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         OVS_CB(packet)->vport = NULL;
 714         OVS_CB(packet)->flow = NULL;
 715         /* execute_actions() will reset tun_id to 0 anyhow. */
 716 #ifdef NEED_CSUM_NORMALIZE
 717         OVS_CB(packet)->ip_summed = OVS_CSUM_NONE;
 718 #endif
 719         vlan_copy_skb_tci(packet);
 720
 721         rcu_read_lock();
 722         dp = get_dp(odp_header->dp_ifindex);
 723         err = -ENODEV;
 724         if (dp)
 725                 err = execute_actions(dp, packet, &key,
 726                                       nla_data(a[ODP_PACKET_ATTR_ACTIONS]),
 727                                       nla_len(a[ODP_PACKET_ATTR_ACTIONS]));
 728         rcu_read_unlock();
 729
 730 exit:
 731         return err;
 732 }
 733
 734 static const struct nla_policy packet_policy[ODP_PACKET_ATTR_MAX + 1] = {
 735         [ODP_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
 736         [ODP_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 737 };
 738
 739 static struct genl_ops dp_packet_genl_ops[] = {
 740         { .cmd = ODP_PACKET_CMD_EXECUTE,
 741           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 742           .policy = packet_policy,
 743           .doit = odp_packet_cmd_execute
 744         }
 745 };
 746
 747 static void get_dp_stats(struct datapath *dp, struct odp_stats *stats)
 748 {
 749         int i;
 750
 751         stats->n_frags = stats->n_hit = stats->n_missed = stats->n_lost = 0;
 752         for_each_possible_cpu(i) {
 753                 const struct dp_stats_percpu *percpu_stats;
 754                 struct dp_stats_percpu local_stats;
 755                 unsigned seqcount;
 756
 757                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 758
 759                 do {
 760                         seqcount = read_seqcount_begin(&percpu_stats->seqlock);
 761                         local_stats = *percpu_stats;
 762                 } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount));
 763
 764                 stats->n_frags += local_stats.n_frags;
 765                 stats->n_hit += local_stats.n_hit;
 766                 stats->n_missed += local_stats.n_missed;
 767                 stats->n_lost += local_stats.n_lost;
 768         }
 769 }
 770
 771 /* MTU of the dp pseudo-device: ETH_DATA_LEN or the minimum of the ports.
 772  * Called with RTNL lock.
 773  */
 774 int dp_min_mtu(const struct datapath *dp)
 775 {
 776         struct vport *p;
 777         int mtu = 0;
 778
 779         ASSERT_RTNL();
 780
 781         list_for_each_entry (p, &dp->port_list, node) {
 782                 int dev_mtu;
 783
 784                 /* Skip any internal ports, since that's what we're trying to
 785                  * set. */
 786                 if (is_internal_vport(p))
 787                         continue;
 788
 789                 dev_mtu = vport_get_mtu(p);
 790                 if (!dev_mtu)
 791                         continue;
 792                 if (!mtu || dev_mtu < mtu)
 793                         mtu = dev_mtu;
 794         }
 795
 796         return mtu ? mtu : ETH_DATA_LEN;
 797 }
 798
 799 /* Sets the MTU of all datapath devices to the minimum of the ports
 800  * Called with RTNL lock.
 801  */
 802 void set_internal_devs_mtu(const struct datapath *dp)
 803 {
 804         struct vport *p;
 805         int mtu;
 806
 807         ASSERT_RTNL();
 808
 809         mtu = dp_min_mtu(dp);
 810
 811         list_for_each_entry (p, &dp->port_list, node) {
 812                 if (is_internal_vport(p))
 813                         vport_set_mtu(p, mtu);
 814         }
 815 }
 816
 817 static const struct nla_policy flow_policy[ODP_FLOW_ATTR_MAX + 1] = {
 818         [ODP_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
 819         [ODP_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
 820         [ODP_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
 821 };
 822
 823 static struct genl_family dp_flow_genl_family = {
 824         .id = GENL_ID_GENERATE,
 825         .hdrsize = sizeof(struct odp_header),
 826         .name = ODP_FLOW_FAMILY,
 827         .version = 1,
 828         .maxattr = ODP_FLOW_ATTR_MAX
 829 };
 830
 831 static struct genl_multicast_group dp_flow_multicast_group = {
 832         .name = ODP_FLOW_MCGROUP
 833 };
 834
 835 /* Called with genl_lock. */
 836 static int odp_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
 837                                   struct sk_buff *skb, u32 pid, u32 seq, u32 flags, u8 cmd)
 838 {
 839         const int skb_orig_len = skb->len;
 840         const struct sw_flow_actions *sf_acts;
 841         struct odp_flow_stats stats;
 842         struct odp_header *odp_header;
 843         struct nlattr *nla;
 844         unsigned long used;
 845         u8 tcp_flags;
 846         int err;
 847
 848         sf_acts = rcu_dereference_protected(flow->sf_acts,
 849                                             lockdep_genl_is_held());
 850
 851         odp_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags, cmd);
 852         if (!odp_header)
 853                 return -EMSGSIZE;
 854
 855         odp_header->dp_ifindex = dp->dp_ifindex;
 856
 857         nla = nla_nest_start(skb, ODP_FLOW_ATTR_KEY);
 858         if (!nla)
 859                 goto nla_put_failure;
 860         err = flow_to_nlattrs(&flow->key, skb);
 861         if (err)
 862                 goto error;
 863         nla_nest_end(skb, nla);
 864
 865         spin_lock_bh(&flow->lock);
 866         used = flow->used;
 867         stats.n_packets = flow->packet_count;
 868         stats.n_bytes = flow->byte_count;
 869         tcp_flags = flow->tcp_flags;
 870         spin_unlock_bh(&flow->lock);
 871
 872         if (used)
 873                 NLA_PUT_U64(skb, ODP_FLOW_ATTR_USED, flow_used_time(used));
 874
 875         if (stats.n_packets)
 876                 NLA_PUT(skb, ODP_FLOW_ATTR_STATS, sizeof(struct odp_flow_stats), &stats);
 877
 878         if (tcp_flags)
 879                 NLA_PUT_U8(skb, ODP_FLOW_ATTR_TCP_FLAGS, tcp_flags);
 880
 881         /* If ODP_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 882          * this is the first flow to be dumped into 'skb'.  This is unusual for
 883          * Netlink but individual action lists can be longer than
 884          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 885          * The userspace caller can always fetch the actions separately if it
 886          * really wants them.  (Most userspace callers in fact don't care.)
 887          *
 888          * This can only fail for dump operations because the skb is always
 889          * properly sized for single flows.
 890          */
 891         err = nla_put(skb, ODP_FLOW_ATTR_ACTIONS, sf_acts->actions_len,
 892                       sf_acts->actions);
 893         if (err < 0 && skb_orig_len)
 894                 goto error;
 895
 896         return genlmsg_end(skb, odp_header);
 897
 898 nla_put_failure:
 899         err = -EMSGSIZE;
 900 error:
 901         genlmsg_cancel(skb, odp_header);
 902         return err;
 903 }
 904
 905 static struct sk_buff *odp_flow_cmd_alloc_info(struct sw_flow *flow)
 906 {
 907         const struct sw_flow_actions *sf_acts;
 908         int len;
 909
 910         sf_acts = rcu_dereference_protected(flow->sf_acts,
 911                                             lockdep_genl_is_held());
 912
 913         len = nla_total_size(FLOW_BUFSIZE); /* ODP_FLOW_ATTR_KEY */
 914         len += nla_total_size(sf_acts->actions_len); /* ODP_FLOW_ATTR_ACTIONS */
 915         len += nla_total_size(sizeof(struct odp_flow_stats)); /* ODP_FLOW_ATTR_STATS */
 916         len += nla_total_size(1); /* ODP_FLOW_ATTR_TCP_FLAGS */
 917         len += nla_total_size(8); /* ODP_FLOW_ATTR_USED */
 918         return genlmsg_new(NLMSG_ALIGN(sizeof(struct odp_header)) + len, GFP_KERNEL);
 919 }
 920
 921 static struct sk_buff *odp_flow_cmd_build_info(struct sw_flow *flow, struct datapath *dp,
 922                                                u32 pid, u32 seq, u8 cmd)
 923 {
 924         struct sk_buff *skb;
 925         int retval;
 926
 927         skb = odp_flow_cmd_alloc_info(flow);
 928         if (!skb)
 929                 return ERR_PTR(-ENOMEM);
 930
 931         retval = odp_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd);
 932         BUG_ON(retval < 0);
 933         return skb;
 934 }
 935
 936 static int odp_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
 937 {
 938         struct nlattr **a = info->attrs;
 939         struct odp_header *odp_header = info->userhdr;
 940         struct tbl_node *flow_node;
 941         struct sw_flow_key key;
 942         struct sw_flow *flow;
 943         struct sk_buff *reply;
 944         struct datapath *dp;
 945         struct tbl *table;
 946         u32 hash;
 947         int error;
 948
 949         /* Extract key. */
 950         error = -EINVAL;
 951         if (!a[ODP_FLOW_ATTR_KEY])
 952                 goto error;
 953         error = flow_from_nlattrs(&key, a[ODP_FLOW_ATTR_KEY]);
 954         if (error)
 955                 goto error;
 956
 957         /* Validate actions. */
 958         if (a[ODP_FLOW_ATTR_ACTIONS]) {
 959                 error = validate_actions(a[ODP_FLOW_ATTR_ACTIONS]);
 960                 if (error)
 961                         goto error;
 962         } else if (info->genlhdr->cmd == ODP_FLOW_CMD_NEW) {
 963                 error = -EINVAL;
 964                 goto error;
 965         }
 966
 967         dp = get_dp(odp_header->dp_ifindex);
 968         error = -ENODEV;
 969         if (!dp)
 970                 goto error;
 971
 972         hash = flow_hash(&key);
 973         table = get_table_protected(dp);
 974         flow_node = tbl_lookup(table, &key, hash, flow_cmp);
 975         if (!flow_node) {
 976                 struct sw_flow_actions *acts;
 977
 978                 /* Bail out if we're not allowed to create a new flow. */
 979                 error = -ENOENT;
 980                 if (info->genlhdr->cmd == ODP_FLOW_CMD_SET)
 981                         goto error;
 982
 983                 /* Expand table, if necessary, to make room. */
 984                 if (tbl_count(table) >= tbl_n_buckets(table)) {
 985                         error = expand_table(dp);
 986                         if (error)
 987                                 goto error;
 988                         table = get_table_protected(dp);
 989                 }
 990
 991                 /* Allocate flow. */
 992                 flow = flow_alloc();
 993                 if (IS_ERR(flow)) {
 994                         error = PTR_ERR(flow);
 995                         goto error;
 996                 }
 997                 flow->key = key;
 998                 clear_stats(flow);
 999
1000                 /* Obtain actions. */
1001                 acts = flow_actions_alloc(a[ODP_FLOW_ATTR_ACTIONS]);
1002                 error = PTR_ERR(acts);
1003                 if (IS_ERR(acts))
1004                         goto error_free_flow;
1005                 rcu_assign_pointer(flow->sf_acts, acts);
1006
1007                 /* Put flow in bucket. */
1008                 error = tbl_insert(table, &flow->tbl_node, hash);
1009                 if (error)
1010                         goto error_free_flow;
1011
1012                 reply = odp_flow_cmd_build_info(flow, dp, info->snd_pid,
1013                                                 info->snd_seq, ODP_FLOW_CMD_NEW);
1014         } else {
1015                 /* We found a matching flow. */
1016                 struct sw_flow_actions *old_acts;
1017
1018                 /* Bail out if we're not allowed to modify an existing flow.
1019                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1020                  * because Generic Netlink treats the latter as a dump
1021                  * request.  We also accept NLM_F_EXCL in case that bug ever
1022                  * gets fixed.
1023                  */
1024                 error = -EEXIST;
1025                 if (info->genlhdr->cmd == ODP_FLOW_CMD_NEW &&
1026                     info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1027                         goto error;
1028
1029                 /* Update actions. */
1030                 flow = flow_cast(flow_node);
1031                 old_acts = rcu_dereference_protected(flow->sf_acts,
1032                                                      lockdep_genl_is_held());
1033                 if (a[ODP_FLOW_ATTR_ACTIONS] &&
1034                     (old_acts->actions_len != nla_len(a[ODP_FLOW_ATTR_ACTIONS]) ||
1035                      memcmp(old_acts->actions, nla_data(a[ODP_FLOW_ATTR_ACTIONS]),
1036                             old_acts->actions_len))) {
1037                         struct sw_flow_actions *new_acts;
1038
1039                         new_acts = flow_actions_alloc(a[ODP_FLOW_ATTR_ACTIONS]);
1040                         error = PTR_ERR(new_acts);
1041                         if (IS_ERR(new_acts))
1042                                 goto error;
1043
1044                         rcu_assign_pointer(flow->sf_acts, new_acts);
1045                         flow_deferred_free_acts(old_acts);
1046                 }
1047
1048                 reply = odp_flow_cmd_build_info(flow, dp, info->snd_pid,
1049                                                 info->snd_seq, ODP_FLOW_CMD_NEW);
1050
1051                 /* Clear stats. */
1052                 if (a[ODP_FLOW_ATTR_CLEAR]) {
1053                         spin_lock_bh(&flow->lock);
1054                         clear_stats(flow);
1055                         spin_unlock_bh(&flow->lock);
1056                 }
1057         }
1058
1059         if (!IS_ERR(reply))
1060                 genl_notify(reply, genl_info_net(info), info->snd_pid,
1061                             dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1062         else
1063                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1064                                 dp_flow_multicast_group.id, PTR_ERR(reply));
1065         return 0;
1066
1067 error_free_flow:
1068         flow_put(flow);
1069 error:
1070         return error;
1071 }
1072
1073 static int odp_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1074 {
1075         struct nlattr **a = info->attrs;
1076         struct odp_header *odp_header = info->userhdr;
1077         struct sw_flow_key key;
1078         struct tbl_node *flow_node;
1079         struct sk_buff *reply;
1080         struct sw_flow *flow;
1081         struct datapath *dp;
1082         struct tbl *table;
1083         int err;
1084
1085         if (!a[ODP_FLOW_ATTR_KEY])
1086                 return -EINVAL;
1087         err = flow_from_nlattrs(&key, a[ODP_FLOW_ATTR_KEY]);
1088         if (err)
1089                 return err;
1090
1091         dp = get_dp(odp_header->dp_ifindex);
1092         if (!dp)
1093                 return -ENODEV;
1094
1095         table = get_table_protected(dp);
1096         flow_node = tbl_lookup(table, &key, flow_hash(&key), flow_cmp);
1097         if (!flow_node)
1098                 return -ENOENT;
1099
1100         flow = flow_cast(flow_node);
1101         reply = odp_flow_cmd_build_info(flow, dp, info->snd_pid, info->snd_seq, ODP_FLOW_CMD_NEW);
1102         if (IS_ERR(reply))
1103                 return PTR_ERR(reply);
1104
1105         return genlmsg_reply(reply, info);
1106 }
1107
1108 static int odp_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1109 {
1110         struct nlattr **a = info->attrs;
1111         struct odp_header *odp_header = info->userhdr;
1112         struct sw_flow_key key;
1113         struct tbl_node *flow_node;
1114         struct sk_buff *reply;
1115         struct sw_flow *flow;
1116         struct datapath *dp;
1117         struct tbl *table;
1118         int err;
1119
1120         if (!a[ODP_FLOW_ATTR_KEY])
1121                 return flush_flows(odp_header->dp_ifindex);
1122         err = flow_from_nlattrs(&key, a[ODP_FLOW_ATTR_KEY]);
1123         if (err)
1124                 return err;
1125
1126         dp = get_dp(odp_header->dp_ifindex);
1127         if (!dp)
1128                 return -ENODEV;
1129
1130         table = get_table_protected(dp);
1131         flow_node = tbl_lookup(table, &key, flow_hash(&key), flow_cmp);
1132         if (!flow_node)
1133                 return -ENOENT;
1134         flow = flow_cast(flow_node);
1135
1136         reply = odp_flow_cmd_alloc_info(flow);
1137         if (!reply)
1138                 return -ENOMEM;
1139
1140         err = tbl_remove(table, flow_node);
1141         if (err) {
1142                 kfree_skb(reply);
1143                 return err;
1144         }
1145
1146         err = odp_flow_cmd_fill_info(flow, dp, reply, info->snd_pid,
1147                                      info->snd_seq, 0, ODP_FLOW_CMD_DEL);
1148         BUG_ON(err < 0);
1149
1150         flow_deferred_free(flow);
1151
1152         genl_notify(reply, genl_info_net(info), info->snd_pid,
1153                     dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1154         return 0;
1155 }
1156
1157 static int odp_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1158 {
1159         struct odp_header *odp_header = genlmsg_data(nlmsg_data(cb->nlh));
1160         struct datapath *dp;
1161
1162         dp = get_dp(odp_header->dp_ifindex);
1163         if (!dp)
1164                 return -ENODEV;
1165
1166         for (;;) {
1167                 struct tbl_node *flow_node;
1168                 struct sw_flow *flow;
1169                 u32 bucket, obj;
1170
1171                 bucket = cb->args[0];
1172                 obj = cb->args[1];
1173                 flow_node = tbl_next(get_table_protected(dp), &bucket, &obj);
1174                 if (!flow_node)
1175                         break;
1176
1177                 flow = flow_cast(flow_node);
1178                 if (odp_flow_cmd_fill_info(flow, dp, skb, NETLINK_CB(cb->skb).pid,
1179                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1180                                            ODP_FLOW_CMD_NEW) < 0)
1181                         break;
1182
1183                 cb->args[0] = bucket;
1184                 cb->args[1] = obj;
1185         }
1186         return skb->len;
1187 }
1188
1189 static struct genl_ops dp_flow_genl_ops[] = {
1190         { .cmd = ODP_FLOW_CMD_NEW,
1191           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1192           .policy = flow_policy,
1193           .doit = odp_flow_cmd_new_or_set
1194         },
1195         { .cmd = ODP_FLOW_CMD_DEL,
1196           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1197           .policy = flow_policy,
1198           .doit = odp_flow_cmd_del
1199         },
1200         { .cmd = ODP_FLOW_CMD_GET,
1201           .flags = 0,               /* OK for unprivileged users. */
1202           .policy = flow_policy,
1203           .doit = odp_flow_cmd_get,
1204           .dumpit = odp_flow_cmd_dump
1205         },
1206         { .cmd = ODP_FLOW_CMD_SET,
1207           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1208           .policy = flow_policy,
1209           .doit = odp_flow_cmd_new_or_set,
1210         },
1211 };
1212
1213 static const struct nla_policy datapath_policy[ODP_DP_ATTR_MAX + 1] = {
1214 #ifdef HAVE_NLA_NUL_STRING
1215         [ODP_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1216 #endif
1217         [ODP_DP_ATTR_IPV4_FRAGS] = { .type = NLA_U32 },
1218         [ODP_DP_ATTR_SAMPLING] = { .type = NLA_U32 },
1219 };
1220
1221 static struct genl_family dp_datapath_genl_family = {
1222         .id = GENL_ID_GENERATE,
1223         .hdrsize = sizeof(struct odp_header),
1224         .name = ODP_DATAPATH_FAMILY,
1225         .version = 1,
1226         .maxattr = ODP_DP_ATTR_MAX
1227 };
1228
1229 static struct genl_multicast_group dp_datapath_multicast_group = {
1230         .name = ODP_DATAPATH_MCGROUP
1231 };
1232
1233 static int odp_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1234                                 u32 pid, u32 seq, u32 flags, u8 cmd)
1235 {
1236         struct odp_header *odp_header;
1237         struct nlattr *nla;
1238         int err;
1239
1240         odp_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family,
1241                                    flags, cmd);
1242         if (!odp_header)
1243                 goto error;
1244
1245         odp_header->dp_ifindex = dp->dp_ifindex;
1246
1247         rcu_read_lock();
1248         err = nla_put_string(skb, ODP_DP_ATTR_NAME, dp_name(dp));
1249         rcu_read_unlock();
1250         if (err)
1251                 goto nla_put_failure;
1252
1253         nla = nla_reserve(skb, ODP_DP_ATTR_STATS, sizeof(struct odp_stats));
1254         if (!nla)
1255                 goto nla_put_failure;
1256         get_dp_stats(dp, nla_data(nla));
1257
1258         NLA_PUT_U32(skb, ODP_DP_ATTR_IPV4_FRAGS,
1259                     dp->drop_frags ? ODP_DP_FRAG_DROP : ODP_DP_FRAG_ZERO);
1260
1261         if (dp->sflow_probability)
1262                 NLA_PUT_U32(skb, ODP_DP_ATTR_SAMPLING, dp->sflow_probability);
1263
1264         nla = nla_nest_start(skb, ODP_DP_ATTR_MCGROUPS);
1265         if (!nla)
1266                 goto nla_put_failure;
1267         NLA_PUT_U32(skb, ODP_PACKET_CMD_MISS, packet_mc_group(dp, ODP_PACKET_CMD_MISS));
1268         NLA_PUT_U32(skb, ODP_PACKET_CMD_ACTION, packet_mc_group(dp, ODP_PACKET_CMD_ACTION));
1269         NLA_PUT_U32(skb, ODP_PACKET_CMD_SAMPLE, packet_mc_group(dp, ODP_PACKET_CMD_SAMPLE));
1270         nla_nest_end(skb, nla);
1271
1272         return genlmsg_end(skb, odp_header);
1273
1274 nla_put_failure:
1275         genlmsg_cancel(skb, odp_header);
1276 error:
1277         return -EMSGSIZE;
1278 }
1279
1280 static struct sk_buff *odp_dp_cmd_build_info(struct datapath *dp, u32 pid,
1281                                              u32 seq, u8 cmd)
1282 {
1283         struct sk_buff *skb;
1284         int retval;
1285
1286         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1287         if (!skb)
1288                 return ERR_PTR(-ENOMEM);
1289
1290         retval = odp_dp_cmd_fill_info(dp, skb, pid, seq, 0, cmd);
1291         if (retval < 0) {
1292                 kfree_skb(skb);
1293                 return ERR_PTR(retval);
1294         }
1295         return skb;
1296 }
1297
1298 static int odp_dp_cmd_validate(struct nlattr *a[ODP_DP_ATTR_MAX + 1])
1299 {
1300         if (a[ODP_DP_ATTR_IPV4_FRAGS]) {
1301                 u32 frags = nla_get_u32(a[ODP_DP_ATTR_IPV4_FRAGS]);
1302
1303                 if (frags != ODP_DP_FRAG_ZERO && frags != ODP_DP_FRAG_DROP)
1304                         return -EINVAL;
1305         }
1306
1307         return VERIFY_NUL_STRING(a[ODP_DP_ATTR_NAME], IFNAMSIZ - 1);
1308 }
1309
1310 /* Called with genl_mutex and optionally with RTNL lock also. */
1311 static struct datapath *lookup_datapath(struct odp_header *odp_header, struct nlattr *a[ODP_DP_ATTR_MAX + 1])
1312 {
1313         struct datapath *dp;
1314
1315         if (!a[ODP_DP_ATTR_NAME])
1316                 dp = get_dp(odp_header->dp_ifindex);
1317         else {
1318                 struct vport *vport;
1319
1320                 rcu_read_lock();
1321                 vport = vport_locate(nla_data(a[ODP_DP_ATTR_NAME]));
1322                 dp = vport && vport->port_no == ODPP_LOCAL ? vport->dp : NULL;
1323                 rcu_read_unlock();
1324         }
1325         return dp ? dp : ERR_PTR(-ENODEV);
1326 }
1327
1328 /* Called with genl_mutex. */
1329 static void change_datapath(struct datapath *dp, struct nlattr *a[ODP_DP_ATTR_MAX + 1])
1330 {
1331         if (a[ODP_DP_ATTR_IPV4_FRAGS])
1332                 dp->drop_frags = nla_get_u32(a[ODP_DP_ATTR_IPV4_FRAGS]) == ODP_DP_FRAG_DROP;
1333         if (a[ODP_DP_ATTR_SAMPLING])
1334                 dp->sflow_probability = nla_get_u32(a[ODP_DP_ATTR_SAMPLING]);
1335 }
1336
1337 static int odp_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1338 {
1339         struct nlattr **a = info->attrs;
1340         struct vport_parms parms;
1341         struct sk_buff *reply;
1342         struct datapath *dp;
1343         struct vport *vport;
1344         int err;
1345
1346         err = -EINVAL;
1347         if (!a[ODP_DP_ATTR_NAME])
1348                 goto err;
1349
1350         err = odp_dp_cmd_validate(a);
1351         if (err)
1352                 goto err;
1353
1354         rtnl_lock();
1355         err = -ENODEV;
1356         if (!try_module_get(THIS_MODULE))
1357                 goto err_unlock_rtnl;
1358
1359         err = -ENOMEM;
1360         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1361         if (dp == NULL)
1362                 goto err_put_module;
1363         INIT_LIST_HEAD(&dp->port_list);
1364
1365         /* Initialize kobject for bridge.  This will be added as
1366          * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
1367         dp->ifobj.kset = NULL;
1368         kobject_init(&dp->ifobj, &dp_ktype);
1369
1370         /* Allocate table. */
1371         err = -ENOMEM;
1372         rcu_assign_pointer(dp->table, tbl_create(TBL_MIN_BUCKETS));
1373         if (!dp->table)
1374                 goto err_free_dp;
1375
1376         /* Set up our datapath device. */
1377         parms.name = nla_data(a[ODP_DP_ATTR_NAME]);
1378         parms.type = ODP_VPORT_TYPE_INTERNAL;
1379         parms.options = NULL;
1380         parms.dp = dp;
1381         parms.port_no = ODPP_LOCAL;
1382         vport = new_vport(&parms);
1383         if (IS_ERR(vport)) {
1384                 err = PTR_ERR(vport);
1385                 if (err == -EBUSY)
1386                         err = -EEXIST;
1387
1388                 goto err_destroy_table;
1389         }
1390         dp->dp_ifindex = vport_get_ifindex(vport);
1391
1392         dp->drop_frags = 0;
1393         dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1394         if (!dp->stats_percpu) {
1395                 err = -ENOMEM;
1396                 goto err_destroy_local_port;
1397         }
1398
1399         change_datapath(dp, a);
1400
1401         reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_NEW);
1402         err = PTR_ERR(reply);
1403         if (IS_ERR(reply))
1404                 goto err_destroy_local_port;
1405
1406         list_add_tail(&dp->list_node, &dps);
1407         dp_sysfs_add_dp(dp);
1408
1409         rtnl_unlock();
1410
1411         genl_notify(reply, genl_info_net(info), info->snd_pid,
1412                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1413         return 0;
1414
1415 err_destroy_local_port:
1416         dp_detach_port(get_vport_protected(dp, ODPP_LOCAL));
1417 err_destroy_table:
1418         tbl_destroy(get_table_protected(dp), NULL);
1419 err_free_dp:
1420         kfree(dp);
1421 err_put_module:
1422         module_put(THIS_MODULE);
1423 err_unlock_rtnl:
1424         rtnl_unlock();
1425 err:
1426         return err;
1427 }
1428
1429 static int odp_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1430 {
1431         struct vport *vport, *next_vport;
1432         struct sk_buff *reply;
1433         struct datapath *dp;
1434         int err;
1435
1436         err = odp_dp_cmd_validate(info->attrs);
1437         if (err)
1438                 goto exit;
1439
1440         rtnl_lock();
1441         dp = lookup_datapath(info->userhdr, info->attrs);
1442         err = PTR_ERR(dp);
1443         if (IS_ERR(dp))
1444                 goto exit_unlock;
1445
1446         reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_DEL);
1447         err = PTR_ERR(reply);
1448         if (IS_ERR(reply))
1449                 goto exit_unlock;
1450
1451         list_for_each_entry_safe (vport, next_vport, &dp->port_list, node)
1452                 if (vport->port_no != ODPP_LOCAL)
1453                         dp_detach_port(vport);
1454
1455         dp_sysfs_del_dp(dp);
1456         list_del(&dp->list_node);
1457         dp_detach_port(get_vport_protected(dp, ODPP_LOCAL));
1458
1459         call_rcu(&dp->rcu, destroy_dp_rcu);
1460         module_put(THIS_MODULE);
1461
1462         genl_notify(reply, genl_info_net(info), info->snd_pid,
1463                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1464         err = 0;
1465
1466 exit_unlock:
1467         rtnl_unlock();
1468 exit:
1469         return err;
1470 }
1471
1472 static int odp_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1473 {
1474         struct sk_buff *reply;
1475         struct datapath *dp;
1476         int err;
1477
1478         err = odp_dp_cmd_validate(info->attrs);
1479         if (err)
1480                 return err;
1481
1482         dp = lookup_datapath(info->userhdr, info->attrs);
1483         if (IS_ERR(dp))
1484                 return PTR_ERR(dp);
1485
1486         change_datapath(dp, info->attrs);
1487
1488         reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_NEW);
1489         if (IS_ERR(reply)) {
1490                 err = PTR_ERR(reply);
1491                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1492                                 dp_datapath_multicast_group.id, err);
1493                 return 0;
1494         }
1495
1496         genl_notify(reply, genl_info_net(info), info->snd_pid,
1497                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1498         return 0;
1499 }
1500
1501 static int odp_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1502 {
1503         struct sk_buff *reply;
1504         struct datapath *dp;
1505         int err;
1506
1507         err = odp_dp_cmd_validate(info->attrs);
1508         if (err)
1509                 return err;
1510
1511         dp = lookup_datapath(info->userhdr, info->attrs);
1512         if (IS_ERR(dp))
1513                 return PTR_ERR(dp);
1514
1515         reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_NEW);
1516         if (IS_ERR(reply))
1517                 return PTR_ERR(reply);
1518
1519         return genlmsg_reply(reply, info);
1520 }
1521
1522 static int odp_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1523 {
1524         struct datapath *dp;
1525         int skip = cb->args[0];
1526         int i = 0;
1527
1528         list_for_each_entry (dp, &dps, list_node) {
1529                 if (i < skip)
1530                         continue;
1531                 if (odp_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).pid,
1532                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1533                                          ODP_DP_CMD_NEW) < 0)
1534                         break;
1535                 i++;
1536         }
1537
1538         cb->args[0] = i;
1539
1540         return skb->len;
1541 }
1542
1543 static struct genl_ops dp_datapath_genl_ops[] = {
1544         { .cmd = ODP_DP_CMD_NEW,
1545           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1546           .policy = datapath_policy,
1547           .doit = odp_dp_cmd_new
1548         },
1549         { .cmd = ODP_DP_CMD_DEL,
1550           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1551           .policy = datapath_policy,
1552           .doit = odp_dp_cmd_del
1553         },
1554         { .cmd = ODP_DP_CMD_GET,
1555           .flags = 0,               /* OK for unprivileged users. */
1556           .policy = datapath_policy,
1557           .doit = odp_dp_cmd_get,
1558           .dumpit = odp_dp_cmd_dump
1559         },
1560         { .cmd = ODP_DP_CMD_SET,
1561           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1562           .policy = datapath_policy,
1563           .doit = odp_dp_cmd_set,
1564         },
1565 };
1566
1567 static const struct nla_policy vport_policy[ODP_VPORT_ATTR_MAX + 1] = {
1568 #ifdef HAVE_NLA_NUL_STRING
1569         [ODP_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1570         [ODP_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1571         [ODP_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1572         [ODP_VPORT_ATTR_STATS] = { .len = sizeof(struct rtnl_link_stats64) },
1573         [ODP_VPORT_ATTR_ADDRESS] = { .len = ETH_ALEN },
1574 #else
1575         [ODP_VPORT_ATTR_STATS] = { .minlen = sizeof(struct rtnl_link_stats64) },
1576         [ODP_VPORT_ATTR_ADDRESS] = { .minlen = ETH_ALEN },
1577 #endif
1578         [ODP_VPORT_ATTR_MTU] = { .type = NLA_U32 },
1579         [ODP_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1580 };
1581
1582 static struct genl_family dp_vport_genl_family = {
1583         .id = GENL_ID_GENERATE,
1584         .hdrsize = sizeof(struct odp_header),
1585         .name = ODP_VPORT_FAMILY,
1586         .version = 1,
1587         .maxattr = ODP_VPORT_ATTR_MAX
1588 };
1589
1590 static struct genl_multicast_group dp_vport_multicast_group = {
1591         .name = ODP_VPORT_MCGROUP
1592 };
1593
1594 /* Called with RTNL lock or RCU read lock. */
1595 static int odp_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1596                                    u32 pid, u32 seq, u32 flags, u8 cmd)
1597 {
1598         struct odp_header *odp_header;
1599         struct nlattr *nla;
1600         int ifindex, iflink;
1601         int mtu;
1602         int err;
1603
1604         odp_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family,
1605                                  flags, cmd);
1606         if (!odp_header)
1607                 return -EMSGSIZE;
1608
1609         odp_header->dp_ifindex = vport->dp->dp_ifindex;
1610
1611         NLA_PUT_U32(skb, ODP_VPORT_ATTR_PORT_NO, vport->port_no);
1612         NLA_PUT_U32(skb, ODP_VPORT_ATTR_TYPE, vport_get_type(vport));
1613         NLA_PUT_STRING(skb, ODP_VPORT_ATTR_NAME, vport_get_name(vport));
1614
1615         nla = nla_reserve(skb, ODP_VPORT_ATTR_STATS, sizeof(struct rtnl_link_stats64));
1616         if (!nla)
1617                 goto nla_put_failure;
1618         if (vport_get_stats(vport, nla_data(nla)))
1619                 __skb_trim(skb, skb->len - nla->nla_len);
1620
1621         NLA_PUT(skb, ODP_VPORT_ATTR_ADDRESS, ETH_ALEN, vport_get_addr(vport));
1622
1623         mtu = vport_get_mtu(vport);
1624         if (mtu)
1625                 NLA_PUT_U32(skb, ODP_VPORT_ATTR_MTU, mtu);
1626
1627         err = vport_get_options(vport, skb);
1628         if (err == -EMSGSIZE)
1629                 goto error;
1630
1631         ifindex = vport_get_ifindex(vport);
1632         if (ifindex > 0)
1633                 NLA_PUT_U32(skb, ODP_VPORT_ATTR_IFINDEX, ifindex);
1634
1635         iflink = vport_get_iflink(vport);
1636         if (iflink > 0)
1637                 NLA_PUT_U32(skb, ODP_VPORT_ATTR_IFLINK, iflink);
1638
1639         return genlmsg_end(skb, odp_header);
1640
1641 nla_put_failure:
1642         err = -EMSGSIZE;
1643 error:
1644         genlmsg_cancel(skb, odp_header);
1645         return err;
1646 }
1647
1648 /* Called with RTNL lock or RCU read lock. */
1649 static struct sk_buff *odp_vport_cmd_build_info(struct vport *vport, u32 pid,
1650                                                 u32 seq, u8 cmd)
1651 {
1652         struct sk_buff *skb;
1653         int retval;
1654
1655         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1656         if (!skb)
1657                 return ERR_PTR(-ENOMEM);
1658
1659         retval = odp_vport_cmd_fill_info(vport, skb, pid, seq, 0, cmd);
1660         if (retval < 0) {
1661                 kfree_skb(skb);
1662                 return ERR_PTR(retval);
1663         }
1664         return skb;
1665 }
1666
1667 static int odp_vport_cmd_validate(struct nlattr *a[ODP_VPORT_ATTR_MAX + 1])
1668 {
1669         return VERIFY_NUL_STRING(a[ODP_VPORT_ATTR_NAME], IFNAMSIZ - 1);
1670 }
1671
1672 /* Called with RTNL lock or RCU read lock. */
1673 static struct vport *lookup_vport(struct odp_header *odp_header,
1674                                   struct nlattr *a[ODP_VPORT_ATTR_MAX + 1])
1675 {
1676         struct datapath *dp;
1677         struct vport *vport;
1678
1679         if (a[ODP_VPORT_ATTR_NAME]) {
1680                 vport = vport_locate(nla_data(a[ODP_VPORT_ATTR_NAME]));
1681                 if (!vport)
1682                         return ERR_PTR(-ENODEV);
1683                 return vport;
1684         } else if (a[ODP_VPORT_ATTR_PORT_NO]) {
1685                 u32 port_no = nla_get_u32(a[ODP_VPORT_ATTR_PORT_NO]);
1686
1687                 if (port_no >= DP_MAX_PORTS)
1688                         return ERR_PTR(-EFBIG);
1689
1690                 dp = get_dp(odp_header->dp_ifindex);
1691                 if (!dp)
1692                         return ERR_PTR(-ENODEV);
1693
1694                 vport = get_vport_protected(dp, port_no);
1695                 if (!vport)
1696                         return ERR_PTR(-ENOENT);
1697                 return vport;
1698         } else
1699                 return ERR_PTR(-EINVAL);
1700 }
1701
1702 /* Called with RTNL lock. */
1703 static int change_vport(struct vport *vport, struct nlattr *a[ODP_VPORT_ATTR_MAX + 1])
1704 {
1705         int err = 0;
1706         if (a[ODP_VPORT_ATTR_STATS])
1707                 err = vport_set_stats(vport, nla_data(a[ODP_VPORT_ATTR_STATS]));
1708         if (!err && a[ODP_VPORT_ATTR_ADDRESS])
1709                 err = vport_set_addr(vport, nla_data(a[ODP_VPORT_ATTR_ADDRESS]));
1710         if (!err && a[ODP_VPORT_ATTR_MTU])
1711                 err = vport_set_mtu(vport, nla_get_u32(a[ODP_VPORT_ATTR_MTU]));
1712         return err;
1713 }
1714
1715 static int odp_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1716 {
1717         struct nlattr **a = info->attrs;
1718         struct odp_header *odp_header = info->userhdr;
1719         struct vport_parms parms;
1720         struct sk_buff *reply;
1721         struct vport *vport;
1722         struct datapath *dp;
1723         u32 port_no;
1724         int err;
1725
1726         err = -EINVAL;
1727         if (!a[ODP_VPORT_ATTR_NAME] || !a[ODP_VPORT_ATTR_TYPE])
1728                 goto exit;
1729
1730         err = odp_vport_cmd_validate(a);
1731         if (err)
1732                 goto exit;
1733
1734         rtnl_lock();
1735         dp = get_dp(odp_header->dp_ifindex);
1736         err = -ENODEV;
1737         if (!dp)
1738                 goto exit_unlock;
1739
1740         if (a[ODP_VPORT_ATTR_PORT_NO]) {
1741                 port_no = nla_get_u32(a[ODP_VPORT_ATTR_PORT_NO]);
1742
1743                 err = -EFBIG;
1744                 if (port_no >= DP_MAX_PORTS)
1745                         goto exit_unlock;
1746
1747                 vport = get_vport_protected(dp, port_no);
1748                 err = -EBUSY;
1749                 if (vport)
1750                         goto exit_unlock;
1751         } else {
1752                 for (port_no = 1; ; port_no++) {
1753                         if (port_no >= DP_MAX_PORTS) {
1754                                 err = -EFBIG;
1755                                 goto exit_unlock;
1756                         }
1757                         vport = get_vport_protected(dp, port_no);
1758                         if (!vport)
1759                                 break;
1760                 }
1761         }
1762
1763         parms.name = nla_data(a[ODP_VPORT_ATTR_NAME]);
1764         parms.type = nla_get_u32(a[ODP_VPORT_ATTR_TYPE]);
1765         parms.options = a[ODP_VPORT_ATTR_OPTIONS];
1766         parms.dp = dp;
1767         parms.port_no = port_no;
1768
1769         vport = new_vport(&parms);
1770         err = PTR_ERR(vport);
1771         if (IS_ERR(vport))
1772                 goto exit_unlock;
1773
1774         set_internal_devs_mtu(dp);
1775         dp_sysfs_add_if(vport);
1776
1777         err = change_vport(vport, a);
1778         if (!err) {
1779                 reply = odp_vport_cmd_build_info(vport, info->snd_pid,
1780                                                  info->snd_seq, ODP_VPORT_CMD_NEW);
1781                 if (IS_ERR(reply))
1782                         err = PTR_ERR(reply);
1783         }
1784         if (err) {
1785                 dp_detach_port(vport);
1786                 goto exit_unlock;
1787         }
1788         genl_notify(reply, genl_info_net(info), info->snd_pid,
1789                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1790
1791
1792 exit_unlock:
1793         rtnl_unlock();
1794 exit:
1795         return err;
1796 }
1797
1798 static int odp_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1799 {
1800         struct nlattr **a = info->attrs;
1801         struct sk_buff *reply;
1802         struct vport *vport;
1803         int err;
1804
1805         err = odp_vport_cmd_validate(a);
1806         if (err)
1807                 goto exit;
1808
1809         rtnl_lock();
1810         vport = lookup_vport(info->userhdr, a);
1811         err = PTR_ERR(vport);
1812         if (IS_ERR(vport))
1813                 goto exit_unlock;
1814
1815         err = 0;
1816         if (a[ODP_VPORT_ATTR_OPTIONS])
1817                 err = vport_set_options(vport, a[ODP_VPORT_ATTR_OPTIONS]);
1818         if (!err)
1819                 err = change_vport(vport, a);
1820
1821         reply = odp_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1822                                          ODP_VPORT_CMD_NEW);
1823         if (IS_ERR(reply)) {
1824                 err = PTR_ERR(reply);
1825                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1826                                 dp_vport_multicast_group.id, err);
1827                 return 0;
1828         }
1829
1830         genl_notify(reply, genl_info_net(info), info->snd_pid,
1831                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1832
1833 exit_unlock:
1834         rtnl_unlock();
1835 exit:
1836         return err;
1837 }
1838
1839 static int odp_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1840 {
1841         struct nlattr **a = info->attrs;
1842         struct sk_buff *reply;
1843         struct vport *vport;
1844         int err;
1845
1846         err = odp_vport_cmd_validate(a);
1847         if (err)
1848                 goto exit;
1849
1850         rtnl_lock();
1851         vport = lookup_vport(info->userhdr, a);
1852         err = PTR_ERR(vport);
1853         if (IS_ERR(vport))
1854                 goto exit_unlock;
1855
1856         if (vport->port_no == ODPP_LOCAL) {
1857                 err = -EINVAL;
1858                 goto exit_unlock;
1859         }
1860
1861         reply = odp_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1862                                          ODP_VPORT_CMD_DEL);
1863         err = PTR_ERR(reply);
1864         if (IS_ERR(reply))
1865                 goto exit_unlock;
1866
1867         err = dp_detach_port(vport);
1868
1869         genl_notify(reply, genl_info_net(info), info->snd_pid,
1870                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1871
1872 exit_unlock:
1873         rtnl_unlock();
1874 exit:
1875         return err;
1876 }
1877
1878 static int odp_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1879 {
1880         struct nlattr **a = info->attrs;
1881         struct odp_header *odp_header = info->userhdr;
1882         struct sk_buff *reply;
1883         struct vport *vport;
1884         int err;
1885
1886         err = odp_vport_cmd_validate(a);
1887         if (err)
1888                 goto exit;
1889
1890         rcu_read_lock();
1891         vport = lookup_vport(odp_header, a);
1892         err = PTR_ERR(vport);
1893         if (IS_ERR(vport))
1894                 goto exit_unlock;
1895
1896         reply = odp_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1897                                          ODP_VPORT_CMD_NEW);
1898         err = PTR_ERR(reply);
1899         if (IS_ERR(reply))
1900                 goto exit_unlock;
1901
1902         err = genlmsg_reply(reply, info);
1903
1904 exit_unlock:
1905         rcu_read_unlock();
1906 exit:
1907         return err;
1908 }
1909
1910 static int odp_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1911 {
1912         struct odp_header *odp_header = genlmsg_data(nlmsg_data(cb->nlh));
1913         struct datapath *dp;
1914         u32 port_no;
1915         int retval;
1916
1917         dp = get_dp(odp_header->dp_ifindex);
1918         if (!dp)
1919                 return -ENODEV;
1920
1921         rcu_read_lock();
1922         for (port_no = cb->args[0]; port_no < DP_MAX_PORTS; port_no++) {
1923                 struct vport *vport;
1924
1925                 vport = get_vport_protected(dp, port_no);
1926                 if (!vport)
1927                         continue;
1928
1929                 if (odp_vport_cmd_fill_info(vport, skb, NETLINK_CB(cb->skb).pid,
1930                                             cb->nlh->nlmsg_seq, NLM_F_MULTI,
1931                                             ODP_VPORT_CMD_NEW) < 0)
1932                         break;
1933         }
1934         rcu_read_unlock();
1935
1936         cb->args[0] = port_no;
1937         retval = skb->len;
1938
1939         return retval;
1940 }
1941
1942 static struct genl_ops dp_vport_genl_ops[] = {
1943         { .cmd = ODP_VPORT_CMD_NEW,
1944           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1945           .policy = vport_policy,
1946           .doit = odp_vport_cmd_new
1947         },
1948         { .cmd = ODP_VPORT_CMD_DEL,
1949           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1950           .policy = vport_policy,
1951           .doit = odp_vport_cmd_del
1952         },
1953         { .cmd = ODP_VPORT_CMD_GET,
1954           .flags = 0,               /* OK for unprivileged users. */
1955           .policy = vport_policy,
1956           .doit = odp_vport_cmd_get,
1957           .dumpit = odp_vport_cmd_dump
1958         },
1959         { .cmd = ODP_VPORT_CMD_SET,
1960           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1961           .policy = vport_policy,
1962           .doit = odp_vport_cmd_set,
1963         },
1964 };
1965
1966 struct genl_family_and_ops {
1967         struct genl_family *family;
1968         struct genl_ops *ops;
1969         int n_ops;
1970         struct genl_multicast_group *group;
1971 };
1972
1973 static const struct genl_family_and_ops dp_genl_families[] = {
1974         { &dp_datapath_genl_family,
1975           dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1976           &dp_datapath_multicast_group },
1977         { &dp_vport_genl_family,
1978           dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1979           &dp_vport_multicast_group },
1980         { &dp_flow_genl_family,
1981           dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1982           &dp_flow_multicast_group },
1983         { &dp_packet_genl_family,
1984           dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1985           NULL },
1986 };
1987
1988 static void dp_unregister_genl(int n_families)
1989 {
1990         int i;
1991
1992         for (i = 0; i < n_families; i++) {
1993                 genl_unregister_family(dp_genl_families[i].family);
1994         }
1995 }
1996
1997 static int dp_register_genl(void)
1998 {
1999         int n_registered;
2000         int err;
2001         int i;
2002
2003         n_registered = 0;
2004         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2005                 const struct genl_family_and_ops *f = &dp_genl_families[i];
2006
2007                 err = genl_register_family_with_ops(f->family, f->ops,
2008                                                     f->n_ops);
2009                 if (err)
2010                         goto error;
2011                 n_registered++;
2012
2013                 if (f->group) {
2014                         err = genl_register_mc_group(f->family, f->group);
2015                         if (err)
2016                                 goto error;
2017                 }
2018         }
2019
2020         err = packet_register_mc_groups();
2021         if (err)
2022                 goto error;
2023         return 0;
2024
2025 error:
2026         dp_unregister_genl(n_registered);
2027         return err;
2028 }
2029
2030 static int __init dp_init(void)
2031 {
2032         struct sk_buff *dummy_skb;
2033         int err;
2034
2035         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
2036
2037         printk("Open vSwitch %s, built "__DATE__" "__TIME__"\n", VERSION BUILDNR);
2038
2039         err = flow_init();
2040         if (err)
2041                 goto error;
2042
2043         err = vport_init();
2044         if (err)
2045                 goto error_flow_exit;
2046
2047         err = register_netdevice_notifier(&dp_device_notifier);
2048         if (err)
2049                 goto error_vport_exit;
2050
2051         err = dp_register_genl();
2052         if (err < 0)
2053                 goto error_unreg_notifier;
2054
2055         return 0;
2056
2057 error_unreg_notifier:
2058         unregister_netdevice_notifier(&dp_device_notifier);
2059 error_vport_exit:
2060         vport_exit();
2061 error_flow_exit:
2062         flow_exit();
2063 error:
2064         return err;
2065 }
2066
2067 static void dp_cleanup(void)
2068 {
2069         rcu_barrier();
2070         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2071         unregister_netdevice_notifier(&dp_device_notifier);
2072         vport_exit();
2073         flow_exit();
2074 }
2075
2076 module_init(dp_init);
2077 module_exit(dp_cleanup);
2078
2079 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2080 MODULE_LICENSE("GPL");