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