pintos-os.org Git - openvswitch/blob - 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         return queue_control_packets(dp, skb, upcall_info);
 445
 446 err_kfree_skb:
 447         kfree_skb(skb);
 448 err:
 449         local_bh_disable();
 450         stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
 451
 452         write_seqcount_begin(&stats->seqlock);
 453         stats->n_lost++;
 454         write_seqcount_end(&stats->seqlock);
 455
 456         local_bh_enable();
 457
 458         return err;
 459 }
 460
 461 /* Send each packet in the 'skb' list to userspace for 'dp' as directed by
 462  * 'upcall_info'.  There will be only one packet unless we broke up a GSO
 463  * packet.
 464  */
 465 static int queue_control_packets(struct datapath *dp, struct sk_buff *skb,
 466                                  const struct dp_upcall_info *upcall_info)
 467 {
 468         u32 group = packet_mc_group(dp, upcall_info->cmd);
 469         struct sk_buff *nskb;
 470         int port_no;
 471         int err;
 472
 473         if (OVS_CB(skb)->vport)
 474                 port_no = OVS_CB(skb)->vport->port_no;
 475         else
 476                 port_no = ODPP_LOCAL;
 477
 478         do {
 479                 struct odp_header *upcall;
 480                 struct sk_buff *user_skb; /* to be queued to userspace */
 481                 struct nlattr *nla;
 482                 unsigned int len;
 483
 484                 nskb = skb->next;
 485                 skb->next = NULL;
 486
 487                 err = vlan_deaccel_tag(skb);
 488                 if (unlikely(err))
 489                         goto err_kfree_skbs;
 490
 491                 len = sizeof(struct odp_header);
 492                 len += nla_total_size(skb->len);
 493                 len += nla_total_size(FLOW_BUFSIZE);
 494                 if (upcall_info->userdata)
 495                         len += nla_total_size(8);
 496                 if (upcall_info->sample_pool)
 497                         len += nla_total_size(4);
 498                 if (upcall_info->actions_len)
 499                         len += nla_total_size(upcall_info->actions_len);
 500
 501                 user_skb = genlmsg_new(len, GFP_ATOMIC);
 502                 if (!user_skb) {
 503                         netlink_set_err(INIT_NET_GENL_SOCK, 0, group, -ENOBUFS);
 504                         goto err_kfree_skbs;
 505                 }
 506
 507                 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family, 0, upcall_info->cmd);
 508                 upcall->dp_ifindex = dp->dp_ifindex;
 509
 510                 nla = nla_nest_start(user_skb, ODP_PACKET_ATTR_KEY);
 511                 flow_to_nlattrs(upcall_info->key, user_skb);
 512                 nla_nest_end(user_skb, nla);
 513
 514                 if (upcall_info->userdata)
 515                         nla_put_u64(user_skb, ODP_PACKET_ATTR_USERDATA, upcall_info->userdata);
 516                 if (upcall_info->sample_pool)
 517                         nla_put_u32(user_skb, ODP_PACKET_ATTR_SAMPLE_POOL, upcall_info->sample_pool);
 518                 if (upcall_info->actions_len) {
 519                         const struct nlattr *actions = upcall_info->actions;
 520                         u32 actions_len = upcall_info->actions_len;
 521
 522                         nla = nla_nest_start(user_skb, ODP_PACKET_ATTR_ACTIONS);
 523                         memcpy(__skb_put(user_skb, actions_len), actions, actions_len);
 524                         nla_nest_end(user_skb, nla);
 525                 }
 526
 527                 nla = __nla_reserve(user_skb, ODP_PACKET_ATTR_PACKET, skb->len);
 528                 if (skb->ip_summed == CHECKSUM_PARTIAL)
 529                         copy_and_csum_skb(skb, nla_data(nla));
 530                 else
 531                         skb_copy_bits(skb, 0, nla_data(nla), skb->len);
 532
 533                 err = genlmsg_multicast(user_skb, 0, group, GFP_ATOMIC);
 534                 if (err)
 535                         goto err_kfree_skbs;
 536
 537                 kfree_skb(skb);
 538                 skb = nskb;
 539         } while (skb);
 540         return 0;
 541
 542 err_kfree_skbs:
 543         kfree_skb(skb);
 544         while ((skb = nskb) != NULL) {
 545                 nskb = skb->next;
 546                 kfree_skb(skb);
 547         }
 548         return err;
 549 }
 550
 551 /* Called with genl_mutex. */
 552 static int flush_flows(int dp_ifindex)
 553 {
 554         struct tbl *old_table;
 555         struct tbl *new_table;
 556         struct datapath *dp;
 557
 558         dp = get_dp(dp_ifindex);
 559         if (!dp)
 560                 return -ENODEV;
 561
 562         old_table = get_table_protected(dp);
 563         new_table = tbl_create(TBL_MIN_BUCKETS);
 564         if (!new_table)
 565                 return -ENOMEM;
 566
 567         rcu_assign_pointer(dp->table, new_table);
 568
 569         tbl_deferred_destroy(old_table, flow_free_tbl);
 570
 571         return 0;
 572 }
 573
 574 static int validate_actions(const struct nlattr *attr)
 575 {
 576         const struct nlattr *a;
 577         int rem;
 578
 579         nla_for_each_nested(a, attr, rem) {
 580                 static const u32 action_lens[ODP_ACTION_ATTR_MAX + 1] = {
 581                         [ODP_ACTION_ATTR_OUTPUT] = 4,
 582                         [ODP_ACTION_ATTR_CONTROLLER] = 8,
 583                         [ODP_ACTION_ATTR_SET_DL_TCI] = 2,
 584                         [ODP_ACTION_ATTR_STRIP_VLAN] = 0,
 585                         [ODP_ACTION_ATTR_SET_DL_SRC] = ETH_ALEN,
 586                         [ODP_ACTION_ATTR_SET_DL_DST] = ETH_ALEN,
 587                         [ODP_ACTION_ATTR_SET_NW_SRC] = 4,
 588                         [ODP_ACTION_ATTR_SET_NW_DST] = 4,
 589                         [ODP_ACTION_ATTR_SET_NW_TOS] = 1,
 590                         [ODP_ACTION_ATTR_SET_TP_SRC] = 2,
 591                         [ODP_ACTION_ATTR_SET_TP_DST] = 2,
 592                         [ODP_ACTION_ATTR_SET_TUNNEL] = 8,
 593                         [ODP_ACTION_ATTR_SET_PRIORITY] = 4,
 594                         [ODP_ACTION_ATTR_POP_PRIORITY] = 0,
 595                         [ODP_ACTION_ATTR_DROP_SPOOFED_ARP] = 0,
 596                 };
 597                 int type = nla_type(a);
 598
 599                 if (type > ODP_ACTION_ATTR_MAX || nla_len(a) != action_lens[type])
 600                         return -EINVAL;
 601
 602                 switch (type) {
 603                 case ODP_ACTION_ATTR_UNSPEC:
 604                         return -EINVAL;
 605
 606                 case ODP_ACTION_ATTR_CONTROLLER:
 607                 case ODP_ACTION_ATTR_STRIP_VLAN:
 608                 case ODP_ACTION_ATTR_SET_DL_SRC:
 609                 case ODP_ACTION_ATTR_SET_DL_DST:
 610                 case ODP_ACTION_ATTR_SET_NW_SRC:
 611                 case ODP_ACTION_ATTR_SET_NW_DST:
 612                 case ODP_ACTION_ATTR_SET_TP_SRC:
 613                 case ODP_ACTION_ATTR_SET_TP_DST:
 614                 case ODP_ACTION_ATTR_SET_TUNNEL:
 615                 case ODP_ACTION_ATTR_SET_PRIORITY:
 616                 case ODP_ACTION_ATTR_POP_PRIORITY:
 617                 case ODP_ACTION_ATTR_DROP_SPOOFED_ARP:
 618                         /* No validation needed. */
 619                         break;
 620
 621                 case ODP_ACTION_ATTR_OUTPUT:
 622                         if (nla_get_u32(a) >= DP_MAX_PORTS)
 623                                 return -EINVAL;
 624                         break;
 625
 626                 case ODP_ACTION_ATTR_SET_DL_TCI:
 627                         if (nla_get_be16(a) & htons(VLAN_CFI_MASK))
 628                                 return -EINVAL;
 629                         break;
 630
 631                 case ODP_ACTION_ATTR_SET_NW_TOS:
 632                         if (nla_get_u8(a) & INET_ECN_MASK)
 633                                 return -EINVAL;
 634                         break;
 635
 636                 default:
 637                         return -EOPNOTSUPP;
 638                 }
 639         }
 640
 641         if (rem > 0)
 642                 return -EINVAL;
 643
 644         return 0;
 645 }
 646 static void clear_stats(struct sw_flow *flow)
 647 {
 648         flow->used = 0;
 649         flow->tcp_flags = 0;
 650         flow->packet_count = 0;
 651         flow->byte_count = 0;
 652 }
 653
 654 /* Called with genl_mutex. */
 655 static int expand_table(struct datapath *dp)
 656 {
 657         struct tbl *old_table = get_table_protected(dp);
 658         struct tbl *new_table;
 659
 660         new_table = tbl_expand(old_table);
 661         if (IS_ERR(new_table))
 662                 return PTR_ERR(new_table);
 663
 664         rcu_assign_pointer(dp->table, new_table);
 665         tbl_deferred_destroy(old_table, NULL);
 666
 667         return 0;
 668 }
 669
 670 static int odp_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 671 {
 672         struct odp_header *odp_header = info->userhdr;
 673         struct nlattr **a = info->attrs;
 674         struct sk_buff *packet;
 675         struct sw_flow_key key;
 676         struct datapath *dp;
 677         struct ethhdr *eth;
 678         bool is_frag;
 679         int err;
 680
 681         err = -EINVAL;
 682         if (!a[ODP_PACKET_ATTR_PACKET] || !a[ODP_PACKET_ATTR_ACTIONS] ||
 683             nla_len(a[ODP_PACKET_ATTR_PACKET]) < ETH_HLEN)
 684                 goto exit;
 685
 686         err = validate_actions(a[ODP_PACKET_ATTR_ACTIONS]);
 687         if (err)
 688                 goto exit;
 689
 690         packet = skb_clone(skb, GFP_KERNEL);
 691         err = -ENOMEM;
 692         if (!packet)
 693                 goto exit;
 694         packet->data = nla_data(a[ODP_PACKET_ATTR_PACKET]);
 695         packet->len = nla_len(a[ODP_PACKET_ATTR_PACKET]);
 696
 697         skb_reset_mac_header(packet);
 698         eth = eth_hdr(packet);
 699
 700         /* Normally, setting the skb 'protocol' field would be handled by a
 701          * call to eth_type_trans(), but it assumes there's a sending
 702          * device, which we may not have. */
 703         if (ntohs(eth->h_proto) >= 1536)
 704                 packet->protocol = eth->h_proto;
 705         else
 706                 packet->protocol = htons(ETH_P_802_2);
 707
 708         err = flow_extract(packet, -1, &key, &is_frag);
 709         if (err)
 710                 goto exit;
 711
 712         /* Initialize OVS_CB (it came from Netlink so might not be zeroed). */
 713         memset(OVS_CB(packet), 0, sizeof(struct ovs_skb_cb));
 714
 715         rcu_read_lock();
 716         dp = get_dp(odp_header->dp_ifindex);
 717         err = -ENODEV;
 718         if (dp)
 719                 err = execute_actions(dp, packet, &key,
 720                                       nla_data(a[ODP_PACKET_ATTR_ACTIONS]),
 721                                       nla_len(a[ODP_PACKET_ATTR_ACTIONS]));
 722         rcu_read_unlock();
 723
 724 exit:
 725         return err;
 726 }
 727
 728 static const struct nla_policy packet_policy[ODP_PACKET_ATTR_MAX + 1] = {
 729         [ODP_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
 730         [ODP_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 731 };
 732
 733 static struct genl_ops dp_packet_genl_ops[] = {
 734         { .cmd = ODP_PACKET_CMD_EXECUTE,
 735           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 736           .policy = packet_policy,
 737           .doit = odp_packet_cmd_execute
 738         }
 739 };
 740
 741 static void get_dp_stats(struct datapath *dp, struct odp_stats *stats)
 742 {
 743         int i;
 744
 745         stats->n_frags = stats->n_hit = stats->n_missed = stats->n_lost = 0;
 746         for_each_possible_cpu(i) {
 747                 const struct dp_stats_percpu *percpu_stats;
 748                 struct dp_stats_percpu local_stats;
 749                 unsigned seqcount;
 750
 751                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 752
 753                 do {
 754                         seqcount = read_seqcount_begin(&percpu_stats->seqlock);
 755                         local_stats = *percpu_stats;
 756                 } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount));
 757
 758                 stats->n_frags += local_stats.n_frags;
 759                 stats->n_hit += local_stats.n_hit;
 760                 stats->n_missed += local_stats.n_missed;
 761                 stats->n_lost += local_stats.n_lost;
 762         }
 763 }
 764
 765 /* MTU of the dp pseudo-device: ETH_DATA_LEN or the minimum of the ports.
 766  * Called with RTNL lock.
 767  */
 768 int dp_min_mtu(const struct datapath *dp)
 769 {
 770         struct vport *p;
 771         int mtu = 0;
 772
 773         ASSERT_RTNL();
 774
 775         list_for_each_entry (p, &dp->port_list, node) {
 776                 int dev_mtu;
 777
 778                 /* Skip any internal ports, since that's what we're trying to
 779                  * set. */
 780                 if (is_internal_vport(p))
 781                         continue;
 782
 783                 dev_mtu = vport_get_mtu(p);
 784                 if (!dev_mtu)
 785                         continue;
 786                 if (!mtu || dev_mtu < mtu)
 787                         mtu = dev_mtu;
 788         }
 789
 790         return mtu ? mtu : ETH_DATA_LEN;
 791 }
 792
 793 /* Sets the MTU of all datapath devices to the minimum of the ports
 794  * Called with RTNL lock.
 795  */
 796 void set_internal_devs_mtu(const struct datapath *dp)
 797 {
 798         struct vport *p;
 799         int mtu;
 800
 801         ASSERT_RTNL();
 802
 803         mtu = dp_min_mtu(dp);
 804
 805         list_for_each_entry (p, &dp->port_list, node) {
 806                 if (is_internal_vport(p))
 807                         vport_set_mtu(p, mtu);
 808         }
 809 }
 810
 811 static const struct nla_policy flow_policy[ODP_FLOW_ATTR_MAX + 1] = {
 812         [ODP_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
 813         [ODP_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
 814         [ODP_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
 815 };
 816
 817 static struct genl_family dp_flow_genl_family = {
 818         .id = GENL_ID_GENERATE,
 819         .hdrsize = sizeof(struct odp_header),
 820         .name = ODP_FLOW_FAMILY,
 821         .version = 1,
 822         .maxattr = ODP_FLOW_ATTR_MAX
 823 };
 824
 825 static struct genl_multicast_group dp_flow_multicast_group = {
 826         .name = ODP_FLOW_MCGROUP
 827 };
 828
 829 /* Called with genl_lock. */
 830 static int odp_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
 831                                   struct sk_buff *skb, u32 pid, u32 seq, u32 flags, u8 cmd)
 832 {
 833         const int skb_orig_len = skb->len;
 834         const struct sw_flow_actions *sf_acts;
 835         struct odp_flow_stats stats;
 836         struct odp_header *odp_header;
 837         struct nlattr *nla;
 838         unsigned long used;
 839         u8 tcp_flags;
 840         int err;
 841
 842         sf_acts = rcu_dereference_protected(flow->sf_acts,
 843                                             lockdep_genl_is_held());
 844
 845         odp_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags, cmd);
 846         if (!odp_header)
 847                 return -EMSGSIZE;
 848
 849         odp_header->dp_ifindex = dp->dp_ifindex;
 850
 851         nla = nla_nest_start(skb, ODP_FLOW_ATTR_KEY);
 852         if (!nla)
 853                 goto nla_put_failure;
 854         err = flow_to_nlattrs(&flow->key, skb);
 855         if (err)
 856                 goto error;
 857         nla_nest_end(skb, nla);
 858
 859         spin_lock_bh(&flow->lock);
 860         used = flow->used;
 861         stats.n_packets = flow->packet_count;
 862         stats.n_bytes = flow->byte_count;
 863         tcp_flags = flow->tcp_flags;
 864         spin_unlock_bh(&flow->lock);
 865
 866         if (used)
 867                 NLA_PUT_U64(skb, ODP_FLOW_ATTR_USED, flow_used_time(used));
 868
 869         if (stats.n_packets)
 870                 NLA_PUT(skb, ODP_FLOW_ATTR_STATS, sizeof(struct odp_flow_stats), &stats);
 871
 872         if (tcp_flags)
 873                 NLA_PUT_U8(skb, ODP_FLOW_ATTR_TCP_FLAGS, tcp_flags);
 874
 875         /* If ODP_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 876          * this is the first flow to be dumped into 'skb'.  This is unusual for
 877          * Netlink but individual action lists can be longer than
 878          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 879          * The userspace caller can always fetch the actions separately if it
 880          * really wants them.  (Most userspace callers in fact don't care.)
 881          *
 882          * This can only fail for dump operations because the skb is always
 883          * properly sized for single flows.
 884          */
 885         err = nla_put(skb, ODP_FLOW_ATTR_ACTIONS, sf_acts->actions_len,
 886                       sf_acts->actions);
 887         if (err < 0 && skb_orig_len)
 888                 goto error;
 889
 890         return genlmsg_end(skb, odp_header);
 891
 892 nla_put_failure:
 893         err = -EMSGSIZE;
 894 error:
 895         genlmsg_cancel(skb, odp_header);
 896         return err;
 897 }
 898
 899 static struct sk_buff *odp_flow_cmd_alloc_info(struct sw_flow *flow)
 900 {
 901         const struct sw_flow_actions *sf_acts;
 902         int len;
 903
 904         sf_acts = rcu_dereference_protected(flow->sf_acts,
 905                                             lockdep_genl_is_held());
 906
 907         len = nla_total_size(FLOW_BUFSIZE); /* ODP_FLOW_ATTR_KEY */
 908         len += nla_total_size(sf_acts->actions_len); /* ODP_FLOW_ATTR_ACTIONS */
 909         len += nla_total_size(sizeof(struct odp_flow_stats)); /* ODP_FLOW_ATTR_STATS */
 910         len += nla_total_size(1); /* ODP_FLOW_ATTR_TCP_FLAGS */
 911         len += nla_total_size(8); /* ODP_FLOW_ATTR_USED */
 912         return genlmsg_new(NLMSG_ALIGN(sizeof(struct odp_header)) + len, GFP_KERNEL);
 913 }
 914
 915 static struct sk_buff *odp_flow_cmd_build_info(struct sw_flow *flow, struct datapath *dp,
 916                                                u32 pid, u32 seq, u8 cmd)
 917 {
 918         struct sk_buff *skb;
 919         int retval;
 920
 921         skb = odp_flow_cmd_alloc_info(flow);
 922         if (!skb)
 923                 return ERR_PTR(-ENOMEM);
 924
 925         retval = odp_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd);
 926         BUG_ON(retval < 0);
 927         return skb;
 928 }
 929
 930 static int odp_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
 931 {
 932         struct nlattr **a = info->attrs;
 933         struct odp_header *odp_header = info->userhdr;
 934         struct tbl_node *flow_node;
 935         struct sw_flow_key key;
 936         struct sw_flow *flow;
 937         struct sk_buff *reply;
 938         struct datapath *dp;
 939         struct tbl *table;
 940         u32 hash;
 941         int error;
 942
 943         /* Extract key. */
 944         error = -EINVAL;
 945         if (!a[ODP_FLOW_ATTR_KEY])
 946                 goto error;
 947         error = flow_from_nlattrs(&key, a[ODP_FLOW_ATTR_KEY]);
 948         if (error)
 949                 goto error;
 950
 951         /* Validate actions. */
 952         if (a[ODP_FLOW_ATTR_ACTIONS]) {
 953                 error = validate_actions(a[ODP_FLOW_ATTR_ACTIONS]);
 954                 if (error)
 955                         goto error;
 956         } else if (info->genlhdr->cmd == ODP_FLOW_CMD_NEW) {
 957                 error = -EINVAL;
 958                 goto error;
 959         }
 960
 961         dp = get_dp(odp_header->dp_ifindex);
 962         error = -ENODEV;
 963         if (!dp)
 964                 goto error;
 965
 966         hash = flow_hash(&key);
 967         table = get_table_protected(dp);
 968         flow_node = tbl_lookup(table, &key, hash, flow_cmp);
 969         if (!flow_node) {
 970                 struct sw_flow_actions *acts;
 971
 972                 /* Bail out if we're not allowed to create a new flow. */
 973                 error = -ENOENT;
 974                 if (info->genlhdr->cmd == ODP_FLOW_CMD_SET)
 975                         goto error;
 976
 977                 /* Expand table, if necessary, to make room. */
 978                 if (tbl_count(table) >= tbl_n_buckets(table)) {
 979                         error = expand_table(dp);
 980                         if (error)
 981                                 goto error;
 982                         table = get_table_protected(dp);
 983                 }
 984
 985                 /* Allocate flow. */
 986                 flow = flow_alloc();
 987                 if (IS_ERR(flow)) {
 988                         error = PTR_ERR(flow);
 989                         goto error;
 990                 }
 991                 flow->key = key;
 992                 clear_stats(flow);
 993
 994                 /* Obtain actions. */
 995                 acts = flow_actions_alloc(a[ODP_FLOW_ATTR_ACTIONS]);
 996                 error = PTR_ERR(acts);
 997                 if (IS_ERR(acts))
 998                         goto error_free_flow;
 999                 rcu_assign_pointer(flow->sf_acts, acts);
1000
1001                 /* Put flow in bucket. */
1002                 error = tbl_insert(table, &flow->tbl_node, hash);
1003                 if (error)
1004                         goto error_free_flow;
1005
1006                 reply = odp_flow_cmd_build_info(flow, dp, info->snd_pid,
1007                                                 info->snd_seq, ODP_FLOW_CMD_NEW);
1008         } else {
1009                 /* We found a matching flow. */
1010                 struct sw_flow_actions *old_acts;
1011
1012                 /* Bail out if we're not allowed to modify an existing flow.
1013                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1014                  * because Generic Netlink treats the latter as a dump
1015                  * request.  We also accept NLM_F_EXCL in case that bug ever
1016                  * gets fixed.
1017                  */
1018                 error = -EEXIST;
1019                 if (info->genlhdr->cmd == ODP_FLOW_CMD_NEW &&
1020                     info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1021                         goto error;
1022
1023                 /* Update actions. */
1024                 flow = flow_cast(flow_node);
1025                 old_acts = rcu_dereference_protected(flow->sf_acts,
1026                                                      lockdep_genl_is_held());
1027                 if (a[ODP_FLOW_ATTR_ACTIONS] &&
1028                     (old_acts->actions_len != nla_len(a[ODP_FLOW_ATTR_ACTIONS]) ||
1029                      memcmp(old_acts->actions, nla_data(a[ODP_FLOW_ATTR_ACTIONS]),
1030                             old_acts->actions_len))) {
1031                         struct sw_flow_actions *new_acts;
1032
1033                         new_acts = flow_actions_alloc(a[ODP_FLOW_ATTR_ACTIONS]);
1034                         error = PTR_ERR(new_acts);
1035                         if (IS_ERR(new_acts))
1036                                 goto error;
1037
1038                         rcu_assign_pointer(flow->sf_acts, new_acts);
1039                         flow_deferred_free_acts(old_acts);
1040                 }
1041
1042                 reply = odp_flow_cmd_build_info(flow, dp, info->snd_pid,
1043                                                 info->snd_seq, ODP_FLOW_CMD_NEW);
1044
1045                 /* Clear stats. */
1046                 if (a[ODP_FLOW_ATTR_CLEAR]) {
1047                         spin_lock_bh(&flow->lock);
1048                         clear_stats(flow);
1049                         spin_unlock_bh(&flow->lock);
1050                 }
1051         }
1052
1053         if (!IS_ERR(reply))
1054                 genl_notify(reply, genl_info_net(info), info->snd_pid,
1055                             dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1056         else
1057                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1058                                 dp_flow_multicast_group.id, PTR_ERR(reply));
1059         return 0;
1060
1061 error_free_flow:
1062         flow_put(flow);
1063 error:
1064         return error;
1065 }
1066
1067 static int odp_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1068 {
1069         struct nlattr **a = info->attrs;
1070         struct odp_header *odp_header = info->userhdr;
1071         struct sw_flow_key key;
1072         struct tbl_node *flow_node;
1073         struct sk_buff *reply;
1074         struct sw_flow *flow;
1075         struct datapath *dp;
1076         struct tbl *table;
1077         int err;
1078
1079         if (!a[ODP_FLOW_ATTR_KEY])
1080                 return -EINVAL;
1081         err = flow_from_nlattrs(&key, a[ODP_FLOW_ATTR_KEY]);
1082         if (err)
1083                 return err;
1084
1085         dp = get_dp(odp_header->dp_ifindex);
1086         if (!dp)
1087                 return -ENODEV;
1088
1089         table = get_table_protected(dp);
1090         flow_node = tbl_lookup(table, &key, flow_hash(&key), flow_cmp);
1091         if (!flow_node)
1092                 return -ENOENT;
1093
1094         flow = flow_cast(flow_node);
1095         reply = odp_flow_cmd_build_info(flow, dp, info->snd_pid, info->snd_seq, ODP_FLOW_CMD_NEW);
1096         if (IS_ERR(reply))
1097                 return PTR_ERR(reply);
1098
1099         return genlmsg_reply(reply, info);
1100 }
1101
1102 static int odp_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1103 {
1104         struct nlattr **a = info->attrs;
1105         struct odp_header *odp_header = info->userhdr;
1106         struct sw_flow_key key;
1107         struct tbl_node *flow_node;
1108         struct sk_buff *reply;
1109         struct sw_flow *flow;
1110         struct datapath *dp;
1111         struct tbl *table;
1112         int err;
1113
1114         if (!a[ODP_FLOW_ATTR_KEY])
1115                 return flush_flows(odp_header->dp_ifindex);
1116         err = flow_from_nlattrs(&key, a[ODP_FLOW_ATTR_KEY]);
1117         if (err)
1118                 return err;
1119
1120         dp = get_dp(odp_header->dp_ifindex);
1121         if (!dp)
1122                 return -ENODEV;
1123
1124         table = get_table_protected(dp);
1125         flow_node = tbl_lookup(table, &key, flow_hash(&key), flow_cmp);
1126         if (!flow_node)
1127                 return -ENOENT;
1128         flow = flow_cast(flow_node);
1129
1130         reply = odp_flow_cmd_alloc_info(flow);
1131         if (!reply)
1132                 return -ENOMEM;
1133
1134         err = tbl_remove(table, flow_node);
1135         if (err) {
1136                 kfree_skb(reply);
1137                 return err;
1138         }
1139
1140         err = odp_flow_cmd_fill_info(flow, dp, reply, info->snd_pid,
1141                                      info->snd_seq, 0, ODP_FLOW_CMD_DEL);
1142         BUG_ON(err < 0);
1143
1144         flow_deferred_free(flow);
1145
1146         genl_notify(reply, genl_info_net(info), info->snd_pid,
1147                     dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1148         return 0;
1149 }
1150
1151 static int odp_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1152 {
1153         struct odp_header *odp_header = genlmsg_data(nlmsg_data(cb->nlh));
1154         struct datapath *dp;
1155
1156         dp = get_dp(odp_header->dp_ifindex);
1157         if (!dp)
1158                 return -ENODEV;
1159
1160         for (;;) {
1161                 struct tbl_node *flow_node;
1162                 struct sw_flow *flow;
1163                 u32 bucket, obj;
1164
1165                 bucket = cb->args[0];
1166                 obj = cb->args[1];
1167                 flow_node = tbl_next(get_table_protected(dp), &bucket, &obj);
1168                 if (!flow_node)
1169                         break;
1170
1171                 flow = flow_cast(flow_node);
1172                 if (odp_flow_cmd_fill_info(flow, dp, skb, NETLINK_CB(cb->skb).pid,
1173                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1174                                            ODP_FLOW_CMD_NEW) < 0)
1175                         break;
1176
1177                 cb->args[0] = bucket;
1178                 cb->args[1] = obj;
1179         }
1180         return skb->len;
1181 }
1182
1183 static struct genl_ops dp_flow_genl_ops[] = {
1184         { .cmd = ODP_FLOW_CMD_NEW,
1185           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1186           .policy = flow_policy,
1187           .doit = odp_flow_cmd_new_or_set
1188         },
1189         { .cmd = ODP_FLOW_CMD_DEL,
1190           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1191           .policy = flow_policy,
1192           .doit = odp_flow_cmd_del
1193         },
1194         { .cmd = ODP_FLOW_CMD_GET,
1195           .flags = 0,               /* OK for unprivileged users. */
1196           .policy = flow_policy,
1197           .doit = odp_flow_cmd_get,
1198           .dumpit = odp_flow_cmd_dump
1199         },
1200         { .cmd = ODP_FLOW_CMD_SET,
1201           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1202           .policy = flow_policy,
1203           .doit = odp_flow_cmd_new_or_set,
1204         },
1205 };
1206
1207 static const struct nla_policy datapath_policy[ODP_DP_ATTR_MAX + 1] = {
1208 #ifdef HAVE_NLA_NUL_STRING
1209         [ODP_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1210 #endif
1211         [ODP_DP_ATTR_IPV4_FRAGS] = { .type = NLA_U32 },
1212         [ODP_DP_ATTR_SAMPLING] = { .type = NLA_U32 },
1213 };
1214
1215 static struct genl_family dp_datapath_genl_family = {
1216         .id = GENL_ID_GENERATE,
1217         .hdrsize = sizeof(struct odp_header),
1218         .name = ODP_DATAPATH_FAMILY,
1219         .version = 1,
1220         .maxattr = ODP_DP_ATTR_MAX
1221 };
1222
1223 static struct genl_multicast_group dp_datapath_multicast_group = {
1224         .name = ODP_DATAPATH_MCGROUP
1225 };
1226
1227 static int odp_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1228                                 u32 pid, u32 seq, u32 flags, u8 cmd)
1229 {
1230         struct odp_header *odp_header;
1231         struct nlattr *nla;
1232         int err;
1233
1234         odp_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family,
1235                                    flags, cmd);
1236         if (!odp_header)
1237                 goto error;
1238
1239         odp_header->dp_ifindex = dp->dp_ifindex;
1240
1241         rcu_read_lock();
1242         err = nla_put_string(skb, ODP_DP_ATTR_NAME, dp_name(dp));
1243         rcu_read_unlock();
1244         if (err)
1245                 goto nla_put_failure;
1246
1247         nla = nla_reserve(skb, ODP_DP_ATTR_STATS, sizeof(struct odp_stats));
1248         if (!nla)
1249                 goto nla_put_failure;
1250         get_dp_stats(dp, nla_data(nla));
1251
1252         NLA_PUT_U32(skb, ODP_DP_ATTR_IPV4_FRAGS,
1253                     dp->drop_frags ? ODP_DP_FRAG_DROP : ODP_DP_FRAG_ZERO);
1254
1255         if (dp->sflow_probability)
1256                 NLA_PUT_U32(skb, ODP_DP_ATTR_SAMPLING, dp->sflow_probability);
1257
1258         nla = nla_nest_start(skb, ODP_DP_ATTR_MCGROUPS);
1259         if (!nla)
1260                 goto nla_put_failure;
1261         NLA_PUT_U32(skb, ODP_PACKET_CMD_MISS, packet_mc_group(dp, ODP_PACKET_CMD_MISS));
1262         NLA_PUT_U32(skb, ODP_PACKET_CMD_ACTION, packet_mc_group(dp, ODP_PACKET_CMD_ACTION));
1263         NLA_PUT_U32(skb, ODP_PACKET_CMD_SAMPLE, packet_mc_group(dp, ODP_PACKET_CMD_SAMPLE));
1264         nla_nest_end(skb, nla);
1265
1266         return genlmsg_end(skb, odp_header);
1267
1268 nla_put_failure:
1269         genlmsg_cancel(skb, odp_header);
1270 error:
1271         return -EMSGSIZE;
1272 }
1273
1274 static struct sk_buff *odp_dp_cmd_build_info(struct datapath *dp, u32 pid,
1275                                              u32 seq, u8 cmd)
1276 {
1277         struct sk_buff *skb;
1278         int retval;
1279
1280         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1281         if (!skb)
1282                 return ERR_PTR(-ENOMEM);
1283
1284         retval = odp_dp_cmd_fill_info(dp, skb, pid, seq, 0, cmd);
1285         if (retval < 0) {
1286                 kfree_skb(skb);
1287                 return ERR_PTR(retval);
1288         }
1289         return skb;
1290 }
1291
1292 static int odp_dp_cmd_validate(struct nlattr *a[ODP_DP_ATTR_MAX + 1])
1293 {
1294         if (a[ODP_DP_ATTR_IPV4_FRAGS]) {
1295                 u32 frags = nla_get_u32(a[ODP_DP_ATTR_IPV4_FRAGS]);
1296
1297                 if (frags != ODP_DP_FRAG_ZERO && frags != ODP_DP_FRAG_DROP)
1298                         return -EINVAL;
1299         }
1300
1301         return VERIFY_NUL_STRING(a[ODP_DP_ATTR_NAME], IFNAMSIZ - 1);
1302 }
1303
1304 /* Called with genl_mutex and optionally with RTNL lock also. */
1305 static struct datapath *lookup_datapath(struct odp_header *odp_header, struct nlattr *a[ODP_DP_ATTR_MAX + 1])
1306 {
1307         struct datapath *dp;
1308
1309         if (!a[ODP_DP_ATTR_NAME])
1310                 dp = get_dp(odp_header->dp_ifindex);
1311         else {
1312                 struct vport *vport;
1313
1314                 rcu_read_lock();
1315                 vport = vport_locate(nla_data(a[ODP_DP_ATTR_NAME]));
1316                 dp = vport && vport->port_no == ODPP_LOCAL ? vport->dp : NULL;
1317                 rcu_read_unlock();
1318         }
1319         return dp ? dp : ERR_PTR(-ENODEV);
1320 }
1321
1322 /* Called with genl_mutex. */
1323 static void change_datapath(struct datapath *dp, struct nlattr *a[ODP_DP_ATTR_MAX + 1])
1324 {
1325         if (a[ODP_DP_ATTR_IPV4_FRAGS])
1326                 dp->drop_frags = nla_get_u32(a[ODP_DP_ATTR_IPV4_FRAGS]) == ODP_DP_FRAG_DROP;
1327         if (a[ODP_DP_ATTR_SAMPLING])
1328                 dp->sflow_probability = nla_get_u32(a[ODP_DP_ATTR_SAMPLING]);
1329 }
1330
1331 static int odp_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1332 {
1333         struct nlattr **a = info->attrs;
1334         struct vport_parms parms;
1335         struct sk_buff *reply;
1336         struct datapath *dp;
1337         struct vport *vport;
1338         int err;
1339
1340         err = -EINVAL;
1341         if (!a[ODP_DP_ATTR_NAME])
1342                 goto err;
1343
1344         err = odp_dp_cmd_validate(a);
1345         if (err)
1346                 goto err;
1347
1348         rtnl_lock();
1349         err = -ENODEV;
1350         if (!try_module_get(THIS_MODULE))
1351                 goto err_unlock_rtnl;
1352
1353         err = -ENOMEM;
1354         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1355         if (dp == NULL)
1356                 goto err_put_module;
1357         INIT_LIST_HEAD(&dp->port_list);
1358
1359         /* Initialize kobject for bridge.  This will be added as
1360          * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
1361         dp->ifobj.kset = NULL;
1362         kobject_init(&dp->ifobj, &dp_ktype);
1363
1364         /* Allocate table. */
1365         err = -ENOMEM;
1366         rcu_assign_pointer(dp->table, tbl_create(TBL_MIN_BUCKETS));
1367         if (!dp->table)
1368                 goto err_free_dp;
1369
1370         /* Set up our datapath device. */
1371         parms.name = nla_data(a[ODP_DP_ATTR_NAME]);
1372         parms.type = ODP_VPORT_TYPE_INTERNAL;
1373         parms.options = NULL;
1374         parms.dp = dp;
1375         parms.port_no = ODPP_LOCAL;
1376         vport = new_vport(&parms);
1377         if (IS_ERR(vport)) {
1378                 err = PTR_ERR(vport);
1379                 if (err == -EBUSY)
1380                         err = -EEXIST;
1381
1382                 goto err_destroy_table;
1383         }
1384         dp->dp_ifindex = vport_get_ifindex(vport);
1385
1386         dp->drop_frags = 0;
1387         dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1388         if (!dp->stats_percpu) {
1389                 err = -ENOMEM;
1390                 goto err_destroy_local_port;
1391         }
1392
1393         change_datapath(dp, a);
1394
1395         reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_NEW);
1396         err = PTR_ERR(reply);
1397         if (IS_ERR(reply))
1398                 goto err_destroy_local_port;
1399
1400         list_add_tail(&dp->list_node, &dps);
1401         dp_sysfs_add_dp(dp);
1402
1403         rtnl_unlock();
1404
1405         genl_notify(reply, genl_info_net(info), info->snd_pid,
1406                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1407         return 0;
1408
1409 err_destroy_local_port:
1410         dp_detach_port(get_vport_protected(dp, ODPP_LOCAL));
1411 err_destroy_table:
1412         tbl_destroy(get_table_protected(dp), NULL);
1413 err_free_dp:
1414         kfree(dp);
1415 err_put_module:
1416         module_put(THIS_MODULE);
1417 err_unlock_rtnl:
1418         rtnl_unlock();
1419 err:
1420         return err;
1421 }
1422
1423 static int odp_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1424 {
1425         struct vport *vport, *next_vport;
1426         struct sk_buff *reply;
1427         struct datapath *dp;
1428         int err;
1429
1430         err = odp_dp_cmd_validate(info->attrs);
1431         if (err)
1432                 goto exit;
1433
1434         rtnl_lock();
1435         dp = lookup_datapath(info->userhdr, info->attrs);
1436         err = PTR_ERR(dp);
1437         if (IS_ERR(dp))
1438                 goto exit_unlock;
1439
1440         reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_DEL);
1441         err = PTR_ERR(reply);
1442         if (IS_ERR(reply))
1443                 goto exit_unlock;
1444
1445         list_for_each_entry_safe (vport, next_vport, &dp->port_list, node)
1446                 if (vport->port_no != ODPP_LOCAL)
1447                         dp_detach_port(vport);
1448
1449         dp_sysfs_del_dp(dp);
1450         list_del(&dp->list_node);
1451         dp_detach_port(get_vport_protected(dp, ODPP_LOCAL));
1452
1453         /* rtnl_unlock() will wait until all the references to devices that
1454          * are pending unregistration have been dropped.  We do it here to
1455          * ensure that any internal devices (which contain DP pointers) are
1456          * fully destroyed before freeing the datapath.
1457          */
1458         rtnl_unlock();
1459
1460         call_rcu(&dp->rcu, destroy_dp_rcu);
1461         module_put(THIS_MODULE);
1462
1463         genl_notify(reply, genl_info_net(info), info->snd_pid,
1464                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1465
1466         return 0;
1467
1468 exit_unlock:
1469         rtnl_unlock();
1470 exit:
1471         return err;
1472 }
1473
1474 static int odp_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1475 {
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                 return err;
1483
1484         dp = lookup_datapath(info->userhdr, info->attrs);
1485         if (IS_ERR(dp))
1486                 return PTR_ERR(dp);
1487
1488         change_datapath(dp, info->attrs);
1489
1490         reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_NEW);
1491         if (IS_ERR(reply)) {
1492                 err = PTR_ERR(reply);
1493                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1494                                 dp_datapath_multicast_group.id, err);
1495                 return 0;
1496         }
1497
1498         genl_notify(reply, genl_info_net(info), info->snd_pid,
1499                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1500         return 0;
1501 }
1502
1503 static int odp_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1504 {
1505         struct sk_buff *reply;
1506         struct datapath *dp;
1507         int err;
1508
1509         err = odp_dp_cmd_validate(info->attrs);
1510         if (err)
1511                 return err;
1512
1513         dp = lookup_datapath(info->userhdr, info->attrs);
1514         if (IS_ERR(dp))
1515                 return PTR_ERR(dp);
1516
1517         reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_NEW);
1518         if (IS_ERR(reply))
1519                 return PTR_ERR(reply);
1520
1521         return genlmsg_reply(reply, info);
1522 }
1523
1524 static int odp_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1525 {
1526         struct datapath *dp;
1527         int skip = cb->args[0];
1528         int i = 0;
1529
1530         list_for_each_entry (dp, &dps, list_node) {
1531                 if (i < skip)
1532                         continue;
1533                 if (odp_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).pid,
1534                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1535                                          ODP_DP_CMD_NEW) < 0)
1536                         break;
1537                 i++;
1538         }
1539
1540         cb->args[0] = i;
1541
1542         return skb->len;
1543 }
1544
1545 static struct genl_ops dp_datapath_genl_ops[] = {
1546         { .cmd = ODP_DP_CMD_NEW,
1547           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1548           .policy = datapath_policy,
1549           .doit = odp_dp_cmd_new
1550         },
1551         { .cmd = ODP_DP_CMD_DEL,
1552           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1553           .policy = datapath_policy,
1554           .doit = odp_dp_cmd_del
1555         },
1556         { .cmd = ODP_DP_CMD_GET,
1557           .flags = 0,               /* OK for unprivileged users. */
1558           .policy = datapath_policy,
1559           .doit = odp_dp_cmd_get,
1560           .dumpit = odp_dp_cmd_dump
1561         },
1562         { .cmd = ODP_DP_CMD_SET,
1563           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1564           .policy = datapath_policy,
1565           .doit = odp_dp_cmd_set,
1566         },
1567 };
1568
1569 static const struct nla_policy vport_policy[ODP_VPORT_ATTR_MAX + 1] = {
1570 #ifdef HAVE_NLA_NUL_STRING
1571         [ODP_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1572         [ODP_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1573         [ODP_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1574         [ODP_VPORT_ATTR_STATS] = { .len = sizeof(struct rtnl_link_stats64) },
1575         [ODP_VPORT_ATTR_ADDRESS] = { .len = ETH_ALEN },
1576 #else
1577         [ODP_VPORT_ATTR_STATS] = { .minlen = sizeof(struct rtnl_link_stats64) },
1578         [ODP_VPORT_ATTR_ADDRESS] = { .minlen = ETH_ALEN },
1579 #endif
1580         [ODP_VPORT_ATTR_MTU] = { .type = NLA_U32 },
1581         [ODP_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1582 };
1583
1584 static struct genl_family dp_vport_genl_family = {
1585         .id = GENL_ID_GENERATE,
1586         .hdrsize = sizeof(struct odp_header),
1587         .name = ODP_VPORT_FAMILY,
1588         .version = 1,
1589         .maxattr = ODP_VPORT_ATTR_MAX
1590 };
1591
1592 static struct genl_multicast_group dp_vport_multicast_group = {
1593         .name = ODP_VPORT_MCGROUP
1594 };
1595
1596 /* Called with RTNL lock or RCU read lock. */
1597 static int odp_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1598                                    u32 pid, u32 seq, u32 flags, u8 cmd)
1599 {
1600         struct odp_header *odp_header;
1601         struct nlattr *nla;
1602         int ifindex, iflink;
1603         int mtu;
1604         int err;
1605
1606         odp_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family,
1607                                  flags, cmd);
1608         if (!odp_header)
1609                 return -EMSGSIZE;
1610
1611         odp_header->dp_ifindex = vport->dp->dp_ifindex;
1612
1613         NLA_PUT_U32(skb, ODP_VPORT_ATTR_PORT_NO, vport->port_no);
1614         NLA_PUT_U32(skb, ODP_VPORT_ATTR_TYPE, vport_get_type(vport));
1615         NLA_PUT_STRING(skb, ODP_VPORT_ATTR_NAME, vport_get_name(vport));
1616
1617         nla = nla_reserve(skb, ODP_VPORT_ATTR_STATS, sizeof(struct rtnl_link_stats64));
1618         if (!nla)
1619                 goto nla_put_failure;
1620         if (vport_get_stats(vport, nla_data(nla)))
1621                 __skb_trim(skb, skb->len - nla->nla_len);
1622
1623         NLA_PUT(skb, ODP_VPORT_ATTR_ADDRESS, ETH_ALEN, vport_get_addr(vport));
1624
1625         mtu = vport_get_mtu(vport);
1626         if (mtu)
1627                 NLA_PUT_U32(skb, ODP_VPORT_ATTR_MTU, mtu);
1628
1629         err = vport_get_options(vport, skb);
1630         if (err == -EMSGSIZE)
1631                 goto error;
1632
1633         ifindex = vport_get_ifindex(vport);
1634         if (ifindex > 0)
1635                 NLA_PUT_U32(skb, ODP_VPORT_ATTR_IFINDEX, ifindex);
1636
1637         iflink = vport_get_iflink(vport);
1638         if (iflink > 0)
1639                 NLA_PUT_U32(skb, ODP_VPORT_ATTR_IFLINK, iflink);
1640
1641         return genlmsg_end(skb, odp_header);
1642
1643 nla_put_failure:
1644         err = -EMSGSIZE;
1645 error:
1646         genlmsg_cancel(skb, odp_header);
1647         return err;
1648 }
1649
1650 /* Called with RTNL lock or RCU read lock. */
1651 static struct sk_buff *odp_vport_cmd_build_info(struct vport *vport, u32 pid,
1652                                                 u32 seq, u8 cmd)
1653 {
1654         struct sk_buff *skb;
1655         int retval;
1656
1657         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1658         if (!skb)
1659                 return ERR_PTR(-ENOMEM);
1660
1661         retval = odp_vport_cmd_fill_info(vport, skb, pid, seq, 0, cmd);
1662         if (retval < 0) {
1663                 kfree_skb(skb);
1664                 return ERR_PTR(retval);
1665         }
1666         return skb;
1667 }
1668
1669 static int odp_vport_cmd_validate(struct nlattr *a[ODP_VPORT_ATTR_MAX + 1])
1670 {
1671         return VERIFY_NUL_STRING(a[ODP_VPORT_ATTR_NAME], IFNAMSIZ - 1);
1672 }
1673
1674 /* Called with RTNL lock or RCU read lock. */
1675 static struct vport *lookup_vport(struct odp_header *odp_header,
1676                                   struct nlattr *a[ODP_VPORT_ATTR_MAX + 1])
1677 {
1678         struct datapath *dp;
1679         struct vport *vport;
1680
1681         if (a[ODP_VPORT_ATTR_NAME]) {
1682                 vport = vport_locate(nla_data(a[ODP_VPORT_ATTR_NAME]));
1683                 if (!vport)
1684                         return ERR_PTR(-ENODEV);
1685                 return vport;
1686         } else if (a[ODP_VPORT_ATTR_PORT_NO]) {
1687                 u32 port_no = nla_get_u32(a[ODP_VPORT_ATTR_PORT_NO]);
1688
1689                 if (port_no >= DP_MAX_PORTS)
1690                         return ERR_PTR(-EFBIG);
1691
1692                 dp = get_dp(odp_header->dp_ifindex);
1693                 if (!dp)
1694                         return ERR_PTR(-ENODEV);
1695
1696                 vport = get_vport_protected(dp, port_no);
1697                 if (!vport)
1698                         return ERR_PTR(-ENOENT);
1699                 return vport;
1700         } else
1701                 return ERR_PTR(-EINVAL);
1702 }
1703
1704 /* Called with RTNL lock. */
1705 static int change_vport(struct vport *vport, struct nlattr *a[ODP_VPORT_ATTR_MAX + 1])
1706 {
1707         int err = 0;
1708         if (a[ODP_VPORT_ATTR_STATS])
1709                 err = vport_set_stats(vport, nla_data(a[ODP_VPORT_ATTR_STATS]));
1710         if (!err && a[ODP_VPORT_ATTR_ADDRESS])
1711                 err = vport_set_addr(vport, nla_data(a[ODP_VPORT_ATTR_ADDRESS]));
1712         if (!err && a[ODP_VPORT_ATTR_MTU])
1713                 err = vport_set_mtu(vport, nla_get_u32(a[ODP_VPORT_ATTR_MTU]));
1714         return err;
1715 }
1716
1717 static int odp_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1718 {
1719         struct nlattr **a = info->attrs;
1720         struct odp_header *odp_header = info->userhdr;
1721         struct vport_parms parms;
1722         struct sk_buff *reply;
1723         struct vport *vport;
1724         struct datapath *dp;
1725         u32 port_no;
1726         int err;
1727
1728         err = -EINVAL;
1729         if (!a[ODP_VPORT_ATTR_NAME] || !a[ODP_VPORT_ATTR_TYPE])
1730                 goto exit;
1731
1732         err = odp_vport_cmd_validate(a);
1733         if (err)
1734                 goto exit;
1735
1736         rtnl_lock();
1737         dp = get_dp(odp_header->dp_ifindex);
1738         err = -ENODEV;
1739         if (!dp)
1740                 goto exit_unlock;
1741
1742         if (a[ODP_VPORT_ATTR_PORT_NO]) {
1743                 port_no = nla_get_u32(a[ODP_VPORT_ATTR_PORT_NO]);
1744
1745                 err = -EFBIG;
1746                 if (port_no >= DP_MAX_PORTS)
1747                         goto exit_unlock;
1748
1749                 vport = get_vport_protected(dp, port_no);
1750                 err = -EBUSY;
1751                 if (vport)
1752                         goto exit_unlock;
1753         } else {
1754                 for (port_no = 1; ; port_no++) {
1755                         if (port_no >= DP_MAX_PORTS) {
1756                                 err = -EFBIG;
1757                                 goto exit_unlock;
1758                         }
1759                         vport = get_vport_protected(dp, port_no);
1760                         if (!vport)
1761                                 break;
1762                 }
1763         }
1764
1765         parms.name = nla_data(a[ODP_VPORT_ATTR_NAME]);
1766         parms.type = nla_get_u32(a[ODP_VPORT_ATTR_TYPE]);
1767         parms.options = a[ODP_VPORT_ATTR_OPTIONS];
1768         parms.dp = dp;
1769         parms.port_no = port_no;
1770
1771         vport = new_vport(&parms);
1772         err = PTR_ERR(vport);
1773         if (IS_ERR(vport))
1774                 goto exit_unlock;
1775
1776         set_internal_devs_mtu(dp);
1777         dp_sysfs_add_if(vport);
1778
1779         err = change_vport(vport, a);
1780         if (!err) {
1781                 reply = odp_vport_cmd_build_info(vport, info->snd_pid,
1782                                                  info->snd_seq, ODP_VPORT_CMD_NEW);
1783                 if (IS_ERR(reply))
1784                         err = PTR_ERR(reply);
1785         }
1786         if (err) {
1787                 dp_detach_port(vport);
1788                 goto exit_unlock;
1789         }
1790         genl_notify(reply, genl_info_net(info), info->snd_pid,
1791                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1792
1793
1794 exit_unlock:
1795         rtnl_unlock();
1796 exit:
1797         return err;
1798 }
1799
1800 static int odp_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1801 {
1802         struct nlattr **a = info->attrs;
1803         struct sk_buff *reply;
1804         struct vport *vport;
1805         int err;
1806
1807         err = odp_vport_cmd_validate(a);
1808         if (err)
1809                 goto exit;
1810
1811         rtnl_lock();
1812         vport = lookup_vport(info->userhdr, a);
1813         err = PTR_ERR(vport);
1814         if (IS_ERR(vport))
1815                 goto exit_unlock;
1816
1817         err = 0;
1818         if (a[ODP_VPORT_ATTR_OPTIONS])
1819                 err = vport_set_options(vport, a[ODP_VPORT_ATTR_OPTIONS]);
1820         if (!err)
1821                 err = change_vport(vport, a);
1822
1823         reply = odp_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1824                                          ODP_VPORT_CMD_NEW);
1825         if (IS_ERR(reply)) {
1826                 err = PTR_ERR(reply);
1827                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1828                                 dp_vport_multicast_group.id, err);
1829                 return 0;
1830         }
1831
1832         genl_notify(reply, genl_info_net(info), info->snd_pid,
1833                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1834
1835 exit_unlock:
1836         rtnl_unlock();
1837 exit:
1838         return err;
1839 }
1840
1841 static int odp_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1842 {
1843         struct nlattr **a = info->attrs;
1844         struct sk_buff *reply;
1845         struct vport *vport;
1846         int err;
1847
1848         err = odp_vport_cmd_validate(a);
1849         if (err)
1850                 goto exit;
1851
1852         rtnl_lock();
1853         vport = lookup_vport(info->userhdr, a);
1854         err = PTR_ERR(vport);
1855         if (IS_ERR(vport))
1856                 goto exit_unlock;
1857
1858         if (vport->port_no == ODPP_LOCAL) {
1859                 err = -EINVAL;
1860                 goto exit_unlock;
1861         }
1862
1863         reply = odp_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1864                                          ODP_VPORT_CMD_DEL);
1865         err = PTR_ERR(reply);
1866         if (IS_ERR(reply))
1867                 goto exit_unlock;
1868
1869         err = dp_detach_port(vport);
1870
1871         genl_notify(reply, genl_info_net(info), info->snd_pid,
1872                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1873
1874 exit_unlock:
1875         rtnl_unlock();
1876 exit:
1877         return err;
1878 }
1879
1880 static int odp_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1881 {
1882         struct nlattr **a = info->attrs;
1883         struct odp_header *odp_header = info->userhdr;
1884         struct sk_buff *reply;
1885         struct vport *vport;
1886         int err;
1887
1888         err = odp_vport_cmd_validate(a);
1889         if (err)
1890                 goto exit;
1891
1892         rcu_read_lock();
1893         vport = lookup_vport(odp_header, a);
1894         err = PTR_ERR(vport);
1895         if (IS_ERR(vport))
1896                 goto exit_unlock;
1897
1898         reply = odp_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1899                                          ODP_VPORT_CMD_NEW);
1900         err = PTR_ERR(reply);
1901         if (IS_ERR(reply))
1902                 goto exit_unlock;
1903
1904         err = genlmsg_reply(reply, info);
1905
1906 exit_unlock:
1907         rcu_read_unlock();
1908 exit:
1909         return err;
1910 }
1911
1912 static int odp_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1913 {
1914         struct odp_header *odp_header = genlmsg_data(nlmsg_data(cb->nlh));
1915         struct datapath *dp;
1916         u32 port_no;
1917         int retval;
1918
1919         dp = get_dp(odp_header->dp_ifindex);
1920         if (!dp)
1921                 return -ENODEV;
1922
1923         rcu_read_lock();
1924         for (port_no = cb->args[0]; port_no < DP_MAX_PORTS; port_no++) {
1925                 struct vport *vport;
1926
1927                 vport = get_vport_protected(dp, port_no);
1928                 if (!vport)
1929                         continue;
1930
1931                 if (odp_vport_cmd_fill_info(vport, skb, NETLINK_CB(cb->skb).pid,
1932                                             cb->nlh->nlmsg_seq, NLM_F_MULTI,
1933                                             ODP_VPORT_CMD_NEW) < 0)
1934                         break;
1935         }
1936         rcu_read_unlock();
1937
1938         cb->args[0] = port_no;
1939         retval = skb->len;
1940
1941         return retval;
1942 }
1943
1944 static struct genl_ops dp_vport_genl_ops[] = {
1945         { .cmd = ODP_VPORT_CMD_NEW,
1946           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1947           .policy = vport_policy,
1948           .doit = odp_vport_cmd_new
1949         },
1950         { .cmd = ODP_VPORT_CMD_DEL,
1951           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1952           .policy = vport_policy,
1953           .doit = odp_vport_cmd_del
1954         },
1955         { .cmd = ODP_VPORT_CMD_GET,
1956           .flags = 0,               /* OK for unprivileged users. */
1957           .policy = vport_policy,
1958           .doit = odp_vport_cmd_get,
1959           .dumpit = odp_vport_cmd_dump
1960         },
1961         { .cmd = ODP_VPORT_CMD_SET,
1962           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1963           .policy = vport_policy,
1964           .doit = odp_vport_cmd_set,
1965         },
1966 };
1967
1968 struct genl_family_and_ops {
1969         struct genl_family *family;
1970         struct genl_ops *ops;
1971         int n_ops;
1972         struct genl_multicast_group *group;
1973 };
1974
1975 static const struct genl_family_and_ops dp_genl_families[] = {
1976         { &dp_datapath_genl_family,
1977           dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1978           &dp_datapath_multicast_group },
1979         { &dp_vport_genl_family,
1980           dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1981           &dp_vport_multicast_group },
1982         { &dp_flow_genl_family,
1983           dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1984           &dp_flow_multicast_group },
1985         { &dp_packet_genl_family,
1986           dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1987           NULL },
1988 };
1989
1990 static void dp_unregister_genl(int n_families)
1991 {
1992         int i;
1993
1994         for (i = 0; i < n_families; i++) {
1995                 genl_unregister_family(dp_genl_families[i].family);
1996         }
1997 }
1998
1999 static int dp_register_genl(void)
2000 {
2001         int n_registered;
2002         int err;
2003         int i;
2004
2005         n_registered = 0;
2006         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2007                 const struct genl_family_and_ops *f = &dp_genl_families[i];
2008
2009                 err = genl_register_family_with_ops(f->family, f->ops,
2010                                                     f->n_ops);
2011                 if (err)
2012                         goto error;
2013                 n_registered++;
2014
2015                 if (f->group) {
2016                         err = genl_register_mc_group(f->family, f->group);
2017                         if (err)
2018                                 goto error;
2019                 }
2020         }
2021
2022         err = packet_register_mc_groups();
2023         if (err)
2024                 goto error;
2025         return 0;
2026
2027 error:
2028         dp_unregister_genl(n_registered);
2029         return err;
2030 }
2031
2032 static int __init dp_init(void)
2033 {
2034         struct sk_buff *dummy_skb;
2035         int err;
2036
2037         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
2038
2039         printk("Open vSwitch %s, built "__DATE__" "__TIME__"\n", VERSION BUILDNR);
2040
2041         err = flow_init();
2042         if (err)
2043                 goto error;
2044
2045         err = vport_init();
2046         if (err)
2047                 goto error_flow_exit;
2048
2049         err = register_netdevice_notifier(&dp_device_notifier);
2050         if (err)
2051                 goto error_vport_exit;
2052
2053         err = dp_register_genl();
2054         if (err < 0)
2055                 goto error_unreg_notifier;
2056
2057         return 0;
2058
2059 error_unreg_notifier:
2060         unregister_netdevice_notifier(&dp_device_notifier);
2061 error_vport_exit:
2062         vport_exit();
2063 error_flow_exit:
2064         flow_exit();
2065 error:
2066         return err;
2067 }
2068
2069 static void dp_cleanup(void)
2070 {
2071         rcu_barrier();
2072         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2073         unregister_netdevice_notifier(&dp_device_notifier);
2074         vport_exit();
2075         flow_exit();
2076 }
2077
2078 module_init(dp_init);
2079 module_exit(dp_cleanup);
2080
2081 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2082 MODULE_LICENSE("GPL");