datapath/datapath.c

   1 /*
   2  * Copyright (c) 2007-2014 Nicira, Inc.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of version 2 of the GNU General Public
   6  * License as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program; if not, write to the Free Software
  15  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
  16  * 02110-1301, USA
  17  */
  18
  19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  20
  21 #include <linux/init.h>
  22 #include <linux/module.h>
  23 #include <linux/if_arp.h>
  24 #include <linux/if_vlan.h>
  25 #include <linux/in.h>
  26 #include <linux/ip.h>
  27 #include <linux/jhash.h>
  28 #include <linux/delay.h>
  29 #include <linux/time.h>
  30 #include <linux/etherdevice.h>
  31 #include <linux/genetlink.h>
  32 #include <linux/kernel.h>
  33 #include <linux/kthread.h>
  34 #include <linux/mutex.h>
  35 #include <linux/percpu.h>
  36 #include <linux/rcupdate.h>
  37 #include <linux/tcp.h>
  38 #include <linux/udp.h>
  39 #include <linux/version.h>
  40 #include <linux/ethtool.h>
  41 #include <linux/wait.h>
  42 #include <asm/div64.h>
  43 #include <linux/highmem.h>
  44 #include <linux/netfilter_bridge.h>
  45 #include <linux/netfilter_ipv4.h>
  46 #include <linux/inetdevice.h>
  47 #include <linux/list.h>
  48 #include <linux/openvswitch.h>
  49 #include <linux/rculist.h>
  50 #include <linux/dmi.h>
  51 #include <linux/genetlink.h>
  52 #include <net/genetlink.h>
  53 #include <net/genetlink.h>
  54 #include <net/net_namespace.h>
  55 #include <net/netns/generic.h>
  56
  57 #include "datapath.h"
  58 #include "flow.h"
  59 #include "flow_table.h"
  60 #include "flow_netlink.h"
  61 #include "vlan.h"
  62 #include "vport-internal_dev.h"
  63 #include "vport-netdev.h"
  64
  65 int ovs_net_id __read_mostly;
  66
  67 static struct genl_family dp_packet_genl_family;
  68 static struct genl_family dp_flow_genl_family;
  69 static struct genl_family dp_datapath_genl_family;
  70
  71 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
  72         .name = OVS_FLOW_MCGROUP
  73 };
  74
  75 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
  76         .name = OVS_DATAPATH_MCGROUP
  77 };
  78
  79 struct genl_multicast_group ovs_dp_vport_multicast_group = {
  80         .name = OVS_VPORT_MCGROUP
  81 };
  82
  83 /* Check if need to build a reply message.
  84  * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
  85 static bool ovs_must_notify(struct genl_info *info,
  86                             const struct genl_multicast_group *grp)
  87 {
  88         return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
  89                 netlink_has_listeners(genl_info_net(info)->genl_sock, GROUP_ID(grp));
  90 }
  91
  92 static void ovs_notify(struct genl_family *family, struct genl_multicast_group *grp,
  93                        struct sk_buff *skb, struct genl_info *info)
  94 {
  95         genl_notify(family, skb, genl_info_net(info),
  96                     info->snd_portid, GROUP_ID(grp), info->nlhdr, GFP_KERNEL);
  97 }
  98
  99 /**
 100  * DOC: Locking:
 101  *
 102  * All writes e.g. Writes to device state (add/remove datapath, port, set
 103  * operations on vports, etc.), Writes to other state (flow table
 104  * modifications, set miscellaneous datapath parameters, etc.) are protected
 105  * by ovs_lock.
 106  *
 107  * Reads are protected by RCU.
 108  *
 109  * There are a few special cases (mostly stats) that have their own
 110  * synchronization but they nest under all of above and don't interact with
 111  * each other.
 112  *
 113  * The RTNL lock nests inside ovs_mutex.
 114  */
 115
 116 static DEFINE_MUTEX(ovs_mutex);
 117
 118 void ovs_lock(void)
 119 {
 120         mutex_lock(&ovs_mutex);
 121 }
 122
 123 void ovs_unlock(void)
 124 {
 125         mutex_unlock(&ovs_mutex);
 126 }
 127
 128 #ifdef CONFIG_LOCKDEP
 129 int lockdep_ovsl_is_held(void)
 130 {
 131         if (debug_locks)
 132                 return lockdep_is_held(&ovs_mutex);
 133         else
 134                 return 1;
 135 }
 136 #endif
 137
 138 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
 139                              const struct dp_upcall_info *);
 140 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
 141                                   const struct dp_upcall_info *);
 142
 143 /* Must be called with rcu_read_lock. */
 144 static struct datapath *get_dp_rcu(struct net *net, int dp_ifindex)
 145 {
 146         struct net_device *dev = dev_get_by_index_rcu(net, dp_ifindex);
 147
 148         if (dev) {
 149                 struct vport *vport = ovs_internal_dev_get_vport(dev);
 150                 if (vport)
 151                         return vport->dp;
 152         }
 153
 154         return NULL;
 155 }
 156
 157 /* The caller must hold either ovs_mutex or rcu_read_lock to keep the
 158  * returned dp pointer valid. */
 159 static inline struct datapath *get_dp(struct net *net, int dp_ifindex)
 160 {
 161         struct datapath *dp;
 162
 163         WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_ovsl_is_held());
 164         rcu_read_lock();
 165         dp = get_dp_rcu(net, dp_ifindex);
 166         rcu_read_unlock();
 167
 168         return dp;
 169 }
 170
 171 /* Must be called with rcu_read_lock or ovs_mutex. */
 172 const char *ovs_dp_name(const struct datapath *dp)
 173 {
 174         struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
 175         return vport->ops->get_name(vport);
 176 }
 177
 178 static int get_dpifindex(struct datapath *dp)
 179 {
 180         struct vport *local;
 181         int ifindex;
 182
 183         rcu_read_lock();
 184
 185         local = ovs_vport_rcu(dp, OVSP_LOCAL);
 186         if (local)
 187                 ifindex = netdev_vport_priv(local)->dev->ifindex;
 188         else
 189                 ifindex = 0;
 190
 191         rcu_read_unlock();
 192
 193         return ifindex;
 194 }
 195
 196 static void destroy_dp_rcu(struct rcu_head *rcu)
 197 {
 198         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 199
 200         ovs_flow_tbl_destroy(&dp->table);
 201         free_percpu(dp->stats_percpu);
 202         release_net(ovs_dp_get_net(dp));
 203         kfree(dp->ports);
 204         kfree(dp);
 205 }
 206
 207 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
 208                                             u16 port_no)
 209 {
 210         return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
 211 }
 212
 213 /* Called with ovs_mutex or RCU read lock. */
 214 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
 215 {
 216         struct vport *vport;
 217         struct hlist_head *head;
 218
 219         head = vport_hash_bucket(dp, port_no);
 220         hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
 221                 if (vport->port_no == port_no)
 222                         return vport;
 223         }
 224         return NULL;
 225 }
 226
 227 /* Called with ovs_mutex. */
 228 static struct vport *new_vport(const struct vport_parms *parms)
 229 {
 230         struct vport *vport;
 231
 232         vport = ovs_vport_add(parms);
 233         if (!IS_ERR(vport)) {
 234                 struct datapath *dp = parms->dp;
 235                 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
 236
 237                 hlist_add_head_rcu(&vport->dp_hash_node, head);
 238         }
 239         return vport;
 240 }
 241
 242 void ovs_dp_detach_port(struct vport *p)
 243 {
 244         ASSERT_OVSL();
 245
 246         /* First drop references to device. */
 247         hlist_del_rcu(&p->dp_hash_node);
 248
 249         /* Then destroy it. */
 250         ovs_vport_del(p);
 251 }
 252
 253 void ovs_dp_process_packet_with_key(struct sk_buff *skb,
 254                                     struct sw_flow_key *pkt_key,
 255                                     bool recirc)
 256 {
 257         const struct vport *p = OVS_CB(skb)->input_vport;
 258         struct datapath *dp = p->dp;
 259         struct sw_flow *flow;
 260         struct dp_stats_percpu *stats;
 261         u64 *stats_counter;
 262         u32 n_mask_hit;
 263
 264         stats = this_cpu_ptr(dp->stats_percpu);
 265         OVS_CB(skb)->pkt_key = pkt_key;
 266
 267         /* Look up flow. */
 268         flow = ovs_flow_tbl_lookup_stats(&dp->table, pkt_key, skb_get_hash(skb),
 269                                          &n_mask_hit);
 270         if (unlikely(!flow)) {
 271                 struct dp_upcall_info upcall;
 272
 273                 upcall.cmd = OVS_PACKET_CMD_MISS;
 274                 upcall.userdata = NULL;
 275                 upcall.portid = ovs_vport_find_upcall_portid(p, skb);
 276                 ovs_dp_upcall(dp, skb, &upcall);
 277                 consume_skb(skb);
 278                 stats_counter = &stats->n_missed;
 279                 goto out;
 280         }
 281
 282         OVS_CB(skb)->flow = flow;
 283
 284         ovs_flow_stats_update(OVS_CB(skb)->flow, pkt_key->tp.flags, skb);
 285         ovs_execute_actions(dp, skb, recirc);
 286         stats_counter = &stats->n_hit;
 287
 288 out:
 289         /* Update datapath statistics. */
 290         u64_stats_update_begin(&stats->sync);
 291         (*stats_counter)++;
 292         stats->n_mask_hit += n_mask_hit;
 293         u64_stats_update_end(&stats->sync);
 294 }
 295
 296 /* Must be called with rcu_read_lock. */
 297 void ovs_dp_process_received_packet(struct sk_buff *skb)
 298 {
 299         int error;
 300         struct sw_flow_key key;
 301
 302         /* Extract flow from 'skb' into 'key'. */
 303         error = ovs_flow_extract(skb, OVS_CB(skb)->input_vport->port_no, &key);
 304         if (unlikely(error)) {
 305                 kfree_skb(skb);
 306                 return;
 307         }
 308
 309         ovs_dp_process_packet_with_key(skb, &key, false);
 310 }
 311
 312 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
 313                   const struct dp_upcall_info *upcall_info)
 314 {
 315         struct dp_stats_percpu *stats;
 316         int err;
 317
 318         BUG_ON(!OVS_CB(skb)->pkt_key);
 319
 320         if (upcall_info->portid == 0) {
 321                 err = -ENOTCONN;
 322                 goto err;
 323         }
 324
 325         if (!skb_is_gso(skb))
 326                 err = queue_userspace_packet(dp, skb, upcall_info);
 327         else
 328                 err = queue_gso_packets(dp, skb, upcall_info);
 329         if (err)
 330                 goto err;
 331
 332         return 0;
 333
 334 err:
 335         stats = this_cpu_ptr(dp->stats_percpu);
 336
 337         u64_stats_update_begin(&stats->sync);
 338         stats->n_lost++;
 339         u64_stats_update_end(&stats->sync);
 340
 341         return err;
 342 }
 343
 344 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
 345                              const struct dp_upcall_info *upcall_info)
 346 {
 347         unsigned short gso_type = skb_shinfo(skb)->gso_type;
 348         struct sw_flow_key later_key;
 349         struct sk_buff *segs, *nskb;
 350         int err;
 351
 352         segs = __skb_gso_segment(skb, NETIF_F_SG, false);
 353         if (IS_ERR(segs))
 354                 return PTR_ERR(segs);
 355
 356         if (gso_type & SKB_GSO_UDP) {
 357                 /* The initial flow key extracted by ovs_flow_extract()
 358                  * in this case is for a first fragment, so we need to
 359                  * properly mark later fragments.
 360                  */
 361                 later_key = *OVS_CB(skb)->pkt_key;
 362                 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
 363         }
 364
 365         /* Queue all of the segments. */
 366         skb = segs;
 367         do {
 368                 if (gso_type & SKB_GSO_UDP && skb != segs)
 369                         OVS_CB(skb)->pkt_key = &later_key;
 370
 371                 err = queue_userspace_packet(dp, skb, upcall_info);
 372                 if (err)
 373                         break;
 374
 375         } while ((skb = skb->next));
 376
 377         /* Free all of the segments. */
 378         skb = segs;
 379         do {
 380                 nskb = skb->next;
 381                 if (err)
 382                         kfree_skb(skb);
 383                 else
 384                         consume_skb(skb);
 385         } while ((skb = nskb));
 386         return err;
 387 }
 388
 389 static size_t key_attr_size(void)
 390 {
 391         /* Whenever adding new OVS_KEY_ FIELDS, we should consider
 392          * updating this function.  */
 393         BUILD_BUG_ON(OVS_KEY_ATTR_TUNNEL_INFO != 22);
 394
 395         return    nla_total_size(4)   /* OVS_KEY_ATTR_PRIORITY */
 396                 + nla_total_size(0)   /* OVS_KEY_ATTR_TUNNEL */
 397                   + nla_total_size(8)   /* OVS_TUNNEL_KEY_ATTR_ID */
 398                   + nla_total_size(4)   /* OVS_TUNNEL_KEY_ATTR_IPV4_SRC */
 399                   + nla_total_size(4)   /* OVS_TUNNEL_KEY_ATTR_IPV4_DST */
 400                   + nla_total_size(1)   /* OVS_TUNNEL_KEY_ATTR_TOS */
 401                   + nla_total_size(1)   /* OVS_TUNNEL_KEY_ATTR_TTL */
 402                   + nla_total_size(0)   /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
 403                   + nla_total_size(0)   /* OVS_TUNNEL_KEY_ATTR_CSUM */
 404                   + nla_total_size(0)   /* OVS_TUNNEL_KEY_ATTR_OAM */
 405                   + nla_total_size(256) /* OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS */
 406                 + nla_total_size(4)   /* OVS_KEY_ATTR_IN_PORT */
 407                 + nla_total_size(4)   /* OVS_KEY_ATTR_SKB_MARK */
 408                 + nla_total_size(4)   /* OVS_KEY_ATTR_DP_HASH */
 409                 + nla_total_size(4)   /* OVS_KEY_ATTR_RECIRC_ID */
 410                 + nla_total_size(12)  /* OVS_KEY_ATTR_ETHERNET */
 411                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 412                 + nla_total_size(4)   /* OVS_KEY_ATTR_8021Q */
 413                 + nla_total_size(0)   /* OVS_KEY_ATTR_ENCAP */
 414                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 415                 + nla_total_size(40)  /* OVS_KEY_ATTR_IPV6 */
 416                 + nla_total_size(2)   /* OVS_KEY_ATTR_ICMPV6 */
 417                 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
 418 }
 419
 420 static size_t upcall_msg_size(const struct nlattr *userdata,
 421                               unsigned int hdrlen)
 422 {
 423         size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
 424                 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
 425                 + nla_total_size(key_attr_size()); /* OVS_PACKET_ATTR_KEY */
 426
 427         /* OVS_PACKET_ATTR_USERDATA */
 428         if (userdata)
 429                 size += NLA_ALIGN(userdata->nla_len);
 430
 431         return size;
 432 }
 433
 434 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
 435                                   const struct dp_upcall_info *upcall_info)
 436 {
 437         struct ovs_header *upcall;
 438         struct sk_buff *nskb = NULL;
 439         struct sk_buff *user_skb; /* to be queued to userspace */
 440         struct sw_flow_key *pkt_key = OVS_CB(skb)->pkt_key;
 441         struct nlattr *nla;
 442         struct genl_info info = {
 443 #if LINUX_VERSION_CODE >= KERNEL_VERSION(3,14,0)
 444                 .dst_sk = ovs_dp_get_net(dp)->genl_sock,
 445 #endif
 446                 .snd_portid = upcall_info->portid,
 447         };
 448         size_t len;
 449         unsigned int hlen;
 450         int err, dp_ifindex;
 451
 452         dp_ifindex = get_dpifindex(dp);
 453         if (!dp_ifindex)
 454                 return -ENODEV;
 455
 456         if (vlan_tx_tag_present(skb)) {
 457                 nskb = skb_clone(skb, GFP_ATOMIC);
 458                 if (!nskb)
 459                         return -ENOMEM;
 460
 461                 nskb = __vlan_put_tag(nskb, nskb->vlan_proto, vlan_tx_tag_get(nskb));
 462                 if (!nskb)
 463                         return -ENOMEM;
 464
 465                 vlan_set_tci(nskb, 0);
 466
 467                 skb = nskb;
 468         }
 469
 470         if (nla_attr_size(skb->len) > USHRT_MAX) {
 471                 err = -EFBIG;
 472                 goto out;
 473         }
 474
 475         /* Complete checksum if needed */
 476         if (skb->ip_summed == CHECKSUM_PARTIAL &&
 477             (err = skb_checksum_help(skb)))
 478                 goto out;
 479
 480         /* Older versions of OVS user space enforce alignment of the last
 481          * Netlink attribute to NLA_ALIGNTO which would require extensive
 482          * padding logic. Only perform zerocopy if padding is not required.
 483          */
 484         if (dp->user_features & OVS_DP_F_UNALIGNED)
 485                 hlen = skb_zerocopy_headlen(skb);
 486         else
 487                 hlen = skb->len;
 488
 489         len = upcall_msg_size(upcall_info->userdata, hlen);
 490         user_skb = genlmsg_new_unicast(len, &info, GFP_ATOMIC);
 491         if (!user_skb) {
 492                 err = -ENOMEM;
 493                 goto out;
 494         }
 495
 496         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 497                              0, upcall_info->cmd);
 498         upcall->dp_ifindex = dp_ifindex;
 499
 500         nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
 501         err = ovs_nla_put_flow(dp, pkt_key, pkt_key, user_skb);
 502         BUG_ON(err);
 503         nla_nest_end(user_skb, nla);
 504
 505         if (upcall_info->userdata)
 506                 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
 507                           nla_len(upcall_info->userdata),
 508                           nla_data(upcall_info->userdata));
 509
 510         /* Only reserve room for attribute header, packet data is added
 511          * in skb_zerocopy() */
 512         if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
 513                 err = -ENOBUFS;
 514                 goto out;
 515         }
 516         nla->nla_len = nla_attr_size(skb->len);
 517
 518         err = skb_zerocopy(user_skb, skb, skb->len, hlen);
 519         if (err)
 520                 goto out;
 521
 522         /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
 523         if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
 524                 size_t plen = NLA_ALIGN(user_skb->len) - user_skb->len;
 525
 526                 if (plen > 0)
 527                         memset(skb_put(user_skb, plen), 0, plen);
 528         }
 529
 530         ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
 531
 532         err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
 533 out:
 534         if (err)
 535                 skb_tx_error(skb);
 536         kfree_skb(nskb);
 537         return err;
 538 }
 539
 540 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 541 {
 542         struct ovs_header *ovs_header = info->userhdr;
 543         struct nlattr **a = info->attrs;
 544         struct sw_flow_actions *acts;
 545         struct sk_buff *packet;
 546         struct sw_flow *flow;
 547         struct datapath *dp;
 548         struct ethhdr *eth;
 549         struct vport *input_vport;
 550         int len;
 551         int err;
 552
 553         err = -EINVAL;
 554         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 555             !a[OVS_PACKET_ATTR_ACTIONS])
 556                 goto err;
 557
 558         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 559         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 560         err = -ENOMEM;
 561         if (!packet)
 562                 goto err;
 563         skb_reserve(packet, NET_IP_ALIGN);
 564
 565         nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
 566
 567         skb_reset_mac_header(packet);
 568         eth = eth_hdr(packet);
 569
 570         /* Normally, setting the skb 'protocol' field would be handled by a
 571          * call to eth_type_trans(), but it assumes there's a sending
 572          * device, which we may not have. */
 573         if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
 574                 packet->protocol = eth->h_proto;
 575         else
 576                 packet->protocol = htons(ETH_P_802_2);
 577
 578         /* Build an sw_flow for sending this packet. */
 579         flow = ovs_flow_alloc();
 580         err = PTR_ERR(flow);
 581         if (IS_ERR(flow))
 582                 goto err_kfree_skb;
 583
 584         err = ovs_flow_extract(packet, -1, &flow->key);
 585         if (err)
 586                 goto err_flow_free;
 587
 588         err = ovs_nla_get_flow_metadata(flow, a[OVS_PACKET_ATTR_KEY]);
 589         if (err)
 590                 goto err_flow_free;
 591         acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_PACKET_ATTR_ACTIONS]));
 592         err = PTR_ERR(acts);
 593         if (IS_ERR(acts))
 594                 goto err_flow_free;
 595
 596         err = ovs_nla_copy_actions(a[OVS_PACKET_ATTR_ACTIONS],
 597                                    &flow->key, &acts);
 598         rcu_assign_pointer(flow->sf_acts, acts);
 599         if (err)
 600                 goto err_flow_free;
 601
 602         OVS_CB(packet)->flow = flow;
 603         OVS_CB(packet)->pkt_key = &flow->key;
 604         packet->priority = flow->key.phy.priority;
 605         packet->mark = flow->key.phy.skb_mark;
 606
 607         rcu_read_lock();
 608         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
 609         err = -ENODEV;
 610         if (!dp)
 611                 goto err_unlock;
 612
 613         input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
 614         if (!input_vport)
 615                 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
 616
 617         if (!input_vport)
 618                 goto err_unlock;
 619
 620         OVS_CB(packet)->input_vport = input_vport;
 621
 622         local_bh_disable();
 623         err = ovs_execute_actions(dp, packet, false);
 624         local_bh_enable();
 625         rcu_read_unlock();
 626
 627         ovs_flow_free(flow, false);
 628         return err;
 629
 630 err_unlock:
 631         rcu_read_unlock();
 632 err_flow_free:
 633         ovs_flow_free(flow, false);
 634 err_kfree_skb:
 635         kfree_skb(packet);
 636 err:
 637         return err;
 638 }
 639
 640 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 641         [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
 642         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 643         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 644 };
 645
 646 static struct genl_ops dp_packet_genl_ops[] = {
 647         { .cmd = OVS_PACKET_CMD_EXECUTE,
 648           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 649           .policy = packet_policy,
 650           .doit = ovs_packet_cmd_execute
 651         }
 652 };
 653
 654 static struct genl_family dp_packet_genl_family = {
 655         .id = GENL_ID_GENERATE,
 656         .hdrsize = sizeof(struct ovs_header),
 657         .name = OVS_PACKET_FAMILY,
 658         .version = OVS_PACKET_VERSION,
 659         .maxattr = OVS_PACKET_ATTR_MAX,
 660         .netnsok = true,
 661         .parallel_ops = true,
 662         .ops = dp_packet_genl_ops,
 663         .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
 664 };
 665
 666 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats,
 667                          struct ovs_dp_megaflow_stats *mega_stats)
 668 {
 669         int i;
 670
 671         memset(mega_stats, 0, sizeof(*mega_stats));
 672
 673         stats->n_flows = ovs_flow_tbl_count(&dp->table);
 674         mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
 675
 676         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 677
 678         for_each_possible_cpu(i) {
 679                 const struct dp_stats_percpu *percpu_stats;
 680                 struct dp_stats_percpu local_stats;
 681                 unsigned int start;
 682
 683                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 684
 685                 do {
 686                         start = u64_stats_fetch_begin_irq(&percpu_stats->sync);
 687                         local_stats = *percpu_stats;
 688                 } while (u64_stats_fetch_retry_irq(&percpu_stats->sync, start));
 689
 690                 stats->n_hit += local_stats.n_hit;
 691                 stats->n_missed += local_stats.n_missed;
 692                 stats->n_lost += local_stats.n_lost;
 693                 mega_stats->n_mask_hit += local_stats.n_mask_hit;
 694         }
 695 }
 696
 697 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
 698 {
 699         return NLMSG_ALIGN(sizeof(struct ovs_header))
 700                 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
 701                 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_MASK */
 702                 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
 703                 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
 704                 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
 705                 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
 706 }
 707
 708 /* Called with ovs_mutex or RCU read lock. */
 709 static int ovs_flow_cmd_fill_match(struct datapath *dp,
 710                                    const struct sw_flow *flow,
 711                                    struct sk_buff *skb)
 712 {
 713         struct nlattr *nla;
 714         int err;
 715
 716         /* Fill flow key. */
 717         nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
 718         if (!nla)
 719                 return -EMSGSIZE;
 720
 721         err = ovs_nla_put_flow(dp, &flow->unmasked_key,
 722                                &flow->unmasked_key, skb);
 723         if (err)
 724                 return err;
 725         nla_nest_end(skb, nla);
 726
 727         /* Fill flow mask. */
 728         nla = nla_nest_start(skb, OVS_FLOW_ATTR_MASK);
 729         if (!nla)
 730                 return -EMSGSIZE;
 731
 732         err = ovs_nla_put_flow(dp, &flow->key, &flow->mask->key, skb);
 733         if (err)
 734                 return err;
 735         nla_nest_end(skb, nla);
 736
 737         return 0;
 738 }
 739
 740 /* Called with ovs_mutex or RCU read lock. */
 741 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
 742                                    struct sk_buff *skb)
 743 {
 744         struct ovs_flow_stats stats;
 745         __be16 tcp_flags;
 746         unsigned long used;
 747
 748         ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
 749
 750         if (used &&
 751             nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
 752                 return -EMSGSIZE;
 753
 754         if (stats.n_packets &&
 755             nla_put(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats))
 756                 return -EMSGSIZE;
 757
 758         if ((u8)ntohs(tcp_flags) &&
 759              nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
 760                 return -EMSGSIZE;
 761
 762         return 0;
 763 }
 764
 765 /* Called with ovs_mutex or RCU read lock. */
 766 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
 767                                      struct sk_buff *skb, int skb_orig_len)
 768 {
 769         struct nlattr *start;
 770         int err;
 771
 772         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 773          * this is the first flow to be dumped into 'skb'.  This is unusual for
 774          * Netlink but individual action lists can be longer than
 775          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 776          * The userspace caller can always fetch the actions separately if it
 777          * really wants them.  (Most userspace callers in fact don't care.)
 778          *
 779          * This can only fail for dump operations because the skb is always
 780          * properly sized for single flows.
 781          */
 782         start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
 783         if (start) {
 784                 const struct sw_flow_actions *sf_acts;
 785
 786                 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
 787                 err = ovs_nla_put_actions(sf_acts->actions,
 788                                           sf_acts->actions_len, skb);
 789
 790                 if (!err)
 791                         nla_nest_end(skb, start);
 792                 else {
 793                         if (skb_orig_len)
 794                                 return err;
 795
 796                         nla_nest_cancel(skb, start);
 797                 }
 798         } else if (skb_orig_len) {
 799                 return -EMSGSIZE;
 800         }
 801
 802         return 0;
 803 }
 804
 805 /* Called with ovs_mutex or RCU read lock. */
 806 static int ovs_flow_cmd_fill_info(struct datapath *dp,
 807                                   const struct sw_flow *flow, int dp_ifindex,
 808                                   struct sk_buff *skb, u32 portid,
 809                                   u32 seq, u32 flags, u8 cmd)
 810 {
 811         const int skb_orig_len = skb->len;
 812         struct ovs_header *ovs_header;
 813         int err;
 814
 815         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
 816         if (!ovs_header)
 817                 return -EMSGSIZE;
 818         ovs_header->dp_ifindex = dp_ifindex;
 819
 820         err = ovs_flow_cmd_fill_match(dp, flow, skb);
 821         if (err)
 822                 goto error;
 823
 824         err = ovs_flow_cmd_fill_stats(flow, skb);
 825         if (err)
 826                 goto error;
 827
 828         err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
 829         if (err)
 830                 goto error;
 831
 832         return genlmsg_end(skb, ovs_header);
 833
 834 error:
 835         genlmsg_cancel(skb, ovs_header);
 836         return err;
 837 }
 838
 839 /* May not be called with RCU read lock. */
 840 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
 841                                                struct genl_info *info,
 842                                                bool always)
 843 {
 844         struct sk_buff *skb;
 845
 846         if (!always && !ovs_must_notify(info, &ovs_dp_flow_multicast_group))
 847                 return NULL;
 848
 849         skb = genlmsg_new_unicast(ovs_flow_cmd_msg_size(acts), info, GFP_KERNEL);
 850
 851         if (!skb)
 852                 return ERR_PTR(-ENOMEM);
 853
 854         return skb;
 855 }
 856
 857 /* Called with ovs_mutex. */
 858 static struct sk_buff *ovs_flow_cmd_build_info(struct datapath *dp,
 859                                                const struct sw_flow *flow,
 860                                                int dp_ifindex,
 861                                                struct genl_info *info, u8 cmd,
 862                                                bool always)
 863 {
 864         struct sk_buff *skb;
 865         int retval;
 866
 867         skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts), info,
 868                                       always);
 869         if (!skb || IS_ERR(skb))
 870                 return skb;
 871
 872         retval = ovs_flow_cmd_fill_info(dp, flow, dp_ifindex, skb,
 873                                         info->snd_portid, info->snd_seq, 0,
 874                                         cmd);
 875         BUG_ON(retval < 0);
 876         return skb;
 877 }
 878
 879 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
 880 {
 881         struct nlattr **a = info->attrs;
 882         struct ovs_header *ovs_header = info->userhdr;
 883         struct sw_flow *flow, *new_flow;
 884         struct sw_flow_mask mask;
 885         struct sk_buff *reply;
 886         struct datapath *dp;
 887         struct sw_flow_actions *acts;
 888         struct sw_flow_match match;
 889         int error;
 890
 891         /* Must have key and actions. */
 892         error = -EINVAL;
 893         if (!a[OVS_FLOW_ATTR_KEY]) {
 894                 OVS_NLERR("Flow key attribute not present in new flow.\n");
 895                 goto error;
 896         }
 897         if (!a[OVS_FLOW_ATTR_ACTIONS]) {
 898                 OVS_NLERR("Flow actions attribute not present in new flow.\n");
 899                 goto error;
 900         }
 901
 902         /* Most of the time we need to allocate a new flow, do it before
 903          * locking. */
 904         new_flow = ovs_flow_alloc();
 905         if (IS_ERR(new_flow)) {
 906                 error = PTR_ERR(new_flow);
 907                 goto error;
 908         }
 909
 910         /* Extract key. */
 911         ovs_match_init(&match, &new_flow->unmasked_key, &mask);
 912         error = ovs_nla_get_match(&match,
 913                                   a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
 914         if (error)
 915                 goto err_kfree_flow;
 916
 917         ovs_flow_mask_key(&new_flow->key, &new_flow->unmasked_key, &mask);
 918
 919         /* Validate actions. */
 920         acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
 921         error = PTR_ERR(acts);
 922         if (IS_ERR(acts))
 923                 goto err_kfree_flow;
 924
 925         error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS], &new_flow->key,
 926                                      &acts);
 927         if (error) {
 928                 OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
 929                 goto err_kfree_acts;
 930         }
 931
 932         reply = ovs_flow_cmd_alloc_info(acts, info, false);
 933         if (IS_ERR(reply)) {
 934                 error = PTR_ERR(reply);
 935                 goto err_kfree_acts;
 936         }
 937
 938         ovs_lock();
 939         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 940         if (unlikely(!dp)) {
 941                 error = -ENODEV;
 942                 goto err_unlock_ovs;
 943         }
 944         /* Check if this is a duplicate flow */
 945         flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->unmasked_key);
 946         if (likely(!flow)) {
 947                 rcu_assign_pointer(new_flow->sf_acts, acts);
 948
 949                 /* Put flow in bucket. */
 950                 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
 951                 if (unlikely(error)) {
 952                         acts = NULL;
 953                         goto err_unlock_ovs;
 954                 }
 955
 956                 if (unlikely(reply)) {
 957                         error = ovs_flow_cmd_fill_info(dp, new_flow,
 958                                                        ovs_header->dp_ifindex,
 959                                                        reply, info->snd_portid,
 960                                                        info->snd_seq, 0,
 961                                                        OVS_FLOW_CMD_NEW);
 962                         BUG_ON(error < 0);
 963                 }
 964                 ovs_unlock();
 965         } else {
 966                 struct sw_flow_actions *old_acts;
 967
 968                 /* Bail out if we're not allowed to modify an existing flow.
 969                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
 970                  * because Generic Netlink treats the latter as a dump
 971                  * request.  We also accept NLM_F_EXCL in case that bug ever
 972                  * gets fixed.
 973                  */
 974                 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
 975                                                          | NLM_F_EXCL))) {
 976                         error = -EEXIST;
 977                         goto err_unlock_ovs;
 978                 }
 979                 /* The unmasked key has to be the same for flow updates. */
 980                 if (unlikely(!ovs_flow_cmp_unmasked_key(flow, &match))) {
 981                         /* Look for any overlapping flow. */
 982                         flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
 983                         if (!flow) {
 984                                 error = -ENOENT;
 985                                 goto err_unlock_ovs;
 986                         }
 987                 }
 988                 /* Update actions. */
 989                 old_acts = ovsl_dereference(flow->sf_acts);
 990                 rcu_assign_pointer(flow->sf_acts, acts);
 991
 992                 if (unlikely(reply)) {
 993                         error = ovs_flow_cmd_fill_info(dp, flow,
 994                                                        ovs_header->dp_ifindex,
 995                                                        reply, info->snd_portid,
 996                                                        info->snd_seq, 0,
 997                                                        OVS_FLOW_CMD_NEW);
 998                         BUG_ON(error < 0);
 999                 }
1000                 ovs_unlock();
1001
1002                 ovs_nla_free_flow_actions(old_acts);
1003                 ovs_flow_free(new_flow, false);
1004         }
1005
1006         if (reply)
1007                 ovs_notify(&dp_flow_genl_family, &ovs_dp_flow_multicast_group, reply, info);
1008         return 0;
1009
1010 err_unlock_ovs:
1011         ovs_unlock();
1012         kfree_skb(reply);
1013 err_kfree_acts:
1014         kfree(acts);
1015 err_kfree_flow:
1016         ovs_flow_free(new_flow, false);
1017 error:
1018         return error;
1019 }
1020
1021 static struct sw_flow_actions *get_flow_actions(const struct nlattr *a,
1022                                                 const struct sw_flow_key *key,
1023                                                 const struct sw_flow_mask *mask)
1024 {
1025         struct sw_flow_actions *acts;
1026         struct sw_flow_key masked_key;
1027         int error;
1028
1029         acts = ovs_nla_alloc_flow_actions(nla_len(a));
1030         if (IS_ERR(acts))
1031                 return acts;
1032
1033         ovs_flow_mask_key(&masked_key, key, mask);
1034         error = ovs_nla_copy_actions(a, &masked_key, &acts);
1035         if (error) {
1036                 OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
1037                 kfree(acts);
1038                 return ERR_PTR(error);
1039         }
1040
1041         return acts;
1042 }
1043
1044 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1045 {
1046         struct nlattr **a = info->attrs;
1047         struct ovs_header *ovs_header = info->userhdr;
1048         struct sw_flow_key key;
1049         struct sw_flow *flow;
1050         struct sw_flow_mask mask;
1051         struct sk_buff *reply = NULL;
1052         struct datapath *dp;
1053         struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1054         struct sw_flow_match match;
1055         int error;
1056
1057         /* Extract key. */
1058         error = -EINVAL;
1059         if (!a[OVS_FLOW_ATTR_KEY]) {
1060                 OVS_NLERR("Flow key attribute not present in set flow.\n");
1061                 goto error;
1062         }
1063
1064         ovs_match_init(&match, &key, &mask);
1065         error = ovs_nla_get_match(&match,
1066                                   a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
1067         if (error)
1068                 goto error;
1069
1070         /* Validate actions. */
1071         if (a[OVS_FLOW_ATTR_ACTIONS]) {
1072                 acts = get_flow_actions(a[OVS_FLOW_ATTR_ACTIONS], &key, &mask);
1073                 if (IS_ERR(acts)) {
1074                         error = PTR_ERR(acts);
1075                         goto error;
1076                 }
1077         }
1078
1079         /* Can allocate before locking if have acts. */
1080         if (acts) {
1081                 reply = ovs_flow_cmd_alloc_info(acts, info, false);
1082                 if (IS_ERR(reply)) {
1083                         error = PTR_ERR(reply);
1084                         goto err_kfree_acts;
1085                 }
1086         }
1087
1088         ovs_lock();
1089         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1090         if (unlikely(!dp)) {
1091                 error = -ENODEV;
1092                 goto err_unlock_ovs;
1093         }
1094         /* Check that the flow exists. */
1095         flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1096         if (unlikely(!flow)) {
1097                 error = -ENOENT;
1098                 goto err_unlock_ovs;
1099         }
1100
1101         /* Update actions, if present. */
1102         if (likely(acts)) {
1103                 old_acts = ovsl_dereference(flow->sf_acts);
1104                 rcu_assign_pointer(flow->sf_acts, acts);
1105
1106                 if (unlikely(reply)) {
1107                         error = ovs_flow_cmd_fill_info(dp, flow,
1108                                                        ovs_header->dp_ifindex,
1109                                                        reply, info->snd_portid,
1110                                                        info->snd_seq, 0,
1111                                                        OVS_FLOW_CMD_NEW);
1112                         BUG_ON(error < 0);
1113                 }
1114         } else {
1115                 /* Could not alloc without acts before locking. */
1116                 reply = ovs_flow_cmd_build_info(dp, flow,
1117                                                 ovs_header->dp_ifindex,
1118                                                 info, OVS_FLOW_CMD_NEW, false);
1119                 if (unlikely(IS_ERR(reply))) {
1120                         error = PTR_ERR(reply);
1121                         goto err_unlock_ovs;
1122                 }
1123         }
1124
1125         /* Clear stats. */
1126         if (a[OVS_FLOW_ATTR_CLEAR])
1127                 ovs_flow_stats_clear(flow);
1128         ovs_unlock();
1129
1130         if (reply)
1131                 ovs_notify(&dp_flow_genl_family, &ovs_dp_flow_multicast_group, reply, info);
1132         if (old_acts)
1133                 ovs_nla_free_flow_actions(old_acts);
1134         return 0;
1135
1136 err_unlock_ovs:
1137         ovs_unlock();
1138         kfree_skb(reply);
1139 err_kfree_acts:
1140         kfree(acts);
1141 error:
1142         return error;
1143 }
1144
1145 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1146 {
1147         struct nlattr **a = info->attrs;
1148         struct ovs_header *ovs_header = info->userhdr;
1149         struct sw_flow_key key;
1150         struct sk_buff *reply;
1151         struct sw_flow *flow;
1152         struct datapath *dp;
1153         struct sw_flow_match match;
1154         int err;
1155
1156         if (!a[OVS_FLOW_ATTR_KEY]) {
1157                 OVS_NLERR("Flow get message rejected, Key attribute missing.\n");
1158                 return -EINVAL;
1159         }
1160
1161         ovs_match_init(&match, &key, NULL);
1162         err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
1163         if (err)
1164                 return err;
1165
1166         ovs_lock();
1167         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1168         if (!dp) {
1169                 err = -ENODEV;
1170                 goto unlock;
1171         }
1172
1173         flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1174         if (!flow) {
1175                 err = -ENOENT;
1176                 goto unlock;
1177         }
1178
1179         reply = ovs_flow_cmd_build_info(dp, flow, ovs_header->dp_ifindex, info,
1180                                         OVS_FLOW_CMD_NEW, true);
1181         if (IS_ERR(reply)) {
1182                 err = PTR_ERR(reply);
1183                 goto unlock;
1184         }
1185
1186         ovs_unlock();
1187         return genlmsg_reply(reply, info);
1188 unlock:
1189         ovs_unlock();
1190         return err;
1191 }
1192
1193 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1194 {
1195         struct nlattr **a = info->attrs;
1196         struct ovs_header *ovs_header = info->userhdr;
1197         struct sw_flow_key key;
1198         struct sk_buff *reply;
1199         struct sw_flow *flow;
1200         struct datapath *dp;
1201         struct sw_flow_match match;
1202         int err;
1203
1204         if (likely(a[OVS_FLOW_ATTR_KEY])) {
1205                 ovs_match_init(&match, &key, NULL);
1206                 err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
1207                 if (unlikely(err))
1208                         return err;
1209         }
1210
1211         ovs_lock();
1212         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1213         if (unlikely(!dp)) {
1214                 err = -ENODEV;
1215                 goto unlock;
1216         }
1217         if (unlikely(!a[OVS_FLOW_ATTR_KEY])) {
1218                 err = ovs_flow_tbl_flush(&dp->table);
1219                 goto unlock;
1220         }
1221         flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1222         if (unlikely(!flow)) {
1223                 err = -ENOENT;
1224                 goto unlock;
1225         }
1226
1227         ovs_flow_tbl_remove(&dp->table, flow);
1228         ovs_unlock();
1229
1230         reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *)flow->sf_acts,
1231                                         info, false);
1232
1233         if (likely(reply)) {
1234                 if (likely(!IS_ERR(reply))) {
1235                         rcu_read_lock(); /* Keep RCU checker happy. */
1236                         err = ovs_flow_cmd_fill_info(dp, flow,
1237                                                      ovs_header->dp_ifindex,
1238                                                      reply, info->snd_portid,
1239                                                      info->snd_seq, 0,
1240                                                      OVS_FLOW_CMD_DEL);
1241                         rcu_read_unlock();
1242                         BUG_ON(err < 0);
1243                         ovs_notify(&dp_flow_genl_family, &ovs_dp_flow_multicast_group, reply, info);
1244                 } else {
1245                         genl_set_err(&dp_flow_genl_family, sock_net(skb->sk), 0,
1246                                      GROUP_ID(&ovs_dp_flow_multicast_group), PTR_ERR(reply));
1247
1248                 }
1249         }
1250
1251         ovs_flow_free(flow, true);
1252         return 0;
1253 unlock:
1254         ovs_unlock();
1255         return err;
1256 }
1257
1258 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1259 {
1260         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1261         struct table_instance *ti;
1262         struct datapath *dp;
1263
1264         rcu_read_lock();
1265         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1266         if (!dp) {
1267                 rcu_read_unlock();
1268                 return -ENODEV;
1269         }
1270
1271         ti = rcu_dereference(dp->table.ti);
1272         for (;;) {
1273                 struct sw_flow *flow;
1274                 u32 bucket, obj;
1275
1276                 bucket = cb->args[0];
1277                 obj = cb->args[1];
1278                 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1279                 if (!flow)
1280                         break;
1281
1282                 if (ovs_flow_cmd_fill_info(dp, flow, ovs_header->dp_ifindex, skb,
1283                                            NETLINK_CB(cb->skb).portid,
1284                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1285                                            OVS_FLOW_CMD_NEW) < 0)
1286                         break;
1287
1288                 cb->args[0] = bucket;
1289                 cb->args[1] = obj;
1290         }
1291         rcu_read_unlock();
1292         return skb->len;
1293 }
1294
1295 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1296         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1297         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1298         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1299 };
1300
1301 static struct genl_ops dp_flow_genl_ops[] = {
1302         { .cmd = OVS_FLOW_CMD_NEW,
1303           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1304           .policy = flow_policy,
1305           .doit = ovs_flow_cmd_new
1306         },
1307         { .cmd = OVS_FLOW_CMD_DEL,
1308           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1309           .policy = flow_policy,
1310           .doit = ovs_flow_cmd_del
1311         },
1312         { .cmd = OVS_FLOW_CMD_GET,
1313           .flags = 0,               /* OK for unprivileged users. */
1314           .policy = flow_policy,
1315           .doit = ovs_flow_cmd_get,
1316           .dumpit = ovs_flow_cmd_dump
1317         },
1318         { .cmd = OVS_FLOW_CMD_SET,
1319           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1320           .policy = flow_policy,
1321           .doit = ovs_flow_cmd_set,
1322         },
1323 };
1324
1325 static struct genl_family dp_flow_genl_family = {
1326         .id = GENL_ID_GENERATE,
1327         .hdrsize = sizeof(struct ovs_header),
1328         .name = OVS_FLOW_FAMILY,
1329         .version = OVS_FLOW_VERSION,
1330         .maxattr = OVS_FLOW_ATTR_MAX,
1331         .netnsok = true,
1332         .parallel_ops = true,
1333         .ops = dp_flow_genl_ops,
1334         .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1335         .mcgrps = &ovs_dp_flow_multicast_group,
1336         .n_mcgrps = 1,
1337 };
1338
1339 static size_t ovs_dp_cmd_msg_size(void)
1340 {
1341         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1342
1343         msgsize += nla_total_size(IFNAMSIZ);
1344         msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1345         msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1346         msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1347
1348         return msgsize;
1349 }
1350
1351 /* Called with ovs_mutex or RCU read lock. */
1352 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1353                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1354 {
1355         struct ovs_header *ovs_header;
1356         struct ovs_dp_stats dp_stats;
1357         struct ovs_dp_megaflow_stats dp_megaflow_stats;
1358         int err;
1359
1360         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1361                                    flags, cmd);
1362         if (!ovs_header)
1363                 goto error;
1364
1365         ovs_header->dp_ifindex = get_dpifindex(dp);
1366
1367         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1368         if (err)
1369                 goto nla_put_failure;
1370
1371         get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1372         if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1373                         &dp_stats))
1374                 goto nla_put_failure;
1375
1376         if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1377                         sizeof(struct ovs_dp_megaflow_stats),
1378                         &dp_megaflow_stats))
1379                 goto nla_put_failure;
1380
1381         if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1382                 goto nla_put_failure;
1383
1384         return genlmsg_end(skb, ovs_header);
1385
1386 nla_put_failure:
1387         genlmsg_cancel(skb, ovs_header);
1388 error:
1389         return -EMSGSIZE;
1390 }
1391
1392 static struct sk_buff *ovs_dp_cmd_alloc_info(struct genl_info *info)
1393 {
1394         return genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info, GFP_KERNEL);
1395 }
1396
1397 /* Called with rcu_read_lock or ovs_mutex. */
1398 static struct datapath *lookup_datapath(struct net *net,
1399                                         struct ovs_header *ovs_header,
1400                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1401 {
1402         struct datapath *dp;
1403
1404         if (!a[OVS_DP_ATTR_NAME])
1405                 dp = get_dp(net, ovs_header->dp_ifindex);
1406         else {
1407                 struct vport *vport;
1408
1409                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1410                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1411         }
1412         return dp ? dp : ERR_PTR(-ENODEV);
1413 }
1414
1415 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1416 {
1417         struct datapath *dp;
1418
1419         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1420         if (IS_ERR(dp))
1421                 return;
1422
1423         WARN(dp->user_features, "Dropping previously announced user features\n");
1424         dp->user_features = 0;
1425 }
1426
1427 static void ovs_dp_change(struct datapath *dp, struct nlattr **a)
1428 {
1429         if (a[OVS_DP_ATTR_USER_FEATURES])
1430                 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1431 }
1432
1433 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1434 {
1435         struct nlattr **a = info->attrs;
1436         struct vport_parms parms;
1437         struct sk_buff *reply;
1438         struct datapath *dp;
1439         struct vport *vport;
1440         struct ovs_net *ovs_net;
1441         int err, i;
1442
1443         err = -EINVAL;
1444         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1445                 goto err;
1446
1447         reply = ovs_dp_cmd_alloc_info(info);
1448         if (!reply)
1449                 return -ENOMEM;
1450
1451         err = -ENOMEM;
1452         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1453         if (dp == NULL)
1454                 goto err_free_reply;
1455
1456         ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1457
1458         /* Allocate table. */
1459         err = ovs_flow_tbl_init(&dp->table);
1460         if (err)
1461                 goto err_free_dp;
1462
1463         dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1464         if (!dp->stats_percpu) {
1465                 err = -ENOMEM;
1466                 goto err_destroy_table;
1467         }
1468
1469         for_each_possible_cpu(i) {
1470                 struct dp_stats_percpu *dpath_stats;
1471                 dpath_stats = per_cpu_ptr(dp->stats_percpu, i);
1472                 u64_stats_init(&dpath_stats->sync);
1473         }
1474
1475         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1476                             GFP_KERNEL);
1477         if (!dp->ports) {
1478                 err = -ENOMEM;
1479                 goto err_destroy_percpu;
1480         }
1481
1482         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1483                 INIT_HLIST_HEAD(&dp->ports[i]);
1484
1485         /* Set up our datapath device. */
1486         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1487         parms.type = OVS_VPORT_TYPE_INTERNAL;
1488         parms.options = NULL;
1489         parms.dp = dp;
1490         parms.port_no = OVSP_LOCAL;
1491         parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1492
1493         ovs_dp_change(dp, a);
1494
1495         /* So far only local changes have been made, now need the lock. */
1496         ovs_lock();
1497
1498         vport = new_vport(&parms);
1499         if (IS_ERR(vport)) {
1500                 err = PTR_ERR(vport);
1501                 if (err == -EBUSY)
1502                         err = -EEXIST;
1503
1504                 if (err == -EEXIST) {
1505                         /* An outdated user space instance that does not understand
1506                          * the concept of user_features has attempted to create a new
1507                          * datapath and is likely to reuse it. Drop all user features.
1508                          */
1509                         if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1510                                 ovs_dp_reset_user_features(skb, info);
1511                 }
1512
1513                 goto err_destroy_ports_array;
1514         }
1515
1516         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1517                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1518         BUG_ON(err < 0);
1519
1520         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1521         list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1522
1523         ovs_unlock();
1524
1525         ovs_notify(&dp_datapath_genl_family, &ovs_dp_datapath_multicast_group, reply, info);
1526         return 0;
1527
1528 err_destroy_ports_array:
1529         ovs_unlock();
1530         kfree(dp->ports);
1531 err_destroy_percpu:
1532         free_percpu(dp->stats_percpu);
1533 err_destroy_table:
1534         ovs_flow_tbl_destroy(&dp->table);
1535 err_free_dp:
1536         release_net(ovs_dp_get_net(dp));
1537         kfree(dp);
1538 err_free_reply:
1539         kfree_skb(reply);
1540 err:
1541         return err;
1542 }
1543
1544 /* Called with ovs_mutex. */
1545 static void __dp_destroy(struct datapath *dp)
1546 {
1547         int i;
1548
1549         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1550                 struct vport *vport;
1551                 struct hlist_node *n;
1552
1553                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1554                         if (vport->port_no != OVSP_LOCAL)
1555                                 ovs_dp_detach_port(vport);
1556         }
1557
1558         list_del_rcu(&dp->list_node);
1559
1560         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1561          * all ports in datapath are destroyed first before freeing datapath.
1562          */
1563         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1564
1565         /* RCU destroy the flow table */
1566         call_rcu(&dp->rcu, destroy_dp_rcu);
1567 }
1568
1569 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1570 {
1571         struct sk_buff *reply;
1572         struct datapath *dp;
1573         int err;
1574
1575         reply = ovs_dp_cmd_alloc_info(info);
1576         if (!reply)
1577                 return -ENOMEM;
1578
1579         ovs_lock();
1580         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1581         err = PTR_ERR(dp);
1582         if (IS_ERR(dp))
1583                 goto err_unlock_free;
1584
1585         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1586                                    info->snd_seq, 0, OVS_DP_CMD_DEL);
1587         BUG_ON(err < 0);
1588
1589         __dp_destroy(dp);
1590
1591         ovs_unlock();
1592         ovs_notify(&dp_datapath_genl_family, &ovs_dp_datapath_multicast_group, reply, info);
1593         return 0;
1594
1595 err_unlock_free:
1596         ovs_unlock();
1597         kfree_skb(reply);
1598         return err;
1599 }
1600
1601 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1602 {
1603         struct sk_buff *reply;
1604         struct datapath *dp;
1605         int err;
1606
1607         reply = ovs_dp_cmd_alloc_info(info);
1608         if (!reply)
1609                 return -ENOMEM;
1610
1611         ovs_lock();
1612         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1613         err = PTR_ERR(dp);
1614         if (IS_ERR(dp))
1615                 goto err_unlock_free;
1616
1617         ovs_dp_change(dp, info->attrs);
1618
1619         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1620                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1621         BUG_ON(err < 0);
1622
1623         ovs_unlock();
1624         ovs_notify(&dp_datapath_genl_family, &ovs_dp_datapath_multicast_group, reply, info);
1625         return 0;
1626
1627 err_unlock_free:
1628         ovs_unlock();
1629         kfree_skb(reply);
1630         return err;
1631 }
1632
1633 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1634 {
1635         struct sk_buff *reply;
1636         struct datapath *dp;
1637         int err;
1638
1639         reply = ovs_dp_cmd_alloc_info(info);
1640         if (!reply)
1641                 return -ENOMEM;
1642
1643         rcu_read_lock();
1644         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1645         if (IS_ERR(dp)) {
1646                 err = PTR_ERR(dp);
1647                 goto err_unlock_free;
1648         }
1649         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1650                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1651         BUG_ON(err < 0);
1652         rcu_read_unlock();
1653
1654         return genlmsg_reply(reply, info);
1655
1656 err_unlock_free:
1657         rcu_read_unlock();
1658         kfree_skb(reply);
1659         return err;
1660 }
1661
1662 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1663 {
1664         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1665         struct datapath *dp;
1666         int skip = cb->args[0];
1667         int i = 0;
1668
1669         rcu_read_lock();
1670         list_for_each_entry_rcu(dp, &ovs_net->dps, list_node) {
1671                 if (i >= skip &&
1672                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1673                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1674                                          OVS_DP_CMD_NEW) < 0)
1675                         break;
1676                 i++;
1677         }
1678         rcu_read_unlock();
1679
1680         cb->args[0] = i;
1681
1682         return skb->len;
1683 }
1684
1685 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1686         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1687         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1688         [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1689 };
1690
1691 static struct genl_ops dp_datapath_genl_ops[] = {
1692         { .cmd = OVS_DP_CMD_NEW,
1693           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1694           .policy = datapath_policy,
1695           .doit = ovs_dp_cmd_new
1696         },
1697         { .cmd = OVS_DP_CMD_DEL,
1698           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1699           .policy = datapath_policy,
1700           .doit = ovs_dp_cmd_del
1701         },
1702         { .cmd = OVS_DP_CMD_GET,
1703           .flags = 0,               /* OK for unprivileged users. */
1704           .policy = datapath_policy,
1705           .doit = ovs_dp_cmd_get,
1706           .dumpit = ovs_dp_cmd_dump
1707         },
1708         { .cmd = OVS_DP_CMD_SET,
1709           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1710           .policy = datapath_policy,
1711           .doit = ovs_dp_cmd_set,
1712         },
1713 };
1714
1715 static struct genl_family dp_datapath_genl_family = {
1716         .id = GENL_ID_GENERATE,
1717         .hdrsize = sizeof(struct ovs_header),
1718         .name = OVS_DATAPATH_FAMILY,
1719         .version = OVS_DATAPATH_VERSION,
1720         .maxattr = OVS_DP_ATTR_MAX,
1721         .netnsok = true,
1722         .parallel_ops = true,
1723         .ops = dp_datapath_genl_ops,
1724         .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1725         .mcgrps = &ovs_dp_datapath_multicast_group,
1726         .n_mcgrps = 1,
1727 };
1728
1729 /* Called with ovs_mutex or RCU read lock. */
1730 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1731                                    u32 portid, u32 seq, u32 flags, u8 cmd)
1732 {
1733         struct ovs_header *ovs_header;
1734         struct ovs_vport_stats vport_stats;
1735         int err;
1736
1737         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1738                                  flags, cmd);
1739         if (!ovs_header)
1740                 return -EMSGSIZE;
1741
1742         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1743
1744         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1745             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1746             nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)))
1747                 goto nla_put_failure;
1748
1749         ovs_vport_get_stats(vport, &vport_stats);
1750         if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1751                     &vport_stats))
1752                 goto nla_put_failure;
1753
1754         if (ovs_vport_get_upcall_portids(vport, skb))
1755                 goto nla_put_failure;
1756
1757         err = ovs_vport_get_options(vport, skb);
1758         if (err == -EMSGSIZE)
1759                 goto error;
1760
1761         return genlmsg_end(skb, ovs_header);
1762
1763 nla_put_failure:
1764         err = -EMSGSIZE;
1765 error:
1766         genlmsg_cancel(skb, ovs_header);
1767         return err;
1768 }
1769
1770 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1771 {
1772         return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1773 }
1774
1775 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1776 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1777                                          u32 seq, u8 cmd)
1778 {
1779         struct sk_buff *skb;
1780         int retval;
1781
1782         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1783         if (!skb)
1784                 return ERR_PTR(-ENOMEM);
1785
1786         retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1787         BUG_ON(retval < 0);
1788
1789         return skb;
1790 }
1791
1792 /* Called with ovs_mutex or RCU read lock. */
1793 static struct vport *lookup_vport(struct net *net,
1794                                   struct ovs_header *ovs_header,
1795                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1796 {
1797         struct datapath *dp;
1798         struct vport *vport;
1799
1800         if (a[OVS_VPORT_ATTR_NAME]) {
1801                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1802                 if (!vport)
1803                         return ERR_PTR(-ENODEV);
1804                 if (ovs_header->dp_ifindex &&
1805                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1806                         return ERR_PTR(-ENODEV);
1807                 return vport;
1808         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1809                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1810
1811                 if (port_no >= DP_MAX_PORTS)
1812                         return ERR_PTR(-EFBIG);
1813
1814                 dp = get_dp(net, ovs_header->dp_ifindex);
1815                 if (!dp)
1816                         return ERR_PTR(-ENODEV);
1817
1818                 vport = ovs_vport_ovsl_rcu(dp, port_no);
1819                 if (!vport)
1820                         return ERR_PTR(-ENODEV);
1821                 return vport;
1822         } else
1823                 return ERR_PTR(-EINVAL);
1824 }
1825
1826 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1827 {
1828         struct nlattr **a = info->attrs;
1829         struct ovs_header *ovs_header = info->userhdr;
1830         struct vport_parms parms;
1831         struct sk_buff *reply;
1832         struct vport *vport;
1833         struct datapath *dp;
1834         u32 port_no;
1835         int err;
1836
1837         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1838             !a[OVS_VPORT_ATTR_UPCALL_PID])
1839                 return -EINVAL;
1840
1841         port_no = a[OVS_VPORT_ATTR_PORT_NO]
1842                 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
1843         if (port_no >= DP_MAX_PORTS)
1844                 return -EFBIG;
1845
1846         reply = ovs_vport_cmd_alloc_info();
1847         if (!reply)
1848                 return -ENOMEM;
1849
1850         ovs_lock();
1851         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1852         err = -ENODEV;
1853         if (!dp)
1854                 goto exit_unlock_free;
1855
1856         if (port_no) {
1857                 vport = ovs_vport_ovsl(dp, port_no);
1858                 err = -EBUSY;
1859                 if (vport)
1860                         goto exit_unlock_free;
1861         } else {
1862                 for (port_no = 1; ; port_no++) {
1863                         if (port_no >= DP_MAX_PORTS) {
1864                                 err = -EFBIG;
1865                                 goto exit_unlock_free;
1866                         }
1867                         vport = ovs_vport_ovsl(dp, port_no);
1868                         if (!vport)
1869                                 break;
1870                 }
1871         }
1872
1873         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1874         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1875         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1876         parms.dp = dp;
1877         parms.port_no = port_no;
1878         parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
1879
1880         vport = new_vport(&parms);
1881         err = PTR_ERR(vport);
1882         if (IS_ERR(vport))
1883                 goto exit_unlock_free;
1884
1885         err = 0;
1886         if (a[OVS_VPORT_ATTR_STATS])
1887                 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1888
1889         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1890                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1891         BUG_ON(err < 0);
1892         ovs_unlock();
1893
1894         ovs_notify(&dp_vport_genl_family, &ovs_dp_vport_multicast_group, reply, info);
1895         return 0;
1896
1897 exit_unlock_free:
1898         ovs_unlock();
1899         kfree_skb(reply);
1900         return err;
1901 }
1902
1903 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1904 {
1905         struct nlattr **a = info->attrs;
1906         struct sk_buff *reply;
1907         struct vport *vport;
1908         int err;
1909
1910         reply = ovs_vport_cmd_alloc_info();
1911         if (!reply)
1912                 return -ENOMEM;
1913
1914         ovs_lock();
1915         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1916         err = PTR_ERR(vport);
1917         if (IS_ERR(vport))
1918                 goto exit_unlock_free;
1919
1920         if (a[OVS_VPORT_ATTR_TYPE] &&
1921             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
1922                 err = -EINVAL;
1923                 goto exit_unlock_free;
1924         }
1925
1926         if (a[OVS_VPORT_ATTR_OPTIONS]) {
1927                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1928                 if (err)
1929                         goto exit_unlock_free;
1930         }
1931
1932         if (a[OVS_VPORT_ATTR_STATS])
1933                 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1934
1935
1936         if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
1937                 err = ovs_vport_set_upcall_portids(vport,
1938                                                    a[OVS_VPORT_ATTR_UPCALL_PID]);
1939                 if (err)
1940                         goto exit_unlock_free;
1941         }
1942
1943         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1944                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1945         BUG_ON(err < 0);
1946         ovs_unlock();
1947
1948         ovs_notify(&dp_vport_genl_family, &ovs_dp_vport_multicast_group, reply, info);
1949         return 0;
1950
1951 exit_unlock_free:
1952         ovs_unlock();
1953         kfree_skb(reply);
1954         return err;
1955 }
1956
1957 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1958 {
1959         struct nlattr **a = info->attrs;
1960         struct sk_buff *reply;
1961         struct vport *vport;
1962         int err;
1963
1964         reply = ovs_vport_cmd_alloc_info();
1965         if (!reply)
1966                 return -ENOMEM;
1967
1968         ovs_lock();
1969         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1970         err = PTR_ERR(vport);
1971         if (IS_ERR(vport))
1972                 goto exit_unlock_free;
1973
1974         if (vport->port_no == OVSP_LOCAL) {
1975                 err = -EINVAL;
1976                 goto exit_unlock_free;
1977         }
1978
1979         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1980                                       info->snd_seq, 0, OVS_VPORT_CMD_DEL);
1981         BUG_ON(err < 0);
1982         ovs_dp_detach_port(vport);
1983         ovs_unlock();
1984
1985         ovs_notify(&dp_vport_genl_family, &ovs_dp_vport_multicast_group, reply, info);
1986         return 0;
1987
1988 exit_unlock_free:
1989         ovs_unlock();
1990         kfree_skb(reply);
1991         return err;
1992 }
1993
1994 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1995 {
1996         struct nlattr **a = info->attrs;
1997         struct ovs_header *ovs_header = info->userhdr;
1998         struct sk_buff *reply;
1999         struct vport *vport;
2000         int err;
2001
2002         reply = ovs_vport_cmd_alloc_info();
2003         if (!reply)
2004                 return -ENOMEM;
2005
2006         rcu_read_lock();
2007         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2008         err = PTR_ERR(vport);
2009         if (IS_ERR(vport))
2010                 goto exit_unlock_free;
2011         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2012                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2013         BUG_ON(err < 0);
2014         rcu_read_unlock();
2015
2016         return genlmsg_reply(reply, info);
2017
2018 exit_unlock_free:
2019         rcu_read_unlock();
2020         kfree_skb(reply);
2021         return err;
2022 }
2023
2024 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2025 {
2026         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2027         struct datapath *dp;
2028         int bucket = cb->args[0], skip = cb->args[1];
2029         int i, j = 0;
2030
2031         rcu_read_lock();
2032         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2033         if (!dp) {
2034                 rcu_read_unlock();
2035                 return -ENODEV;
2036         }
2037         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2038                 struct vport *vport;
2039
2040                 j = 0;
2041                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2042                         if (j >= skip &&
2043                             ovs_vport_cmd_fill_info(vport, skb,
2044                                                     NETLINK_CB(cb->skb).portid,
2045                                                     cb->nlh->nlmsg_seq,
2046                                                     NLM_F_MULTI,
2047                                                     OVS_VPORT_CMD_NEW) < 0)
2048                                 goto out;
2049
2050                         j++;
2051                 }
2052                 skip = 0;
2053         }
2054 out:
2055         rcu_read_unlock();
2056
2057         cb->args[0] = i;
2058         cb->args[1] = j;
2059
2060         return skb->len;
2061 }
2062
2063 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2064         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2065         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2066         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2067         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2068         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
2069         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2070 };
2071
2072 static struct genl_ops dp_vport_genl_ops[] = {
2073         { .cmd = OVS_VPORT_CMD_NEW,
2074           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2075           .policy = vport_policy,
2076           .doit = ovs_vport_cmd_new
2077         },
2078         { .cmd = OVS_VPORT_CMD_DEL,
2079           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2080           .policy = vport_policy,
2081           .doit = ovs_vport_cmd_del
2082         },
2083         { .cmd = OVS_VPORT_CMD_GET,
2084           .flags = 0,               /* OK for unprivileged users. */
2085           .policy = vport_policy,
2086           .doit = ovs_vport_cmd_get,
2087           .dumpit = ovs_vport_cmd_dump
2088         },
2089         { .cmd = OVS_VPORT_CMD_SET,
2090           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2091           .policy = vport_policy,
2092           .doit = ovs_vport_cmd_set,
2093         },
2094 };
2095
2096 struct genl_family dp_vport_genl_family = {
2097         .id = GENL_ID_GENERATE,
2098         .hdrsize = sizeof(struct ovs_header),
2099         .name = OVS_VPORT_FAMILY,
2100         .version = OVS_VPORT_VERSION,
2101         .maxattr = OVS_VPORT_ATTR_MAX,
2102         .netnsok = true,
2103         .parallel_ops = true,
2104         .ops = dp_vport_genl_ops,
2105         .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2106         .mcgrps = &ovs_dp_vport_multicast_group,
2107         .n_mcgrps = 1,
2108 };
2109
2110 static struct genl_family *dp_genl_families[] = {
2111         &dp_datapath_genl_family,
2112         &dp_vport_genl_family,
2113         &dp_flow_genl_family,
2114         &dp_packet_genl_family,
2115 };
2116
2117 static void dp_unregister_genl(int n_families)
2118 {
2119         int i;
2120
2121         for (i = 0; i < n_families; i++)
2122                 genl_unregister_family(dp_genl_families[i]);
2123 }
2124
2125 static int dp_register_genl(void)
2126 {
2127         int err;
2128         int i;
2129
2130         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2131
2132                 err = genl_register_family(dp_genl_families[i]);
2133                 if (err)
2134                         goto error;
2135         }
2136
2137         return 0;
2138
2139 error:
2140         dp_unregister_genl(i);
2141         return err;
2142 }
2143
2144 static int __net_init ovs_init_net(struct net *net)
2145 {
2146         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2147
2148         INIT_LIST_HEAD(&ovs_net->dps);
2149         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2150         return 0;
2151 }
2152
2153 static void __net_exit ovs_exit_net(struct net *net)
2154 {
2155         struct datapath *dp, *dp_next;
2156         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2157
2158         ovs_lock();
2159         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2160                 __dp_destroy(dp);
2161         ovs_unlock();
2162
2163         cancel_work_sync(&ovs_net->dp_notify_work);
2164 }
2165
2166 static struct pernet_operations ovs_net_ops = {
2167         .init = ovs_init_net,
2168         .exit = ovs_exit_net,
2169         .id   = &ovs_net_id,
2170         .size = sizeof(struct ovs_net),
2171 };
2172
2173 DEFINE_COMPAT_PNET_REG_FUNC(device);
2174
2175 static int __init dp_init(void)
2176 {
2177         int err;
2178
2179         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2180
2181         pr_info("Open vSwitch switching datapath %s, built "__DATE__" "__TIME__"\n",
2182                 VERSION);
2183
2184         err = ovs_flow_init();
2185         if (err)
2186                 goto error;
2187
2188         err = ovs_vport_init();
2189         if (err)
2190                 goto error_flow_exit;
2191
2192         err = register_pernet_device(&ovs_net_ops);
2193         if (err)
2194                 goto error_vport_exit;
2195
2196         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2197         if (err)
2198                 goto error_netns_exit;
2199
2200         err = dp_register_genl();
2201         if (err < 0)
2202                 goto error_unreg_notifier;
2203
2204         return 0;
2205
2206 error_unreg_notifier:
2207         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2208 error_netns_exit:
2209         unregister_pernet_device(&ovs_net_ops);
2210 error_vport_exit:
2211         ovs_vport_exit();
2212 error_flow_exit:
2213         ovs_flow_exit();
2214 error:
2215         return err;
2216 }
2217
2218 static void dp_cleanup(void)
2219 {
2220         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2221         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2222         unregister_pernet_device(&ovs_net_ops);
2223         rcu_barrier();
2224         ovs_vport_exit();
2225         ovs_flow_exit();
2226 }
2227
2228 module_init(dp_init);
2229 module_exit(dp_cleanup);
2230
2231 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2232 MODULE_LICENSE("GPL");
2233 MODULE_VERSION(VERSION);