761 err_kfree_skb:
 762         kfree_skb(packet);
 763 err:
 764         return err;
 765 }
 766
 767 static const struct nla_policy packet_policy[ODP_PACKET_ATTR_MAX + 1] = {
 768         [ODP_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
 769         [ODP_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 770         [ODP_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 771 };
 772
 773 static struct genl_ops dp_packet_genl_ops[] = {
 774         { .cmd = ODP_PACKET_CMD_EXECUTE,
 775           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 776           .policy = packet_policy,
 777           .doit = odp_packet_cmd_execute
 778         }
 779 };
 780
 781 static void get_dp_stats(struct datapath *dp, struct odp_stats *stats)
 782 {
 783         int i;
 784
 785         stats->n_frags = stats->n_hit = stats->n_missed = stats->n_lost = 0;
 786         for_each_possible_cpu(i) {
 787                 const struct dp_stats_percpu *percpu_stats;
 788                 struct dp_stats_percpu local_stats;
 789                 unsigned seqcount;
 790
 791                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 792
 793                 do {
 794                         seqcount = read_seqcount_begin(&percpu_stats->seqlock);
 795                         local_stats = *percpu_stats;
 796                 } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount));
 797
 798                 stats->n_frags += local_stats.n_frags;
 799                 stats->n_hit += local_stats.n_hit;
 800                 stats->n_missed += local_stats.n_missed;
 801                 stats->n_lost += local_stats.n_lost;
 802         }
 803 }
 804
 805 /* MTU of the dp pseudo-device: ETH_DATA_LEN or the minimum of the ports.
 806  * Called with RTNL lock.
 807  */
 808 int dp_min_mtu(const struct datapath *dp)
 809 {
 810         struct vport *p;
 811         int mtu = 0;
 812
 813         ASSERT_RTNL();
 814
 815         list_for_each_entry (p, &dp->port_list, node) {
 816                 int dev_mtu;
 817
 818                 /* Skip any internal ports, since that's what we're trying to
 819                  * set. */
 820                 if (is_internal_vport(p))
 821                         continue;
 822
 823                 dev_mtu = vport_get_mtu(p);
 824                 if (!dev_mtu)
 825                         continue;
 826                 if (!mtu || dev_mtu < mtu)
 827                         mtu = dev_mtu;
 828         }
 829
 830         return mtu ? mtu : ETH_DATA_LEN;
 831 }
 832
 833 /* Sets the MTU of all datapath devices to the minimum of the ports
 834  * Called with RTNL lock.
 835  */
 836 void set_internal_devs_mtu(const struct datapath *dp)
 837 {
 838         struct vport *p;
 839         int mtu;
 840
 841         ASSERT_RTNL();
 842
 843         mtu = dp_min_mtu(dp);
 844
 845         list_for_each_entry (p, &dp->port_list, node) {
 846                 if (is_internal_vport(p))
 847                         vport_set_mtu(p, mtu);
 848         }
 849 }
 850
 851 static const struct nla_policy flow_policy[ODP_FLOW_ATTR_MAX + 1] = {
 852         [ODP_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
 853         [ODP_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
 854         [ODP_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
 855 };
 856
 857 static struct genl_family dp_flow_genl_family = {
 858         .id = GENL_ID_GENERATE,
 859         .hdrsize = sizeof(struct odp_header),
 860         .name = ODP_FLOW_FAMILY,
 861         .version = 1,
 862         .maxattr = ODP_FLOW_ATTR_MAX
 863 };
 864
 865 static struct genl_multicast_group dp_flow_multicast_group = {
 866         .name = ODP_FLOW_MCGROUP
 867 };
 868
 869 /* Called with genl_lock. */
 870 static int odp_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
 871                                   struct sk_buff *skb, u32 pid, u32 seq, u32 flags, u8 cmd)
 872 {
 873         const int skb_orig_len = skb->len;
 874         const struct sw_flow_actions *sf_acts;
 875         struct odp_flow_stats stats;
 876         struct odp_header *odp_header;
 877         struct nlattr *nla;
 878         unsigned long used;
 879         u8 tcp_flags;
 880         int err;
 881
 882         sf_acts = rcu_dereference_protected(flow->sf_acts,
 883                                             lockdep_genl_is_held());
 884
 885         odp_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags, cmd);
 886         if (!odp_header)
 887                 return -EMSGSIZE;
 888
 889         odp_header->dp_ifindex = get_dpifindex(dp);
 890
 891         nla = nla_nest_start(skb, ODP_FLOW_ATTR_KEY);
 892         if (!nla)
 893                 goto nla_put_failure;
 894         err = flow_to_nlattrs(&flow->key, skb);
 895         if (err)
 896                 goto error;
 897         nla_nest_end(skb, nla);
 898
 899         spin_lock_bh(&flow->lock);
 900         used = flow->used;
 901         stats.n_packets = flow->packet_count;
 902         stats.n_bytes = flow->byte_count;
 903         tcp_flags = flow->tcp_flags;
 904         spin_unlock_bh(&flow->lock);
 905
 906         if (used)
 907                 NLA_PUT_U64(skb, ODP_FLOW_ATTR_USED, flow_used_time(used));
 908
 909         if (stats.n_packets)
 910                 NLA_PUT(skb, ODP_FLOW_ATTR_STATS, sizeof(struct odp_flow_stats), &stats);
 911
 912         if (tcp_flags)
 913                 NLA_PUT_U8(skb, ODP_FLOW_ATTR_TCP_FLAGS, tcp_flags);
 914
 915         /* If ODP_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 916          * this is the first flow to be dumped into 'skb'.  This is unusual for
 917          * Netlink but individual action lists can be longer than
 918          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 919          * The userspace caller can always fetch the actions separately if it
 920          * really wants them.  (Most userspace callers in fact don't care.)
 921          *
 922          * This can only fail for dump operations because the skb is always
 923          * properly sized for single flows.
 924          */
 925         err = nla_put(skb, ODP_FLOW_ATTR_ACTIONS, sf_acts->actions_len,
 926                       sf_acts->actions);
 927         if (err < 0 && skb_orig_len)
 928                 goto error;
 929
 930         return genlmsg_end(skb, odp_header);
 931
 932 nla_put_failure:
 933         err = -EMSGSIZE;
 934 error:
 935         genlmsg_cancel(skb, odp_header);
 936         return err;
 937 }
 938
 939 static struct sk_buff *odp_flow_cmd_alloc_info(struct sw_flow *flow)
 940 {
 941         const struct sw_flow_actions *sf_acts;
 942         int len;
 943
 944         sf_acts = rcu_dereference_protected(flow->sf_acts,
 945                                             lockdep_genl_is_held());
 946
 947         len = nla_total_size(FLOW_BUFSIZE); /* ODP_FLOW_ATTR_KEY */
 948         len += nla_total_size(sf_acts->actions_len); /* ODP_FLOW_ATTR_ACTIONS */
 949         len += nla_total_size(sizeof(struct odp_flow_stats)); /* ODP_FLOW_ATTR_STATS */
 950         len += nla_total_size(1); /* ODP_FLOW_ATTR_TCP_FLAGS */
 951         len += nla_total_size(8); /* ODP_FLOW_ATTR_USED */
 952         return genlmsg_new(NLMSG_ALIGN(sizeof(struct odp_header)) + len, GFP_KERNEL);
 953 }
 954
 955 static struct sk_buff *odp_flow_cmd_build_info(struct sw_flow *flow, struct datapath *dp,
 956                                                u32 pid, u32 seq, u8 cmd)
 957 {
 958         struct sk_buff *skb;
 959         int retval;
 960
 961         skb = odp_flow_cmd_alloc_info(flow);
 962         if (!skb)
 963                 return ERR_PTR(-ENOMEM);
 964
 965         retval = odp_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd);
 966         BUG_ON(retval < 0);
 967         return skb;
 968 }
 969
 970 static int odp_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
 971 {
 972         struct nlattr **a = info->attrs;
 973         struct odp_header *odp_header = info->userhdr;
 974         struct tbl_node *flow_node;
 975         struct sw_flow_key key;
 976         struct sw_flow *flow;
 977         struct sk_buff *reply;
 978         struct datapath *dp;
 979         struct tbl *table;
 980         u32 hash;
 981         int error;
 982         int key_len;
 983
 984         /* Extract key. */
 985         error = -EINVAL;
 986         if (!a[ODP_FLOW_ATTR_KEY])
 987                 goto error;
 988         error = flow_from_nlattrs(&key, &key_len, a[ODP_FLOW_ATTR_KEY]);
 989         if (error)
 990                 goto error;
 991
 992         /* Validate actions. */
 993         if (a[ODP_FLOW_ATTR_ACTIONS]) {
 994                 error = validate_actions(a[ODP_FLOW_ATTR_ACTIONS]);
 995                 if (error)
 996                         goto error;
 997         } else if (info->genlhdr->cmd == ODP_FLOW_CMD_NEW) {
 998                 error = -EINVAL;
 999                 goto error;
1000         }
1001
1002         dp = get_dp(odp_header->dp_ifindex);
1003         error = -ENODEV;
1004         if (!dp)
1005                 goto error;
1006
1007         hash = flow_hash(&key, key_len);
1008         table = get_table_protected(dp);
1009         flow_node = tbl_lookup(table, &key, key_len, hash, flow_cmp);
1010         if (!flow_node) {
1011                 struct sw_flow_actions *acts;
1012
1013                 /* Bail out if we're not allowed to create a new flow. */
1014                 error = -ENOENT;
1015                 if (info->genlhdr->cmd == ODP_FLOW_CMD_SET)
1016                         goto error;
1017
1018                 /* Expand table, if necessary, to make room. */
1019                 if (tbl_count(table) >= tbl_n_buckets(table)) {
1020                         error = expand_table(dp);
1021                         if (error)
1022                                 goto error;
1023                         table = get_table_protected(dp);
1024                 }
1025
1026                 /* Allocate flow. */
1027                 flow = flow_alloc();
1028                 if (IS_ERR(flow)) {
1029                         error = PTR_ERR(flow);
1030                         goto error;
1031                 }
1032                 flow->key = key;
1033                 clear_stats(flow);
1034
1035                 /* Obtain actions. */
1036                 acts = flow_actions_alloc(a[ODP_FLOW_ATTR_ACTIONS]);
1037                 error = PTR_ERR(acts);
1038                 if (IS_ERR(acts))
1039                         goto error_free_flow;
1040                 rcu_assign_pointer(flow->sf_acts, acts);
1041
1042                 /* Put flow in bucket. */
1043                 error = tbl_insert(table, &flow->tbl_node, hash);
1044                 if (error)
1045                         goto error_free_flow;
1046
1047                 reply = odp_flow_cmd_build_info(flow, dp, info->snd_pid,
1048                                                 info->snd_seq, ODP_FLOW_CMD_NEW);
1049         } else {
1050                 /* We found a matching flow. */
1051                 struct sw_flow_actions *old_acts;
1052
1053                 /* Bail out if we're not allowed to modify an existing flow.
1054                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1055                  * because Generic Netlink treats the latter as a dump
1056                  * request.  We also accept NLM_F_EXCL in case that bug ever
1057                  * gets fixed.
1058                  */
1059                 error = -EEXIST;
1060                 if (info->genlhdr->cmd == ODP_FLOW_CMD_NEW &&
1061                     info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1062                         goto error;
1063
1064                 /* Update actions. */
1065                 flow = flow_cast(flow_node);
1066                 old_acts = rcu_dereference_protected(flow->sf_acts,
1067                                                      lockdep_genl_is_held());
1068                 if (a[ODP_FLOW_ATTR_ACTIONS] &&
1069                     (old_acts->actions_len != nla_len(a[ODP_FLOW_ATTR_ACTIONS]) ||
1070                      memcmp(old_acts->actions, nla_data(a[ODP_FLOW_ATTR_ACTIONS]),
1071                             old_acts->actions_len))) {
1072                         struct sw_flow_actions *new_acts;
1073
1074                         new_acts = flow_actions_alloc(a[ODP_FLOW_ATTR_ACTIONS]);
1075                         error = PTR_ERR(new_acts);
1076                         if (IS_ERR(new_acts))
1077                                 goto error;
1078
1079                         rcu_assign_pointer(flow->sf_acts, new_acts);
1080                         flow_deferred_free_acts(old_acts);
1081                 }
1082
1083                 reply = odp_flow_cmd_build_info(flow, dp, info->snd_pid,
1084                                                 info->snd_seq, ODP_FLOW_CMD_NEW);
1085
1086                 /* Clear stats. */
1087                 if (a[ODP_FLOW_ATTR_CLEAR]) {
1088                         spin_lock_bh(&flow->lock);
1089                         clear_stats(flow);
1090                         spin_unlock_bh(&flow->lock);
1091                 }
1092         }
1093
1094         if (!IS_ERR(reply))
1095                 genl_notify(reply, genl_info_net(info), info->snd_pid,
1096                             dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1097         else
1098                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1099                                 dp_flow_multicast_group.id, PTR_ERR(reply));
1100         return 0;
1101
1102 error_free_flow:
1103         flow_put(flow);
1104 error:
1105         return error;
1106 }
1107
1108 static int odp_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1109 {
1110         struct nlattr **a = info->attrs;
1111         struct odp_header *odp_header = info->userhdr;
1112         struct sw_flow_key key;
1113         struct tbl_node *flow_node;
1114         struct sk_buff *reply;
1115         struct sw_flow *flow;
1116         struct datapath *dp;
1117         struct tbl *table;
1118         int err;
1119         int key_len;
1120
1121         if (!a[ODP_FLOW_ATTR_KEY])
1122                 return -EINVAL;
1123         err = flow_from_nlattrs(&key, &key_len, a[ODP_FLOW_ATTR_KEY]);
1124         if (err)
1125                 return err;
1126
1127         dp = get_dp(odp_header->dp_ifindex);
1128         if (!dp)
1129                 return -ENODEV;
1130
1131         table = get_table_protected(dp);
1132         flow_node = tbl_lookup(table, &key, key_len, flow_hash(&key, key_len),
1133                                flow_cmp);
1134         if (!flow_node)
1135                 return -ENOENT;
1136
1137         flow = flow_cast(flow_node);
1138         reply = odp_flow_cmd_build_info(flow, dp, info->snd_pid, info->snd_seq, ODP_FLOW_CMD_NEW);
1139         if (IS_ERR(reply))
1140                 return PTR_ERR(reply);
1141
1142         return genlmsg_reply(reply, info);
1143 }
1144
1145 static int odp_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1146 {
1147         struct nlattr **a = info->attrs;
1148         struct odp_header *odp_header = info->userhdr;
1149         struct sw_flow_key key;
1150         struct tbl_node *flow_node;
1151         struct sk_buff *reply;
1152         struct sw_flow *flow;
1153         struct datapath *dp;
1154         struct tbl *table;
1155         int err;
1156         int key_len;
1157
1158         if (!a[ODP_FLOW_ATTR_KEY])
1159                 return flush_flows(odp_header->dp_ifindex);
1160         err = flow_from_nlattrs(&key, &key_len, a[ODP_FLOW_ATTR_KEY]);
1161         if (err)
1162                 return err;
1163
1164         dp = get_dp(odp_header->dp_ifindex);
1165         if (!dp)
1166                 return -ENODEV;
1167
1168         table = get_table_protected(dp);
1169         flow_node = tbl_lookup(table, &key, key_len, flow_hash(&key, key_len),
1170                                flow_cmp);
1171         if (!flow_node)
1172                 return -ENOENT;
1173         flow = flow_cast(flow_node);
1174
1175         reply = odp_flow_cmd_alloc_info(flow);
1176         if (!reply)
1177                 return -ENOMEM;
1178
1179         err = tbl_remove(table, flow_node);
1180         if (err) {
1181                 kfree_skb(reply);
1182                 return err;
1183         }
1184
1185         err = odp_flow_cmd_fill_info(flow, dp, reply, info->snd_pid,
1186                                      info->snd_seq, 0, ODP_FLOW_CMD_DEL);
1187         BUG_ON(err < 0);
1188
1189         flow_deferred_free(flow);
1190
1191         genl_notify(reply, genl_info_net(info), info->snd_pid,
1192                     dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1193         return 0;
1194 }
1195
1196 static int odp_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1197 {
1198         struct odp_header *odp_header = genlmsg_data(nlmsg_data(cb->nlh));
1199         struct datapath *dp;
1200
1201         dp = get_dp(odp_header->dp_ifindex);
1202         if (!dp)
1203                 return -ENODEV;
1204
1205         for (;;) {
1206                 struct tbl_node *flow_node;
1207                 struct sw_flow *flow;
1208                 u32 bucket, obj;
1209
1210                 bucket = cb->args[0];
1211                 obj = cb->args[1];
1212                 flow_node = tbl_next(get_table_protected(dp), &bucket, &obj);
1213                 if (!flow_node)
1214                         break;
1215
1216                 flow = flow_cast(flow_node);
1217                 if (odp_flow_cmd_fill_info(flow, dp, skb, NETLINK_CB(cb->skb).pid,
1218                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1219                                            ODP_FLOW_CMD_NEW) < 0)
1220                         break;
1221
1222                 cb->args[0] = bucket;
1223                 cb->args[1] = obj;
1224         }
1225         return skb->len;
1226 }
1227
1228 static struct genl_ops dp_flow_genl_ops[] = {
1229         { .cmd = ODP_FLOW_CMD_NEW,
1230           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1231           .policy = flow_policy,
1232           .doit = odp_flow_cmd_new_or_set
1233         },
1234         { .cmd = ODP_FLOW_CMD_DEL,
1235           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1236           .policy = flow_policy,
1237           .doit = odp_flow_cmd_del
1238         },
1239         { .cmd = ODP_FLOW_CMD_GET,
1240           .flags = 0,               /* OK for unprivileged users. */
1241           .policy = flow_policy,
1242           .doit = odp_flow_cmd_get,
1243           .dumpit = odp_flow_cmd_dump
1244         },
1245         { .cmd = ODP_FLOW_CMD_SET,
1246           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1247           .policy = flow_policy,
1248           .doit = odp_flow_cmd_new_or_set,
1249         },
1250 };
1251
1252 static const struct nla_policy datapath_policy[ODP_DP_ATTR_MAX + 1] = {
1253 #ifdef HAVE_NLA_NUL_STRING
1254         [ODP_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1255 #endif
1256         [ODP_DP_ATTR_IPV4_FRAGS] = { .type = NLA_U32 },
1257         [ODP_DP_ATTR_SAMPLING] = { .type = NLA_U32 },
1258 };
1259
1260 static struct genl_family dp_datapath_genl_family = {
1261         .id = GENL_ID_GENERATE,
1262         .hdrsize = sizeof(struct odp_header),
1263         .name = ODP_DATAPATH_FAMILY,
1264         .version = 1,
1265         .maxattr = ODP_DP_ATTR_MAX
1266 };
1267
1268 static struct genl_multicast_group dp_datapath_multicast_group = {
1269         .name = ODP_DATAPATH_MCGROUP
1270 };
1271
1272 static int odp_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1273                                 u32 pid, u32 seq, u32 flags, u8 cmd)
1274 {
1275         struct odp_header *odp_header;
1276         struct nlattr *nla;
1277         int err;
1278         int dp_ifindex = get_dpifindex(dp);
1279
1280         odp_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family,
1281                                    flags, cmd);
1282         if (!odp_header)
1283                 goto error;
1284
1285         odp_header->dp_ifindex = dp_ifindex;
1286
1287         rcu_read_lock();
1288         err = nla_put_string(skb, ODP_DP_ATTR_NAME, dp_name(dp));
1289         rcu_read_unlock();
1290         if (err)
1291                 goto nla_put_failure;
1292
1293         nla = nla_reserve(skb, ODP_DP_ATTR_STATS, sizeof(struct odp_stats));
1294         if (!nla)
1295                 goto nla_put_failure;
1296         get_dp_stats(dp, nla_data(nla));
1297
1298         NLA_PUT_U32(skb, ODP_DP_ATTR_IPV4_FRAGS,
1299                     dp->drop_frags ? ODP_DP_FRAG_DROP : ODP_DP_FRAG_ZERO);
1300
1301         if (dp->sflow_probability)
1302                 NLA_PUT_U32(skb, ODP_DP_ATTR_SAMPLING, dp->sflow_probability);
1303
1304         nla = nla_nest_start(skb, ODP_DP_ATTR_MCGROUPS);
1305         if (!nla)
1306                 goto nla_put_failure;
1307         NLA_PUT_U32(skb, ODP_PACKET_CMD_MISS,
1308                         packet_mc_group(dp_ifindex, ODP_PACKET_CMD_MISS));
1309         NLA_PUT_U32(skb, ODP_PACKET_CMD_ACTION,
1310                         packet_mc_group(dp_ifindex, ODP_PACKET_CMD_ACTION));
1311         NLA_PUT_U32(skb, ODP_PACKET_CMD_SAMPLE,
1312                         packet_mc_group(dp_ifindex, ODP_PACKET_CMD_SAMPLE));
1313         nla_nest_end(skb, nla);
1314
1315         return genlmsg_end(skb, odp_header);
1316
1317 nla_put_failure:
1318         genlmsg_cancel(skb, odp_header);
1319 error:
1320         return -EMSGSIZE;
1321 }
1322
1323 static struct sk_buff *odp_dp_cmd_build_info(struct datapath *dp, u32 pid,
1324                                              u32 seq, u8 cmd)
1325 {
1326         struct sk_buff *skb;
1327         int retval;
1328
1329         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1330         if (!skb)
1331                 return ERR_PTR(-ENOMEM);
1332
1333         retval = odp_dp_cmd_fill_info(dp, skb, pid, seq, 0, cmd);
1334         if (retval < 0) {
1335                 kfree_skb(skb);
1336                 return ERR_PTR(retval);
1337         }
1338         return skb;
1339 }
1340
1341 static int odp_dp_cmd_validate(struct nlattr *a[ODP_DP_ATTR_MAX + 1])
1342 {
1343         if (a[ODP_DP_ATTR_IPV4_FRAGS]) {
1344                 u32 frags = nla_get_u32(a[ODP_DP_ATTR_IPV4_FRAGS]);
1345
1346                 if (frags != ODP_DP_FRAG_ZERO && frags != ODP_DP_FRAG_DROP)
1347                         return -EINVAL;
1348         }
1349
1350         return CHECK_NUL_STRING(a[ODP_DP_ATTR_NAME], IFNAMSIZ - 1);
1351 }
1352
1353 /* Called with genl_mutex and optionally with RTNL lock also. */
1354 static struct datapath *lookup_datapath(struct odp_header *odp_header, struct nlattr *a[ODP_DP_ATTR_MAX + 1])
1355 {
1356         struct datapath *dp;
1357
1358         if (!a[ODP_DP_ATTR_NAME])
1359                 dp = get_dp(odp_header->dp_ifindex);
1360         else {
1361                 struct vport *vport;
1362
1363                 rcu_read_lock();
1364                 vport = vport_locate(nla_data(a[ODP_DP_ATTR_NAME]));
1365                 dp = vport && vport->port_no == ODPP_LOCAL ? vport->dp : NULL;
1366                 rcu_read_unlock();
1367         }
1368         return dp ? dp : ERR_PTR(-ENODEV);
1369 }
1370
1371 /* Called with genl_mutex. */
1372 static void change_datapath(struct datapath *dp, struct nlattr *a[ODP_DP_ATTR_MAX + 1])
1373 {
1374         if (a[ODP_DP_ATTR_IPV4_FRAGS])
1375                 dp->drop_frags = nla_get_u32(a[ODP_DP_ATTR_IPV4_FRAGS]) == ODP_DP_FRAG_DROP;
1376         if (a[ODP_DP_ATTR_SAMPLING])
1377                 dp->sflow_probability = nla_get_u32(a[ODP_DP_ATTR_SAMPLING]);
1378 }
1379
1380 static int odp_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1381 {
1382         struct nlattr **a = info->attrs;
1383         struct vport_parms parms;
1384         struct sk_buff *reply;
1385         struct datapath *dp;
1386         struct vport *vport;
1387         int err;
1388
1389         err = -EINVAL;
1390         if (!a[ODP_DP_ATTR_NAME])
1391                 goto err;
1392
1393         err = odp_dp_cmd_validate(a);
1394         if (err)
1395                 goto err;
1396
1397         rtnl_lock();
1398         err = -ENODEV;
1399         if (!try_module_get(THIS_MODULE))
1400                 goto err_unlock_rtnl;
1401
1402         err = -ENOMEM;
1403         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1404         if (dp == NULL)
1405                 goto err_put_module;
1406         INIT_LIST_HEAD(&dp->port_list);
1407
1408         /* Initialize kobject for bridge.  This will be added as
1409          * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
1410         dp->ifobj.kset = NULL;
1411         kobject_init(&dp->ifobj, &dp_ktype);
1412
1413         /* Allocate table. */
1414         err = -ENOMEM;
1415         rcu_assign_pointer(dp->table, tbl_create(TBL_MIN_BUCKETS));
1416         if (!dp->table)
1417                 goto err_free_dp;
1418
1419         dp->drop_frags = 0;
1420         dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1421         if (!dp->stats_percpu) {
1422                 err = -ENOMEM;
1423                 goto err_destroy_table;
1424         }
1425
1426         change_datapath(dp, a);
1427
1428         /* Set up our datapath device. */
1429         parms.name = nla_data(a[ODP_DP_ATTR_NAME]);
1430         parms.type = ODP_VPORT_TYPE_INTERNAL;
1431         parms.options = NULL;
1432         parms.dp = dp;
1433         parms.port_no = ODPP_LOCAL;
1434         vport = new_vport(&parms);
1435         if (IS_ERR(vport)) {
1436                 err = PTR_ERR(vport);
1437                 if (err == -EBUSY)
1438                         err = -EEXIST;
1439
1440                 goto err_destroy_percpu;
1441         }
1442
1443         reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_NEW);
1444         err = PTR_ERR(reply);
1445         if (IS_ERR(reply))
1446                 goto err_destroy_local_port;
1447
1448         list_add_tail(&dp->list_node, &dps);
1449         dp_sysfs_add_dp(dp);
1450
1451         rtnl_unlock();
1452
1453         genl_notify(reply, genl_info_net(info), info->snd_pid,
1454                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1455         return 0;
1456
1457 err_destroy_local_port:
1458         dp_detach_port(get_vport_protected(dp, ODPP_LOCAL));
1459 err_destroy_percpu:
1460         free_percpu(dp->stats_percpu);
1461 err_destroy_table:
1462         tbl_destroy(get_table_protected(dp), NULL);
1463 err_free_dp:
1464         kfree(dp);
1465 err_put_module:
1466         module_put(THIS_MODULE);
1467 err_unlock_rtnl:
1468         rtnl_unlock();
1469 err:
1470         return err;
1471 }
1472
1473 static int odp_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1474 {
1475         struct vport *vport, *next_vport;
1476         struct sk_buff *reply;
1477         struct datapath *dp;
1478         int err;
1479
1480         err = odp_dp_cmd_validate(info->attrs);
1481         if (err)
1482                 goto exit;
1483
1484         rtnl_lock();
1485         dp = lookup_datapath(info->userhdr, info->attrs);
1486         err = PTR_ERR(dp);
1487         if (IS_ERR(dp))
1488                 goto exit_unlock;
1489
1490         reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_DEL);
1491         err = PTR_ERR(reply);
1492         if (IS_ERR(reply))
1493                 goto exit_unlock;
1494
1495         list_for_each_entry_safe (vport, next_vport, &dp->port_list, node)
1496                 if (vport->port_no != ODPP_LOCAL)
1497                         dp_detach_port(vport);
1498
1499         dp_sysfs_del_dp(dp);
1500         list_del(&dp->list_node);
1501         dp_detach_port(get_vport_protected(dp, ODPP_LOCAL));
1502
1503         /* rtnl_unlock() will wait until all the references to devices that
1504          * are pending unregistration have been dropped.  We do it here to
1505          * ensure that any internal devices (which contain DP pointers) are
1506          * fully destroyed before freeing the datapath.
1507          */
1508         rtnl_unlock();
1509
1510         call_rcu(&dp->rcu, destroy_dp_rcu);
1511         module_put(THIS_MODULE);
1512
1513         genl_notify(reply, genl_info_net(info), info->snd_pid,
1514                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1515
1516         return 0;
1517
1518 exit_unlock:
1519         rtnl_unlock();
1520 exit:
1521         return err;
1522 }
1523
1524 static int odp_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1525 {
1526         struct sk_buff *reply;
1527         struct datapath *dp;
1528         int err;
1529
1530         err = odp_dp_cmd_validate(info->attrs);
1531         if (err)
1532                 return err;
1533
1534         dp = lookup_datapath(info->userhdr, info->attrs);
1535         if (IS_ERR(dp))
1536                 return PTR_ERR(dp);
1537
1538         change_datapath(dp, info->attrs);
1539
1540         reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_NEW);
1541         if (IS_ERR(reply)) {
1542                 err = PTR_ERR(reply);
1543                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1544                                 dp_datapath_multicast_group.id, err);
1545                 return 0;
1546         }
1547
1548         genl_notify(reply, genl_info_net(info), info->snd_pid,
1549                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1550         return 0;
1551 }
1552
1553 static int odp_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1554 {
1555         struct sk_buff *reply;
1556         struct datapath *dp;
1557         int err;
1558
1559         err = odp_dp_cmd_validate(info->attrs);
1560         if (err)
1561                 return err;
1562
1563         dp = lookup_datapath(info->userhdr, info->attrs);
1564         if (IS_ERR(dp))
1565                 return PTR_ERR(dp);
1566
1567         reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_NEW);
1568         if (IS_ERR(reply))
1569                 return PTR_ERR(reply);
1570
1571         return genlmsg_reply(reply, info);
1572 }
1573
1574 static int odp_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1575 {
1576         struct datapath *dp;
1577         int skip = cb->args[0];
1578         int i = 0;
1579
1580         list_for_each_entry (dp, &dps, list_node) {
1581                 if (i < skip)
1582                         continue;
1583                 if (odp_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).pid,
1584                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1585                                          ODP_DP_CMD_NEW) < 0)
1586                         break;
1587                 i++;
1588         }
1589
1590         cb->args[0] = i;
1591
1592         return skb->len;
1593 }
1594
1595 static struct genl_ops dp_datapath_genl_ops[] = {
1596         { .cmd = ODP_DP_CMD_NEW,
1597           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1598           .policy = datapath_policy,
1599           .doit = odp_dp_cmd_new
1600         },
1601         { .cmd = ODP_DP_CMD_DEL,
1602           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1603           .policy = datapath_policy,
1604           .doit = odp_dp_cmd_del
1605         },
1606         { .cmd = ODP_DP_CMD_GET,
1607           .flags = 0,               /* OK for unprivileged users. */
1608           .policy = datapath_policy,
1609           .doit = odp_dp_cmd_get,
1610           .dumpit = odp_dp_cmd_dump
1611         },
1612         { .cmd = ODP_DP_CMD_SET,
1613           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1614           .policy = datapath_policy,
1615           .doit = odp_dp_cmd_set,
1616         },
1617 };
1618
1619 static const struct nla_policy vport_policy[ODP_VPORT_ATTR_MAX + 1] = {
1620 #ifdef HAVE_NLA_NUL_STRING
1621         [ODP_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1622         [ODP_VPORT_ATTR_STATS] = { .len = sizeof(struct rtnl_link_stats64) },
1623         [ODP_VPORT_ATTR_ADDRESS] = { .len = ETH_ALEN },
1624 #else
1625         [ODP_VPORT_ATTR_STATS] = { .minlen = sizeof(struct rtnl_link_stats64) },
1626         [ODP_VPORT_ATTR_ADDRESS] = { .minlen = ETH_ALEN },
1627 #endif
1628         [ODP_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1629         [ODP_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1630         [ODP_VPORT_ATTR_MTU] = { .type = NLA_U32 },
1631         [ODP_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1632 };
1633
1634 static struct genl_family dp_vport_genl_family = {
1635         .id = GENL_ID_GENERATE,
1636         .hdrsize = sizeof(struct odp_header),
1637         .name = ODP_VPORT_FAMILY,
1638         .version = 1,
1639         .maxattr = ODP_VPORT_ATTR_MAX
1640 };
1641
1642 static struct genl_multicast_group dp_vport_multicast_group = {
1643         .name = ODP_VPORT_MCGROUP
1644 };
1645
1646 /* Called with RTNL lock or RCU read lock. */
1647 static int odp_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1648                                    u32 pid, u32 seq, u32 flags, u8 cmd)
1649 {
1650         struct odp_header *odp_header;
1651         struct nlattr *nla;
1652         int ifindex, iflink;
1653         int mtu;
1654         int err;
1655
1656         odp_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family,
1657                                  flags, cmd);
1658         if (!odp_header)
1659                 return -EMSGSIZE;
1660
1661         odp_header->dp_ifindex = get_dpifindex(vport->dp);
1662
1663         NLA_PUT_U32(skb, ODP_VPORT_ATTR_PORT_NO, vport->port_no);
1664         NLA_PUT_U32(skb, ODP_VPORT_ATTR_TYPE, vport_get_type(vport));
1665         NLA_PUT_STRING(skb, ODP_VPORT_ATTR_NAME, vport_get_name(vport));
1666
1667         nla = nla_reserve(skb, ODP_VPORT_ATTR_STATS, sizeof(struct rtnl_link_stats64));
1668         if (!nla)
1669                 goto nla_put_failure;
1670         if (vport_get_stats(vport, nla_data(nla)))
1671                 __skb_trim(skb, skb->len - nla->nla_len);
1672
1673         NLA_PUT(skb, ODP_VPORT_ATTR_ADDRESS, ETH_ALEN, vport_get_addr(vport));
1674
1675         mtu = vport_get_mtu(vport);
1676         if (mtu)
1677                 NLA_PUT_U32(skb, ODP_VPORT_ATTR_MTU, mtu);
1678
1679         err = vport_get_options(vport, skb);
1680         if (err == -EMSGSIZE)
1681                 goto error;
1682
1683         ifindex = vport_get_ifindex(vport);
1684         if (ifindex > 0)
1685                 NLA_PUT_U32(skb, ODP_VPORT_ATTR_IFINDEX, ifindex);
1686
1687         iflink = vport_get_iflink(vport);
1688         if (iflink > 0)
1689                 NLA_PUT_U32(skb, ODP_VPORT_ATTR_IFLINK, iflink);
1690
1691         return genlmsg_end(skb, odp_header);
1692
1693 nla_put_failure:
1694         err = -EMSGSIZE;
1695 error:
1696         genlmsg_cancel(skb, odp_header);
1697         return err;
1698 }
1699
1700 /* Called with RTNL lock or RCU read lock. */
1701 static struct sk_buff *odp_vport_cmd_build_info(struct vport *vport, u32 pid,
1702                                                 u32 seq, u8 cmd)
1703 {
1704         struct sk_buff *skb;
1705         int retval;
1706
1707         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1708         if (!skb)
1709                 return ERR_PTR(-ENOMEM);
1710
1711         retval = odp_vport_cmd_fill_info(vport, skb, pid, seq, 0, cmd);
1712         if (retval < 0) {
1713                 kfree_skb(skb);
1714                 return ERR_PTR(retval);
1715         }
1716         return skb;
1717 }
1718
1719 static int odp_vport_cmd_validate(struct nlattr *a[ODP_VPORT_ATTR_MAX + 1])
1720 {
1721         return CHECK_NUL_STRING(a[ODP_VPORT_ATTR_NAME], IFNAMSIZ - 1);
1722 }
1723
1724 /* Called with RTNL lock or RCU read lock. */
1725 static struct vport *lookup_vport(struct odp_header *odp_header,
1726                                   struct nlattr *a[ODP_VPORT_ATTR_MAX + 1])
1727 {
1728         struct datapath *dp;
1729         struct vport *vport;
1730
1731         if (a[ODP_VPORT_ATTR_NAME]) {
1732                 vport = vport_locate(nla_data(a[ODP_VPORT_ATTR_NAME]));
1733                 if (!vport)
1734                         return ERR_PTR(-ENODEV);
1735                 return vport;
1736         } else if (a[ODP_VPORT_ATTR_PORT_NO]) {
1737                 u32 port_no = nla_get_u32(a[ODP_VPORT_ATTR_PORT_NO]);
1738
1739                 if (port_no >= DP_MAX_PORTS)
1740                         return ERR_PTR(-EFBIG);
1741
1742                 dp = get_dp(odp_header->dp_ifindex);
1743                 if (!dp)
1744                         return ERR_PTR(-ENODEV);
1745
1746                 vport = get_vport_protected(dp, port_no);
1747                 if (!vport)
1748                         return ERR_PTR(-ENOENT);
1749                 return vport;
1750         } else
1751                 return ERR_PTR(-EINVAL);
1752 }
1753
1754 /* Called with RTNL lock. */
1755 static int change_vport(struct vport *vport, struct nlattr *a[ODP_VPORT_ATTR_MAX + 1])
1756 {
1757         int err = 0;
1758         if (a[ODP_VPORT_ATTR_STATS])
1759                 err = vport_set_stats(vport, nla_data(a[ODP_VPORT_ATTR_STATS]));
1760         if (!err && a[ODP_VPORT_ATTR_ADDRESS])
1761                 err = vport_set_addr(vport, nla_data(a[ODP_VPORT_ATTR_ADDRESS]));
1762         if (!err && a[ODP_VPORT_ATTR_MTU])
1763                 err = vport_set_mtu(vport, nla_get_u32(a[ODP_VPORT_ATTR_MTU]));
1764         return err;
1765 }
1766
1767 static int odp_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1768 {
1769         struct nlattr **a = info->attrs;
1770         struct odp_header *odp_header = info->userhdr;
1771         struct vport_parms parms;
1772         struct sk_buff *reply;
1773         struct vport *vport;
1774         struct datapath *dp;
1775         u32 port_no;
1776         int err;
1777
1778         err = -EINVAL;
1779         if (!a[ODP_VPORT_ATTR_NAME] || !a[ODP_VPORT_ATTR_TYPE])
1780                 goto exit;
1781
1782         err = odp_vport_cmd_validate(a);
1783         if (err)
1784                 goto exit;
1785
1786         rtnl_lock();
1787         dp = get_dp(odp_header->dp_ifindex);
1788         err = -ENODEV;
1789         if (!dp)
1790                 goto exit_unlock;
1791
1792         if (a[ODP_VPORT_ATTR_PORT_NO]) {
1793                 port_no = nla_get_u32(a[ODP_VPORT_ATTR_PORT_NO]);
1794
1795                 err = -EFBIG;
1796                 if (port_no >= DP_MAX_PORTS)
1797                         goto exit_unlock;
1798
1799                 vport = get_vport_protected(dp, port_no);
1800                 err = -EBUSY;
1801                 if (vport)
1802                         goto exit_unlock;
1803         } else {
1804                 for (port_no = 1; ; port_no++) {
1805                         if (port_no >= DP_MAX_PORTS) {
1806                                 err = -EFBIG;
1807                                 goto exit_unlock;
1808                         }
1809                         vport = get_vport_protected(dp, port_no);
1810                         if (!vport)
1811                                 break;
1812                 }
1813         }
1814
1815         parms.name = nla_data(a[ODP_VPORT_ATTR_NAME]);
1816         parms.type = nla_get_u32(a[ODP_VPORT_ATTR_TYPE]);
1817         parms.options = a[ODP_VPORT_ATTR_OPTIONS];
1818         parms.dp = dp;
1819         parms.port_no = port_no;
1820
1821         vport = new_vport(&parms);
1822         err = PTR_ERR(vport);
1823         if (IS_ERR(vport))
1824                 goto exit_unlock;
1825
1826         set_internal_devs_mtu(dp);
1827         dp_sysfs_add_if(vport);
1828
1829         err = change_vport(vport, a);
1830         if (!err) {
1831                 reply = odp_vport_cmd_build_info(vport, info->snd_pid,
1832                                                  info->snd_seq, ODP_VPORT_CMD_NEW);
1833                 if (IS_ERR(reply))
1834                         err = PTR_ERR(reply);
1835         }
1836         if (err) {
1837                 dp_detach_port(vport);
1838                 goto exit_unlock;
1839         }
1840         genl_notify(reply, genl_info_net(info), info->snd_pid,
1841                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1842
1843
1844 exit_unlock:
1845         rtnl_unlock();
1846 exit:
1847         return err;
1848 }
1849
1850 static int odp_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1851 {
1852         struct nlattr **a = info->attrs;
1853         struct sk_buff *reply;
1854         struct vport *vport;
1855         int err;
1856
1857         err = odp_vport_cmd_validate(a);
1858         if (err)
1859                 goto exit;
1860
1861         rtnl_lock();
1862         vport = lookup_vport(info->userhdr, a);
1863         err = PTR_ERR(vport);
1864         if (IS_ERR(vport))
1865                 goto exit_unlock;
1866
1867         err = 0;
1868         if (a[ODP_VPORT_ATTR_OPTIONS])
1869                 err = vport_set_options(vport, a[ODP_VPORT_ATTR_OPTIONS]);
1870         if (!err)
1871                 err = change_vport(vport, a);
1872
1873         reply = odp_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1874                                          ODP_VPORT_CMD_NEW);
1875         if (IS_ERR(reply)) {
1876                 err = PTR_ERR(reply);
1877                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1878                                 dp_vport_multicast_group.id, err);
1879                 return 0;
1880         }
1881
1882         genl_notify(reply, genl_info_net(info), info->snd_pid,
1883                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1884
1885 exit_unlock:
1886         rtnl_unlock();
1887 exit:
1888         return err;
1889 }
1890
1891 static int odp_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1892 {
1893         struct nlattr **a = info->attrs;
1894         struct sk_buff *reply;
1895         struct vport *vport;
1896         int err;
1897
1898         err = odp_vport_cmd_validate(a);
1899         if (err)
1900                 goto exit;
1901
1902         rtnl_lock();
1903         vport = lookup_vport(info->userhdr, a);
1904         err = PTR_ERR(vport);
1905         if (IS_ERR(vport))
1906                 goto exit_unlock;
1907
1908         if (vport->port_no == ODPP_LOCAL) {
1909                 err = -EINVAL;
1910                 goto exit_unlock;
1911         }
1912
1913         reply = odp_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1914                                          ODP_VPORT_CMD_DEL);
1915         err = PTR_ERR(reply);
1916         if (IS_ERR(reply))
1917                 goto exit_unlock;
1918
1919         err = dp_detach_port(vport);
1920
1921         genl_notify(reply, genl_info_net(info), info->snd_pid,
1922                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1923
1924 exit_unlock:
1925         rtnl_unlock();
1926 exit:
1927         return err;
1928 }
1929
1930 static int odp_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1931 {
1932         struct nlattr **a = info->attrs;
1933         struct odp_header *odp_header = info->userhdr;
1934         struct sk_buff *reply;
1935         struct vport *vport;
1936         int err;
1937
1938         err = odp_vport_cmd_validate(a);
1939         if (err)
1940                 goto exit;
1941
1942         rcu_read_lock();
1943         vport = lookup_vport(odp_header, a);
1944         err = PTR_ERR(vport);
1945         if (IS_ERR(vport))
1946                 goto exit_unlock;
1947
1948         reply = odp_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1949                                          ODP_VPORT_CMD_NEW);
1950         err = PTR_ERR(reply);
1951         if (IS_ERR(reply))
1952                 goto exit_unlock;
1953
1954         rcu_read_unlock();
1955
1956         return genlmsg_reply(reply, info);
1957
1958 exit_unlock:
1959         rcu_read_unlock();
1960 exit:
1961         return err;
1962 }
1963
1964 static int odp_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1965 {
1966         struct odp_header *odp_header = genlmsg_data(nlmsg_data(cb->nlh));
1967         struct datapath *dp;
1968         u32 port_no;
1969         int retval;
1970
1971         dp = get_dp(odp_header->dp_ifindex);
1972         if (!dp)
1973                 return -ENODEV;
1974
1975         rcu_read_lock();
1976         for (port_no = cb->args[0]; port_no < DP_MAX_PORTS; port_no++) {
1977                 struct vport *vport;
1978
1979                 vport = get_vport_protected(dp, port_no);
1980                 if (!vport)
1981                         continue;
1982
1983                 if (odp_vport_cmd_fill_info(vport, skb, NETLINK_CB(cb->skb).pid,
1984                                             cb->nlh->nlmsg_seq, NLM_F_MULTI,
1985                                             ODP_VPORT_CMD_NEW) < 0)
1986                         break;
1987         }
1988         rcu_read_unlock();
1989
1990         cb->args[0] = port_no;
1991         retval = skb->len;
1992
1993         return retval;
1994 }
1995
1996 static struct genl_ops dp_vport_genl_ops[] = {
1997         { .cmd = ODP_VPORT_CMD_NEW,
1998           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1999           .policy = vport_policy,
2000           .doit = odp_vport_cmd_new
2001         },
2002         { .cmd = ODP_VPORT_CMD_DEL,
2003           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2004           .policy = vport_policy,
2005           .doit = odp_vport_cmd_del
2006         },
2007         { .cmd = ODP_VPORT_CMD_GET,
2008           .flags = 0,               /* OK for unprivileged users. */
2009           .policy = vport_policy,
2010           .doit = odp_vport_cmd_get,
2011           .dumpit = odp_vport_cmd_dump
2012         },
2013         { .cmd = ODP_VPORT_CMD_SET,
2014           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2015           .policy = vport_policy,
2016           .doit = odp_vport_cmd_set,
2017         },
2018 };
2019
2020 struct genl_family_and_ops {
2021         struct genl_family *family;
2022         struct genl_ops *ops;
2023         int n_ops;
2024         struct genl_multicast_group *group;
2025 };
2026
2027 static const struct genl_family_and_ops dp_genl_families[] = {
2028         { &dp_datapath_genl_family,
2029           dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
2030           &dp_datapath_multicast_group },
2031         { &dp_vport_genl_family,
2032           dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
2033           &dp_vport_multicast_group },
2034         { &dp_flow_genl_family,
2035           dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
2036           &dp_flow_multicast_group },
2037         { &dp_packet_genl_family,
2038           dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
2039           NULL },
2040 };
2041
2042 static void dp_unregister_genl(int n_families)
2043 {
2044         int i;
2045
2046         for (i = 0; i < n_families; i++) {
2047                 genl_unregister_family(dp_genl_families[i].family);
2048         }
2049 }
2050
2051 static int dp_register_genl(void)
2052 {
2053         int n_registered;
2054         int err;
2055         int i;
2056
2057         n_registered = 0;
2058         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2059                 const struct genl_family_and_ops *f = &dp_genl_families[i];
2060
2061                 err = genl_register_family_with_ops(f->family, f->ops,
2062                                                     f->n_ops);
2063                 if (err)
2064                         goto error;
2065                 n_registered++;
2066
2067                 if (f->group) {
2068                         err = genl_register_mc_group(f->family, f->group);
2069                         if (err)
2070                                 goto error;
2071                 }
2072         }
2073
2074         err = packet_register_mc_groups();
2075         if (err)
2076                 goto error;
2077         return 0;
2078
2079 error:
2080         dp_unregister_genl(n_registered);
2081         return err;
2082 }
2083
2084 static int __init dp_init(void)
2085 {
2086         struct sk_buff *dummy_skb;
2087         int err;
2088
2089         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
2090
2091         printk("Open vSwitch %s, built "__DATE__" "__TIME__"\n", VERSION BUILDNR);
2092
2093         err = flow_init();
2094         if (err)
2095                 goto error;
2096
2097         err = vport_init();
2098         if (err)
2099                 goto error_flow_exit;
2100
2101         err = register_netdevice_notifier(&dp_device_notifier);
2102         if (err)
2103                 goto error_vport_exit;
2104
2105         err = dp_register_genl();
2106         if (err < 0)
2107                 goto error_unreg_notifier;
2108
2109         return 0;
2110
2111 error_unreg_notifier:
2112         unregister_netdevice_notifier(&dp_device_notifier);
2113 error_vport_exit:
2114         vport_exit();
2115 error_flow_exit:
2116         flow_exit();
2117 error:
2118         return err;
2119 }
2120
2121 static void dp_cleanup(void)
2122 {
2123         rcu_barrier();
2124         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2125         unregister_netdevice_notifier(&dp_device_notifier);
2126         vport_exit();
2127         flow_exit();
2128 }
2129
2130 module_init(dp_init);
2131 module_exit(dp_cleanup);
2132
2133 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2134 MODULE_LICENSE("GPL");