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