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