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 "table.h"
  53 #include "vlan.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         int error;
 270
 271         OVS_CB(skb)->vport = p;
 272
 273         if (!OVS_CB(skb)->flow) {
 274                 struct sw_flow_key key;
 275                 struct tbl_node *flow_node;
 276                 int key_len;
 277                 bool is_frag;
 278
 279                 /* Extract flow from 'skb' into 'key'. */
 280                 error = flow_extract(skb, p->port_no, &key, &key_len, &is_frag);
 281                 if (unlikely(error)) {
 282                         kfree_skb(skb);
 283                         return;
 284                 }
 285
 286                 if (is_frag && dp->drop_frags) {
 287                         kfree_skb(skb);
 288                         stats_counter_off = offsetof(struct dp_stats_percpu, n_frags);
 289                         goto out;
 290                 }
 291
 292                 /* Look up flow. */
 293                 flow_node = tbl_lookup(rcu_dereference(dp->table), &key, key_len,
 294                                        flow_hash(&key, key_len), flow_cmp);
 295                 if (unlikely(!flow_node)) {
 296                         struct dp_upcall_info upcall;
 297
 298                         upcall.cmd = ODP_PACKET_CMD_MISS;
 299                         upcall.key = &key;
 300                         upcall.userdata = 0;
 301                         upcall.sample_pool = 0;
 302                         upcall.actions = NULL;
 303                         upcall.actions_len = 0;
 304                         dp_upcall(dp, skb, &upcall);
 305                         stats_counter_off = offsetof(struct dp_stats_percpu, n_missed);
 306                         goto out;
 307                 }
 308
 309                 OVS_CB(skb)->flow = flow_cast(flow_node);
 310         }
 311
 312         stats_counter_off = offsetof(struct dp_stats_percpu, n_hit);
 313         flow_used(OVS_CB(skb)->flow, skb);
 314         execute_actions(dp, skb);
 315
 316 out:
 317         /* Update datapath statistics. */
 318         local_bh_disable();
 319         stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
 320
 321         write_seqcount_begin(&stats->seqlock);
 322         (*(u64 *)((u8 *)stats + stats_counter_off))++;
 323         write_seqcount_end(&stats->seqlock);
 324
 325         local_bh_enable();
 326 }
 327
 328 static void copy_and_csum_skb(struct sk_buff *skb, void *to)
 329 {
 330         u16 csum_start, csum_offset;
 331         __wsum csum;
 332
 333         get_skb_csum_pointers(skb, &csum_start, &csum_offset);
 334         csum_start -= skb_headroom(skb);
 335
 336         skb_copy_bits(skb, 0, to, csum_start);
 337
 338         csum = skb_copy_and_csum_bits(skb, csum_start, to + csum_start,
 339                                       skb->len - csum_start, 0);
 340         *(__sum16 *)(to + csum_start + csum_offset) = csum_fold(csum);
 341 }
 342
 343 static struct genl_family dp_packet_genl_family = {
 344         .id = GENL_ID_GENERATE,
 345         .hdrsize = sizeof(struct odp_header),
 346         .name = ODP_PACKET_FAMILY,
 347         .version = 1,
 348         .maxattr = ODP_PACKET_ATTR_MAX
 349 };
 350
 351 /* Generic Netlink multicast groups for upcalls.
 352  *
 353  * We really want three unique multicast groups per datapath, but we can't even
 354  * get one, because genl_register_mc_group() takes genl_lock, which is also
 355  * held during Generic Netlink message processing, so trying to acquire
 356  * multicast groups during ODP_DP_NEW processing deadlocks.  Instead, we
 357  * preallocate a few groups and use them round-robin for datapaths.  Collision
 358  * isn't fatal--multicast listeners should check that the family is the one
 359  * that they want and discard others--but it wastes time and memory to receive
 360  * unwanted messages.
 361  */
 362 #define PACKET_N_MC_GROUPS 16
 363 static struct genl_multicast_group packet_mc_groups[PACKET_N_MC_GROUPS];
 364
 365 static u32 packet_mc_group(struct datapath *dp, u8 cmd)
 366 {
 367         u32 idx;
 368         BUILD_BUG_ON_NOT_POWER_OF_2(PACKET_N_MC_GROUPS);
 369
 370         idx = jhash_2words(dp->dp_ifindex, cmd, 0) & (PACKET_N_MC_GROUPS - 1);
 371         return packet_mc_groups[idx].id;
 372 }
 373
 374 static int packet_register_mc_groups(void)
 375 {
 376         int i;
 377
 378         for (i = 0; i < PACKET_N_MC_GROUPS; i++) {
 379                 struct genl_multicast_group *group = &packet_mc_groups[i];
 380                 int error;
 381
 382                 sprintf(group->name, "packet%d", i);
 383                 error = genl_register_mc_group(&dp_packet_genl_family, group);
 384                 if (error)
 385                         return error;
 386         }
 387         return 0;
 388 }
 389
 390 int dp_upcall(struct datapath *dp, struct sk_buff *skb, const struct dp_upcall_info *upcall_info)
 391 {
 392         struct dp_stats_percpu *stats;
 393         int err;
 394
 395         WARN_ON_ONCE(skb_shared(skb));
 396
 397         forward_ip_summed(skb);
 398
 399         err = vswitch_skb_checksum_setup(skb);
 400         if (err)
 401                 goto err_kfree_skb;
 402
 403         /* Break apart GSO packets into their component pieces.  Otherwise
 404          * userspace may try to stuff a 64kB packet into a 1500-byte MTU. */
 405         if (skb_is_gso(skb)) {
 406                 struct sk_buff *nskb = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
 407
 408                 kfree_skb(skb);
 409                 skb = nskb;
 410                 if (IS_ERR(skb)) {
 411                         err = PTR_ERR(skb);
 412                         goto err;
 413                 }
 414         }
 415
 416         err = queue_control_packets(dp, skb, upcall_info);
 417         if (err)
 418                 goto err;
 419
 420         return 0;
 421
 422 err_kfree_skb:
 423         kfree_skb(skb);
 424 err:
 425         local_bh_disable();
 426         stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
 427
 428         write_seqcount_begin(&stats->seqlock);
 429         stats->n_lost++;
 430         write_seqcount_end(&stats->seqlock);
 431
 432         local_bh_enable();
 433
 434         return err;
 435 }
 436
 437 /* Send each packet in the 'skb' list to userspace for 'dp' as directed by
 438  * 'upcall_info'.  There will be only one packet unless we broke up a GSO
 439  * packet.
 440  */
 441 static int queue_control_packets(struct datapath *dp, struct sk_buff *skb,
 442                                  const struct dp_upcall_info *upcall_info)
 443 {
 444         u32 group = packet_mc_group(dp, upcall_info->cmd);
 445         struct sk_buff *nskb;
 446         int err;
 447
 448         do {
 449                 struct odp_header *upcall;
 450                 struct sk_buff *user_skb; /* to be queued to userspace */
 451                 struct nlattr *nla;
 452                 unsigned int len;
 453
 454                 nskb = skb->next;
 455                 skb->next = NULL;
 456
 457                 err = vlan_deaccel_tag(skb);
 458                 if (unlikely(err))
 459                         goto err_kfree_skbs;
 460
 461                 if (nla_attr_size(skb->len) > USHRT_MAX)
 462                         goto err_kfree_skbs;
 463
 464                 len = sizeof(struct odp_header);
 465                 len += nla_total_size(skb->len);
 466                 len += nla_total_size(FLOW_BUFSIZE);
 467                 if (upcall_info->userdata)
 468                         len += nla_total_size(8);
 469                 if (upcall_info->sample_pool)
 470                         len += nla_total_size(4);
 471                 if (upcall_info->actions_len)
 472                         len += nla_total_size(upcall_info->actions_len);
 473
 474                 user_skb = genlmsg_new(len, GFP_ATOMIC);
 475                 if (!user_skb) {
 476                         netlink_set_err(INIT_NET_GENL_SOCK, 0, group, -ENOBUFS);
 477                         goto err_kfree_skbs;
 478                 }
 479
 480                 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family, 0, upcall_info->cmd);
 481                 upcall->dp_ifindex = dp->dp_ifindex;
 482
 483                 nla = nla_nest_start(user_skb, ODP_PACKET_ATTR_KEY);
 484                 flow_to_nlattrs(upcall_info->key, user_skb);
 485                 nla_nest_end(user_skb, nla);
 486
 487                 if (upcall_info->userdata)
 488                         nla_put_u64(user_skb, ODP_PACKET_ATTR_USERDATA, upcall_info->userdata);
 489                 if (upcall_info->sample_pool)
 490                         nla_put_u32(user_skb, ODP_PACKET_ATTR_SAMPLE_POOL, upcall_info->sample_pool);
 491                 if (upcall_info->actions_len) {
 492                         const struct nlattr *actions = upcall_info->actions;
 493                         u32 actions_len = upcall_info->actions_len;
 494
 495                         nla = nla_nest_start(user_skb, ODP_PACKET_ATTR_ACTIONS);
 496                         memcpy(__skb_put(user_skb, actions_len), actions, actions_len);
 497                         nla_nest_end(user_skb, nla);
 498                 }
 499
 500                 nla = __nla_reserve(user_skb, ODP_PACKET_ATTR_PACKET, skb->len);
 501                 if (skb->ip_summed == CHECKSUM_PARTIAL)
 502                         copy_and_csum_skb(skb, nla_data(nla));
 503                 else
 504                         skb_copy_bits(skb, 0, nla_data(nla), skb->len);
 505
 506                 err = genlmsg_multicast(user_skb, 0, group, GFP_ATOMIC);
 507                 if (err)
 508                         goto err_kfree_skbs;
 509
 510                 kfree_skb(skb);
 511                 skb = nskb;
 512         } while (skb);
 513         return 0;
 514
 515 err_kfree_skbs:
 516         kfree_skb(skb);
 517         while ((skb = nskb) != NULL) {
 518                 nskb = skb->next;
 519                 kfree_skb(skb);
 520         }
 521         return err;
 522 }
 523
 524 /* Called with genl_mutex. */
 525 static int flush_flows(int dp_ifindex)
 526 {
 527         struct tbl *old_table;
 528         struct tbl *new_table;
 529         struct datapath *dp;
 530
 531         dp = get_dp(dp_ifindex);
 532         if (!dp)
 533                 return -ENODEV;
 534
 535         old_table = get_table_protected(dp);
 536         new_table = tbl_create(TBL_MIN_BUCKETS);
 537         if (!new_table)
 538                 return -ENOMEM;
 539
 540         rcu_assign_pointer(dp->table, new_table);
 541
 542         tbl_deferred_destroy(old_table, flow_free_tbl);
 543
 544         return 0;
 545 }
 546
 547 static int validate_actions(const struct nlattr *attr)
 548 {
 549         const struct nlattr *a;
 550         int rem;
 551
 552         nla_for_each_nested(a, attr, rem) {
 553                 static const u32 action_lens[ODP_ACTION_ATTR_MAX + 1] = {
 554                         [ODP_ACTION_ATTR_OUTPUT] = 4,
 555                         [ODP_ACTION_ATTR_CONTROLLER] = 8,
 556                         [ODP_ACTION_ATTR_SET_DL_TCI] = 2,
 557                         [ODP_ACTION_ATTR_STRIP_VLAN] = 0,
 558                         [ODP_ACTION_ATTR_SET_DL_SRC] = ETH_ALEN,
 559                         [ODP_ACTION_ATTR_SET_DL_DST] = ETH_ALEN,
 560                         [ODP_ACTION_ATTR_SET_NW_SRC] = 4,
 561                         [ODP_ACTION_ATTR_SET_NW_DST] = 4,
 562                         [ODP_ACTION_ATTR_SET_NW_TOS] = 1,
 563                         [ODP_ACTION_ATTR_SET_TP_SRC] = 2,
 564                         [ODP_ACTION_ATTR_SET_TP_DST] = 2,
 565                         [ODP_ACTION_ATTR_SET_TUNNEL] = 8,
 566                         [ODP_ACTION_ATTR_SET_PRIORITY] = 4,
 567                         [ODP_ACTION_ATTR_POP_PRIORITY] = 0,
 568                         [ODP_ACTION_ATTR_DROP_SPOOFED_ARP] = 0,
 569                 };
 570                 int type = nla_type(a);
 571
 572                 if (type > ODP_ACTION_ATTR_MAX || nla_len(a) != action_lens[type])
 573                         return -EINVAL;
 574
 575                 switch (type) {
 576                 case ODP_ACTION_ATTR_UNSPEC:
 577                         return -EINVAL;
 578
 579                 case ODP_ACTION_ATTR_CONTROLLER:
 580                 case ODP_ACTION_ATTR_STRIP_VLAN:
 581                 case ODP_ACTION_ATTR_SET_DL_SRC:
 582                 case ODP_ACTION_ATTR_SET_DL_DST:
 583                 case ODP_ACTION_ATTR_SET_NW_SRC:
 584                 case ODP_ACTION_ATTR_SET_NW_DST:
 585                 case ODP_ACTION_ATTR_SET_TP_SRC:
 586                 case ODP_ACTION_ATTR_SET_TP_DST:
 587                 case ODP_ACTION_ATTR_SET_TUNNEL:
 588                 case ODP_ACTION_ATTR_SET_PRIORITY:
 589                 case ODP_ACTION_ATTR_POP_PRIORITY:
 590                 case ODP_ACTION_ATTR_DROP_SPOOFED_ARP:
 591                         /* No validation needed. */
 592                         break;
 593
 594                 case ODP_ACTION_ATTR_OUTPUT:
 595                         if (nla_get_u32(a) >= DP_MAX_PORTS)
 596                                 return -EINVAL;
 597                         break;
 598
 599                 case ODP_ACTION_ATTR_SET_DL_TCI:
 600                         if (nla_get_be16(a) & htons(VLAN_CFI_MASK))
 601                                 return -EINVAL;
 602                         break;
 603
 604                 case ODP_ACTION_ATTR_SET_NW_TOS:
 605                         if (nla_get_u8(a) & INET_ECN_MASK)
 606                                 return -EINVAL;
 607                         break;
 608
 609                 default:
 610                         return -EOPNOTSUPP;
 611                 }
 612         }
 613
 614         if (rem > 0)
 615                 return -EINVAL;
 616
 617         return 0;
 618 }
 619 static void clear_stats(struct sw_flow *flow)
 620 {
 621         flow->used = 0;
 622         flow->tcp_flags = 0;
 623         flow->packet_count = 0;
 624         flow->byte_count = 0;
 625 }
 626
 627 /* Called with genl_mutex. */
 628 static int expand_table(struct datapath *dp)
 629 {
 630         struct tbl *old_table = get_table_protected(dp);
 631         struct tbl *new_table;
 632
 633         new_table = tbl_expand(old_table);
 634         if (IS_ERR(new_table))
 635                 return PTR_ERR(new_table);
 636
 637         rcu_assign_pointer(dp->table, new_table);
 638         tbl_deferred_destroy(old_table, NULL);
 639
 640         return 0;
 641 }
 642
 643 static int odp_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 644 {
 645         struct odp_header *odp_header = info->userhdr;
 646         struct nlattr **a = info->attrs;
 647         struct sw_flow_actions *acts;
 648         struct sk_buff *packet;
 649         struct sw_flow *flow;
 650         struct datapath *dp;
 651         struct ethhdr *eth;
 652         bool is_frag;
 653         int len;
 654         int err;
 655         int key_len;
 656
 657         err = -EINVAL;
 658         if (!a[ODP_PACKET_ATTR_PACKET] || !a[ODP_PACKET_ATTR_ACTIONS] ||
 659             nla_len(a[ODP_PACKET_ATTR_PACKET]) < ETH_HLEN)
 660                 goto err;
 661
 662         err = validate_actions(a[ODP_PACKET_ATTR_ACTIONS]);
 663         if (err)
 664                 goto err;
 665
 666         len = nla_len(a[ODP_PACKET_ATTR_PACKET]);
 667         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 668         err = -ENOMEM;
 669         if (!packet)
 670                 goto err;
 671         skb_reserve(packet, NET_IP_ALIGN);
 672
 673         memcpy(__skb_put(packet, len), nla_data(a[ODP_PACKET_ATTR_PACKET]), len);
 674
 675         skb_reset_mac_header(packet);
 676         eth = eth_hdr(packet);
 677
 678         /* Normally, setting the skb 'protocol' field would be handled by a
 679          * call to eth_type_trans(), but it assumes there's a sending
 680          * device, which we may not have. */
 681         if (ntohs(eth->h_proto) >= 1536)
 682                 packet->protocol = eth->h_proto;
 683         else
 684                 packet->protocol = htons(ETH_P_802_2);
 685
 686         /* Build an sw_flow for sending this packet. */
 687         flow = flow_alloc();
 688         err = PTR_ERR(flow);
 689         if (IS_ERR(flow))
 690                 goto err_kfree_skb;
 691
 692         err = flow_extract(packet, -1, &flow->key, &key_len, &is_frag);
 693         if (err)
 694                 goto err_flow_put;
 695         flow->tbl_node.hash = flow_hash(&flow->key, key_len);
 696
 697         acts = flow_actions_alloc(a[ODP_PACKET_ATTR_ACTIONS]);
 698         err = PTR_ERR(acts);
 699         if (IS_ERR(acts))
 700                 goto err_flow_put;
 701         rcu_assign_pointer(flow->sf_acts, acts);
 702
 703         OVS_CB(packet)->flow = flow;
 704
 705         rcu_read_lock();
 706         dp = get_dp(odp_header->dp_ifindex);
 707         err = -ENODEV;
 708         if (!dp)
 709                 goto err_unlock;
 710         err = execute_actions(dp, packet);
 711         rcu_read_unlock();
 712
 713         flow_put(flow);
 714         return err;
 715
 716 err_unlock:
 717         rcu_read_unlock();
 718 err_flow_put:
 719         flow_put(flow);
 720 err_kfree_skb:
 721         kfree_skb(packet);
 722 err:
 723         return err;
 724 }
 725
 726 static const struct nla_policy packet_policy[ODP_PACKET_ATTR_MAX + 1] = {
 727         [ODP_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
 728         [ODP_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 729 };
 730
 731 static struct genl_ops dp_packet_genl_ops[] = {
 732         { .cmd = ODP_PACKET_CMD_EXECUTE,
 733           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 734           .policy = packet_policy,
 735           .doit = odp_packet_cmd_execute
 736         }
 737 };
 738
 739 static void get_dp_stats(struct datapath *dp, struct odp_stats *stats)
 740 {
 741         int i;
 742
 743         stats->n_frags = stats->n_hit = stats->n_missed = stats->n_lost = 0;
 744         for_each_possible_cpu(i) {
 745                 const struct dp_stats_percpu *percpu_stats;
 746                 struct dp_stats_percpu local_stats;
 747                 unsigned seqcount;
 748
 749                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 750
 751                 do {
 752                         seqcount = read_seqcount_begin(&percpu_stats->seqlock);
 753                         local_stats = *percpu_stats;
 754                 } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount));
 755
 756                 stats->n_frags += local_stats.n_frags;
 757                 stats->n_hit += local_stats.n_hit;
 758                 stats->n_missed += local_stats.n_missed;
 759                 stats->n_lost += local_stats.n_lost;
 760         }
 761 }
 762
 763 /* MTU of the dp pseudo-device: ETH_DATA_LEN or the minimum of the ports.
 764  * Called with RTNL lock.
 765  */
 766 int dp_min_mtu(const struct datapath *dp)
 767 {
 768         struct vport *p;
 769         int mtu = 0;
 770
 771         ASSERT_RTNL();
 772
 773         list_for_each_entry (p, &dp->port_list, node) {
 774                 int dev_mtu;
 775
 776                 /* Skip any internal ports, since that's what we're trying to
 777                  * set. */
 778                 if (is_internal_vport(p))
 779                         continue;
 780
 781                 dev_mtu = vport_get_mtu(p);
 782                 if (!dev_mtu)
 783                         continue;
 784                 if (!mtu || dev_mtu < mtu)
 785                         mtu = dev_mtu;
 786         }
 787
 788         return mtu ? mtu : ETH_DATA_LEN;
 789 }
 790
 791 /* Sets the MTU of all datapath devices to the minimum of the ports
 792  * Called with RTNL lock.
 793  */
 794 void set_internal_devs_mtu(const struct datapath *dp)
 795 {
 796         struct vport *p;
 797         int mtu;
 798
 799         ASSERT_RTNL();
 800
 801         mtu = dp_min_mtu(dp);
 802
 803         list_for_each_entry (p, &dp->port_list, node) {
 804                 if (is_internal_vport(p))
 805                         vport_set_mtu(p, mtu);
 806         }
 807 }
 808
 809 static const struct nla_policy flow_policy[ODP_FLOW_ATTR_MAX + 1] = {
 810         [ODP_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
 811         [ODP_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
 812         [ODP_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
 813 };
 814
 815 static struct genl_family dp_flow_genl_family = {
 816         .id = GENL_ID_GENERATE,
 817         .hdrsize = sizeof(struct odp_header),
 818         .name = ODP_FLOW_FAMILY,
 819         .version = 1,
 820         .maxattr = ODP_FLOW_ATTR_MAX
 821 };
 822
 823 static struct genl_multicast_group dp_flow_multicast_group = {
 824         .name = ODP_FLOW_MCGROUP
 825 };
 826
 827 /* Called with genl_lock. */
 828 static int odp_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
 829                                   struct sk_buff *skb, u32 pid, u32 seq, u32 flags, u8 cmd)
 830 {
 831         const int skb_orig_len = skb->len;
 832         const struct sw_flow_actions *sf_acts;
 833         struct odp_flow_stats stats;
 834         struct odp_header *odp_header;
 835         struct nlattr *nla;
 836         unsigned long used;
 837         u8 tcp_flags;
 838         int err;
 839
 840         sf_acts = rcu_dereference_protected(flow->sf_acts,
 841                                             lockdep_genl_is_held());
 842
 843         odp_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags, cmd);
 844         if (!odp_header)
 845                 return -EMSGSIZE;
 846
 847         odp_header->dp_ifindex = dp->dp_ifindex;
 848
 849         nla = nla_nest_start(skb, ODP_FLOW_ATTR_KEY);
 850         if (!nla)
 851                 goto nla_put_failure;
 852         err = flow_to_nlattrs(&flow->key, skb);
 853         if (err)
 854                 goto error;
 855         nla_nest_end(skb, nla);
 856
 857         spin_lock_bh(&flow->lock);
 858         used = flow->used;
 859         stats.n_packets = flow->packet_count;
 860         stats.n_bytes = flow->byte_count;
 861         tcp_flags = flow->tcp_flags;
 862         spin_unlock_bh(&flow->lock);
 863
 864         if (used)
 865                 NLA_PUT_U64(skb, ODP_FLOW_ATTR_USED, flow_used_time(used));
 866
 867         if (stats.n_packets)
 868                 NLA_PUT(skb, ODP_FLOW_ATTR_STATS, sizeof(struct odp_flow_stats), &stats);
 869
 870         if (tcp_flags)
 871                 NLA_PUT_U8(skb, ODP_FLOW_ATTR_TCP_FLAGS, tcp_flags);
 872
 873         /* If ODP_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 874          * this is the first flow to be dumped into 'skb'.  This is unusual for
 875          * Netlink but individual action lists can be longer than
 876          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 877          * The userspace caller can always fetch the actions separately if it
 878          * really wants them.  (Most userspace callers in fact don't care.)
 879          *
 880          * This can only fail for dump operations because the skb is always
 881          * properly sized for single flows.
 882          */
 883         err = nla_put(skb, ODP_FLOW_ATTR_ACTIONS, sf_acts->actions_len,
 884                       sf_acts->actions);
 885         if (err < 0 && skb_orig_len)
 886                 goto error;
 887
 888         return genlmsg_end(skb, odp_header);
 889
 890 nla_put_failure:
 891         err = -EMSGSIZE;
 892 error:
 893         genlmsg_cancel(skb, odp_header);
 894         return err;
 895 }
 896
 897 static struct sk_buff *odp_flow_cmd_alloc_info(struct sw_flow *flow)
 898 {
 899         const struct sw_flow_actions *sf_acts;
 900         int len;
 901
 902         sf_acts = rcu_dereference_protected(flow->sf_acts,
 903                                             lockdep_genl_is_held());
 904
 905         len = nla_total_size(FLOW_BUFSIZE); /* ODP_FLOW_ATTR_KEY */
 906         len += nla_total_size(sf_acts->actions_len); /* ODP_FLOW_ATTR_ACTIONS */
 907         len += nla_total_size(sizeof(struct odp_flow_stats)); /* ODP_FLOW_ATTR_STATS */
 908         len += nla_total_size(1); /* ODP_FLOW_ATTR_TCP_FLAGS */
 909         len += nla_total_size(8); /* ODP_FLOW_ATTR_USED */
 910         return genlmsg_new(NLMSG_ALIGN(sizeof(struct odp_header)) + len, GFP_KERNEL);
 911 }
 912
 913 static struct sk_buff *odp_flow_cmd_build_info(struct sw_flow *flow, struct datapath *dp,
 914                                                u32 pid, u32 seq, u8 cmd)
 915 {
 916         struct sk_buff *skb;
 917         int retval;
 918
 919         skb = odp_flow_cmd_alloc_info(flow);
 920         if (!skb)
 921                 return ERR_PTR(-ENOMEM);
 922
 923         retval = odp_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd);
 924         BUG_ON(retval < 0);
 925         return skb;
 926 }
 927
 928 static int odp_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
 929 {
 930         struct nlattr **a = info->attrs;
 931         struct odp_header *odp_header = info->userhdr;
 932         struct tbl_node *flow_node;
 933         struct sw_flow_key key;
 934         struct sw_flow *flow;
 935         struct sk_buff *reply;
 936         struct datapath *dp;
 937         struct tbl *table;
 938         u32 hash;
 939         int error;
 940         int key_len;
 941
 942         /* Extract key. */
 943         error = -EINVAL;
 944         if (!a[ODP_FLOW_ATTR_KEY])
 945                 goto error;
 946         error = flow_from_nlattrs(&key, &key_len, a[ODP_FLOW_ATTR_KEY]);
 947         if (error)
 948                 goto error;
 949
 950         /* Validate actions. */
 951         if (a[ODP_FLOW_ATTR_ACTIONS]) {
 952                 error = validate_actions(a[ODP_FLOW_ATTR_ACTIONS]);
 953                 if (error)
 954                         goto error;
 955         } else if (info->genlhdr->cmd == ODP_FLOW_CMD_NEW) {
 956                 error = -EINVAL;
 957                 goto error;
 958         }
 959
 960         dp = get_dp(odp_header->dp_ifindex);
 961         error = -ENODEV;
 962         if (!dp)
 963                 goto error;
 964
 965         hash = flow_hash(&key, key_len);
 966         table = get_table_protected(dp);
 967         flow_node = tbl_lookup(table, &key, key_len, hash, flow_cmp);
 968         if (!flow_node) {
 969                 struct sw_flow_actions *acts;
 970
 971                 /* Bail out if we're not allowed to create a new flow. */
 972                 error = -ENOENT;
 973                 if (info->genlhdr->cmd == ODP_FLOW_CMD_SET)
 974                         goto error;
 975
 976                 /* Expand table, if necessary, to make room. */
 977                 if (tbl_count(table) >= tbl_n_buckets(table)) {
 978                         error = expand_table(dp);
 979                         if (error)
 980                                 goto error;
 981                         table = get_table_protected(dp);
 982                 }
 983
 984                 /* Allocate flow. */
 985                 flow = flow_alloc();
 986                 if (IS_ERR(flow)) {
 987                         error = PTR_ERR(flow);
 988                         goto error;
 989                 }
 990                 flow->key = key;
 991                 clear_stats(flow);
 992
 993                 /* Obtain actions. */
 994                 acts = flow_actions_alloc(a[ODP_FLOW_ATTR_ACTIONS]);
 995                 error = PTR_ERR(acts);
 996                 if (IS_ERR(acts))
 997                         goto error_free_flow;
 998                 rcu_assign_pointer(flow->sf_acts, acts);
 999
1000                 /* Put flow in bucket. */
1001                 error = tbl_insert(table, &flow->tbl_node, hash);
1002                 if (error)
1003                         goto error_free_flow;
1004
1005                 reply = odp_flow_cmd_build_info(flow, dp, info->snd_pid,
1006                                                 info->snd_seq, ODP_FLOW_CMD_NEW);
1007         } else {
1008                 /* We found a matching flow. */
1009                 struct sw_flow_actions *old_acts;
1010
1011                 /* Bail out if we're not allowed to modify an existing flow.
1012                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1013                  * because Generic Netlink treats the latter as a dump
1014                  * request.  We also accept NLM_F_EXCL in case that bug ever
1015                  * gets fixed.
1016                  */
1017                 error = -EEXIST;
1018                 if (info->genlhdr->cmd == ODP_FLOW_CMD_NEW &&
1019                     info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1020                         goto error;
1021
1022                 /* Update actions. */
1023                 flow = flow_cast(flow_node);
1024                 old_acts = rcu_dereference_protected(flow->sf_acts,
1025                                                      lockdep_genl_is_held());
1026                 if (a[ODP_FLOW_ATTR_ACTIONS] &&
1027                     (old_acts->actions_len != nla_len(a[ODP_FLOW_ATTR_ACTIONS]) ||
1028                      memcmp(old_acts->actions, nla_data(a[ODP_FLOW_ATTR_ACTIONS]),
1029                             old_acts->actions_len))) {
1030                         struct sw_flow_actions *new_acts;
1031
1032                         new_acts = flow_actions_alloc(a[ODP_FLOW_ATTR_ACTIONS]);
1033                         error = PTR_ERR(new_acts);
1034                         if (IS_ERR(new_acts))
1035                                 goto error;
1036
1037                         rcu_assign_pointer(flow->sf_acts, new_acts);
1038                         flow_deferred_free_acts(old_acts);
1039                 }
1040
1041                 reply = odp_flow_cmd_build_info(flow, dp, info->snd_pid,
1042                                                 info->snd_seq, ODP_FLOW_CMD_NEW);
1043
1044                 /* Clear stats. */
1045                 if (a[ODP_FLOW_ATTR_CLEAR]) {
1046                         spin_lock_bh(&flow->lock);
1047                         clear_stats(flow);
1048                         spin_unlock_bh(&flow->lock);
1049                 }
1050         }
1051
1052         if (!IS_ERR(reply))
1053                 genl_notify(reply, genl_info_net(info), info->snd_pid,
1054                             dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1055         else
1056                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1057                                 dp_flow_multicast_group.id, PTR_ERR(reply));
1058         return 0;
1059
1060 error_free_flow:
1061         flow_put(flow);
1062 error:
1063         return error;
1064 }
1065
1066 static int odp_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1067 {
1068         struct nlattr **a = info->attrs;
1069         struct odp_header *odp_header = info->userhdr;
1070         struct sw_flow_key key;
1071         struct tbl_node *flow_node;
1072         struct sk_buff *reply;
1073         struct sw_flow *flow;
1074         struct datapath *dp;
1075         struct tbl *table;
1076         int err;
1077         int key_len;
1078
1079         if (!a[ODP_FLOW_ATTR_KEY])
1080                 return -EINVAL;
1081         err = flow_from_nlattrs(&key, &key_len, a[ODP_FLOW_ATTR_KEY]);
1082         if (err)
1083                 return err;
1084
1085         dp = get_dp(odp_header->dp_ifindex);
1086         if (!dp)
1087                 return -ENODEV;
1088
1089         table = get_table_protected(dp);
1090         flow_node = tbl_lookup(table, &key, key_len, flow_hash(&key, key_len),
1091                                flow_cmp);
1092         if (!flow_node)
1093                 return -ENOENT;
1094
1095         flow = flow_cast(flow_node);
1096         reply = odp_flow_cmd_build_info(flow, dp, info->snd_pid, info->snd_seq, ODP_FLOW_CMD_NEW);
1097         if (IS_ERR(reply))
1098                 return PTR_ERR(reply);
1099
1100         return genlmsg_reply(reply, info);
1101 }
1102
1103 static int odp_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1104 {
1105         struct nlattr **a = info->attrs;
1106         struct odp_header *odp_header = info->userhdr;
1107         struct sw_flow_key key;
1108         struct tbl_node *flow_node;
1109         struct sk_buff *reply;
1110         struct sw_flow *flow;
1111         struct datapath *dp;
1112         struct tbl *table;
1113         int err;
1114         int key_len;
1115
1116         if (!a[ODP_FLOW_ATTR_KEY])
1117                 return flush_flows(odp_header->dp_ifindex);
1118         err = flow_from_nlattrs(&key, &key_len, a[ODP_FLOW_ATTR_KEY]);
1119         if (err)
1120                 return err;
1121
1122         dp = get_dp(odp_header->dp_ifindex);
1123         if (!dp)
1124                 return -ENODEV;
1125
1126         table = get_table_protected(dp);
1127         flow_node = tbl_lookup(table, &key, key_len, flow_hash(&key, key_len),
1128                                flow_cmp);
1129         if (!flow_node)
1130                 return -ENOENT;
1131         flow = flow_cast(flow_node);
1132
1133         reply = odp_flow_cmd_alloc_info(flow);
1134         if (!reply)
1135                 return -ENOMEM;
1136
1137         err = tbl_remove(table, flow_node);
1138         if (err) {
1139                 kfree_skb(reply);
1140                 return err;
1141         }
1142
1143         err = odp_flow_cmd_fill_info(flow, dp, reply, info->snd_pid,
1144                                      info->snd_seq, 0, ODP_FLOW_CMD_DEL);
1145         BUG_ON(err < 0);
1146
1147         flow_deferred_free(flow);
1148
1149         genl_notify(reply, genl_info_net(info), info->snd_pid,
1150                     dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1151         return 0;
1152 }
1153
1154 static int odp_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1155 {
1156         struct odp_header *odp_header = genlmsg_data(nlmsg_data(cb->nlh));
1157         struct datapath *dp;
1158
1159         dp = get_dp(odp_header->dp_ifindex);
1160         if (!dp)
1161                 return -ENODEV;
1162
1163         for (;;) {
1164                 struct tbl_node *flow_node;
1165                 struct sw_flow *flow;
1166                 u32 bucket, obj;
1167
1168                 bucket = cb->args[0];
1169                 obj = cb->args[1];
1170                 flow_node = tbl_next(get_table_protected(dp), &bucket, &obj);
1171                 if (!flow_node)
1172                         break;
1173
1174                 flow = flow_cast(flow_node);
1175                 if (odp_flow_cmd_fill_info(flow, dp, skb, NETLINK_CB(cb->skb).pid,
1176                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1177                                            ODP_FLOW_CMD_NEW) < 0)
1178                         break;
1179
1180                 cb->args[0] = bucket;
1181                 cb->args[1] = obj;
1182         }
1183         return skb->len;
1184 }
1185
1186 static struct genl_ops dp_flow_genl_ops[] = {
1187         { .cmd = ODP_FLOW_CMD_NEW,
1188           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1189           .policy = flow_policy,
1190           .doit = odp_flow_cmd_new_or_set
1191         },
1192         { .cmd = ODP_FLOW_CMD_DEL,
1193           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1194           .policy = flow_policy,
1195           .doit = odp_flow_cmd_del
1196         },
1197         { .cmd = ODP_FLOW_CMD_GET,
1198           .flags = 0,               /* OK for unprivileged users. */
1199           .policy = flow_policy,
1200           .doit = odp_flow_cmd_get,
1201           .dumpit = odp_flow_cmd_dump
1202         },
1203         { .cmd = ODP_FLOW_CMD_SET,
1204           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1205           .policy = flow_policy,
1206           .doit = odp_flow_cmd_new_or_set,
1207         },
1208 };
1209
1210 static const struct nla_policy datapath_policy[ODP_DP_ATTR_MAX + 1] = {
1211 #ifdef HAVE_NLA_NUL_STRING
1212         [ODP_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1213 #endif
1214         [ODP_DP_ATTR_IPV4_FRAGS] = { .type = NLA_U32 },
1215         [ODP_DP_ATTR_SAMPLING] = { .type = NLA_U32 },
1216 };
1217
1218 static struct genl_family dp_datapath_genl_family = {
1219         .id = GENL_ID_GENERATE,
1220         .hdrsize = sizeof(struct odp_header),
1221         .name = ODP_DATAPATH_FAMILY,
1222         .version = 1,
1223         .maxattr = ODP_DP_ATTR_MAX
1224 };
1225
1226 static struct genl_multicast_group dp_datapath_multicast_group = {
1227         .name = ODP_DATAPATH_MCGROUP
1228 };
1229
1230 static int odp_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1231                                 u32 pid, u32 seq, u32 flags, u8 cmd)
1232 {
1233         struct odp_header *odp_header;
1234         struct nlattr *nla;
1235         int err;
1236
1237         odp_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family,
1238                                    flags, cmd);
1239         if (!odp_header)
1240                 goto error;
1241
1242         odp_header->dp_ifindex = dp->dp_ifindex;
1243
1244         rcu_read_lock();
1245         err = nla_put_string(skb, ODP_DP_ATTR_NAME, dp_name(dp));
1246         rcu_read_unlock();
1247         if (err)
1248                 goto nla_put_failure;
1249
1250         nla = nla_reserve(skb, ODP_DP_ATTR_STATS, sizeof(struct odp_stats));
1251         if (!nla)
1252                 goto nla_put_failure;
1253         get_dp_stats(dp, nla_data(nla));
1254
1255         NLA_PUT_U32(skb, ODP_DP_ATTR_IPV4_FRAGS,
1256                     dp->drop_frags ? ODP_DP_FRAG_DROP : ODP_DP_FRAG_ZERO);
1257
1258         if (dp->sflow_probability)
1259                 NLA_PUT_U32(skb, ODP_DP_ATTR_SAMPLING, dp->sflow_probability);
1260
1261         nla = nla_nest_start(skb, ODP_DP_ATTR_MCGROUPS);
1262         if (!nla)
1263                 goto nla_put_failure;
1264         NLA_PUT_U32(skb, ODP_PACKET_CMD_MISS, packet_mc_group(dp, ODP_PACKET_CMD_MISS));
1265         NLA_PUT_U32(skb, ODP_PACKET_CMD_ACTION, packet_mc_group(dp, ODP_PACKET_CMD_ACTION));
1266         NLA_PUT_U32(skb, ODP_PACKET_CMD_SAMPLE, packet_mc_group(dp, ODP_PACKET_CMD_SAMPLE));
1267         nla_nest_end(skb, nla);
1268
1269         return genlmsg_end(skb, odp_header);
1270
1271 nla_put_failure:
1272         genlmsg_cancel(skb, odp_header);
1273 error:
1274         return -EMSGSIZE;
1275 }
1276
1277 static struct sk_buff *odp_dp_cmd_build_info(struct datapath *dp, u32 pid,
1278                                              u32 seq, u8 cmd)
1279 {
1280         struct sk_buff *skb;
1281         int retval;
1282
1283         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1284         if (!skb)
1285                 return ERR_PTR(-ENOMEM);
1286
1287         retval = odp_dp_cmd_fill_info(dp, skb, pid, seq, 0, cmd);
1288         if (retval < 0) {
1289                 kfree_skb(skb);
1290                 return ERR_PTR(retval);
1291         }
1292         return skb;
1293 }
1294
1295 static int odp_dp_cmd_validate(struct nlattr *a[ODP_DP_ATTR_MAX + 1])
1296 {
1297         if (a[ODP_DP_ATTR_IPV4_FRAGS]) {
1298                 u32 frags = nla_get_u32(a[ODP_DP_ATTR_IPV4_FRAGS]);
1299
1300                 if (frags != ODP_DP_FRAG_ZERO && frags != ODP_DP_FRAG_DROP)
1301                         return -EINVAL;
1302         }
1303
1304         return CHECK_NUL_STRING(a[ODP_DP_ATTR_NAME], IFNAMSIZ - 1);
1305 }
1306
1307 /* Called with genl_mutex and optionally with RTNL lock also. */
1308 static struct datapath *lookup_datapath(struct odp_header *odp_header, struct nlattr *a[ODP_DP_ATTR_MAX + 1])
1309 {
1310         struct datapath *dp;
1311
1312         if (!a[ODP_DP_ATTR_NAME])
1313                 dp = get_dp(odp_header->dp_ifindex);
1314         else {
1315                 struct vport *vport;
1316
1317                 rcu_read_lock();
1318                 vport = vport_locate(nla_data(a[ODP_DP_ATTR_NAME]));
1319                 dp = vport && vport->port_no == ODPP_LOCAL ? vport->dp : NULL;
1320                 rcu_read_unlock();
1321         }
1322         return dp ? dp : ERR_PTR(-ENODEV);
1323 }
1324
1325 /* Called with genl_mutex. */
1326 static void change_datapath(struct datapath *dp, struct nlattr *a[ODP_DP_ATTR_MAX + 1])
1327 {
1328         if (a[ODP_DP_ATTR_IPV4_FRAGS])
1329                 dp->drop_frags = nla_get_u32(a[ODP_DP_ATTR_IPV4_FRAGS]) == ODP_DP_FRAG_DROP;
1330         if (a[ODP_DP_ATTR_SAMPLING])
1331                 dp->sflow_probability = nla_get_u32(a[ODP_DP_ATTR_SAMPLING]);
1332 }
1333
1334 static int odp_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1335 {
1336         struct nlattr **a = info->attrs;
1337         struct vport_parms parms;
1338         struct sk_buff *reply;
1339         struct datapath *dp;
1340         struct vport *vport;
1341         int err;
1342
1343         err = -EINVAL;
1344         if (!a[ODP_DP_ATTR_NAME])
1345                 goto err;
1346
1347         err = odp_dp_cmd_validate(a);
1348         if (err)
1349                 goto err;
1350
1351         rtnl_lock();
1352         err = -ENODEV;
1353         if (!try_module_get(THIS_MODULE))
1354                 goto err_unlock_rtnl;
1355
1356         err = -ENOMEM;
1357         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1358         if (dp == NULL)
1359                 goto err_put_module;
1360         INIT_LIST_HEAD(&dp->port_list);
1361
1362         /* Initialize kobject for bridge.  This will be added as
1363          * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
1364         dp->ifobj.kset = NULL;
1365         kobject_init(&dp->ifobj, &dp_ktype);
1366
1367         /* Allocate table. */
1368         err = -ENOMEM;
1369         rcu_assign_pointer(dp->table, tbl_create(TBL_MIN_BUCKETS));
1370         if (!dp->table)
1371                 goto err_free_dp;
1372
1373         /* Set up our datapath device. */
1374         parms.name = nla_data(a[ODP_DP_ATTR_NAME]);
1375         parms.type = ODP_VPORT_TYPE_INTERNAL;
1376         parms.options = NULL;
1377         parms.dp = dp;
1378         parms.port_no = ODPP_LOCAL;
1379         vport = new_vport(&parms);
1380         if (IS_ERR(vport)) {
1381                 err = PTR_ERR(vport);
1382                 if (err == -EBUSY)
1383                         err = -EEXIST;
1384
1385                 goto err_destroy_table;
1386         }
1387         dp->dp_ifindex = vport_get_ifindex(vport);
1388
1389         dp->drop_frags = 0;
1390         dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1391         if (!dp->stats_percpu) {
1392                 err = -ENOMEM;
1393                 goto err_destroy_local_port;
1394         }
1395
1396         change_datapath(dp, a);
1397
1398         reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_NEW);
1399         err = PTR_ERR(reply);
1400         if (IS_ERR(reply))
1401                 goto err_destroy_local_port;
1402
1403         list_add_tail(&dp->list_node, &dps);
1404         dp_sysfs_add_dp(dp);
1405
1406         rtnl_unlock();
1407
1408         genl_notify(reply, genl_info_net(info), info->snd_pid,
1409                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1410         return 0;
1411
1412 err_destroy_local_port:
1413         dp_detach_port(get_vport_protected(dp, ODPP_LOCAL));
1414 err_destroy_table:
1415         tbl_destroy(get_table_protected(dp), NULL);
1416 err_free_dp:
1417         kfree(dp);
1418 err_put_module:
1419         module_put(THIS_MODULE);
1420 err_unlock_rtnl:
1421         rtnl_unlock();
1422 err:
1423         return err;
1424 }
1425
1426 static int odp_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1427 {
1428         struct vport *vport, *next_vport;
1429         struct sk_buff *reply;
1430         struct datapath *dp;
1431         int err;
1432
1433         err = odp_dp_cmd_validate(info->attrs);
1434         if (err)
1435                 goto exit;
1436
1437         rtnl_lock();
1438         dp = lookup_datapath(info->userhdr, info->attrs);
1439         err = PTR_ERR(dp);
1440         if (IS_ERR(dp))
1441                 goto exit_unlock;
1442
1443         reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_DEL);
1444         err = PTR_ERR(reply);
1445         if (IS_ERR(reply))
1446                 goto exit_unlock;
1447
1448         list_for_each_entry_safe (vport, next_vport, &dp->port_list, node)
1449                 if (vport->port_no != ODPP_LOCAL)
1450                         dp_detach_port(vport);
1451
1452         dp_sysfs_del_dp(dp);
1453         list_del(&dp->list_node);
1454         dp_detach_port(get_vport_protected(dp, ODPP_LOCAL));
1455
1456         /* rtnl_unlock() will wait until all the references to devices that
1457          * are pending unregistration have been dropped.  We do it here to
1458          * ensure that any internal devices (which contain DP pointers) are
1459          * fully destroyed before freeing the datapath.
1460          */
1461         rtnl_unlock();
1462
1463         call_rcu(&dp->rcu, destroy_dp_rcu);
1464         module_put(THIS_MODULE);
1465
1466         genl_notify(reply, genl_info_net(info), info->snd_pid,
1467                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1468
1469         return 0;
1470
1471 exit_unlock:
1472         rtnl_unlock();
1473 exit:
1474         return err;
1475 }
1476
1477 static int odp_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1478 {
1479         struct sk_buff *reply;
1480         struct datapath *dp;
1481         int err;
1482
1483         err = odp_dp_cmd_validate(info->attrs);
1484         if (err)
1485                 return err;
1486
1487         dp = lookup_datapath(info->userhdr, info->attrs);
1488         if (IS_ERR(dp))
1489                 return PTR_ERR(dp);
1490
1491         change_datapath(dp, info->attrs);
1492
1493         reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_NEW);
1494         if (IS_ERR(reply)) {
1495                 err = PTR_ERR(reply);
1496                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1497                                 dp_datapath_multicast_group.id, err);
1498                 return 0;
1499         }
1500
1501         genl_notify(reply, genl_info_net(info), info->snd_pid,
1502                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1503         return 0;
1504 }
1505
1506 static int odp_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1507 {
1508         struct sk_buff *reply;
1509         struct datapath *dp;
1510         int err;
1511
1512         err = odp_dp_cmd_validate(info->attrs);
1513         if (err)
1514                 return err;
1515
1516         dp = lookup_datapath(info->userhdr, info->attrs);
1517         if (IS_ERR(dp))
1518                 return PTR_ERR(dp);
1519
1520         reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_NEW);
1521         if (IS_ERR(reply))
1522                 return PTR_ERR(reply);
1523
1524         return genlmsg_reply(reply, info);
1525 }
1526
1527 static int odp_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1528 {
1529         struct datapath *dp;
1530         int skip = cb->args[0];
1531         int i = 0;
1532
1533         list_for_each_entry (dp, &dps, list_node) {
1534                 if (i < skip)
1535                         continue;
1536                 if (odp_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).pid,
1537                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1538                                          ODP_DP_CMD_NEW) < 0)
1539                         break;
1540                 i++;
1541         }
1542
1543         cb->args[0] = i;
1544
1545         return skb->len;
1546 }
1547
1548 static struct genl_ops dp_datapath_genl_ops[] = {
1549         { .cmd = ODP_DP_CMD_NEW,
1550           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1551           .policy = datapath_policy,
1552           .doit = odp_dp_cmd_new
1553         },
1554         { .cmd = ODP_DP_CMD_DEL,
1555           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1556           .policy = datapath_policy,
1557           .doit = odp_dp_cmd_del
1558         },
1559         { .cmd = ODP_DP_CMD_GET,
1560           .flags = 0,               /* OK for unprivileged users. */
1561           .policy = datapath_policy,
1562           .doit = odp_dp_cmd_get,
1563           .dumpit = odp_dp_cmd_dump
1564         },
1565         { .cmd = ODP_DP_CMD_SET,
1566           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1567           .policy = datapath_policy,
1568           .doit = odp_dp_cmd_set,
1569         },
1570 };
1571
1572 static const struct nla_policy vport_policy[ODP_VPORT_ATTR_MAX + 1] = {
1573 #ifdef HAVE_NLA_NUL_STRING
1574         [ODP_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1575         [ODP_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1576         [ODP_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1577         [ODP_VPORT_ATTR_STATS] = { .len = sizeof(struct rtnl_link_stats64) },
1578         [ODP_VPORT_ATTR_ADDRESS] = { .len = ETH_ALEN },
1579 #else
1580         [ODP_VPORT_ATTR_STATS] = { .minlen = sizeof(struct rtnl_link_stats64) },
1581         [ODP_VPORT_ATTR_ADDRESS] = { .minlen = ETH_ALEN },
1582 #endif
1583         [ODP_VPORT_ATTR_MTU] = { .type = NLA_U32 },
1584         [ODP_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1585 };
1586
1587 static struct genl_family dp_vport_genl_family = {
1588         .id = GENL_ID_GENERATE,
1589         .hdrsize = sizeof(struct odp_header),
1590         .name = ODP_VPORT_FAMILY,
1591         .version = 1,
1592         .maxattr = ODP_VPORT_ATTR_MAX
1593 };
1594
1595 static struct genl_multicast_group dp_vport_multicast_group = {
1596         .name = ODP_VPORT_MCGROUP
1597 };
1598
1599 /* Called with RTNL lock or RCU read lock. */
1600 static int odp_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1601                                    u32 pid, u32 seq, u32 flags, u8 cmd)
1602 {
1603         struct odp_header *odp_header;
1604         struct nlattr *nla;
1605         int ifindex, iflink;
1606         int mtu;
1607         int err;
1608
1609         odp_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family,
1610                                  flags, cmd);
1611         if (!odp_header)
1612                 return -EMSGSIZE;
1613
1614         odp_header->dp_ifindex = vport->dp->dp_ifindex;
1615
1616         NLA_PUT_U32(skb, ODP_VPORT_ATTR_PORT_NO, vport->port_no);
1617         NLA_PUT_U32(skb, ODP_VPORT_ATTR_TYPE, vport_get_type(vport));
1618         NLA_PUT_STRING(skb, ODP_VPORT_ATTR_NAME, vport_get_name(vport));
1619
1620         nla = nla_reserve(skb, ODP_VPORT_ATTR_STATS, sizeof(struct rtnl_link_stats64));
1621         if (!nla)
1622                 goto nla_put_failure;
1623         if (vport_get_stats(vport, nla_data(nla)))
1624                 __skb_trim(skb, skb->len - nla->nla_len);
1625
1626         NLA_PUT(skb, ODP_VPORT_ATTR_ADDRESS, ETH_ALEN, vport_get_addr(vport));
1627
1628         mtu = vport_get_mtu(vport);
1629         if (mtu)
1630                 NLA_PUT_U32(skb, ODP_VPORT_ATTR_MTU, mtu);
1631
1632         err = vport_get_options(vport, skb);
1633         if (err == -EMSGSIZE)
1634                 goto error;
1635
1636         ifindex = vport_get_ifindex(vport);
1637         if (ifindex > 0)
1638                 NLA_PUT_U32(skb, ODP_VPORT_ATTR_IFINDEX, ifindex);
1639
1640         iflink = vport_get_iflink(vport);
1641         if (iflink > 0)
1642                 NLA_PUT_U32(skb, ODP_VPORT_ATTR_IFLINK, iflink);
1643
1644         return genlmsg_end(skb, odp_header);
1645
1646 nla_put_failure:
1647         err = -EMSGSIZE;
1648 error:
1649         genlmsg_cancel(skb, odp_header);
1650         return err;
1651 }
1652
1653 /* Called with RTNL lock or RCU read lock. */
1654 static struct sk_buff *odp_vport_cmd_build_info(struct vport *vport, u32 pid,
1655                                                 u32 seq, u8 cmd)
1656 {
1657         struct sk_buff *skb;
1658         int retval;
1659
1660         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1661         if (!skb)
1662                 return ERR_PTR(-ENOMEM);
1663
1664         retval = odp_vport_cmd_fill_info(vport, skb, pid, seq, 0, cmd);
1665         if (retval < 0) {
1666                 kfree_skb(skb);
1667                 return ERR_PTR(retval);
1668         }
1669         return skb;
1670 }
1671
1672 static int odp_vport_cmd_validate(struct nlattr *a[ODP_VPORT_ATTR_MAX + 1])
1673 {
1674         return CHECK_NUL_STRING(a[ODP_VPORT_ATTR_NAME], IFNAMSIZ - 1);
1675 }
1676
1677 /* Called with RTNL lock or RCU read lock. */
1678 static struct vport *lookup_vport(struct odp_header *odp_header,
1679                                   struct nlattr *a[ODP_VPORT_ATTR_MAX + 1])
1680 {
1681         struct datapath *dp;
1682         struct vport *vport;
1683
1684         if (a[ODP_VPORT_ATTR_NAME]) {
1685                 vport = vport_locate(nla_data(a[ODP_VPORT_ATTR_NAME]));
1686                 if (!vport)
1687                         return ERR_PTR(-ENODEV);
1688                 return vport;
1689         } else if (a[ODP_VPORT_ATTR_PORT_NO]) {
1690                 u32 port_no = nla_get_u32(a[ODP_VPORT_ATTR_PORT_NO]);
1691
1692                 if (port_no >= DP_MAX_PORTS)
1693                         return ERR_PTR(-EFBIG);
1694
1695                 dp = get_dp(odp_header->dp_ifindex);
1696                 if (!dp)
1697                         return ERR_PTR(-ENODEV);
1698
1699                 vport = get_vport_protected(dp, port_no);
1700                 if (!vport)
1701                         return ERR_PTR(-ENOENT);
1702                 return vport;
1703         } else
1704                 return ERR_PTR(-EINVAL);
1705 }
1706
1707 /* Called with RTNL lock. */
1708 static int change_vport(struct vport *vport, struct nlattr *a[ODP_VPORT_ATTR_MAX + 1])
1709 {
1710         int err = 0;
1711         if (a[ODP_VPORT_ATTR_STATS])
1712                 err = vport_set_stats(vport, nla_data(a[ODP_VPORT_ATTR_STATS]));
1713         if (!err && a[ODP_VPORT_ATTR_ADDRESS])
1714                 err = vport_set_addr(vport, nla_data(a[ODP_VPORT_ATTR_ADDRESS]));
1715         if (!err && a[ODP_VPORT_ATTR_MTU])
1716                 err = vport_set_mtu(vport, nla_get_u32(a[ODP_VPORT_ATTR_MTU]));
1717         return err;
1718 }
1719
1720 static int odp_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1721 {
1722         struct nlattr **a = info->attrs;
1723         struct odp_header *odp_header = info->userhdr;
1724         struct vport_parms parms;
1725         struct sk_buff *reply;
1726         struct vport *vport;
1727         struct datapath *dp;
1728         u32 port_no;
1729         int err;
1730
1731         err = -EINVAL;
1732         if (!a[ODP_VPORT_ATTR_NAME] || !a[ODP_VPORT_ATTR_TYPE])
1733                 goto exit;
1734
1735         err = odp_vport_cmd_validate(a);
1736         if (err)
1737                 goto exit;
1738
1739         rtnl_lock();
1740         dp = get_dp(odp_header->dp_ifindex);
1741         err = -ENODEV;
1742         if (!dp)
1743                 goto exit_unlock;
1744
1745         if (a[ODP_VPORT_ATTR_PORT_NO]) {
1746                 port_no = nla_get_u32(a[ODP_VPORT_ATTR_PORT_NO]);
1747
1748                 err = -EFBIG;
1749                 if (port_no >= DP_MAX_PORTS)
1750                         goto exit_unlock;
1751
1752                 vport = get_vport_protected(dp, port_no);
1753                 err = -EBUSY;
1754                 if (vport)
1755                         goto exit_unlock;
1756         } else {
1757                 for (port_no = 1; ; port_no++) {
1758                         if (port_no >= DP_MAX_PORTS) {
1759                                 err = -EFBIG;
1760                                 goto exit_unlock;
1761                         }
1762                         vport = get_vport_protected(dp, port_no);
1763                         if (!vport)
1764                                 break;
1765                 }
1766         }
1767
1768         parms.name = nla_data(a[ODP_VPORT_ATTR_NAME]);
1769         parms.type = nla_get_u32(a[ODP_VPORT_ATTR_TYPE]);
1770         parms.options = a[ODP_VPORT_ATTR_OPTIONS];
1771         parms.dp = dp;
1772         parms.port_no = port_no;
1773
1774         vport = new_vport(&parms);
1775         err = PTR_ERR(vport);
1776         if (IS_ERR(vport))
1777                 goto exit_unlock;
1778
1779         set_internal_devs_mtu(dp);
1780         dp_sysfs_add_if(vport);
1781
1782         err = change_vport(vport, a);
1783         if (!err) {
1784                 reply = odp_vport_cmd_build_info(vport, info->snd_pid,
1785                                                  info->snd_seq, ODP_VPORT_CMD_NEW);
1786                 if (IS_ERR(reply))
1787                         err = PTR_ERR(reply);
1788         }
1789         if (err) {
1790                 dp_detach_port(vport);
1791                 goto exit_unlock;
1792         }
1793         genl_notify(reply, genl_info_net(info), info->snd_pid,
1794                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1795
1796
1797 exit_unlock:
1798         rtnl_unlock();
1799 exit:
1800         return err;
1801 }
1802
1803 static int odp_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1804 {
1805         struct nlattr **a = info->attrs;
1806         struct sk_buff *reply;
1807         struct vport *vport;
1808         int err;
1809
1810         err = odp_vport_cmd_validate(a);
1811         if (err)
1812                 goto exit;
1813
1814         rtnl_lock();
1815         vport = lookup_vport(info->userhdr, a);
1816         err = PTR_ERR(vport);
1817         if (IS_ERR(vport))
1818                 goto exit_unlock;
1819
1820         err = 0;
1821         if (a[ODP_VPORT_ATTR_OPTIONS])
1822                 err = vport_set_options(vport, a[ODP_VPORT_ATTR_OPTIONS]);
1823         if (!err)
1824                 err = change_vport(vport, a);
1825
1826         reply = odp_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1827                                          ODP_VPORT_CMD_NEW);
1828         if (IS_ERR(reply)) {
1829                 err = PTR_ERR(reply);
1830                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1831                                 dp_vport_multicast_group.id, err);
1832                 return 0;
1833         }
1834
1835         genl_notify(reply, genl_info_net(info), info->snd_pid,
1836                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1837
1838 exit_unlock:
1839         rtnl_unlock();
1840 exit:
1841         return err;
1842 }
1843
1844 static int odp_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1845 {
1846         struct nlattr **a = info->attrs;
1847         struct sk_buff *reply;
1848         struct vport *vport;
1849         int err;
1850
1851         err = odp_vport_cmd_validate(a);
1852         if (err)
1853                 goto exit;
1854
1855         rtnl_lock();
1856         vport = lookup_vport(info->userhdr, a);
1857         err = PTR_ERR(vport);
1858         if (IS_ERR(vport))
1859                 goto exit_unlock;
1860
1861         if (vport->port_no == ODPP_LOCAL) {
1862                 err = -EINVAL;
1863                 goto exit_unlock;
1864         }
1865
1866         reply = odp_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1867                                          ODP_VPORT_CMD_DEL);
1868         err = PTR_ERR(reply);
1869         if (IS_ERR(reply))
1870                 goto exit_unlock;
1871
1872         err = dp_detach_port(vport);
1873
1874         genl_notify(reply, genl_info_net(info), info->snd_pid,
1875                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1876
1877 exit_unlock:
1878         rtnl_unlock();
1879 exit:
1880         return err;
1881 }
1882
1883 static int odp_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1884 {
1885         struct nlattr **a = info->attrs;
1886         struct odp_header *odp_header = info->userhdr;
1887         struct sk_buff *reply;
1888         struct vport *vport;
1889         int err;
1890
1891         err = odp_vport_cmd_validate(a);
1892         if (err)
1893                 goto exit;
1894
1895         rcu_read_lock();
1896         vport = lookup_vport(odp_header, a);
1897         err = PTR_ERR(vport);
1898         if (IS_ERR(vport))
1899                 goto exit_unlock;
1900
1901         reply = odp_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1902                                          ODP_VPORT_CMD_NEW);
1903         err = PTR_ERR(reply);
1904         if (IS_ERR(reply))
1905                 goto exit_unlock;
1906
1907         err = genlmsg_reply(reply, info);
1908
1909 exit_unlock:
1910         rcu_read_unlock();
1911 exit:
1912         return err;
1913 }
1914
1915 static int odp_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1916 {
1917         struct odp_header *odp_header = genlmsg_data(nlmsg_data(cb->nlh));
1918         struct datapath *dp;
1919         u32 port_no;
1920         int retval;
1921
1922         dp = get_dp(odp_header->dp_ifindex);
1923         if (!dp)
1924                 return -ENODEV;
1925
1926         rcu_read_lock();
1927         for (port_no = cb->args[0]; port_no < DP_MAX_PORTS; port_no++) {
1928                 struct vport *vport;
1929
1930                 vport = get_vport_protected(dp, port_no);
1931                 if (!vport)
1932                         continue;
1933
1934                 if (odp_vport_cmd_fill_info(vport, skb, NETLINK_CB(cb->skb).pid,
1935                                             cb->nlh->nlmsg_seq, NLM_F_MULTI,
1936                                             ODP_VPORT_CMD_NEW) < 0)
1937                         break;
1938         }
1939         rcu_read_unlock();
1940
1941         cb->args[0] = port_no;
1942         retval = skb->len;
1943
1944         return retval;
1945 }
1946
1947 static struct genl_ops dp_vport_genl_ops[] = {
1948         { .cmd = ODP_VPORT_CMD_NEW,
1949           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1950           .policy = vport_policy,
1951           .doit = odp_vport_cmd_new
1952         },
1953         { .cmd = ODP_VPORT_CMD_DEL,
1954           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1955           .policy = vport_policy,
1956           .doit = odp_vport_cmd_del
1957         },
1958         { .cmd = ODP_VPORT_CMD_GET,
1959           .flags = 0,               /* OK for unprivileged users. */
1960           .policy = vport_policy,
1961           .doit = odp_vport_cmd_get,
1962           .dumpit = odp_vport_cmd_dump
1963         },
1964         { .cmd = ODP_VPORT_CMD_SET,
1965           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1966           .policy = vport_policy,
1967           .doit = odp_vport_cmd_set,
1968         },
1969 };
1970
1971 struct genl_family_and_ops {
1972         struct genl_family *family;
1973         struct genl_ops *ops;
1974         int n_ops;
1975         struct genl_multicast_group *group;
1976 };
1977
1978 static const struct genl_family_and_ops dp_genl_families[] = {
1979         { &dp_datapath_genl_family,
1980           dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1981           &dp_datapath_multicast_group },
1982         { &dp_vport_genl_family,
1983           dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1984           &dp_vport_multicast_group },
1985         { &dp_flow_genl_family,
1986           dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1987           &dp_flow_multicast_group },
1988         { &dp_packet_genl_family,
1989           dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1990           NULL },
1991 };
1992
1993 static void dp_unregister_genl(int n_families)
1994 {
1995         int i;
1996
1997         for (i = 0; i < n_families; i++) {
1998                 genl_unregister_family(dp_genl_families[i].family);
1999         }
2000 }
2001
2002 static int dp_register_genl(void)
2003 {
2004         int n_registered;
2005         int err;
2006         int i;
2007
2008         n_registered = 0;
2009         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2010                 const struct genl_family_and_ops *f = &dp_genl_families[i];
2011
2012                 err = genl_register_family_with_ops(f->family, f->ops,
2013                                                     f->n_ops);
2014                 if (err)
2015                         goto error;
2016                 n_registered++;
2017
2018                 if (f->group) {
2019                         err = genl_register_mc_group(f->family, f->group);
2020                         if (err)
2021                                 goto error;
2022                 }
2023         }
2024
2025         err = packet_register_mc_groups();
2026         if (err)
2027                 goto error;
2028         return 0;
2029
2030 error:
2031         dp_unregister_genl(n_registered);
2032         return err;
2033 }
2034
2035 static int __init dp_init(void)
2036 {
2037         struct sk_buff *dummy_skb;
2038         int err;
2039
2040         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
2041
2042         printk("Open vSwitch %s, built "__DATE__" "__TIME__"\n", VERSION BUILDNR);
2043
2044         err = flow_init();
2045         if (err)
2046                 goto error;
2047
2048         err = vport_init();
2049         if (err)
2050                 goto error_flow_exit;
2051
2052         err = register_netdevice_notifier(&dp_device_notifier);
2053         if (err)
2054                 goto error_vport_exit;
2055
2056         err = dp_register_genl();
2057         if (err < 0)
2058                 goto error_unreg_notifier;
2059
2060         return 0;
2061
2062 error_unreg_notifier:
2063         unregister_netdevice_notifier(&dp_device_notifier);
2064 error_vport_exit:
2065         vport_exit();
2066 error_flow_exit:
2067         flow_exit();
2068 error:
2069         return err;
2070 }
2071
2072 static void dp_cleanup(void)
2073 {
2074         rcu_barrier();
2075         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2076         unregister_netdevice_notifier(&dp_device_notifier);
2077         vport_exit();
2078         flow_exit();
2079 }
2080
2081 module_init(dp_init);
2082 module_exit(dp_cleanup);
2083
2084 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2085 MODULE_LICENSE("GPL");