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