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 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,18) || \
  57     LINUX_VERSION_CODE >= KERNEL_VERSION(3,1,0)
  58 #error Kernels before 2.6.18 or after 3.0 are not supported by this version of Open vSwitch.
  59 #endif
  60
  61 int (*dp_ioctl_hook)(struct net_device *dev, struct ifreq *rq, int cmd);
  62 EXPORT_SYMBOL(dp_ioctl_hook);
  63
  64 /**
  65  * DOC: Locking:
  66  *
  67  * Writes to device state (add/remove datapath, port, set operations on vports,
  68  * etc.) are protected by RTNL.
  69  *
  70  * Writes to other state (flow table modifications, set miscellaneous datapath
  71  * parameters such as drop frags, etc.) are protected by genl_mutex.  The RTNL
  72  * lock nests inside genl_mutex.
  73  *
  74  * Reads are protected by RCU.
  75  *
  76  * There are a few special cases (mostly stats) that have their own
  77  * synchronization but they nest under all of above and don't interact with
  78  * each other.
  79  */
  80
  81 /* Global list of datapaths to enable dumping them all out.
  82  * Protected by genl_mutex.
  83  */
  84 static LIST_HEAD(dps);
  85
  86 static struct vport *new_vport(const struct vport_parms *);
  87 static int queue_userspace_packets(struct datapath *, struct sk_buff *,
  88                                  const struct dp_upcall_info *);
  89
  90 /* Must be called with rcu_read_lock, genl_mutex, or RTNL lock. */
  91 struct datapath *get_dp(int dp_ifindex)
  92 {
  93         struct datapath *dp = NULL;
  94         struct net_device *dev;
  95
  96         rcu_read_lock();
  97         dev = dev_get_by_index_rcu(&init_net, dp_ifindex);
  98         if (dev) {
  99                 struct vport *vport = internal_dev_get_vport(dev);
 100                 if (vport)
 101                         dp = vport->dp;
 102         }
 103         rcu_read_unlock();
 104
 105         return dp;
 106 }
 107 EXPORT_SYMBOL_GPL(get_dp);
 108
 109 /* Must be called with genl_mutex. */
 110 static struct tbl *get_table_protected(struct datapath *dp)
 111 {
 112         return rcu_dereference_protected(dp->table, lockdep_genl_is_held());
 113 }
 114
 115 /* Must be called with rcu_read_lock or RTNL lock. */
 116 static struct vport *get_vport_protected(struct datapath *dp, u16 port_no)
 117 {
 118         return rcu_dereference_rtnl(dp->ports[port_no]);
 119 }
 120
 121 /* Must be called with rcu_read_lock or RTNL lock. */
 122 const char *dp_name(const struct datapath *dp)
 123 {
 124         return vport_get_name(rcu_dereference_rtnl(dp->ports[OVSP_LOCAL]));
 125 }
 126
 127 static inline size_t br_nlmsg_size(void)
 128 {
 129         return NLMSG_ALIGN(sizeof(struct ifinfomsg))
 130                + nla_total_size(IFNAMSIZ) /* IFLA_IFNAME */
 131                + nla_total_size(MAX_ADDR_LEN) /* IFLA_ADDRESS */
 132                + nla_total_size(4) /* IFLA_MASTER */
 133                + nla_total_size(4) /* IFLA_MTU */
 134                + nla_total_size(4) /* IFLA_LINK */
 135                + nla_total_size(1); /* IFLA_OPERSTATE */
 136 }
 137
 138 /* Caller must hold RTNL lock. */
 139 static int dp_fill_ifinfo(struct sk_buff *skb,
 140                           const struct vport *port,
 141                           int event, unsigned int flags)
 142 {
 143         struct datapath *dp = port->dp;
 144         int ifindex = vport_get_ifindex(port);
 145         int iflink = vport_get_iflink(port);
 146         struct ifinfomsg *hdr;
 147         struct nlmsghdr *nlh;
 148
 149         if (ifindex < 0)
 150                 return ifindex;
 151
 152         if (iflink < 0)
 153                 return iflink;
 154
 155         nlh = nlmsg_put(skb, 0, 0, event, sizeof(*hdr), flags);
 156         if (nlh == NULL)
 157                 return -EMSGSIZE;
 158
 159         hdr = nlmsg_data(nlh);
 160         hdr->ifi_family = AF_BRIDGE;
 161         hdr->__ifi_pad = 0;
 162         hdr->ifi_type = ARPHRD_ETHER;
 163         hdr->ifi_index = ifindex;
 164         hdr->ifi_flags = vport_get_flags(port);
 165         hdr->ifi_change = 0;
 166
 167         NLA_PUT_STRING(skb, IFLA_IFNAME, vport_get_name(port));
 168         NLA_PUT_U32(skb, IFLA_MASTER,
 169                 vport_get_ifindex(get_vport_protected(dp, OVSP_LOCAL)));
 170         NLA_PUT_U32(skb, IFLA_MTU, vport_get_mtu(port));
 171 #ifdef IFLA_OPERSTATE
 172         NLA_PUT_U8(skb, IFLA_OPERSTATE,
 173                    vport_is_running(port)
 174                         ? vport_get_operstate(port)
 175                         : IF_OPER_DOWN);
 176 #endif
 177
 178         NLA_PUT(skb, IFLA_ADDRESS, ETH_ALEN, vport_get_addr(port));
 179
 180         if (ifindex != iflink)
 181                 NLA_PUT_U32(skb, IFLA_LINK,iflink);
 182
 183         return nlmsg_end(skb, nlh);
 184
 185 nla_put_failure:
 186         nlmsg_cancel(skb, nlh);
 187         return -EMSGSIZE;
 188 }
 189
 190 /* Caller must hold RTNL lock. */
 191 static void dp_ifinfo_notify(int event, struct vport *port)
 192 {
 193         struct sk_buff *skb;
 194         int err = -ENOBUFS;
 195
 196         skb = nlmsg_new(br_nlmsg_size(), GFP_KERNEL);
 197         if (skb == NULL)
 198                 goto errout;
 199
 200         err = dp_fill_ifinfo(skb, port, event, 0);
 201         if (err < 0) {
 202                 /* -EMSGSIZE implies BUG in br_nlmsg_size() */
 203                 WARN_ON(err == -EMSGSIZE);
 204                 kfree_skb(skb);
 205                 goto errout;
 206         }
 207         rtnl_notify(skb, &init_net, 0, RTNLGRP_LINK, NULL, GFP_KERNEL);
 208         return;
 209 errout:
 210         if (err < 0)
 211                 rtnl_set_sk_err(&init_net, RTNLGRP_LINK, err);
 212 }
 213
 214 static void release_dp(struct kobject *kobj)
 215 {
 216         struct datapath *dp = container_of(kobj, struct datapath, ifobj);
 217         kfree(dp);
 218 }
 219
 220 static struct kobj_type dp_ktype = {
 221         .release = release_dp
 222 };
 223
 224 static void destroy_dp_rcu(struct rcu_head *rcu)
 225 {
 226         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 227
 228         tbl_destroy((struct tbl __force *)dp->table, flow_free_tbl);
 229         free_percpu(dp->stats_percpu);
 230         kobject_put(&dp->ifobj);
 231 }
 232
 233 /* Called with RTNL lock and genl_lock. */
 234 static struct vport *new_vport(const struct vport_parms *parms)
 235 {
 236         struct vport *vport;
 237
 238         vport = vport_add(parms);
 239         if (!IS_ERR(vport)) {
 240                 struct datapath *dp = parms->dp;
 241
 242                 rcu_assign_pointer(dp->ports[parms->port_no], vport);
 243                 list_add(&vport->node, &dp->port_list);
 244
 245                 dp_ifinfo_notify(RTM_NEWLINK, vport);
 246         }
 247
 248         return vport;
 249 }
 250
 251 /* Called with RTNL lock. */
 252 int dp_detach_port(struct vport *p)
 253 {
 254         ASSERT_RTNL();
 255
 256         if (p->port_no != OVSP_LOCAL)
 257                 dp_sysfs_del_if(p);
 258         dp_ifinfo_notify(RTM_DELLINK, p);
 259
 260         /* First drop references to device. */
 261         list_del(&p->node);
 262         rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
 263
 264         /* Then destroy it. */
 265         return vport_del(p);
 266 }
 267
 268 /* Must be called with rcu_read_lock. */
 269 void dp_process_received_packet(struct vport *p, struct sk_buff *skb)
 270 {
 271         struct datapath *dp = p->dp;
 272         struct dp_stats_percpu *stats;
 273         int stats_counter_off;
 274         int error;
 275
 276         OVS_CB(skb)->vport = p;
 277
 278         if (!OVS_CB(skb)->flow) {
 279                 struct sw_flow_key key;
 280                 struct tbl_node *flow_node;
 281                 int key_len;
 282                 bool is_frag;
 283
 284                 /* Extract flow from 'skb' into 'key'. */
 285                 error = flow_extract(skb, p->port_no, &key, &key_len, &is_frag);
 286                 if (unlikely(error)) {
 287                         kfree_skb(skb);
 288                         return;
 289                 }
 290
 291                 if (is_frag && dp->drop_frags) {
 292                         consume_skb(skb);
 293                         stats_counter_off = offsetof(struct dp_stats_percpu, n_frags);
 294                         goto out;
 295                 }
 296
 297                 /* Look up flow. */
 298                 flow_node = tbl_lookup(rcu_dereference(dp->table), &key, key_len,
 299                                        flow_hash(&key, key_len), flow_cmp);
 300                 if (unlikely(!flow_node)) {
 301                         struct dp_upcall_info upcall;
 302
 303                         upcall.cmd = OVS_PACKET_CMD_MISS;
 304                         upcall.key = &key;
 305                         upcall.userdata = 0;
 306                         upcall.sample_pool = 0;
 307                         upcall.actions = NULL;
 308                         upcall.actions_len = 0;
 309                         dp_upcall(dp, skb, &upcall);
 310                         stats_counter_off = offsetof(struct dp_stats_percpu, n_missed);
 311                         goto out;
 312                 }
 313
 314                 OVS_CB(skb)->flow = flow_cast(flow_node);
 315         }
 316
 317         stats_counter_off = offsetof(struct dp_stats_percpu, n_hit);
 318         flow_used(OVS_CB(skb)->flow, skb);
 319         execute_actions(dp, skb);
 320
 321 out:
 322         /* Update datapath statistics. */
 323         local_bh_disable();
 324         stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
 325
 326         write_seqcount_begin(&stats->seqlock);
 327         (*(u64 *)((u8 *)stats + stats_counter_off))++;
 328         write_seqcount_end(&stats->seqlock);
 329
 330         local_bh_enable();
 331 }
 332
 333 static void copy_and_csum_skb(struct sk_buff *skb, void *to)
 334 {
 335         u16 csum_start, csum_offset;
 336         __wsum csum;
 337
 338         get_skb_csum_pointers(skb, &csum_start, &csum_offset);
 339         csum_start -= skb_headroom(skb);
 340
 341         skb_copy_bits(skb, 0, to, csum_start);
 342
 343         csum = skb_copy_and_csum_bits(skb, csum_start, to + csum_start,
 344                                       skb->len - csum_start, 0);
 345         *(__sum16 *)(to + csum_start + csum_offset) = csum_fold(csum);
 346 }
 347
 348 static struct genl_family dp_packet_genl_family = {
 349         .id = GENL_ID_GENERATE,
 350         .hdrsize = sizeof(struct ovs_header),
 351         .name = OVS_PACKET_FAMILY,
 352         .version = 1,
 353         .maxattr = OVS_PACKET_ATTR_MAX
 354 };
 355
 356 /* Generic Netlink multicast groups for upcalls.
 357  *
 358  * We really want three unique multicast groups per datapath, but we can't even
 359  * get one, because genl_register_mc_group() takes genl_lock, which is also
 360  * held during Generic Netlink message processing, so trying to acquire
 361  * multicast groups during OVS_DP_NEW processing deadlocks.  Instead, we
 362  * preallocate a few groups and use them round-robin for datapaths.  Collision
 363  * isn't fatal--multicast listeners should check that the family is the one
 364  * that they want and discard others--but it wastes time and memory to receive
 365  * unwanted messages.
 366  */
 367 #define PACKET_N_MC_GROUPS 16
 368 static struct genl_multicast_group packet_mc_groups[PACKET_N_MC_GROUPS];
 369
 370 static u32 packet_mc_group(struct datapath *dp, u8 cmd)
 371 {
 372         u32 idx;
 373         BUILD_BUG_ON_NOT_POWER_OF_2(PACKET_N_MC_GROUPS);
 374
 375         idx = jhash_2words(dp->dp_ifindex, cmd, 0) & (PACKET_N_MC_GROUPS - 1);
 376         return packet_mc_groups[idx].id;
 377 }
 378
 379 static int packet_register_mc_groups(void)
 380 {
 381         int i;
 382
 383         for (i = 0; i < PACKET_N_MC_GROUPS; i++) {
 384                 struct genl_multicast_group *group = &packet_mc_groups[i];
 385                 int error;
 386
 387                 sprintf(group->name, "packet%d", i);
 388                 error = genl_register_mc_group(&dp_packet_genl_family, group);
 389                 if (error)
 390                         return error;
 391         }
 392         return 0;
 393 }
 394
 395 int dp_upcall(struct datapath *dp, struct sk_buff *skb, const struct dp_upcall_info *upcall_info)
 396 {
 397         struct dp_stats_percpu *stats;
 398         int err;
 399
 400         WARN_ON_ONCE(skb_shared(skb));
 401
 402         forward_ip_summed(skb, true);
 403
 404         /* Break apart GSO packets into their component pieces.  Otherwise
 405          * userspace may try to stuff a 64kB packet into a 1500-byte MTU. */
 406         if (skb_is_gso(skb)) {
 407                 struct sk_buff *nskb = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
 408
 409                 if (IS_ERR(nskb)) {
 410                         kfree_skb(skb);
 411                         err = PTR_ERR(nskb);
 412                         goto err;
 413                 }
 414                 consume_skb(skb);
 415                 skb = nskb;
 416         }
 417
 418         err = queue_userspace_packets(dp, skb, upcall_info);
 419         if (err)
 420                 goto err;
 421
 422         return 0;
 423
 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_userspace_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 ovs_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 ovs_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, OVS_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, OVS_PACKET_ATTR_USERDATA, upcall_info->userdata);
 489                 if (upcall_info->sample_pool)
 490                         nla_put_u32(user_skb, OVS_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, OVS_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, OVS_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                 consume_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[OVS_ACTION_ATTR_MAX + 1] = {
 554                         [OVS_ACTION_ATTR_OUTPUT] = 4,
 555                         [OVS_ACTION_ATTR_USERSPACE] = 8,
 556                         [OVS_ACTION_ATTR_SET_DL_TCI] = 2,
 557                         [OVS_ACTION_ATTR_STRIP_VLAN] = 0,
 558                         [OVS_ACTION_ATTR_SET_DL_SRC] = ETH_ALEN,
 559                         [OVS_ACTION_ATTR_SET_DL_DST] = ETH_ALEN,
 560                         [OVS_ACTION_ATTR_SET_NW_SRC] = 4,
 561                         [OVS_ACTION_ATTR_SET_NW_DST] = 4,
 562                         [OVS_ACTION_ATTR_SET_NW_TOS] = 1,
 563                         [OVS_ACTION_ATTR_SET_TP_SRC] = 2,
 564                         [OVS_ACTION_ATTR_SET_TP_DST] = 2,
 565                         [OVS_ACTION_ATTR_SET_TUNNEL] = 8,
 566                         [OVS_ACTION_ATTR_SET_PRIORITY] = 4,
 567                         [OVS_ACTION_ATTR_POP_PRIORITY] = 0,
 568                 };
 569                 int type = nla_type(a);
 570
 571                 if (type > OVS_ACTION_ATTR_MAX || nla_len(a) != action_lens[type])
 572                         return -EINVAL;
 573
 574                 switch (type) {
 575                 case OVS_ACTION_ATTR_UNSPEC:
 576                         return -EINVAL;
 577
 578                 case OVS_ACTION_ATTR_USERSPACE:
 579                 case OVS_ACTION_ATTR_STRIP_VLAN:
 580                 case OVS_ACTION_ATTR_SET_DL_SRC:
 581                 case OVS_ACTION_ATTR_SET_DL_DST:
 582                 case OVS_ACTION_ATTR_SET_NW_SRC:
 583                 case OVS_ACTION_ATTR_SET_NW_DST:
 584                 case OVS_ACTION_ATTR_SET_TP_SRC:
 585                 case OVS_ACTION_ATTR_SET_TP_DST:
 586                 case OVS_ACTION_ATTR_SET_TUNNEL:
 587                 case OVS_ACTION_ATTR_SET_PRIORITY:
 588                 case OVS_ACTION_ATTR_POP_PRIORITY:
 589                         /* No validation needed. */
 590                         break;
 591
 592                 case OVS_ACTION_ATTR_OUTPUT:
 593                         if (nla_get_u32(a) >= DP_MAX_PORTS)
 594                                 return -EINVAL;
 595                         break;
 596
 597                 case OVS_ACTION_ATTR_SET_DL_TCI:
 598                         if (nla_get_be16(a) & htons(VLAN_CFI_MASK))
 599                                 return -EINVAL;
 600                         break;
 601
 602                 case OVS_ACTION_ATTR_SET_NW_TOS:
 603                         if (nla_get_u8(a) & INET_ECN_MASK)
 604                                 return -EINVAL;
 605                         break;
 606
 607                 default:
 608                         return -EOPNOTSUPP;
 609                 }
 610         }
 611
 612         if (rem > 0)
 613                 return -EINVAL;
 614
 615         return 0;
 616 }
 617 static void clear_stats(struct sw_flow *flow)
 618 {
 619         flow->used = 0;
 620         flow->tcp_flags = 0;
 621         flow->packet_count = 0;
 622         flow->byte_count = 0;
 623 }
 624
 625 /* Called with genl_mutex. */
 626 static int expand_table(struct datapath *dp)
 627 {
 628         struct tbl *old_table = get_table_protected(dp);
 629         struct tbl *new_table;
 630
 631         new_table = tbl_expand(old_table);
 632         if (IS_ERR(new_table)) {
 633                 if (PTR_ERR(new_table) != -ENOSPC)
 634                         return PTR_ERR(new_table);
 635         } else {
 636                 rcu_assign_pointer(dp->table, new_table);
 637                 tbl_deferred_destroy(old_table, NULL);
 638         }
 639
 640         return 0;
 641 }
 642
 643 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 644 {
 645         struct ovs_header *ovs_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[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 659             !a[OVS_PACKET_ATTR_ACTIONS] ||
 660             nla_len(a[OVS_PACKET_ATTR_PACKET]) < ETH_HLEN)
 661                 goto err;
 662
 663         err = validate_actions(a[OVS_PACKET_ATTR_ACTIONS]);
 664         if (err)
 665                 goto err;
 666
 667         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 668         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 669         err = -ENOMEM;
 670         if (!packet)
 671                 goto err;
 672         skb_reserve(packet, NET_IP_ALIGN);
 673
 674         memcpy(__skb_put(packet, len), nla_data(a[OVS_PACKET_ATTR_PACKET]), len);
 675
 676         skb_reset_mac_header(packet);
 677         eth = eth_hdr(packet);
 678
 679         /* Normally, setting the skb 'protocol' field would be handled by a
 680          * call to eth_type_trans(), but it assumes there's a sending
 681          * device, which we may not have. */
 682         if (ntohs(eth->h_proto) >= 1536)
 683                 packet->protocol = eth->h_proto;
 684         else
 685                 packet->protocol = htons(ETH_P_802_2);
 686
 687         /* Build an sw_flow for sending this packet. */
 688         flow = flow_alloc();
 689         err = PTR_ERR(flow);
 690         if (IS_ERR(flow))
 691                 goto err_kfree_skb;
 692
 693         err = flow_extract(packet, -1, &flow->key, &key_len, &is_frag);
 694         if (err)
 695                 goto err_flow_put;
 696         flow->tbl_node.hash = flow_hash(&flow->key, key_len);
 697
 698         err = flow_metadata_from_nlattrs(&flow->key.eth.in_port,
 699                                          &flow->key.eth.tun_id,
 700                                          a[OVS_PACKET_ATTR_KEY]);
 701         if (err)
 702                 goto err_flow_put;
 703
 704         acts = flow_actions_alloc(a[OVS_PACKET_ATTR_ACTIONS]);
 705         err = PTR_ERR(acts);
 706         if (IS_ERR(acts))
 707                 goto err_flow_put;
 708         rcu_assign_pointer(flow->sf_acts, acts);
 709
 710         OVS_CB(packet)->flow = flow;
 711
 712         rcu_read_lock();
 713         dp = get_dp(ovs_header->dp_ifindex);
 714         err = -ENODEV;
 715         if (!dp)
 716                 goto err_unlock;
 717         err = execute_actions(dp, packet);
 718         rcu_read_unlock();
 719
 720         flow_put(flow);
 721         return err;
 722
 723 err_unlock:
 724         rcu_read_unlock();
 725 err_flow_put:
 726         flow_put(flow);
 727 err_kfree_skb:
 728         kfree_skb(packet);
 729 err:
 730         return err;
 731 }
 732
 733 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 734         [OVS_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
 735         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 736         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 737 };
 738
 739 static struct genl_ops dp_packet_genl_ops[] = {
 740         { .cmd = OVS_PACKET_CMD_EXECUTE,
 741           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 742           .policy = packet_policy,
 743           .doit = ovs_packet_cmd_execute
 744         }
 745 };
 746
 747 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
 748 {
 749         int i;
 750         struct tbl *table = get_table_protected(dp);
 751
 752         stats->n_flows = tbl_count(table);
 753
 754         stats->n_frags = stats->n_hit = stats->n_missed = stats->n_lost = 0;
 755         for_each_possible_cpu(i) {
 756                 const struct dp_stats_percpu *percpu_stats;
 757                 struct dp_stats_percpu local_stats;
 758                 unsigned seqcount;
 759
 760                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 761
 762                 do {
 763                         seqcount = read_seqcount_begin(&percpu_stats->seqlock);
 764                         local_stats = *percpu_stats;
 765                 } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount));
 766
 767                 stats->n_frags += local_stats.n_frags;
 768                 stats->n_hit += local_stats.n_hit;
 769                 stats->n_missed += local_stats.n_missed;
 770                 stats->n_lost += local_stats.n_lost;
 771         }
 772 }
 773
 774 /* MTU of the dp pseudo-device: ETH_DATA_LEN or the minimum of the ports.
 775  * Called with RTNL lock.
 776  */
 777 int dp_min_mtu(const struct datapath *dp)
 778 {
 779         struct vport *p;
 780         int mtu = 0;
 781
 782         ASSERT_RTNL();
 783
 784         list_for_each_entry (p, &dp->port_list, node) {
 785                 int dev_mtu;
 786
 787                 /* Skip any internal ports, since that's what we're trying to
 788                  * set. */
 789                 if (is_internal_vport(p))
 790                         continue;
 791
 792                 dev_mtu = vport_get_mtu(p);
 793                 if (!dev_mtu)
 794                         continue;
 795                 if (!mtu || dev_mtu < mtu)
 796                         mtu = dev_mtu;
 797         }
 798
 799         return mtu ? mtu : ETH_DATA_LEN;
 800 }
 801
 802 /* Sets the MTU of all datapath devices to the minimum of the ports
 803  * Called with RTNL lock.
 804  */
 805 void set_internal_devs_mtu(const struct datapath *dp)
 806 {
 807         struct vport *p;
 808         int mtu;
 809
 810         ASSERT_RTNL();
 811
 812         mtu = dp_min_mtu(dp);
 813
 814         list_for_each_entry (p, &dp->port_list, node) {
 815                 if (is_internal_vport(p))
 816                         vport_set_mtu(p, mtu);
 817         }
 818 }
 819
 820 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
 821         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
 822         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
 823         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
 824 };
 825
 826 static struct genl_family dp_flow_genl_family = {
 827         .id = GENL_ID_GENERATE,
 828         .hdrsize = sizeof(struct ovs_header),
 829         .name = OVS_FLOW_FAMILY,
 830         .version = 1,
 831         .maxattr = OVS_FLOW_ATTR_MAX
 832 };
 833
 834 static struct genl_multicast_group dp_flow_multicast_group = {
 835         .name = OVS_FLOW_MCGROUP
 836 };
 837
 838 /* Called with genl_lock. */
 839 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
 840                                   struct sk_buff *skb, u32 pid, u32 seq, u32 flags, u8 cmd)
 841 {
 842         const int skb_orig_len = skb->len;
 843         const struct sw_flow_actions *sf_acts;
 844         struct ovs_flow_stats stats;
 845         struct ovs_header *ovs_header;
 846         struct nlattr *nla;
 847         unsigned long used;
 848         u8 tcp_flags;
 849         int err;
 850
 851         sf_acts = rcu_dereference_protected(flow->sf_acts,
 852                                             lockdep_genl_is_held());
 853
 854         ovs_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags, cmd);
 855         if (!ovs_header)
 856                 return -EMSGSIZE;
 857
 858         ovs_header->dp_ifindex = dp->dp_ifindex;
 859
 860         nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
 861         if (!nla)
 862                 goto nla_put_failure;
 863         err = flow_to_nlattrs(&flow->key, skb);
 864         if (err)
 865                 goto error;
 866         nla_nest_end(skb, nla);
 867
 868         spin_lock_bh(&flow->lock);
 869         used = flow->used;
 870         stats.n_packets = flow->packet_count;
 871         stats.n_bytes = flow->byte_count;
 872         tcp_flags = flow->tcp_flags;
 873         spin_unlock_bh(&flow->lock);
 874
 875         if (used)
 876                 NLA_PUT_U64(skb, OVS_FLOW_ATTR_USED, flow_used_time(used));
 877
 878         if (stats.n_packets)
 879                 NLA_PUT(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats);
 880
 881         if (tcp_flags)
 882                 NLA_PUT_U8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags);
 883
 884         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 885          * this is the first flow to be dumped into 'skb'.  This is unusual for
 886          * Netlink but individual action lists can be longer than
 887          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 888          * The userspace caller can always fetch the actions separately if it
 889          * really wants them.  (Most userspace callers in fact don't care.)
 890          *
 891          * This can only fail for dump operations because the skb is always
 892          * properly sized for single flows.
 893          */
 894         err = nla_put(skb, OVS_FLOW_ATTR_ACTIONS, sf_acts->actions_len,
 895                       sf_acts->actions);
 896         if (err < 0 && skb_orig_len)
 897                 goto error;
 898
 899         return genlmsg_end(skb, ovs_header);
 900
 901 nla_put_failure:
 902         err = -EMSGSIZE;
 903 error:
 904         genlmsg_cancel(skb, ovs_header);
 905         return err;
 906 }
 907
 908 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
 909 {
 910         const struct sw_flow_actions *sf_acts;
 911         int len;
 912
 913         sf_acts = rcu_dereference_protected(flow->sf_acts,
 914                                             lockdep_genl_is_held());
 915
 916         len = nla_total_size(FLOW_BUFSIZE); /* OVS_FLOW_ATTR_KEY */
 917         len += nla_total_size(sf_acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
 918         len += nla_total_size(sizeof(struct ovs_flow_stats)); /* OVS_FLOW_ATTR_STATS */
 919         len += nla_total_size(1); /* OVS_FLOW_ATTR_TCP_FLAGS */
 920         len += nla_total_size(8); /* OVS_FLOW_ATTR_USED */
 921         return genlmsg_new(NLMSG_ALIGN(sizeof(struct ovs_header)) + len, GFP_KERNEL);
 922 }
 923
 924 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow, struct datapath *dp,
 925                                                u32 pid, u32 seq, u8 cmd)
 926 {
 927         struct sk_buff *skb;
 928         int retval;
 929
 930         skb = ovs_flow_cmd_alloc_info(flow);
 931         if (!skb)
 932                 return ERR_PTR(-ENOMEM);
 933
 934         retval = ovs_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd);
 935         BUG_ON(retval < 0);
 936         return skb;
 937 }
 938
 939 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
 940 {
 941         struct nlattr **a = info->attrs;
 942         struct ovs_header *ovs_header = info->userhdr;
 943         struct tbl_node *flow_node;
 944         struct sw_flow_key key;
 945         struct sw_flow *flow;
 946         struct sk_buff *reply;
 947         struct datapath *dp;
 948         struct tbl *table;
 949         u32 hash;
 950         int error;
 951         int key_len;
 952
 953         /* Extract key. */
 954         error = -EINVAL;
 955         if (!a[OVS_FLOW_ATTR_KEY])
 956                 goto error;
 957         error = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
 958         if (error)
 959                 goto error;
 960
 961         /* Validate actions. */
 962         if (a[OVS_FLOW_ATTR_ACTIONS]) {
 963                 error = validate_actions(a[OVS_FLOW_ATTR_ACTIONS]);
 964                 if (error)
 965                         goto error;
 966         } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
 967                 error = -EINVAL;
 968                 goto error;
 969         }
 970
 971         dp = get_dp(ovs_header->dp_ifindex);
 972         error = -ENODEV;
 973         if (!dp)
 974                 goto error;
 975
 976         hash = flow_hash(&key, key_len);
 977         table = get_table_protected(dp);
 978         flow_node = tbl_lookup(table, &key, key_len, hash, flow_cmp);
 979         if (!flow_node) {
 980                 struct sw_flow_actions *acts;
 981
 982                 /* Bail out if we're not allowed to create a new flow. */
 983                 error = -ENOENT;
 984                 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
 985                         goto error;
 986
 987                 /* Expand table, if necessary, to make room. */
 988                 if (tbl_count(table) >= tbl_n_buckets(table)) {
 989                         error = expand_table(dp);
 990                         if (error)
 991                                 goto error;
 992                         table = get_table_protected(dp);
 993                 }
 994
 995                 /* Allocate flow. */
 996                 flow = flow_alloc();
 997                 if (IS_ERR(flow)) {
 998                         error = PTR_ERR(flow);
 999                         goto error;
1000                 }
1001                 flow->key = key;
1002                 clear_stats(flow);
1003
1004                 /* Obtain actions. */
1005                 acts = flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]);
1006                 error = PTR_ERR(acts);
1007                 if (IS_ERR(acts))
1008                         goto error_free_flow;
1009                 rcu_assign_pointer(flow->sf_acts, acts);
1010
1011                 /* Put flow in bucket. */
1012                 error = tbl_insert(table, &flow->tbl_node, hash);
1013                 if (error)
1014                         goto error_free_flow;
1015
1016                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1017                                                 info->snd_seq, OVS_FLOW_CMD_NEW);
1018         } else {
1019                 /* We found a matching flow. */
1020                 struct sw_flow_actions *old_acts;
1021
1022                 /* Bail out if we're not allowed to modify an existing flow.
1023                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1024                  * because Generic Netlink treats the latter as a dump
1025                  * request.  We also accept NLM_F_EXCL in case that bug ever
1026                  * gets fixed.
1027                  */
1028                 error = -EEXIST;
1029                 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
1030                     info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1031                         goto error;
1032
1033                 /* Update actions. */
1034                 flow = flow_cast(flow_node);
1035                 old_acts = rcu_dereference_protected(flow->sf_acts,
1036                                                      lockdep_genl_is_held());
1037                 if (a[OVS_FLOW_ATTR_ACTIONS] &&
1038                     (old_acts->actions_len != nla_len(a[OVS_FLOW_ATTR_ACTIONS]) ||
1039                      memcmp(old_acts->actions, nla_data(a[OVS_FLOW_ATTR_ACTIONS]),
1040                             old_acts->actions_len))) {
1041                         struct sw_flow_actions *new_acts;
1042
1043                         new_acts = flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]);
1044                         error = PTR_ERR(new_acts);
1045                         if (IS_ERR(new_acts))
1046                                 goto error;
1047
1048                         rcu_assign_pointer(flow->sf_acts, new_acts);
1049                         flow_deferred_free_acts(old_acts);
1050                 }
1051
1052                 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1053                                                 info->snd_seq, OVS_FLOW_CMD_NEW);
1054
1055                 /* Clear stats. */
1056                 if (a[OVS_FLOW_ATTR_CLEAR]) {
1057                         spin_lock_bh(&flow->lock);
1058                         clear_stats(flow);
1059                         spin_unlock_bh(&flow->lock);
1060                 }
1061         }
1062
1063         if (!IS_ERR(reply))
1064                 genl_notify(reply, genl_info_net(info), info->snd_pid,
1065                             dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1066         else
1067                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1068                                 dp_flow_multicast_group.id, PTR_ERR(reply));
1069         return 0;
1070
1071 error_free_flow:
1072         flow_put(flow);
1073 error:
1074         return error;
1075 }
1076
1077 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1078 {
1079         struct nlattr **a = info->attrs;
1080         struct ovs_header *ovs_header = info->userhdr;
1081         struct sw_flow_key key;
1082         struct tbl_node *flow_node;
1083         struct sk_buff *reply;
1084         struct sw_flow *flow;
1085         struct datapath *dp;
1086         struct tbl *table;
1087         int err;
1088         int key_len;
1089
1090         if (!a[OVS_FLOW_ATTR_KEY])
1091                 return -EINVAL;
1092         err = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1093         if (err)
1094                 return err;
1095
1096         dp = get_dp(ovs_header->dp_ifindex);
1097         if (!dp)
1098                 return -ENODEV;
1099
1100         table = get_table_protected(dp);
1101         flow_node = tbl_lookup(table, &key, key_len, flow_hash(&key, key_len),
1102                                flow_cmp);
1103         if (!flow_node)
1104                 return -ENOENT;
1105
1106         flow = flow_cast(flow_node);
1107         reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid, info->snd_seq, OVS_FLOW_CMD_NEW);
1108         if (IS_ERR(reply))
1109                 return PTR_ERR(reply);
1110
1111         return genlmsg_reply(reply, info);
1112 }
1113
1114 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1115 {
1116         struct nlattr **a = info->attrs;
1117         struct ovs_header *ovs_header = info->userhdr;
1118         struct sw_flow_key key;
1119         struct tbl_node *flow_node;
1120         struct sk_buff *reply;
1121         struct sw_flow *flow;
1122         struct datapath *dp;
1123         struct tbl *table;
1124         int err;
1125         int key_len;
1126
1127         if (!a[OVS_FLOW_ATTR_KEY])
1128                 return flush_flows(ovs_header->dp_ifindex);
1129         err = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1130         if (err)
1131                 return err;
1132
1133         dp = get_dp(ovs_header->dp_ifindex);
1134         if (!dp)
1135                 return -ENODEV;
1136
1137         table = get_table_protected(dp);
1138         flow_node = tbl_lookup(table, &key, key_len, flow_hash(&key, key_len),
1139                                flow_cmp);
1140         if (!flow_node)
1141                 return -ENOENT;
1142         flow = flow_cast(flow_node);
1143
1144         reply = ovs_flow_cmd_alloc_info(flow);
1145         if (!reply)
1146                 return -ENOMEM;
1147
1148         err = tbl_remove(table, flow_node);
1149         if (err) {
1150                 kfree_skb(reply);
1151                 return err;
1152         }
1153
1154         err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_pid,
1155                                      info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1156         BUG_ON(err < 0);
1157
1158         flow_deferred_free(flow);
1159
1160         genl_notify(reply, genl_info_net(info), info->snd_pid,
1161                     dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1162         return 0;
1163 }
1164
1165 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1166 {
1167         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1168         struct datapath *dp;
1169
1170         dp = get_dp(ovs_header->dp_ifindex);
1171         if (!dp)
1172                 return -ENODEV;
1173
1174         for (;;) {
1175                 struct tbl_node *flow_node;
1176                 struct sw_flow *flow;
1177                 u32 bucket, obj;
1178
1179                 bucket = cb->args[0];
1180                 obj = cb->args[1];
1181                 flow_node = tbl_next(get_table_protected(dp), &bucket, &obj);
1182                 if (!flow_node)
1183                         break;
1184
1185                 flow = flow_cast(flow_node);
1186                 if (ovs_flow_cmd_fill_info(flow, dp, skb, NETLINK_CB(cb->skb).pid,
1187                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1188                                            OVS_FLOW_CMD_NEW) < 0)
1189                         break;
1190
1191                 cb->args[0] = bucket;
1192                 cb->args[1] = obj;
1193         }
1194         return skb->len;
1195 }
1196
1197 static struct genl_ops dp_flow_genl_ops[] = {
1198         { .cmd = OVS_FLOW_CMD_NEW,
1199           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1200           .policy = flow_policy,
1201           .doit = ovs_flow_cmd_new_or_set
1202         },
1203         { .cmd = OVS_FLOW_CMD_DEL,
1204           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1205           .policy = flow_policy,
1206           .doit = ovs_flow_cmd_del
1207         },
1208         { .cmd = OVS_FLOW_CMD_GET,
1209           .flags = 0,               /* OK for unprivileged users. */
1210           .policy = flow_policy,
1211           .doit = ovs_flow_cmd_get,
1212           .dumpit = ovs_flow_cmd_dump
1213         },
1214         { .cmd = OVS_FLOW_CMD_SET,
1215           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1216           .policy = flow_policy,
1217           .doit = ovs_flow_cmd_new_or_set,
1218         },
1219 };
1220
1221 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1222 #ifdef HAVE_NLA_NUL_STRING
1223         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1224 #endif
1225         [OVS_DP_ATTR_IPV4_FRAGS] = { .type = NLA_U32 },
1226         [OVS_DP_ATTR_SAMPLING] = { .type = NLA_U32 },
1227 };
1228
1229 static struct genl_family dp_datapath_genl_family = {
1230         .id = GENL_ID_GENERATE,
1231         .hdrsize = sizeof(struct ovs_header),
1232         .name = OVS_DATAPATH_FAMILY,
1233         .version = 1,
1234         .maxattr = OVS_DP_ATTR_MAX
1235 };
1236
1237 static struct genl_multicast_group dp_datapath_multicast_group = {
1238         .name = OVS_DATAPATH_MCGROUP
1239 };
1240
1241 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1242                                 u32 pid, u32 seq, u32 flags, u8 cmd)
1243 {
1244         struct ovs_header *ovs_header;
1245         struct nlattr *nla;
1246         int err;
1247
1248         ovs_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family,
1249                                    flags, cmd);
1250         if (!ovs_header)
1251                 goto error;
1252
1253         ovs_header->dp_ifindex = dp->dp_ifindex;
1254
1255         rcu_read_lock();
1256         err = nla_put_string(skb, OVS_DP_ATTR_NAME, dp_name(dp));
1257         rcu_read_unlock();
1258         if (err)
1259                 goto nla_put_failure;
1260
1261         nla = nla_reserve(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats));
1262         if (!nla)
1263                 goto nla_put_failure;
1264         get_dp_stats(dp, nla_data(nla));
1265
1266         NLA_PUT_U32(skb, OVS_DP_ATTR_IPV4_FRAGS,
1267                     dp->drop_frags ? OVS_DP_FRAG_DROP : OVS_DP_FRAG_ZERO);
1268
1269         if (dp->sflow_probability)
1270                 NLA_PUT_U32(skb, OVS_DP_ATTR_SAMPLING, dp->sflow_probability);
1271
1272         nla = nla_nest_start(skb, OVS_DP_ATTR_MCGROUPS);
1273         if (!nla)
1274                 goto nla_put_failure;
1275         NLA_PUT_U32(skb, OVS_PACKET_CMD_MISS, packet_mc_group(dp, OVS_PACKET_CMD_MISS));
1276         NLA_PUT_U32(skb, OVS_PACKET_CMD_ACTION, packet_mc_group(dp, OVS_PACKET_CMD_ACTION));
1277         NLA_PUT_U32(skb, OVS_PACKET_CMD_SAMPLE, packet_mc_group(dp, OVS_PACKET_CMD_SAMPLE));
1278         nla_nest_end(skb, nla);
1279
1280         return genlmsg_end(skb, ovs_header);
1281
1282 nla_put_failure:
1283         genlmsg_cancel(skb, ovs_header);
1284 error:
1285         return -EMSGSIZE;
1286 }
1287
1288 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 pid,
1289                                              u32 seq, u8 cmd)
1290 {
1291         struct sk_buff *skb;
1292         int retval;
1293
1294         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1295         if (!skb)
1296                 return ERR_PTR(-ENOMEM);
1297
1298         retval = ovs_dp_cmd_fill_info(dp, skb, pid, seq, 0, cmd);
1299         if (retval < 0) {
1300                 kfree_skb(skb);
1301                 return ERR_PTR(retval);
1302         }
1303         return skb;
1304 }
1305
1306 static int ovs_dp_cmd_validate(struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1307 {
1308         if (a[OVS_DP_ATTR_IPV4_FRAGS]) {
1309                 u32 frags = nla_get_u32(a[OVS_DP_ATTR_IPV4_FRAGS]);
1310
1311                 if (frags != OVS_DP_FRAG_ZERO && frags != OVS_DP_FRAG_DROP)
1312                         return -EINVAL;
1313         }
1314
1315         return CHECK_NUL_STRING(a[OVS_DP_ATTR_NAME], IFNAMSIZ - 1);
1316 }
1317
1318 /* Called with genl_mutex and optionally with RTNL lock also. */
1319 static struct datapath *lookup_datapath(struct ovs_header *ovs_header, struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1320 {
1321         struct datapath *dp;
1322
1323         if (!a[OVS_DP_ATTR_NAME])
1324                 dp = get_dp(ovs_header->dp_ifindex);
1325         else {
1326                 struct vport *vport;
1327
1328                 rcu_read_lock();
1329                 vport = vport_locate(nla_data(a[OVS_DP_ATTR_NAME]));
1330                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1331                 rcu_read_unlock();
1332         }
1333         return dp ? dp : ERR_PTR(-ENODEV);
1334 }
1335
1336 /* Called with genl_mutex. */
1337 static void change_datapath(struct datapath *dp, struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1338 {
1339         if (a[OVS_DP_ATTR_IPV4_FRAGS])
1340                 dp->drop_frags = nla_get_u32(a[OVS_DP_ATTR_IPV4_FRAGS]) == OVS_DP_FRAG_DROP;
1341         if (a[OVS_DP_ATTR_SAMPLING])
1342                 dp->sflow_probability = nla_get_u32(a[OVS_DP_ATTR_SAMPLING]);
1343 }
1344
1345 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1346 {
1347         struct nlattr **a = info->attrs;
1348         struct vport_parms parms;
1349         struct sk_buff *reply;
1350         struct datapath *dp;
1351         struct vport *vport;
1352         int err;
1353
1354         err = -EINVAL;
1355         if (!a[OVS_DP_ATTR_NAME])
1356                 goto err;
1357
1358         err = ovs_dp_cmd_validate(a);
1359         if (err)
1360                 goto err;
1361
1362         rtnl_lock();
1363         err = -ENODEV;
1364         if (!try_module_get(THIS_MODULE))
1365                 goto err_unlock_rtnl;
1366
1367         err = -ENOMEM;
1368         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1369         if (dp == NULL)
1370                 goto err_put_module;
1371         INIT_LIST_HEAD(&dp->port_list);
1372
1373         /* Initialize kobject for bridge.  This will be added as
1374          * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
1375         dp->ifobj.kset = NULL;
1376         kobject_init(&dp->ifobj, &dp_ktype);
1377
1378         /* Allocate table. */
1379         err = -ENOMEM;
1380         rcu_assign_pointer(dp->table, tbl_create(TBL_MIN_BUCKETS));
1381         if (!dp->table)
1382                 goto err_free_dp;
1383
1384         /* Set up our datapath device. */
1385         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1386         parms.type = OVS_VPORT_TYPE_INTERNAL;
1387         parms.options = NULL;
1388         parms.dp = dp;
1389         parms.port_no = OVSP_LOCAL;
1390         vport = new_vport(&parms);
1391         if (IS_ERR(vport)) {
1392                 err = PTR_ERR(vport);
1393                 if (err == -EBUSY)
1394                         err = -EEXIST;
1395
1396                 goto err_destroy_table;
1397         }
1398         dp->dp_ifindex = vport_get_ifindex(vport);
1399
1400         dp->drop_frags = 0;
1401         dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1402         if (!dp->stats_percpu) {
1403                 err = -ENOMEM;
1404                 goto err_destroy_local_port;
1405         }
1406
1407         change_datapath(dp, a);
1408
1409         reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_NEW);
1410         err = PTR_ERR(reply);
1411         if (IS_ERR(reply))
1412                 goto err_destroy_local_port;
1413
1414         list_add_tail(&dp->list_node, &dps);
1415         dp_sysfs_add_dp(dp);
1416
1417         rtnl_unlock();
1418
1419         genl_notify(reply, genl_info_net(info), info->snd_pid,
1420                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1421         return 0;
1422
1423 err_destroy_local_port:
1424         dp_detach_port(get_vport_protected(dp, OVSP_LOCAL));
1425 err_destroy_table:
1426         tbl_destroy(get_table_protected(dp), NULL);
1427 err_free_dp:
1428         kfree(dp);
1429 err_put_module:
1430         module_put(THIS_MODULE);
1431 err_unlock_rtnl:
1432         rtnl_unlock();
1433 err:
1434         return err;
1435 }
1436
1437 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1438 {
1439         struct vport *vport, *next_vport;
1440         struct sk_buff *reply;
1441         struct datapath *dp;
1442         int err;
1443
1444         err = ovs_dp_cmd_validate(info->attrs);
1445         if (err)
1446                 goto exit;
1447
1448         rtnl_lock();
1449         dp = lookup_datapath(info->userhdr, info->attrs);
1450         err = PTR_ERR(dp);
1451         if (IS_ERR(dp))
1452                 goto exit_unlock;
1453
1454         reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_DEL);
1455         err = PTR_ERR(reply);
1456         if (IS_ERR(reply))
1457                 goto exit_unlock;
1458
1459         list_for_each_entry_safe (vport, next_vport, &dp->port_list, node)
1460                 if (vport->port_no != OVSP_LOCAL)
1461                         dp_detach_port(vport);
1462
1463         dp_sysfs_del_dp(dp);
1464         list_del(&dp->list_node);
1465         dp_detach_port(get_vport_protected(dp, OVSP_LOCAL));
1466
1467         /* rtnl_unlock() will wait until all the references to devices that
1468          * are pending unregistration have been dropped.  We do it here to
1469          * ensure that any internal devices (which contain DP pointers) are
1470          * fully destroyed before freeing the datapath.
1471          */
1472         rtnl_unlock();
1473
1474         call_rcu(&dp->rcu, destroy_dp_rcu);
1475         module_put(THIS_MODULE);
1476
1477         genl_notify(reply, genl_info_net(info), info->snd_pid,
1478                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1479
1480         return 0;
1481
1482 exit_unlock:
1483         rtnl_unlock();
1484 exit:
1485         return err;
1486 }
1487
1488 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1489 {
1490         struct sk_buff *reply;
1491         struct datapath *dp;
1492         int err;
1493
1494         err = ovs_dp_cmd_validate(info->attrs);
1495         if (err)
1496                 return err;
1497
1498         dp = lookup_datapath(info->userhdr, info->attrs);
1499         if (IS_ERR(dp))
1500                 return PTR_ERR(dp);
1501
1502         change_datapath(dp, info->attrs);
1503
1504         reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_NEW);
1505         if (IS_ERR(reply)) {
1506                 err = PTR_ERR(reply);
1507                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1508                                 dp_datapath_multicast_group.id, err);
1509                 return 0;
1510         }
1511
1512         genl_notify(reply, genl_info_net(info), info->snd_pid,
1513                     dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1514         return 0;
1515 }
1516
1517 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1518 {
1519         struct sk_buff *reply;
1520         struct datapath *dp;
1521         int err;
1522
1523         err = ovs_dp_cmd_validate(info->attrs);
1524         if (err)
1525                 return err;
1526
1527         dp = lookup_datapath(info->userhdr, info->attrs);
1528         if (IS_ERR(dp))
1529                 return PTR_ERR(dp);
1530
1531         reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_NEW);
1532         if (IS_ERR(reply))
1533                 return PTR_ERR(reply);
1534
1535         return genlmsg_reply(reply, info);
1536 }
1537
1538 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1539 {
1540         struct datapath *dp;
1541         int skip = cb->args[0];
1542         int i = 0;
1543
1544         list_for_each_entry (dp, &dps, list_node) {
1545                 if (i < skip)
1546                         continue;
1547                 if (ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).pid,
1548                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1549                                          OVS_DP_CMD_NEW) < 0)
1550                         break;
1551                 i++;
1552         }
1553
1554         cb->args[0] = i;
1555
1556         return skb->len;
1557 }
1558
1559 static struct genl_ops dp_datapath_genl_ops[] = {
1560         { .cmd = OVS_DP_CMD_NEW,
1561           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1562           .policy = datapath_policy,
1563           .doit = ovs_dp_cmd_new
1564         },
1565         { .cmd = OVS_DP_CMD_DEL,
1566           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1567           .policy = datapath_policy,
1568           .doit = ovs_dp_cmd_del
1569         },
1570         { .cmd = OVS_DP_CMD_GET,
1571           .flags = 0,               /* OK for unprivileged users. */
1572           .policy = datapath_policy,
1573           .doit = ovs_dp_cmd_get,
1574           .dumpit = ovs_dp_cmd_dump
1575         },
1576         { .cmd = OVS_DP_CMD_SET,
1577           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1578           .policy = datapath_policy,
1579           .doit = ovs_dp_cmd_set,
1580         },
1581 };
1582
1583 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1584 #ifdef HAVE_NLA_NUL_STRING
1585         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1586         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1587         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1588         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct rtnl_link_stats64) },
1589         [OVS_VPORT_ATTR_ADDRESS] = { .len = ETH_ALEN },
1590 #else
1591         [OVS_VPORT_ATTR_STATS] = { .minlen = sizeof(struct rtnl_link_stats64) },
1592         [OVS_VPORT_ATTR_ADDRESS] = { .minlen = ETH_ALEN },
1593 #endif
1594         [OVS_VPORT_ATTR_MTU] = { .type = NLA_U32 },
1595         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1596 };
1597
1598 static struct genl_family dp_vport_genl_family = {
1599         .id = GENL_ID_GENERATE,
1600         .hdrsize = sizeof(struct ovs_header),
1601         .name = OVS_VPORT_FAMILY,
1602         .version = 1,
1603         .maxattr = OVS_VPORT_ATTR_MAX
1604 };
1605
1606 static struct genl_multicast_group dp_vport_multicast_group = {
1607         .name = OVS_VPORT_MCGROUP
1608 };
1609
1610 /* Called with RTNL lock or RCU read lock. */
1611 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1612                                    u32 pid, u32 seq, u32 flags, u8 cmd)
1613 {
1614         struct ovs_header *ovs_header;
1615         struct nlattr *nla;
1616         int ifindex, iflink;
1617         int mtu;
1618         int err;
1619
1620         ovs_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family,
1621                                  flags, cmd);
1622         if (!ovs_header)
1623                 return -EMSGSIZE;
1624
1625         ovs_header->dp_ifindex = vport->dp->dp_ifindex;
1626
1627         NLA_PUT_U32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no);
1628         NLA_PUT_U32(skb, OVS_VPORT_ATTR_TYPE, vport_get_type(vport));
1629         NLA_PUT_STRING(skb, OVS_VPORT_ATTR_NAME, vport_get_name(vport));
1630
1631         nla = nla_reserve(skb, OVS_VPORT_ATTR_STATS, sizeof(struct rtnl_link_stats64));
1632         if (!nla)
1633                 goto nla_put_failure;
1634         if (vport_get_stats(vport, nla_data(nla)))
1635                 __skb_trim(skb, skb->len - nla->nla_len);
1636
1637         NLA_PUT(skb, OVS_VPORT_ATTR_ADDRESS, ETH_ALEN, vport_get_addr(vport));
1638
1639         mtu = vport_get_mtu(vport);
1640         if (mtu)
1641                 NLA_PUT_U32(skb, OVS_VPORT_ATTR_MTU, mtu);
1642
1643         err = vport_get_options(vport, skb);
1644         if (err == -EMSGSIZE)
1645                 goto error;
1646
1647         ifindex = vport_get_ifindex(vport);
1648         if (ifindex > 0)
1649                 NLA_PUT_U32(skb, OVS_VPORT_ATTR_IFINDEX, ifindex);
1650
1651         iflink = vport_get_iflink(vport);
1652         if (iflink > 0)
1653                 NLA_PUT_U32(skb, OVS_VPORT_ATTR_IFLINK, iflink);
1654
1655         return genlmsg_end(skb, ovs_header);
1656
1657 nla_put_failure:
1658         err = -EMSGSIZE;
1659 error:
1660         genlmsg_cancel(skb, ovs_header);
1661         return err;
1662 }
1663
1664 /* Called with RTNL lock or RCU read lock. */
1665 static struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 pid,
1666                                                 u32 seq, u8 cmd)
1667 {
1668         struct sk_buff *skb;
1669         int retval;
1670
1671         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1672         if (!skb)
1673                 return ERR_PTR(-ENOMEM);
1674
1675         retval = ovs_vport_cmd_fill_info(vport, skb, pid, seq, 0, cmd);
1676         if (retval < 0) {
1677                 kfree_skb(skb);
1678                 return ERR_PTR(retval);
1679         }
1680         return skb;
1681 }
1682
1683 static int ovs_vport_cmd_validate(struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1684 {
1685         return CHECK_NUL_STRING(a[OVS_VPORT_ATTR_NAME], IFNAMSIZ - 1);
1686 }
1687
1688 /* Called with RTNL lock or RCU read lock. */
1689 static struct vport *lookup_vport(struct ovs_header *ovs_header,
1690                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1691 {
1692         struct datapath *dp;
1693         struct vport *vport;
1694
1695         if (a[OVS_VPORT_ATTR_NAME]) {
1696                 vport = vport_locate(nla_data(a[OVS_VPORT_ATTR_NAME]));
1697                 if (!vport)
1698                         return ERR_PTR(-ENODEV);
1699                 return vport;
1700         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1701                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1702
1703                 if (port_no >= DP_MAX_PORTS)
1704                         return ERR_PTR(-EFBIG);
1705
1706                 dp = get_dp(ovs_header->dp_ifindex);
1707                 if (!dp)
1708                         return ERR_PTR(-ENODEV);
1709
1710                 vport = get_vport_protected(dp, port_no);
1711                 if (!vport)
1712                         return ERR_PTR(-ENOENT);
1713                 return vport;
1714         } else
1715                 return ERR_PTR(-EINVAL);
1716 }
1717
1718 /* Called with RTNL lock. */
1719 static int change_vport(struct vport *vport, struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1720 {
1721         int err = 0;
1722         if (a[OVS_VPORT_ATTR_STATS])
1723                 err = vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1724         if (!err && a[OVS_VPORT_ATTR_ADDRESS])
1725                 err = vport_set_addr(vport, nla_data(a[OVS_VPORT_ATTR_ADDRESS]));
1726         if (!err && a[OVS_VPORT_ATTR_MTU])
1727                 err = vport_set_mtu(vport, nla_get_u32(a[OVS_VPORT_ATTR_MTU]));
1728         return err;
1729 }
1730
1731 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1732 {
1733         struct nlattr **a = info->attrs;
1734         struct ovs_header *ovs_header = info->userhdr;
1735         struct vport_parms parms;
1736         struct sk_buff *reply;
1737         struct vport *vport;
1738         struct datapath *dp;
1739         u32 port_no;
1740         int err;
1741
1742         err = -EINVAL;
1743         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE])
1744                 goto exit;
1745
1746         err = ovs_vport_cmd_validate(a);
1747         if (err)
1748                 goto exit;
1749
1750         rtnl_lock();
1751         dp = get_dp(ovs_header->dp_ifindex);
1752         err = -ENODEV;
1753         if (!dp)
1754                 goto exit_unlock;
1755
1756         if (a[OVS_VPORT_ATTR_PORT_NO]) {
1757                 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1758
1759                 err = -EFBIG;
1760                 if (port_no >= DP_MAX_PORTS)
1761                         goto exit_unlock;
1762
1763                 vport = get_vport_protected(dp, port_no);
1764                 err = -EBUSY;
1765                 if (vport)
1766                         goto exit_unlock;
1767         } else {
1768                 for (port_no = 1; ; port_no++) {
1769                         if (port_no >= DP_MAX_PORTS) {
1770                                 err = -EFBIG;
1771                                 goto exit_unlock;
1772                         }
1773                         vport = get_vport_protected(dp, port_no);
1774                         if (!vport)
1775                                 break;
1776                 }
1777         }
1778
1779         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1780         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1781         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1782         parms.dp = dp;
1783         parms.port_no = port_no;
1784
1785         vport = new_vport(&parms);
1786         err = PTR_ERR(vport);
1787         if (IS_ERR(vport))
1788                 goto exit_unlock;
1789
1790         set_internal_devs_mtu(dp);
1791         dp_sysfs_add_if(vport);
1792
1793         err = change_vport(vport, a);
1794         if (!err) {
1795                 reply = ovs_vport_cmd_build_info(vport, info->snd_pid,
1796                                                  info->snd_seq, OVS_VPORT_CMD_NEW);
1797                 if (IS_ERR(reply))
1798                         err = PTR_ERR(reply);
1799         }
1800         if (err) {
1801                 dp_detach_port(vport);
1802                 goto exit_unlock;
1803         }
1804         genl_notify(reply, genl_info_net(info), info->snd_pid,
1805                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1806
1807
1808 exit_unlock:
1809         rtnl_unlock();
1810 exit:
1811         return err;
1812 }
1813
1814 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1815 {
1816         struct nlattr **a = info->attrs;
1817         struct sk_buff *reply;
1818         struct vport *vport;
1819         int err;
1820
1821         err = ovs_vport_cmd_validate(a);
1822         if (err)
1823                 goto exit;
1824
1825         rtnl_lock();
1826         vport = lookup_vport(info->userhdr, a);
1827         err = PTR_ERR(vport);
1828         if (IS_ERR(vport))
1829                 goto exit_unlock;
1830
1831         err = 0;
1832         if (a[OVS_VPORT_ATTR_OPTIONS])
1833                 err = vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1834         if (!err)
1835                 err = change_vport(vport, a);
1836
1837         reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1838                                          OVS_VPORT_CMD_NEW);
1839         if (IS_ERR(reply)) {
1840                 err = PTR_ERR(reply);
1841                 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1842                                 dp_vport_multicast_group.id, err);
1843                 return 0;
1844         }
1845
1846         genl_notify(reply, genl_info_net(info), info->snd_pid,
1847                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1848
1849 exit_unlock:
1850         rtnl_unlock();
1851 exit:
1852         return err;
1853 }
1854
1855 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1856 {
1857         struct nlattr **a = info->attrs;
1858         struct sk_buff *reply;
1859         struct vport *vport;
1860         int err;
1861
1862         err = ovs_vport_cmd_validate(a);
1863         if (err)
1864                 goto exit;
1865
1866         rtnl_lock();
1867         vport = lookup_vport(info->userhdr, a);
1868         err = PTR_ERR(vport);
1869         if (IS_ERR(vport))
1870                 goto exit_unlock;
1871
1872         if (vport->port_no == OVSP_LOCAL) {
1873                 err = -EINVAL;
1874                 goto exit_unlock;
1875         }
1876
1877         reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1878                                          OVS_VPORT_CMD_DEL);
1879         err = PTR_ERR(reply);
1880         if (IS_ERR(reply))
1881                 goto exit_unlock;
1882
1883         err = dp_detach_port(vport);
1884
1885         genl_notify(reply, genl_info_net(info), info->snd_pid,
1886                     dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1887
1888 exit_unlock:
1889         rtnl_unlock();
1890 exit:
1891         return err;
1892 }
1893
1894 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1895 {
1896         struct nlattr **a = info->attrs;
1897         struct ovs_header *ovs_header = info->userhdr;
1898         struct sk_buff *reply;
1899         struct vport *vport;
1900         int err;
1901
1902         err = ovs_vport_cmd_validate(a);
1903         if (err)
1904                 goto exit;
1905
1906         rcu_read_lock();
1907         vport = lookup_vport(ovs_header, a);
1908         err = PTR_ERR(vport);
1909         if (IS_ERR(vport))
1910                 goto exit_unlock;
1911
1912         reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1913                                          OVS_VPORT_CMD_NEW);
1914         err = PTR_ERR(reply);
1915         if (IS_ERR(reply))
1916                 goto exit_unlock;
1917
1918         rcu_read_unlock();
1919
1920         return genlmsg_reply(reply, info);
1921
1922 exit_unlock:
1923         rcu_read_unlock();
1924 exit:
1925         return err;
1926 }
1927
1928 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1929 {
1930         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1931         struct datapath *dp;
1932         u32 port_no;
1933         int retval;
1934
1935         dp = get_dp(ovs_header->dp_ifindex);
1936         if (!dp)
1937                 return -ENODEV;
1938
1939         rcu_read_lock();
1940         for (port_no = cb->args[0]; port_no < DP_MAX_PORTS; port_no++) {
1941                 struct vport *vport;
1942
1943                 vport = get_vport_protected(dp, port_no);
1944                 if (!vport)
1945                         continue;
1946
1947                 if (ovs_vport_cmd_fill_info(vport, skb, NETLINK_CB(cb->skb).pid,
1948                                             cb->nlh->nlmsg_seq, NLM_F_MULTI,
1949                                             OVS_VPORT_CMD_NEW) < 0)
1950                         break;
1951         }
1952         rcu_read_unlock();
1953
1954         cb->args[0] = port_no;
1955         retval = skb->len;
1956
1957         return retval;
1958 }
1959
1960 static struct genl_ops dp_vport_genl_ops[] = {
1961         { .cmd = OVS_VPORT_CMD_NEW,
1962           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1963           .policy = vport_policy,
1964           .doit = ovs_vport_cmd_new
1965         },
1966         { .cmd = OVS_VPORT_CMD_DEL,
1967           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1968           .policy = vport_policy,
1969           .doit = ovs_vport_cmd_del
1970         },
1971         { .cmd = OVS_VPORT_CMD_GET,
1972           .flags = 0,               /* OK for unprivileged users. */
1973           .policy = vport_policy,
1974           .doit = ovs_vport_cmd_get,
1975           .dumpit = ovs_vport_cmd_dump
1976         },
1977         { .cmd = OVS_VPORT_CMD_SET,
1978           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1979           .policy = vport_policy,
1980           .doit = ovs_vport_cmd_set,
1981         },
1982 };
1983
1984 struct genl_family_and_ops {
1985         struct genl_family *family;
1986         struct genl_ops *ops;
1987         int n_ops;
1988         struct genl_multicast_group *group;
1989 };
1990
1991 static const struct genl_family_and_ops dp_genl_families[] = {
1992         { &dp_datapath_genl_family,
1993           dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1994           &dp_datapath_multicast_group },
1995         { &dp_vport_genl_family,
1996           dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1997           &dp_vport_multicast_group },
1998         { &dp_flow_genl_family,
1999           dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
2000           &dp_flow_multicast_group },
2001         { &dp_packet_genl_family,
2002           dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
2003           NULL },
2004 };
2005
2006 static void dp_unregister_genl(int n_families)
2007 {
2008         int i;
2009
2010         for (i = 0; i < n_families; i++) {
2011                 genl_unregister_family(dp_genl_families[i].family);
2012         }
2013 }
2014
2015 static int dp_register_genl(void)
2016 {
2017         int n_registered;
2018         int err;
2019         int i;
2020
2021         n_registered = 0;
2022         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2023                 const struct genl_family_and_ops *f = &dp_genl_families[i];
2024
2025                 err = genl_register_family_with_ops(f->family, f->ops,
2026                                                     f->n_ops);
2027                 if (err)
2028                         goto error;
2029                 n_registered++;
2030
2031                 if (f->group) {
2032                         err = genl_register_mc_group(f->family, f->group);
2033                         if (err)
2034                                 goto error;
2035                 }
2036         }
2037
2038         err = packet_register_mc_groups();
2039         if (err)
2040                 goto error;
2041         return 0;
2042
2043 error:
2044         dp_unregister_genl(n_registered);
2045         return err;
2046 }
2047
2048 static int __init dp_init(void)
2049 {
2050         struct sk_buff *dummy_skb;
2051         int err;
2052
2053         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
2054
2055         printk("Open vSwitch %s, built "__DATE__" "__TIME__"\n", VERSION BUILDNR);
2056
2057         err = flow_init();
2058         if (err)
2059                 goto error;
2060
2061         err = vport_init();
2062         if (err)
2063                 goto error_flow_exit;
2064
2065         err = register_netdevice_notifier(&dp_device_notifier);
2066         if (err)
2067                 goto error_vport_exit;
2068
2069         err = dp_register_genl();
2070         if (err < 0)
2071                 goto error_unreg_notifier;
2072
2073         return 0;
2074
2075 error_unreg_notifier:
2076         unregister_netdevice_notifier(&dp_device_notifier);
2077 error_vport_exit:
2078         vport_exit();
2079 error_flow_exit:
2080         flow_exit();
2081 error:
2082         return err;
2083 }
2084
2085 static void dp_cleanup(void)
2086 {
2087         rcu_barrier();
2088         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2089         unregister_netdevice_notifier(&dp_device_notifier);
2090         vport_exit();
2091         flow_exit();
2092 }
2093
2094 module_init(dp_init);
2095 module_exit(dp_cleanup);
2096
2097 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2098 MODULE_LICENSE("GPL");