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