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