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 or ovs_mutex. */
 144 static struct datapath *get_dp(struct net *net, int dp_ifindex)
 145 {
 146         struct datapath *dp = NULL;
 147         struct net_device *dev;
 148
 149         rcu_read_lock();
 150         dev = dev_get_by_index_rcu(net, dp_ifindex);
 151         if (dev) {
 152                 struct vport *vport = ovs_internal_dev_get_vport(dev);
 153                 if (vport)
 154                         dp = vport->dp;
 155         }
 156         rcu_read_unlock();
 157
 158         return dp;
 159 }
 160
 161 /* Must be called with rcu_read_lock or ovs_mutex. */
 162 const char *ovs_dp_name(const struct datapath *dp)
 163 {
 164         struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
 165         return vport->ops->get_name(vport);
 166 }
 167
 168 static int get_dpifindex(struct datapath *dp)
 169 {
 170         struct vport *local;
 171         int ifindex;
 172
 173         rcu_read_lock();
 174
 175         local = ovs_vport_rcu(dp, OVSP_LOCAL);
 176         if (local)
 177                 ifindex = netdev_vport_priv(local)->dev->ifindex;
 178         else
 179                 ifindex = 0;
 180
 181         rcu_read_unlock();
 182
 183         return ifindex;
 184 }
 185
 186 static void destroy_dp_rcu(struct rcu_head *rcu)
 187 {
 188         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 189
 190         ovs_flow_tbl_destroy(&dp->table);
 191         free_percpu(dp->stats_percpu);
 192         release_net(ovs_dp_get_net(dp));
 193         kfree(dp->ports);
 194         kfree(dp);
 195 }
 196
 197 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
 198                                             u16 port_no)
 199 {
 200         return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
 201 }
 202
 203 /* Called with ovs_mutex or RCU read lock. */
 204 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
 205 {
 206         struct vport *vport;
 207         struct hlist_head *head;
 208
 209         head = vport_hash_bucket(dp, port_no);
 210         hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
 211                 if (vport->port_no == port_no)
 212                         return vport;
 213         }
 214         return NULL;
 215 }
 216
 217 /* Called with ovs_mutex. */
 218 static struct vport *new_vport(const struct vport_parms *parms)
 219 {
 220         struct vport *vport;
 221
 222         vport = ovs_vport_add(parms);
 223         if (!IS_ERR(vport)) {
 224                 struct datapath *dp = parms->dp;
 225                 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
 226
 227                 hlist_add_head_rcu(&vport->dp_hash_node, head);
 228         }
 229         return vport;
 230 }
 231
 232 void ovs_dp_detach_port(struct vport *p)
 233 {
 234         ASSERT_OVSL();
 235
 236         /* First drop references to device. */
 237         hlist_del_rcu(&p->dp_hash_node);
 238
 239         /* Then destroy it. */
 240         ovs_vport_del(p);
 241 }
 242
 243 void ovs_dp_process_packet_with_key(struct sk_buff *skb,
 244                                     struct sw_flow_key *pkt_key,
 245                                     bool recirc)
 246 {
 247         const struct vport *p = OVS_CB(skb)->input_vport;
 248         struct datapath *dp = p->dp;
 249         struct sw_flow *flow;
 250         struct dp_stats_percpu *stats;
 251         u64 *stats_counter;
 252         u32 n_mask_hit;
 253
 254         stats = this_cpu_ptr(dp->stats_percpu);
 255
 256         /* Look up flow. */
 257         flow = ovs_flow_tbl_lookup_stats(&dp->table, pkt_key, skb_get_hash(skb),
 258                                          &n_mask_hit);
 259         if (unlikely(!flow)) {
 260                 struct dp_upcall_info upcall;
 261
 262                 upcall.cmd = OVS_PACKET_CMD_MISS;
 263                 upcall.key = pkt_key;
 264                 upcall.userdata = NULL;
 265                 upcall.portid = ovs_vport_find_upcall_portid(p, skb);
 266                 ovs_dp_upcall(dp, skb, &upcall);
 267                 consume_skb(skb);
 268                 stats_counter = &stats->n_missed;
 269                 goto out;
 270         }
 271
 272         OVS_CB(skb)->pkt_key = pkt_key;
 273         OVS_CB(skb)->flow = flow;
 274
 275         ovs_flow_stats_update(OVS_CB(skb)->flow, pkt_key->tp.flags, skb);
 276         ovs_execute_actions(dp, skb, recirc);
 277         stats_counter = &stats->n_hit;
 278
 279 out:
 280         /* Update datapath statistics. */
 281         u64_stats_update_begin(&stats->sync);
 282         (*stats_counter)++;
 283         stats->n_mask_hit += n_mask_hit;
 284         u64_stats_update_end(&stats->sync);
 285 }
 286
 287 /* Must be called with rcu_read_lock. */
 288 void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
 289 {
 290         int error;
 291         struct sw_flow_key key;
 292
 293         OVS_CB(skb)->input_vport = p;
 294
 295         /* Extract flow from 'skb' into 'key'. */
 296         error = ovs_flow_extract(skb, p->port_no, &key);
 297         if (unlikely(error)) {
 298                 kfree_skb(skb);
 299                 return;
 300         }
 301
 302         ovs_dp_process_packet_with_key(skb, &key, false);
 303 }
 304
 305 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
 306                   const struct dp_upcall_info *upcall_info)
 307 {
 308         struct dp_stats_percpu *stats;
 309         int err;
 310
 311         if (upcall_info->portid == 0) {
 312                 err = -ENOTCONN;
 313                 goto err;
 314         }
 315
 316         if (!skb_is_gso(skb))
 317                 err = queue_userspace_packet(dp, skb, upcall_info);
 318         else
 319                 err = queue_gso_packets(dp, skb, upcall_info);
 320         if (err)
 321                 goto err;
 322
 323         return 0;
 324
 325 err:
 326         stats = this_cpu_ptr(dp->stats_percpu);
 327
 328         u64_stats_update_begin(&stats->sync);
 329         stats->n_lost++;
 330         u64_stats_update_end(&stats->sync);
 331
 332         return err;
 333 }
 334
 335 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
 336                              const struct dp_upcall_info *upcall_info)
 337 {
 338         unsigned short gso_type = skb_shinfo(skb)->gso_type;
 339         struct dp_upcall_info later_info;
 340         struct sw_flow_key later_key;
 341         struct sk_buff *segs, *nskb;
 342         int err;
 343
 344         segs = __skb_gso_segment(skb, NETIF_F_SG, false);
 345         if (IS_ERR(segs))
 346                 return PTR_ERR(segs);
 347
 348         /* Queue all of the segments. */
 349         skb = segs;
 350         do {
 351                 err = queue_userspace_packet(dp, skb, upcall_info);
 352                 if (err)
 353                         break;
 354
 355                 if (skb == segs && gso_type & SKB_GSO_UDP) {
 356                         /* The initial flow key extracted by ovs_flow_extract()
 357                          * in this case is for a first fragment, so we need to
 358                          * properly mark later fragments.
 359                          */
 360                         later_key = *upcall_info->key;
 361                         later_key.ip.frag = OVS_FRAG_TYPE_LATER;
 362
 363                         later_info = *upcall_info;
 364                         later_info.key = &later_key;
 365                         upcall_info = &later_info;
 366                 }
 367         } while ((skb = skb->next));
 368
 369         /* Free all of the segments. */
 370         skb = segs;
 371         do {
 372                 nskb = skb->next;
 373                 if (err)
 374                         kfree_skb(skb);
 375                 else
 376                         consume_skb(skb);
 377         } while ((skb = nskb));
 378         return err;
 379 }
 380
 381 static size_t key_attr_size(void)
 382 {
 383         /* Whenever adding new OVS_KEY_ FIELDS, we should consider
 384          * updating this function.  */
 385         BUILD_BUG_ON(OVS_KEY_ATTR_IPV4_TUNNEL != 21);
 386
 387         return    nla_total_size(4)   /* OVS_KEY_ATTR_PRIORITY */
 388                 + nla_total_size(0)   /* OVS_KEY_ATTR_TUNNEL */
 389                   + nla_total_size(8)   /* OVS_TUNNEL_KEY_ATTR_ID */
 390                   + nla_total_size(4)   /* OVS_TUNNEL_KEY_ATTR_IPV4_SRC */
 391                   + nla_total_size(4)   /* OVS_TUNNEL_KEY_ATTR_IPV4_DST */
 392                   + nla_total_size(1)   /* OVS_TUNNEL_KEY_ATTR_TOS */
 393                   + nla_total_size(1)   /* OVS_TUNNEL_KEY_ATTR_TTL */
 394                   + nla_total_size(0)   /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
 395                   + nla_total_size(0)   /* OVS_TUNNEL_KEY_ATTR_CSUM */
 396                 + nla_total_size(4)   /* OVS_KEY_ATTR_IN_PORT */
 397                 + nla_total_size(4)   /* OVS_KEY_ATTR_SKB_MARK */
 398                 + nla_total_size(4)   /* OVS_KEY_ATTR_DP_HASH */
 399                 + nla_total_size(4)   /* OVS_KEY_ATTR_RECIRC_ID */
 400                 + nla_total_size(12)  /* OVS_KEY_ATTR_ETHERNET */
 401                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 402                 + nla_total_size(4)   /* OVS_KEY_ATTR_8021Q */
 403                 + nla_total_size(0)   /* OVS_KEY_ATTR_ENCAP */
 404                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 405                 + nla_total_size(40)  /* OVS_KEY_ATTR_IPV6 */
 406                 + nla_total_size(2)   /* OVS_KEY_ATTR_ICMPV6 */
 407                 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
 408 }
 409
 410 static size_t upcall_msg_size(const struct nlattr *userdata,
 411                               unsigned int hdrlen)
 412 {
 413         size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
 414                 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
 415                 + nla_total_size(key_attr_size()); /* OVS_PACKET_ATTR_KEY */
 416
 417         /* OVS_PACKET_ATTR_USERDATA */
 418         if (userdata)
 419                 size += NLA_ALIGN(userdata->nla_len);
 420
 421         return size;
 422 }
 423
 424 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
 425                                   const struct dp_upcall_info *upcall_info)
 426 {
 427         struct ovs_header *upcall;
 428         struct sk_buff *nskb = NULL;
 429         struct sk_buff *user_skb; /* to be queued to userspace */
 430         struct nlattr *nla;
 431         struct genl_info info = {
 432 #if LINUX_VERSION_CODE >= KERNEL_VERSION(3,14,0)
 433                 .dst_sk = ovs_dp_get_net(dp)->genl_sock,
 434 #endif
 435                 .snd_portid = upcall_info->portid,
 436         };
 437         size_t len;
 438         unsigned int hlen;
 439         int err, dp_ifindex;
 440
 441         dp_ifindex = get_dpifindex(dp);
 442         if (!dp_ifindex)
 443                 return -ENODEV;
 444
 445         if (vlan_tx_tag_present(skb)) {
 446                 nskb = skb_clone(skb, GFP_ATOMIC);
 447                 if (!nskb)
 448                         return -ENOMEM;
 449
 450                 nskb = __vlan_put_tag(nskb, nskb->vlan_proto, vlan_tx_tag_get(nskb));
 451                 if (!nskb)
 452                         return -ENOMEM;
 453
 454                 vlan_set_tci(nskb, 0);
 455
 456                 skb = nskb;
 457         }
 458
 459         if (nla_attr_size(skb->len) > USHRT_MAX) {
 460                 err = -EFBIG;
 461                 goto out;
 462         }
 463
 464         /* Complete checksum if needed */
 465         if (skb->ip_summed == CHECKSUM_PARTIAL &&
 466             (err = skb_checksum_help(skb)))
 467                 goto out;
 468
 469         /* Older versions of OVS user space enforce alignment of the last
 470          * Netlink attribute to NLA_ALIGNTO which would require extensive
 471          * padding logic. Only perform zerocopy if padding is not required.
 472          */
 473         if (dp->user_features & OVS_DP_F_UNALIGNED)
 474                 hlen = skb_zerocopy_headlen(skb);
 475         else
 476                 hlen = skb->len;
 477
 478         len = upcall_msg_size(upcall_info->userdata, hlen);
 479         user_skb = genlmsg_new_unicast(len, &info, GFP_ATOMIC);
 480         if (!user_skb) {
 481                 err = -ENOMEM;
 482                 goto out;
 483         }
 484
 485         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 486                              0, upcall_info->cmd);
 487         upcall->dp_ifindex = dp_ifindex;
 488
 489         nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
 490         ovs_nla_put_flow(upcall_info->key, upcall_info->key, user_skb);
 491         nla_nest_end(user_skb, nla);
 492
 493         if (upcall_info->userdata)
 494                 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
 495                           nla_len(upcall_info->userdata),
 496                           nla_data(upcall_info->userdata));
 497
 498         /* Only reserve room for attribute header, packet data is added
 499          * in skb_zerocopy() */
 500         if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
 501                 err = -ENOBUFS;
 502                 goto out;
 503         }
 504         nla->nla_len = nla_attr_size(skb->len);
 505
 506         err = skb_zerocopy(user_skb, skb, skb->len, hlen);
 507         if (err)
 508                 goto out;
 509
 510         /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
 511         if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
 512                 size_t plen = NLA_ALIGN(user_skb->len) - user_skb->len;
 513
 514                 if (plen > 0)
 515                         memset(skb_put(user_skb, plen), 0, plen);
 516         }
 517
 518         ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
 519
 520         err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
 521 out:
 522         if (err)
 523                 skb_tx_error(skb);
 524         kfree_skb(nskb);
 525         return err;
 526 }
 527
 528 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 529 {
 530         struct ovs_header *ovs_header = info->userhdr;
 531         struct nlattr **a = info->attrs;
 532         struct sw_flow_actions *acts;
 533         struct sk_buff *packet;
 534         struct sw_flow *flow;
 535         struct datapath *dp;
 536         struct ethhdr *eth;
 537         struct vport *input_vport;
 538         int len;
 539         int err;
 540
 541         err = -EINVAL;
 542         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 543             !a[OVS_PACKET_ATTR_ACTIONS])
 544                 goto err;
 545
 546         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 547         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 548         err = -ENOMEM;
 549         if (!packet)
 550                 goto err;
 551         skb_reserve(packet, NET_IP_ALIGN);
 552
 553         nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
 554
 555         skb_reset_mac_header(packet);
 556         eth = eth_hdr(packet);
 557
 558         /* Normally, setting the skb 'protocol' field would be handled by a
 559          * call to eth_type_trans(), but it assumes there's a sending
 560          * device, which we may not have. */
 561         if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
 562                 packet->protocol = eth->h_proto;
 563         else
 564                 packet->protocol = htons(ETH_P_802_2);
 565
 566         /* Build an sw_flow for sending this packet. */
 567         flow = ovs_flow_alloc();
 568         err = PTR_ERR(flow);
 569         if (IS_ERR(flow))
 570                 goto err_kfree_skb;
 571
 572         err = ovs_flow_extract(packet, -1, &flow->key);
 573         if (err)
 574                 goto err_flow_free;
 575
 576         err = ovs_nla_get_flow_metadata(flow, a[OVS_PACKET_ATTR_KEY]);
 577         if (err)
 578                 goto err_flow_free;
 579         acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_PACKET_ATTR_ACTIONS]));
 580         err = PTR_ERR(acts);
 581         if (IS_ERR(acts))
 582                 goto err_flow_free;
 583
 584         err = ovs_nla_copy_actions(a[OVS_PACKET_ATTR_ACTIONS],
 585                                    &flow->key, 0, &acts);
 586         rcu_assign_pointer(flow->sf_acts, acts);
 587         if (err)
 588                 goto err_flow_free;
 589
 590         OVS_CB(packet)->flow = flow;
 591         OVS_CB(packet)->pkt_key = &flow->key;
 592         packet->priority = flow->key.phy.priority;
 593         packet->mark = flow->key.phy.skb_mark;
 594
 595         rcu_read_lock();
 596         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 597         err = -ENODEV;
 598         if (!dp)
 599                 goto err_unlock;
 600
 601         input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
 602         if (!input_vport)
 603                 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
 604
 605         if (!input_vport)
 606                 goto err_unlock;
 607
 608         OVS_CB(packet)->input_vport = input_vport;
 609
 610         local_bh_disable();
 611         err = ovs_execute_actions(dp, packet, false);
 612         local_bh_enable();
 613         rcu_read_unlock();
 614
 615         ovs_flow_free(flow, false);
 616         return err;
 617
 618 err_unlock:
 619         rcu_read_unlock();
 620 err_flow_free:
 621         ovs_flow_free(flow, false);
 622 err_kfree_skb:
 623         kfree_skb(packet);
 624 err:
 625         return err;
 626 }
 627
 628 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 629         [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
 630         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 631         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 632 };
 633
 634 static struct genl_ops dp_packet_genl_ops[] = {
 635         { .cmd = OVS_PACKET_CMD_EXECUTE,
 636           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 637           .policy = packet_policy,
 638           .doit = ovs_packet_cmd_execute
 639         }
 640 };
 641
 642 static struct genl_family dp_packet_genl_family = {
 643         .id = GENL_ID_GENERATE,
 644         .hdrsize = sizeof(struct ovs_header),
 645         .name = OVS_PACKET_FAMILY,
 646         .version = OVS_PACKET_VERSION,
 647         .maxattr = OVS_PACKET_ATTR_MAX,
 648         .netnsok = true,
 649         .parallel_ops = true,
 650         .ops = dp_packet_genl_ops,
 651         .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
 652 };
 653
 654 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats,
 655                          struct ovs_dp_megaflow_stats *mega_stats)
 656 {
 657         int i;
 658
 659         memset(mega_stats, 0, sizeof(*mega_stats));
 660
 661         stats->n_flows = ovs_flow_tbl_count(&dp->table);
 662         mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
 663
 664         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 665
 666         for_each_possible_cpu(i) {
 667                 const struct dp_stats_percpu *percpu_stats;
 668                 struct dp_stats_percpu local_stats;
 669                 unsigned int start;
 670
 671                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 672
 673                 do {
 674                         start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
 675                         local_stats = *percpu_stats;
 676                 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
 677
 678                 stats->n_hit += local_stats.n_hit;
 679                 stats->n_missed += local_stats.n_missed;
 680                 stats->n_lost += local_stats.n_lost;
 681                 mega_stats->n_mask_hit += local_stats.n_mask_hit;
 682         }
 683 }
 684
 685 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
 686 {
 687         return NLMSG_ALIGN(sizeof(struct ovs_header))
 688                 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
 689                 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_MASK */
 690                 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
 691                 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
 692                 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
 693                 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
 694 }
 695
 696 /* Called with ovs_mutex or RCU read lock. */
 697 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
 698                                   struct sk_buff *skb, u32 portid,
 699                                   u32 seq, u32 flags, u8 cmd)
 700 {
 701         const int skb_orig_len = skb->len;
 702         struct nlattr *start;
 703         struct ovs_flow_stats stats;
 704         __be16 tcp_flags;
 705         unsigned long used;
 706         struct ovs_header *ovs_header;
 707         struct nlattr *nla;
 708         int err;
 709
 710         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
 711         if (!ovs_header)
 712                 return -EMSGSIZE;
 713
 714         ovs_header->dp_ifindex = dp_ifindex;
 715
 716         /* Fill flow key. */
 717         nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
 718         if (!nla)
 719                 goto nla_put_failure;
 720
 721         err = ovs_nla_put_flow(&flow->unmasked_key, &flow->unmasked_key, skb);
 722         if (err)
 723                 goto error;
 724         nla_nest_end(skb, nla);
 725
 726         nla = nla_nest_start(skb, OVS_FLOW_ATTR_MASK);
 727         if (!nla)
 728                 goto nla_put_failure;
 729
 730         err = ovs_nla_put_flow(&flow->key, &flow->mask->key, skb);
 731         if (err)
 732                 goto error;
 733
 734         nla_nest_end(skb, nla);
 735
 736         ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
 737
 738         if (used &&
 739             nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
 740                 goto nla_put_failure;
 741
 742         if (stats.n_packets &&
 743             nla_put(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats))
 744                 goto nla_put_failure;
 745
 746         if ((u8)ntohs(tcp_flags) &&
 747              nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
 748                 goto nla_put_failure;
 749
 750         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 751          * this is the first flow to be dumped into 'skb'.  This is unusual for
 752          * Netlink but individual action lists can be longer than
 753          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 754          * The userspace caller can always fetch the actions separately if it
 755          * really wants them.  (Most userspace callers in fact don't care.)
 756          *
 757          * This can only fail for dump operations because the skb is always
 758          * properly sized for single flows.
 759          */
 760         start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
 761         if (start) {
 762                 const struct sw_flow_actions *sf_acts;
 763
 764                 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
 765                 err = ovs_nla_put_actions(sf_acts->actions,
 766                                           sf_acts->actions_len, skb);
 767
 768                 if (!err)
 769                         nla_nest_end(skb, start);
 770                 else {
 771                         if (skb_orig_len)
 772                                 goto error;
 773
 774                         nla_nest_cancel(skb, start);
 775                 }
 776         } else if (skb_orig_len)
 777                 goto nla_put_failure;
 778
 779         return genlmsg_end(skb, ovs_header);
 780
 781 nla_put_failure:
 782         err = -EMSGSIZE;
 783 error:
 784         genlmsg_cancel(skb, ovs_header);
 785         return err;
 786 }
 787
 788 /* May not be called with RCU read lock. */
 789 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
 790                                                struct genl_info *info,
 791                                                bool always)
 792 {
 793         struct sk_buff *skb;
 794
 795         if (!always && !ovs_must_notify(info, &ovs_dp_flow_multicast_group))
 796                 return NULL;
 797
 798         skb = genlmsg_new_unicast(ovs_flow_cmd_msg_size(acts), info, GFP_KERNEL);
 799
 800         if (!skb)
 801                 return ERR_PTR(-ENOMEM);
 802
 803         return skb;
 804 }
 805
 806 /* Called with ovs_mutex. */
 807 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
 808                                                int dp_ifindex,
 809                                                struct genl_info *info, u8 cmd,
 810                                                bool always)
 811 {
 812         struct sk_buff *skb;
 813         int retval;
 814
 815         skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts), info,
 816                                       always);
 817         if (!skb || IS_ERR(skb))
 818                 return skb;
 819
 820         retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
 821                                         info->snd_portid, info->snd_seq, 0,
 822                                         cmd);
 823         BUG_ON(retval < 0);
 824         return skb;
 825 }
 826
 827 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
 828 {
 829         struct nlattr **a = info->attrs;
 830         struct ovs_header *ovs_header = info->userhdr;
 831         struct sw_flow *flow, *new_flow;
 832         struct sw_flow_mask mask;
 833         struct sk_buff *reply;
 834         struct datapath *dp;
 835         struct sw_flow_actions *acts;
 836         struct sw_flow_match match;
 837         int error;
 838
 839         /* Must have key and actions. */
 840         error = -EINVAL;
 841         if (!a[OVS_FLOW_ATTR_KEY])
 842                 goto error;
 843         if (!a[OVS_FLOW_ATTR_ACTIONS])
 844                 goto error;
 845
 846         /* Most of the time we need to allocate a new flow, do it before
 847          * locking. */
 848         new_flow = ovs_flow_alloc();
 849         if (IS_ERR(new_flow)) {
 850                 error = PTR_ERR(new_flow);
 851                 goto error;
 852         }
 853
 854         /* Extract key. */
 855         ovs_match_init(&match, &new_flow->unmasked_key, &mask);
 856         error = ovs_nla_get_match(&match,
 857                                   a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
 858         if (error)
 859                 goto err_kfree_flow;
 860
 861         ovs_flow_mask_key(&new_flow->key, &new_flow->unmasked_key, &mask);
 862
 863         /* Validate actions. */
 864         acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
 865         error = PTR_ERR(acts);
 866         if (IS_ERR(acts))
 867                 goto err_kfree_flow;
 868
 869         error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS], &new_flow->key,
 870                                      0, &acts);
 871         if (error) {
 872                 OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
 873                 goto err_kfree_acts;
 874         }
 875
 876         reply = ovs_flow_cmd_alloc_info(acts, info, false);
 877         if (IS_ERR(reply)) {
 878                 error = PTR_ERR(reply);
 879                 goto err_kfree_acts;
 880         }
 881
 882         ovs_lock();
 883         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 884         if (unlikely(!dp)) {
 885                 error = -ENODEV;
 886                 goto err_unlock_ovs;
 887         }
 888         /* Check if this is a duplicate flow */
 889         flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->unmasked_key);
 890         if (likely(!flow)) {
 891                 rcu_assign_pointer(new_flow->sf_acts, acts);
 892
 893                 /* Put flow in bucket. */
 894                 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
 895                 if (unlikely(error)) {
 896                         acts = NULL;
 897                         goto err_unlock_ovs;
 898                 }
 899
 900                 if (unlikely(reply)) {
 901                         error = ovs_flow_cmd_fill_info(new_flow,
 902                                                        ovs_header->dp_ifindex,
 903                                                        reply, info->snd_portid,
 904                                                        info->snd_seq, 0,
 905                                                        OVS_FLOW_CMD_NEW);
 906                         BUG_ON(error < 0);
 907                 }
 908                 ovs_unlock();
 909         } else {
 910                 struct sw_flow_actions *old_acts;
 911
 912                 /* Bail out if we're not allowed to modify an existing flow.
 913                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
 914                  * because Generic Netlink treats the latter as a dump
 915                  * request.  We also accept NLM_F_EXCL in case that bug ever
 916                  * gets fixed.
 917                  */
 918                 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
 919                                                          | NLM_F_EXCL))) {
 920                         error = -EEXIST;
 921                         goto err_unlock_ovs;
 922                 }
 923                 /* The unmasked key has to be the same for flow updates. */
 924                 if (unlikely(!ovs_flow_cmp_unmasked_key(flow, &match))) {
 925                         error = -EEXIST;
 926                         goto err_unlock_ovs;
 927                 }
 928                 /* Update actions. */
 929                 old_acts = ovsl_dereference(flow->sf_acts);
 930                 rcu_assign_pointer(flow->sf_acts, acts);
 931
 932                 if (unlikely(reply)) {
 933                         error = ovs_flow_cmd_fill_info(flow,
 934                                                        ovs_header->dp_ifindex,
 935                                                        reply, info->snd_portid,
 936                                                        info->snd_seq, 0,
 937                                                        OVS_FLOW_CMD_NEW);
 938                         BUG_ON(error < 0);
 939                 }
 940                 ovs_unlock();
 941
 942                 ovs_nla_free_flow_actions(old_acts);
 943                 ovs_flow_free(new_flow, false);
 944         }
 945
 946         if (reply)
 947                 ovs_notify(&dp_flow_genl_family, &ovs_dp_flow_multicast_group, reply, info);
 948         return 0;
 949
 950 err_unlock_ovs:
 951         ovs_unlock();
 952         kfree_skb(reply);
 953 err_kfree_acts:
 954         kfree(acts);
 955 err_kfree_flow:
 956         ovs_flow_free(new_flow, false);
 957 error:
 958         return error;
 959 }
 960
 961 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
 962 {
 963         struct nlattr **a = info->attrs;
 964         struct ovs_header *ovs_header = info->userhdr;
 965         struct sw_flow_key key, masked_key;
 966         struct sw_flow *flow;
 967         struct sw_flow_mask mask;
 968         struct sk_buff *reply = NULL;
 969         struct datapath *dp;
 970         struct sw_flow_actions *old_acts = NULL, *acts = NULL;
 971         struct sw_flow_match match;
 972         int error;
 973
 974         /* Extract key. */
 975         error = -EINVAL;
 976         if (!a[OVS_FLOW_ATTR_KEY])
 977                 goto error;
 978
 979         ovs_match_init(&match, &key, &mask);
 980         error = ovs_nla_get_match(&match,
 981                                   a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
 982         if (error)
 983                 goto error;
 984
 985         /* Validate actions. */
 986         if (a[OVS_FLOW_ATTR_ACTIONS]) {
 987                 acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
 988                 error = PTR_ERR(acts);
 989                 if (IS_ERR(acts))
 990                         goto error;
 991
 992                 ovs_flow_mask_key(&masked_key, &key, &mask);
 993                 error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS],
 994                                              &masked_key, 0, &acts);
 995                 if (error) {
 996                         OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
 997                         goto err_kfree_acts;
 998                 }
 999         }
1000
1001         /* Can allocate before locking if have acts. */
1002         if (acts) {
1003                 reply = ovs_flow_cmd_alloc_info(acts, info, false);
1004                 if (IS_ERR(reply)) {
1005                         error = PTR_ERR(reply);
1006                         goto err_kfree_acts;
1007                 }
1008         }
1009
1010         ovs_lock();
1011         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1012         if (unlikely(!dp)) {
1013                 error = -ENODEV;
1014                 goto err_unlock_ovs;
1015         }
1016         /* Check that the flow exists. */
1017         flow = ovs_flow_tbl_lookup(&dp->table, &key);
1018         if (unlikely(!flow)) {
1019                 error = -ENOENT;
1020                 goto err_unlock_ovs;
1021         }
1022         /* The unmasked key has to be the same for flow updates. */
1023         if (unlikely(!ovs_flow_cmp_unmasked_key(flow, &match))) {
1024                 error = -EEXIST;
1025                 goto err_unlock_ovs;
1026         }
1027         /* Update actions, if present. */
1028         if (likely(acts)) {
1029                 old_acts = ovsl_dereference(flow->sf_acts);
1030                 rcu_assign_pointer(flow->sf_acts, acts);
1031
1032                 if (unlikely(reply)) {
1033                         error = ovs_flow_cmd_fill_info(flow,
1034                                                        ovs_header->dp_ifindex,
1035                                                        reply, info->snd_portid,
1036                                                        info->snd_seq, 0,
1037                                                        OVS_FLOW_CMD_NEW);
1038                         BUG_ON(error < 0);
1039                 }
1040         } else {
1041                 /* Could not alloc without acts before locking. */
1042                 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1043                                                 info, OVS_FLOW_CMD_NEW, false);
1044                 if (unlikely(IS_ERR(reply))) {
1045                         error = PTR_ERR(reply);
1046                         goto err_unlock_ovs;
1047                 }
1048         }
1049
1050         /* Clear stats. */
1051         if (a[OVS_FLOW_ATTR_CLEAR])
1052                 ovs_flow_stats_clear(flow);
1053         ovs_unlock();
1054
1055         if (reply)
1056                 ovs_notify(&dp_flow_genl_family, &ovs_dp_flow_multicast_group, reply, info);
1057         if (old_acts)
1058                 ovs_nla_free_flow_actions(old_acts);
1059         return 0;
1060
1061 err_unlock_ovs:
1062         ovs_unlock();
1063         kfree_skb(reply);
1064 err_kfree_acts:
1065         kfree(acts);
1066 error:
1067         return error;
1068 }
1069
1070 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1071 {
1072         struct nlattr **a = info->attrs;
1073         struct ovs_header *ovs_header = info->userhdr;
1074         struct sw_flow_key key;
1075         struct sk_buff *reply;
1076         struct sw_flow *flow;
1077         struct datapath *dp;
1078         struct sw_flow_match match;
1079         int err;
1080
1081         if (!a[OVS_FLOW_ATTR_KEY]) {
1082                 OVS_NLERR("Flow get message rejected, Key attribute missing.\n");
1083                 return -EINVAL;
1084         }
1085
1086         ovs_match_init(&match, &key, NULL);
1087         err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
1088         if (err)
1089                 return err;
1090
1091         ovs_lock();
1092         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1093         if (!dp) {
1094                 err = -ENODEV;
1095                 goto unlock;
1096         }
1097
1098         flow = ovs_flow_tbl_lookup(&dp->table, &key);
1099         if (!flow || !ovs_flow_cmp_unmasked_key(flow, &match)) {
1100                 err = -ENOENT;
1101                 goto unlock;
1102         }
1103
1104         reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1105                                         OVS_FLOW_CMD_NEW, true);
1106         if (IS_ERR(reply)) {
1107                 err = PTR_ERR(reply);
1108                 goto unlock;
1109         }
1110
1111         ovs_unlock();
1112         return genlmsg_reply(reply, info);
1113 unlock:
1114         ovs_unlock();
1115         return err;
1116 }
1117
1118 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1119 {
1120         struct nlattr **a = info->attrs;
1121         struct ovs_header *ovs_header = info->userhdr;
1122         struct sw_flow_key key;
1123         struct sk_buff *reply;
1124         struct sw_flow *flow;
1125         struct datapath *dp;
1126         struct sw_flow_match match;
1127         int err;
1128
1129         if (likely(a[OVS_FLOW_ATTR_KEY])) {
1130                 ovs_match_init(&match, &key, NULL);
1131                 err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
1132                 if (unlikely(err))
1133                         return err;
1134         }
1135
1136         ovs_lock();
1137         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1138         if (unlikely(!dp)) {
1139                 err = -ENODEV;
1140                 goto unlock;
1141         }
1142         if (unlikely(!a[OVS_FLOW_ATTR_KEY])) {
1143                 err = ovs_flow_tbl_flush(&dp->table);
1144                 goto unlock;
1145         }
1146         flow = ovs_flow_tbl_lookup(&dp->table, &key);
1147         if (unlikely(!flow || !ovs_flow_cmp_unmasked_key(flow, &match))) {
1148                 err = -ENOENT;
1149                 goto unlock;
1150         }
1151
1152         ovs_flow_tbl_remove(&dp->table, flow);
1153         ovs_unlock();
1154
1155         reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *)flow->sf_acts,
1156                                         info, false);
1157
1158         if (likely(reply)) {
1159                 if (likely(!IS_ERR(reply))) {
1160                         rcu_read_lock(); /* Keep RCU checker happy. */
1161                         err = ovs_flow_cmd_fill_info(flow,
1162                                                      ovs_header->dp_ifindex,
1163                                                      reply, info->snd_portid,
1164                                                      info->snd_seq, 0,
1165                                                      OVS_FLOW_CMD_DEL);
1166                         rcu_read_unlock();
1167                         BUG_ON(err < 0);
1168                         ovs_notify(&dp_flow_genl_family, &ovs_dp_flow_multicast_group, reply, info);
1169                 } else {
1170                         genl_set_err(&dp_flow_genl_family, sock_net(skb->sk), 0,
1171                                      GROUP_ID(&ovs_dp_flow_multicast_group), PTR_ERR(reply));
1172
1173                 }
1174         }
1175
1176         ovs_flow_free(flow, true);
1177         return 0;
1178 unlock:
1179         ovs_unlock();
1180         return err;
1181 }
1182
1183 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1184 {
1185         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1186         struct table_instance *ti;
1187         struct datapath *dp;
1188
1189         rcu_read_lock();
1190         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1191         if (!dp) {
1192                 rcu_read_unlock();
1193                 return -ENODEV;
1194         }
1195
1196         ti = rcu_dereference(dp->table.ti);
1197         for (;;) {
1198                 struct sw_flow *flow;
1199                 u32 bucket, obj;
1200
1201                 bucket = cb->args[0];
1202                 obj = cb->args[1];
1203                 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1204                 if (!flow)
1205                         break;
1206
1207                 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1208                                            NETLINK_CB(cb->skb).portid,
1209                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1210                                            OVS_FLOW_CMD_NEW) < 0)
1211                         break;
1212
1213                 cb->args[0] = bucket;
1214                 cb->args[1] = obj;
1215         }
1216         rcu_read_unlock();
1217         return skb->len;
1218 }
1219
1220 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1221         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1222         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1223         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1224 };
1225
1226 static struct genl_ops dp_flow_genl_ops[] = {
1227         { .cmd = OVS_FLOW_CMD_NEW,
1228           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1229           .policy = flow_policy,
1230           .doit = ovs_flow_cmd_new
1231         },
1232         { .cmd = OVS_FLOW_CMD_DEL,
1233           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1234           .policy = flow_policy,
1235           .doit = ovs_flow_cmd_del
1236         },
1237         { .cmd = OVS_FLOW_CMD_GET,
1238           .flags = 0,               /* OK for unprivileged users. */
1239           .policy = flow_policy,
1240           .doit = ovs_flow_cmd_get,
1241           .dumpit = ovs_flow_cmd_dump
1242         },
1243         { .cmd = OVS_FLOW_CMD_SET,
1244           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1245           .policy = flow_policy,
1246           .doit = ovs_flow_cmd_set,
1247         },
1248 };
1249
1250 static struct genl_family dp_flow_genl_family = {
1251         .id = GENL_ID_GENERATE,
1252         .hdrsize = sizeof(struct ovs_header),
1253         .name = OVS_FLOW_FAMILY,
1254         .version = OVS_FLOW_VERSION,
1255         .maxattr = OVS_FLOW_ATTR_MAX,
1256         .netnsok = true,
1257         .parallel_ops = true,
1258         .ops = dp_flow_genl_ops,
1259         .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1260         .mcgrps = &ovs_dp_flow_multicast_group,
1261         .n_mcgrps = 1,
1262 };
1263
1264 static size_t ovs_dp_cmd_msg_size(void)
1265 {
1266         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1267
1268         msgsize += nla_total_size(IFNAMSIZ);
1269         msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1270         msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1271         msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1272
1273         return msgsize;
1274 }
1275
1276 /* Called with ovs_mutex or RCU read lock. */
1277 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1278                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1279 {
1280         struct ovs_header *ovs_header;
1281         struct ovs_dp_stats dp_stats;
1282         struct ovs_dp_megaflow_stats dp_megaflow_stats;
1283         int err;
1284
1285         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1286                                    flags, cmd);
1287         if (!ovs_header)
1288                 goto error;
1289
1290         ovs_header->dp_ifindex = get_dpifindex(dp);
1291
1292         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1293         if (err)
1294                 goto nla_put_failure;
1295
1296         get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1297         if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1298                         &dp_stats))
1299                 goto nla_put_failure;
1300
1301         if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1302                         sizeof(struct ovs_dp_megaflow_stats),
1303                         &dp_megaflow_stats))
1304                 goto nla_put_failure;
1305
1306         if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1307                 goto nla_put_failure;
1308
1309         return genlmsg_end(skb, ovs_header);
1310
1311 nla_put_failure:
1312         genlmsg_cancel(skb, ovs_header);
1313 error:
1314         return -EMSGSIZE;
1315 }
1316
1317 static struct sk_buff *ovs_dp_cmd_alloc_info(struct genl_info *info)
1318 {
1319         return genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info, GFP_KERNEL);
1320 }
1321
1322 /* Called with rcu_read_lock or ovs_mutex. */
1323 static struct datapath *lookup_datapath(struct net *net,
1324                                         struct ovs_header *ovs_header,
1325                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1326 {
1327         struct datapath *dp;
1328
1329         if (!a[OVS_DP_ATTR_NAME])
1330                 dp = get_dp(net, ovs_header->dp_ifindex);
1331         else {
1332                 struct vport *vport;
1333
1334                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1335                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1336         }
1337         return dp ? dp : ERR_PTR(-ENODEV);
1338 }
1339
1340 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1341 {
1342         struct datapath *dp;
1343
1344         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1345         if (IS_ERR(dp))
1346                 return;
1347
1348         WARN(dp->user_features, "Dropping previously announced user features\n");
1349         dp->user_features = 0;
1350 }
1351
1352 static void ovs_dp_change(struct datapath *dp, struct nlattr **a)
1353 {
1354         if (a[OVS_DP_ATTR_USER_FEATURES])
1355                 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1356 }
1357
1358 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1359 {
1360         struct nlattr **a = info->attrs;
1361         struct vport_parms parms;
1362         struct sk_buff *reply;
1363         struct datapath *dp;
1364         struct vport *vport;
1365         struct ovs_net *ovs_net;
1366         int err, i;
1367
1368         err = -EINVAL;
1369         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1370                 goto err;
1371
1372         reply = ovs_dp_cmd_alloc_info(info);
1373         if (!reply)
1374                 return -ENOMEM;
1375
1376         err = -ENOMEM;
1377         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1378         if (dp == NULL)
1379                 goto err_free_reply;
1380
1381         ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1382
1383         /* Allocate table. */
1384         err = ovs_flow_tbl_init(&dp->table);
1385         if (err)
1386                 goto err_free_dp;
1387
1388         dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1389         if (!dp->stats_percpu) {
1390                 err = -ENOMEM;
1391                 goto err_destroy_table;
1392         }
1393
1394         for_each_possible_cpu(i) {
1395                 struct dp_stats_percpu *dpath_stats;
1396                 dpath_stats = per_cpu_ptr(dp->stats_percpu, i);
1397                 u64_stats_init(&dpath_stats->sync);
1398         }
1399
1400         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1401                             GFP_KERNEL);
1402         if (!dp->ports) {
1403                 err = -ENOMEM;
1404                 goto err_destroy_percpu;
1405         }
1406
1407         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1408                 INIT_HLIST_HEAD(&dp->ports[i]);
1409
1410         /* Set up our datapath device. */
1411         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1412         parms.type = OVS_VPORT_TYPE_INTERNAL;
1413         parms.options = NULL;
1414         parms.dp = dp;
1415         parms.port_no = OVSP_LOCAL;
1416         parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1417
1418         ovs_dp_change(dp, a);
1419
1420         /* So far only local changes have been made, now need the lock. */
1421         ovs_lock();
1422
1423         vport = new_vport(&parms);
1424         if (IS_ERR(vport)) {
1425                 err = PTR_ERR(vport);
1426                 if (err == -EBUSY)
1427                         err = -EEXIST;
1428
1429                 if (err == -EEXIST) {
1430                         /* An outdated user space instance that does not understand
1431                          * the concept of user_features has attempted to create a new
1432                          * datapath and is likely to reuse it. Drop all user features.
1433                          */
1434                         if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1435                                 ovs_dp_reset_user_features(skb, info);
1436                 }
1437
1438                 goto err_destroy_ports_array;
1439         }
1440
1441         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1442                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1443         BUG_ON(err < 0);
1444
1445         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1446         list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1447
1448         ovs_unlock();
1449
1450         ovs_notify(&dp_datapath_genl_family, &ovs_dp_datapath_multicast_group, reply, info);
1451         return 0;
1452
1453 err_destroy_ports_array:
1454         ovs_unlock();
1455         kfree(dp->ports);
1456 err_destroy_percpu:
1457         free_percpu(dp->stats_percpu);
1458 err_destroy_table:
1459         ovs_flow_tbl_destroy(&dp->table);
1460 err_free_dp:
1461         release_net(ovs_dp_get_net(dp));
1462         kfree(dp);
1463 err_free_reply:
1464         kfree_skb(reply);
1465 err:
1466         return err;
1467 }
1468
1469 /* Called with ovs_mutex. */
1470 static void __dp_destroy(struct datapath *dp)
1471 {
1472         int i;
1473
1474         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1475                 struct vport *vport;
1476                 struct hlist_node *n;
1477
1478                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1479                         if (vport->port_no != OVSP_LOCAL)
1480                                 ovs_dp_detach_port(vport);
1481         }
1482
1483         list_del_rcu(&dp->list_node);
1484
1485         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1486          * all ports in datapath are destroyed first before freeing datapath.
1487          */
1488         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1489
1490         /* RCU destroy the flow table */
1491         call_rcu(&dp->rcu, destroy_dp_rcu);
1492 }
1493
1494 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1495 {
1496         struct sk_buff *reply;
1497         struct datapath *dp;
1498         int err;
1499
1500         reply = ovs_dp_cmd_alloc_info(info);
1501         if (!reply)
1502                 return -ENOMEM;
1503
1504         ovs_lock();
1505         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1506         err = PTR_ERR(dp);
1507         if (IS_ERR(dp))
1508                 goto err_unlock_free;
1509
1510         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1511                                    info->snd_seq, 0, OVS_DP_CMD_DEL);
1512         BUG_ON(err < 0);
1513
1514         __dp_destroy(dp);
1515
1516         ovs_unlock();
1517         ovs_notify(&dp_datapath_genl_family, &ovs_dp_datapath_multicast_group, reply, info);
1518         return 0;
1519
1520 err_unlock_free:
1521         ovs_unlock();
1522         kfree_skb(reply);
1523         return err;
1524 }
1525
1526 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1527 {
1528         struct sk_buff *reply;
1529         struct datapath *dp;
1530         int err;
1531
1532         reply = ovs_dp_cmd_alloc_info(info);
1533         if (!reply)
1534                 return -ENOMEM;
1535
1536         ovs_lock();
1537         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1538         err = PTR_ERR(dp);
1539         if (IS_ERR(dp))
1540                 goto err_unlock_free;
1541
1542         ovs_dp_change(dp, info->attrs);
1543
1544         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1545                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1546         BUG_ON(err < 0);
1547
1548         ovs_unlock();
1549         ovs_notify(&dp_datapath_genl_family, &ovs_dp_datapath_multicast_group, reply, info);
1550         return 0;
1551
1552 err_unlock_free:
1553         ovs_unlock();
1554         kfree_skb(reply);
1555         return err;
1556 }
1557
1558 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1559 {
1560         struct sk_buff *reply;
1561         struct datapath *dp;
1562         int err;
1563
1564         reply = ovs_dp_cmd_alloc_info(info);
1565         if (!reply)
1566                 return -ENOMEM;
1567
1568         rcu_read_lock();
1569         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1570         if (IS_ERR(dp)) {
1571                 err = PTR_ERR(dp);
1572                 goto err_unlock_free;
1573         }
1574         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1575                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1576         BUG_ON(err < 0);
1577         rcu_read_unlock();
1578
1579         return genlmsg_reply(reply, info);
1580
1581 err_unlock_free:
1582         rcu_read_unlock();
1583         kfree_skb(reply);
1584         return err;
1585 }
1586
1587 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1588 {
1589         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1590         struct datapath *dp;
1591         int skip = cb->args[0];
1592         int i = 0;
1593
1594         rcu_read_lock();
1595         list_for_each_entry_rcu(dp, &ovs_net->dps, list_node) {
1596                 if (i >= skip &&
1597                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1598                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1599                                          OVS_DP_CMD_NEW) < 0)
1600                         break;
1601                 i++;
1602         }
1603         rcu_read_unlock();
1604
1605         cb->args[0] = i;
1606
1607         return skb->len;
1608 }
1609
1610 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1611         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1612         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1613         [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1614 };
1615
1616 static struct genl_ops dp_datapath_genl_ops[] = {
1617         { .cmd = OVS_DP_CMD_NEW,
1618           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1619           .policy = datapath_policy,
1620           .doit = ovs_dp_cmd_new
1621         },
1622         { .cmd = OVS_DP_CMD_DEL,
1623           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1624           .policy = datapath_policy,
1625           .doit = ovs_dp_cmd_del
1626         },
1627         { .cmd = OVS_DP_CMD_GET,
1628           .flags = 0,               /* OK for unprivileged users. */
1629           .policy = datapath_policy,
1630           .doit = ovs_dp_cmd_get,
1631           .dumpit = ovs_dp_cmd_dump
1632         },
1633         { .cmd = OVS_DP_CMD_SET,
1634           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1635           .policy = datapath_policy,
1636           .doit = ovs_dp_cmd_set,
1637         },
1638 };
1639
1640 static struct genl_family dp_datapath_genl_family = {
1641         .id = GENL_ID_GENERATE,
1642         .hdrsize = sizeof(struct ovs_header),
1643         .name = OVS_DATAPATH_FAMILY,
1644         .version = OVS_DATAPATH_VERSION,
1645         .maxattr = OVS_DP_ATTR_MAX,
1646         .netnsok = true,
1647         .parallel_ops = true,
1648         .ops = dp_datapath_genl_ops,
1649         .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1650         .mcgrps = &ovs_dp_datapath_multicast_group,
1651         .n_mcgrps = 1,
1652 };
1653
1654 /* Called with ovs_mutex or RCU read lock. */
1655 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1656                                    u32 portid, u32 seq, u32 flags, u8 cmd)
1657 {
1658         struct ovs_header *ovs_header;
1659         struct ovs_vport_stats vport_stats;
1660         int err;
1661
1662         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1663                                  flags, cmd);
1664         if (!ovs_header)
1665                 return -EMSGSIZE;
1666
1667         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1668
1669         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1670             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1671             nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)))
1672                 goto nla_put_failure;
1673
1674         ovs_vport_get_stats(vport, &vport_stats);
1675         if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1676                     &vport_stats))
1677                 goto nla_put_failure;
1678
1679         if (ovs_vport_get_upcall_portids(vport, skb))
1680                 goto nla_put_failure;
1681
1682         err = ovs_vport_get_options(vport, skb);
1683         if (err == -EMSGSIZE)
1684                 goto error;
1685
1686         return genlmsg_end(skb, ovs_header);
1687
1688 nla_put_failure:
1689         err = -EMSGSIZE;
1690 error:
1691         genlmsg_cancel(skb, ovs_header);
1692         return err;
1693 }
1694
1695 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1696 {
1697         return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1698 }
1699
1700 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1701 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1702                                          u32 seq, u8 cmd)
1703 {
1704         struct sk_buff *skb;
1705         int retval;
1706
1707         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1708         if (!skb)
1709                 return ERR_PTR(-ENOMEM);
1710
1711         retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1712         BUG_ON(retval < 0);
1713
1714         return skb;
1715 }
1716
1717 /* Called with ovs_mutex or RCU read lock. */
1718 static struct vport *lookup_vport(struct net *net,
1719                                   struct ovs_header *ovs_header,
1720                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1721 {
1722         struct datapath *dp;
1723         struct vport *vport;
1724
1725         if (a[OVS_VPORT_ATTR_NAME]) {
1726                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1727                 if (!vport)
1728                         return ERR_PTR(-ENODEV);
1729                 if (ovs_header->dp_ifindex &&
1730                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1731                         return ERR_PTR(-ENODEV);
1732                 return vport;
1733         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1734                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1735
1736                 if (port_no >= DP_MAX_PORTS)
1737                         return ERR_PTR(-EFBIG);
1738
1739                 dp = get_dp(net, ovs_header->dp_ifindex);
1740                 if (!dp)
1741                         return ERR_PTR(-ENODEV);
1742
1743                 vport = ovs_vport_ovsl_rcu(dp, port_no);
1744                 if (!vport)
1745                         return ERR_PTR(-ENODEV);
1746                 return vport;
1747         } else
1748                 return ERR_PTR(-EINVAL);
1749 }
1750
1751 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1752 {
1753         struct nlattr **a = info->attrs;
1754         struct ovs_header *ovs_header = info->userhdr;
1755         struct vport_parms parms;
1756         struct sk_buff *reply;
1757         struct vport *vport;
1758         struct datapath *dp;
1759         u32 port_no;
1760         int err;
1761
1762         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1763             !a[OVS_VPORT_ATTR_UPCALL_PID])
1764                 return -EINVAL;
1765
1766         port_no = a[OVS_VPORT_ATTR_PORT_NO]
1767                 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
1768         if (port_no >= DP_MAX_PORTS)
1769                 return -EFBIG;
1770
1771         reply = ovs_vport_cmd_alloc_info();
1772         if (!reply)
1773                 return -ENOMEM;
1774
1775         ovs_lock();
1776         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1777         err = -ENODEV;
1778         if (!dp)
1779                 goto exit_unlock_free;
1780
1781         if (port_no) {
1782                 vport = ovs_vport_ovsl(dp, port_no);
1783                 err = -EBUSY;
1784                 if (vport)
1785                         goto exit_unlock_free;
1786         } else {
1787                 for (port_no = 1; ; port_no++) {
1788                         if (port_no >= DP_MAX_PORTS) {
1789                                 err = -EFBIG;
1790                                 goto exit_unlock_free;
1791                         }
1792                         vport = ovs_vport_ovsl(dp, port_no);
1793                         if (!vport)
1794                                 break;
1795                 }
1796         }
1797
1798         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1799         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1800         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1801         parms.dp = dp;
1802         parms.port_no = port_no;
1803         parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
1804
1805         vport = new_vport(&parms);
1806         err = PTR_ERR(vport);
1807         if (IS_ERR(vport))
1808                 goto exit_unlock_free;
1809
1810         err = 0;
1811         if (a[OVS_VPORT_ATTR_STATS])
1812                 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1813
1814         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1815                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1816         BUG_ON(err < 0);
1817         ovs_unlock();
1818
1819         ovs_notify(&dp_vport_genl_family, &ovs_dp_vport_multicast_group, reply, info);
1820         return 0;
1821
1822 exit_unlock_free:
1823         ovs_unlock();
1824         kfree_skb(reply);
1825         return err;
1826 }
1827
1828 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1829 {
1830         struct nlattr **a = info->attrs;
1831         struct sk_buff *reply;
1832         struct vport *vport;
1833         int err;
1834
1835         reply = ovs_vport_cmd_alloc_info();
1836         if (!reply)
1837                 return -ENOMEM;
1838
1839         ovs_lock();
1840         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1841         err = PTR_ERR(vport);
1842         if (IS_ERR(vport))
1843                 goto exit_unlock_free;
1844
1845         if (a[OVS_VPORT_ATTR_TYPE] &&
1846             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
1847                 err = -EINVAL;
1848                 goto exit_unlock_free;
1849         }
1850
1851         if (a[OVS_VPORT_ATTR_OPTIONS]) {
1852                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1853                 if (err)
1854                         goto exit_unlock_free;
1855         }
1856
1857         if (a[OVS_VPORT_ATTR_STATS])
1858                 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1859
1860
1861         if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
1862                 err = ovs_vport_set_upcall_portids(vport,
1863                                                    a[OVS_VPORT_ATTR_UPCALL_PID]);
1864                 if (err)
1865                         goto exit_unlock_free;
1866         }
1867
1868         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1869                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1870         BUG_ON(err < 0);
1871         ovs_unlock();
1872
1873         ovs_notify(&dp_vport_genl_family, &ovs_dp_vport_multicast_group, reply, info);
1874         return 0;
1875
1876 exit_unlock_free:
1877         ovs_unlock();
1878         kfree_skb(reply);
1879         return err;
1880 }
1881
1882 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1883 {
1884         struct nlattr **a = info->attrs;
1885         struct sk_buff *reply;
1886         struct vport *vport;
1887         int err;
1888
1889         reply = ovs_vport_cmd_alloc_info();
1890         if (!reply)
1891                 return -ENOMEM;
1892
1893         ovs_lock();
1894         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1895         err = PTR_ERR(vport);
1896         if (IS_ERR(vport))
1897                 goto exit_unlock_free;
1898
1899         if (vport->port_no == OVSP_LOCAL) {
1900                 err = -EINVAL;
1901                 goto exit_unlock_free;
1902         }
1903
1904         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1905                                       info->snd_seq, 0, OVS_VPORT_CMD_DEL);
1906         BUG_ON(err < 0);
1907         ovs_dp_detach_port(vport);
1908         ovs_unlock();
1909
1910         ovs_notify(&dp_vport_genl_family, &ovs_dp_vport_multicast_group, reply, info);
1911         return 0;
1912
1913 exit_unlock_free:
1914         ovs_unlock();
1915         kfree_skb(reply);
1916         return err;
1917 }
1918
1919 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1920 {
1921         struct nlattr **a = info->attrs;
1922         struct ovs_header *ovs_header = info->userhdr;
1923         struct sk_buff *reply;
1924         struct vport *vport;
1925         int err;
1926
1927         reply = ovs_vport_cmd_alloc_info();
1928         if (!reply)
1929                 return -ENOMEM;
1930
1931         rcu_read_lock();
1932         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1933         err = PTR_ERR(vport);
1934         if (IS_ERR(vport))
1935                 goto exit_unlock_free;
1936         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1937                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1938         BUG_ON(err < 0);
1939         rcu_read_unlock();
1940
1941         return genlmsg_reply(reply, info);
1942
1943 exit_unlock_free:
1944         rcu_read_unlock();
1945         kfree_skb(reply);
1946         return err;
1947 }
1948
1949 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1950 {
1951         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1952         struct datapath *dp;
1953         int bucket = cb->args[0], skip = cb->args[1];
1954         int i, j = 0;
1955
1956         rcu_read_lock();
1957         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1958         if (!dp) {
1959                 rcu_read_unlock();
1960                 return -ENODEV;
1961         }
1962         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
1963                 struct vport *vport;
1964
1965                 j = 0;
1966                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1967                         if (j >= skip &&
1968                             ovs_vport_cmd_fill_info(vport, skb,
1969                                                     NETLINK_CB(cb->skb).portid,
1970                                                     cb->nlh->nlmsg_seq,
1971                                                     NLM_F_MULTI,
1972                                                     OVS_VPORT_CMD_NEW) < 0)
1973                                 goto out;
1974
1975                         j++;
1976                 }
1977                 skip = 0;
1978         }
1979 out:
1980         rcu_read_unlock();
1981
1982         cb->args[0] = i;
1983         cb->args[1] = j;
1984
1985         return skb->len;
1986 }
1987
1988 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1989         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1990         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1991         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1992         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1993         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1994         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1995 };
1996
1997 static struct genl_ops dp_vport_genl_ops[] = {
1998         { .cmd = OVS_VPORT_CMD_NEW,
1999           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2000           .policy = vport_policy,
2001           .doit = ovs_vport_cmd_new
2002         },
2003         { .cmd = OVS_VPORT_CMD_DEL,
2004           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2005           .policy = vport_policy,
2006           .doit = ovs_vport_cmd_del
2007         },
2008         { .cmd = OVS_VPORT_CMD_GET,
2009           .flags = 0,               /* OK for unprivileged users. */
2010           .policy = vport_policy,
2011           .doit = ovs_vport_cmd_get,
2012           .dumpit = ovs_vport_cmd_dump
2013         },
2014         { .cmd = OVS_VPORT_CMD_SET,
2015           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2016           .policy = vport_policy,
2017           .doit = ovs_vport_cmd_set,
2018         },
2019 };
2020
2021 struct genl_family dp_vport_genl_family = {
2022         .id = GENL_ID_GENERATE,
2023         .hdrsize = sizeof(struct ovs_header),
2024         .name = OVS_VPORT_FAMILY,
2025         .version = OVS_VPORT_VERSION,
2026         .maxattr = OVS_VPORT_ATTR_MAX,
2027         .netnsok = true,
2028         .parallel_ops = true,
2029         .ops = dp_vport_genl_ops,
2030         .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2031         .mcgrps = &ovs_dp_vport_multicast_group,
2032         .n_mcgrps = 1,
2033 };
2034
2035 static struct genl_family *dp_genl_families[] = {
2036         &dp_datapath_genl_family,
2037         &dp_vport_genl_family,
2038         &dp_flow_genl_family,
2039         &dp_packet_genl_family,
2040 };
2041
2042 static void dp_unregister_genl(int n_families)
2043 {
2044         int i;
2045
2046         for (i = 0; i < n_families; i++)
2047                 genl_unregister_family(dp_genl_families[i]);
2048 }
2049
2050 static int dp_register_genl(void)
2051 {
2052         int err;
2053         int i;
2054
2055         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2056
2057                 err = genl_register_family(dp_genl_families[i]);
2058                 if (err)
2059                         goto error;
2060         }
2061
2062         return 0;
2063
2064 error:
2065         dp_unregister_genl(i);
2066         return err;
2067 }
2068
2069 static int __net_init ovs_init_net(struct net *net)
2070 {
2071         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2072
2073         INIT_LIST_HEAD(&ovs_net->dps);
2074         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2075         return 0;
2076 }
2077
2078 static void __net_exit ovs_exit_net(struct net *net)
2079 {
2080         struct datapath *dp, *dp_next;
2081         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2082
2083         ovs_lock();
2084         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2085                 __dp_destroy(dp);
2086         ovs_unlock();
2087
2088         cancel_work_sync(&ovs_net->dp_notify_work);
2089 }
2090
2091 static struct pernet_operations ovs_net_ops = {
2092         .init = ovs_init_net,
2093         .exit = ovs_exit_net,
2094         .id   = &ovs_net_id,
2095         .size = sizeof(struct ovs_net),
2096 };
2097
2098 DEFINE_COMPAT_PNET_REG_FUNC(device);
2099
2100 static int __init dp_init(void)
2101 {
2102         int err;
2103
2104         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2105
2106         pr_info("Open vSwitch switching datapath %s, built "__DATE__" "__TIME__"\n",
2107                 VERSION);
2108
2109         err = ovs_flow_init();
2110         if (err)
2111                 goto error;
2112
2113         err = ovs_vport_init();
2114         if (err)
2115                 goto error_flow_exit;
2116
2117         err = register_pernet_device(&ovs_net_ops);
2118         if (err)
2119                 goto error_vport_exit;
2120
2121         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2122         if (err)
2123                 goto error_netns_exit;
2124
2125         err = dp_register_genl();
2126         if (err < 0)
2127                 goto error_unreg_notifier;
2128
2129         return 0;
2130
2131 error_unreg_notifier:
2132         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2133 error_netns_exit:
2134         unregister_pernet_device(&ovs_net_ops);
2135 error_vport_exit:
2136         ovs_vport_exit();
2137 error_flow_exit:
2138         ovs_flow_exit();
2139 error:
2140         return err;
2141 }
2142
2143 static void dp_cleanup(void)
2144 {
2145         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2146         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2147         unregister_pernet_device(&ovs_net_ops);
2148         rcu_barrier();
2149         ovs_vport_exit();
2150         ovs_flow_exit();
2151 }
2152
2153 module_init(dp_init);
2154 module_exit(dp_cleanup);
2155
2156 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2157 MODULE_LICENSE("GPL");
2158 MODULE_VERSION(VERSION);