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