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