datapath/datapath.c

   1 /*
   2  * Copyright (c) 2007-2013 Nicira, Inc.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of version 2 of the GNU General Public
   6  * License as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program; if not, write to the Free Software
  15  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
  16  * 02110-1301, USA
  17  */
  18
  19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  20
  21 #include <linux/init.h>
  22 #include <linux/module.h>
  23 #include <linux/if_arp.h>
  24 #include <linux/if_vlan.h>
  25 #include <linux/in.h>
  26 #include <linux/ip.h>
  27 #include <linux/jhash.h>
  28 #include <linux/delay.h>
  29 #include <linux/time.h>
  30 #include <linux/etherdevice.h>
  31 #include <linux/genetlink.h>
  32 #include <linux/kernel.h>
  33 #include <linux/kthread.h>
  34 #include <linux/mutex.h>
  35 #include <linux/percpu.h>
  36 #include <linux/rcupdate.h>
  37 #include <linux/tcp.h>
  38 #include <linux/udp.h>
  39 #include <linux/version.h>
  40 #include <linux/ethtool.h>
  41 #include <linux/wait.h>
  42 #include <asm/div64.h>
  43 #include <linux/highmem.h>
  44 #include <linux/netfilter_bridge.h>
  45 #include <linux/netfilter_ipv4.h>
  46 #include <linux/inetdevice.h>
  47 #include <linux/list.h>
  48 #include <linux/openvswitch.h>
  49 #include <linux/rculist.h>
  50 #include <linux/dmi.h>
  51 #include <linux/genetlink.h>
  52 #include <net/genetlink.h>
  53 #include <net/genetlink.h>
  54 #include <net/net_namespace.h>
  55 #include <net/netns/generic.h>
  56
  57 #include "datapath.h"
  58 #include "flow.h"
  59 #include "flow_table.h"
  60 #include "flow_netlink.h"
  61 #include "vlan.h"
  62 #include "vport-internal_dev.h"
  63 #include "vport-netdev.h"
  64
  65 int ovs_net_id __read_mostly;
  66
  67 /* Check if need to build a reply message.
  68  * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
  69 static bool ovs_must_notify(struct genl_info *info,
  70                             const struct genl_multicast_group *grp)
  71 {
  72         return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
  73                 netlink_has_listeners(genl_info_net(info)->genl_sock, grp->id);
  74 }
  75
  76 static void ovs_notify(struct sk_buff *skb, struct genl_info *info,
  77                        struct genl_multicast_group *grp)
  78 {
  79         genl_notify(skb, genl_info_net(info), info->snd_portid,
  80                     grp->id, info->nlhdr, GFP_KERNEL);
  81 }
  82
  83 /**
  84  * DOC: Locking:
  85  *
  86  * All writes e.g. Writes to device state (add/remove datapath, port, set
  87  * operations on vports, etc.), Writes to other state (flow table
  88  * modifications, set miscellaneous datapath parameters, etc.) are protected
  89  * by ovs_lock.
  90  *
  91  * Reads are protected by RCU.
  92  *
  93  * There are a few special cases (mostly stats) that have their own
  94  * synchronization but they nest under all of above and don't interact with
  95  * each other.
  96  *
  97  * The RTNL lock nests inside ovs_mutex.
  98  */
  99
 100 static DEFINE_MUTEX(ovs_mutex);
 101
 102 void ovs_lock(void)
 103 {
 104         mutex_lock(&ovs_mutex);
 105 }
 106
 107 void ovs_unlock(void)
 108 {
 109         mutex_unlock(&ovs_mutex);
 110 }
 111
 112 #ifdef CONFIG_LOCKDEP
 113 int lockdep_ovsl_is_held(void)
 114 {
 115         if (debug_locks)
 116                 return lockdep_is_held(&ovs_mutex);
 117         else
 118                 return 1;
 119 }
 120 #endif
 121
 122 static struct vport *new_vport(const struct vport_parms *);
 123 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
 124                              const struct dp_upcall_info *);
 125 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
 126                                   const struct dp_upcall_info *);
 127
 128 /* Must be called with rcu_read_lock or ovs_mutex. */
 129 static struct datapath *get_dp(struct net *net, int dp_ifindex)
 130 {
 131         struct datapath *dp = NULL;
 132         struct net_device *dev;
 133
 134         rcu_read_lock();
 135         dev = dev_get_by_index_rcu(net, dp_ifindex);
 136         if (dev) {
 137                 struct vport *vport = ovs_internal_dev_get_vport(dev);
 138                 if (vport)
 139                         dp = vport->dp;
 140         }
 141         rcu_read_unlock();
 142
 143         return dp;
 144 }
 145
 146 /* Must be called with rcu_read_lock or ovs_mutex. */
 147 const char *ovs_dp_name(const struct datapath *dp)
 148 {
 149         struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
 150         return vport->ops->get_name(vport);
 151 }
 152
 153 static int get_dpifindex(struct datapath *dp)
 154 {
 155         struct vport *local;
 156         int ifindex;
 157
 158         rcu_read_lock();
 159
 160         local = ovs_vport_rcu(dp, OVSP_LOCAL);
 161         if (local)
 162                 ifindex = netdev_vport_priv(local)->dev->ifindex;
 163         else
 164                 ifindex = 0;
 165
 166         rcu_read_unlock();
 167
 168         return ifindex;
 169 }
 170
 171 static void destroy_dp_rcu(struct rcu_head *rcu)
 172 {
 173         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 174
 175         free_percpu(dp->stats_percpu);
 176         release_net(ovs_dp_get_net(dp));
 177         kfree(dp->ports);
 178         kfree(dp);
 179 }
 180
 181 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
 182                                             u16 port_no)
 183 {
 184         return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
 185 }
 186
 187 /* Called with ovs_mutex or RCU read lock. */
 188 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
 189 {
 190         struct vport *vport;
 191         struct hlist_head *head;
 192
 193         head = vport_hash_bucket(dp, port_no);
 194         hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
 195                 if (vport->port_no == port_no)
 196                         return vport;
 197         }
 198         return NULL;
 199 }
 200
 201 /* Called with ovs_mutex. */
 202 static struct vport *new_vport(const struct vport_parms *parms)
 203 {
 204         struct vport *vport;
 205
 206         vport = ovs_vport_add(parms);
 207         if (!IS_ERR(vport)) {
 208                 struct datapath *dp = parms->dp;
 209                 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
 210
 211                 hlist_add_head_rcu(&vport->dp_hash_node, head);
 212         }
 213         return vport;
 214 }
 215
 216 void ovs_dp_detach_port(struct vport *p)
 217 {
 218         ASSERT_OVSL();
 219
 220         /* First drop references to device. */
 221         hlist_del_rcu(&p->dp_hash_node);
 222
 223         /* Then destroy it. */
 224         ovs_vport_del(p);
 225 }
 226
 227 /* Must be called with rcu_read_lock. */
 228 void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
 229 {
 230         struct datapath *dp = p->dp;
 231         struct sw_flow *flow;
 232         struct dp_stats_percpu *stats;
 233         struct sw_flow_key key;
 234         u64 *stats_counter;
 235         u32 n_mask_hit;
 236         int error;
 237
 238         stats = this_cpu_ptr(dp->stats_percpu);
 239
 240         /* Extract flow from 'skb' into 'key'. */
 241         error = ovs_flow_extract(skb, p->port_no, &key);
 242         if (unlikely(error)) {
 243                 kfree_skb(skb);
 244                 return;
 245         }
 246
 247         /* Look up flow. */
 248         flow = ovs_flow_tbl_lookup_stats(&dp->table, &key, &n_mask_hit);
 249         if (unlikely(!flow)) {
 250                 struct dp_upcall_info upcall;
 251
 252                 upcall.cmd = OVS_PACKET_CMD_MISS;
 253                 upcall.key = &key;
 254                 upcall.userdata = NULL;
 255                 upcall.portid = p->upcall_portid;
 256                 ovs_dp_upcall(dp, skb, &upcall);
 257                 consume_skb(skb);
 258                 stats_counter = &stats->n_missed;
 259                 goto out;
 260         }
 261
 262         OVS_CB(skb)->flow = flow;
 263         OVS_CB(skb)->pkt_key = &key;
 264
 265         ovs_flow_stats_update(OVS_CB(skb)->flow, skb);
 266         ovs_execute_actions(dp, skb);
 267         stats_counter = &stats->n_hit;
 268
 269 out:
 270         /* Update datapath statistics. */
 271         u64_stats_update_begin(&stats->sync);
 272         (*stats_counter)++;
 273         stats->n_mask_hit += n_mask_hit;
 274         u64_stats_update_end(&stats->sync);
 275 }
 276
 277 static struct genl_family dp_packet_genl_family = {
 278         .id = GENL_ID_GENERATE,
 279         .hdrsize = sizeof(struct ovs_header),
 280         .name = OVS_PACKET_FAMILY,
 281         .version = OVS_PACKET_VERSION,
 282         .maxattr = OVS_PACKET_ATTR_MAX,
 283         .netnsok = true,
 284          SET_PARALLEL_OPS
 285 };
 286
 287 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
 288                   const struct dp_upcall_info *upcall_info)
 289 {
 290         struct dp_stats_percpu *stats;
 291         int err;
 292
 293         if (upcall_info->portid == 0) {
 294                 err = -ENOTCONN;
 295                 goto err;
 296         }
 297
 298         if (!skb_is_gso(skb))
 299                 err = queue_userspace_packet(dp, skb, upcall_info);
 300         else
 301                 err = queue_gso_packets(dp, skb, upcall_info);
 302         if (err)
 303                 goto err;
 304
 305         return 0;
 306
 307 err:
 308         stats = this_cpu_ptr(dp->stats_percpu);
 309
 310         u64_stats_update_begin(&stats->sync);
 311         stats->n_lost++;
 312         u64_stats_update_end(&stats->sync);
 313
 314         return err;
 315 }
 316
 317 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
 318                              const struct dp_upcall_info *upcall_info)
 319 {
 320         unsigned short gso_type = skb_shinfo(skb)->gso_type;
 321         struct dp_upcall_info later_info;
 322         struct sw_flow_key later_key;
 323         struct sk_buff *segs, *nskb;
 324         int err;
 325
 326         segs = __skb_gso_segment(skb, NETIF_F_SG, false);
 327         if (IS_ERR(segs))
 328                 return PTR_ERR(segs);
 329
 330         /* Queue all of the segments. */
 331         skb = segs;
 332         do {
 333                 err = queue_userspace_packet(dp, skb, upcall_info);
 334                 if (err)
 335                         break;
 336
 337                 if (skb == segs && gso_type & SKB_GSO_UDP) {
 338                         /* The initial flow key extracted by ovs_flow_extract()
 339                          * in this case is for a first fragment, so we need to
 340                          * properly mark later fragments.
 341                          */
 342                         later_key = *upcall_info->key;
 343                         later_key.ip.frag = OVS_FRAG_TYPE_LATER;
 344
 345                         later_info = *upcall_info;
 346                         later_info.key = &later_key;
 347                         upcall_info = &later_info;
 348                 }
 349         } while ((skb = skb->next));
 350
 351         /* Free all of the segments. */
 352         skb = segs;
 353         do {
 354                 nskb = skb->next;
 355                 if (err)
 356                         kfree_skb(skb);
 357                 else
 358                         consume_skb(skb);
 359         } while ((skb = nskb));
 360         return err;
 361 }
 362
 363 static size_t key_attr_size(void)
 364 {
 365         return    nla_total_size(4)   /* OVS_KEY_ATTR_PRIORITY */
 366                 + nla_total_size(0)   /* OVS_KEY_ATTR_TUNNEL */
 367                   + nla_total_size(8)   /* OVS_TUNNEL_KEY_ATTR_ID */
 368                   + nla_total_size(4)   /* OVS_TUNNEL_KEY_ATTR_IPV4_SRC */
 369                   + nla_total_size(4)   /* OVS_TUNNEL_KEY_ATTR_IPV4_DST */
 370                   + nla_total_size(1)   /* OVS_TUNNEL_KEY_ATTR_TOS */
 371                   + nla_total_size(1)   /* OVS_TUNNEL_KEY_ATTR_TTL */
 372                   + nla_total_size(0)   /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
 373                   + nla_total_size(0)   /* OVS_TUNNEL_KEY_ATTR_CSUM */
 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; /* to be queued to userspace */
 406         struct nlattr *nla;
 407         struct genl_info info = {
 408 #if LINUX_VERSION_CODE >= KERNEL_VERSION(3,14,0)
 409                 .dst_sk = ovs_dp_get_net(dp)->genl_sock,
 410 #endif
 411                 .snd_portid = upcall_info->portid,
 412         };
 413         size_t len;
 414         unsigned int hlen;
 415         int err, dp_ifindex;
 416
 417         dp_ifindex = get_dpifindex(dp);
 418         if (!dp_ifindex)
 419                 return -ENODEV;
 420
 421         if (vlan_tx_tag_present(skb)) {
 422                 nskb = skb_clone(skb, GFP_ATOMIC);
 423                 if (!nskb)
 424                         return -ENOMEM;
 425
 426                 nskb = __vlan_put_tag(nskb, nskb->vlan_proto, vlan_tx_tag_get(nskb));
 427                 if (!nskb)
 428                         return -ENOMEM;
 429
 430                 vlan_set_tci(nskb, 0);
 431
 432                 skb = nskb;
 433         }
 434
 435         if (nla_attr_size(skb->len) > USHRT_MAX) {
 436                 err = -EFBIG;
 437                 goto out;
 438         }
 439
 440         /* Complete checksum if needed */
 441         if (skb->ip_summed == CHECKSUM_PARTIAL &&
 442             (err = skb_checksum_help(skb)))
 443                 goto out;
 444
 445         /* Older versions of OVS user space enforce alignment of the last
 446          * Netlink attribute to NLA_ALIGNTO which would require extensive
 447          * padding logic. Only perform zerocopy if padding is not required.
 448          */
 449         if (dp->user_features & OVS_DP_F_UNALIGNED)
 450                 hlen = skb_zerocopy_headlen(skb);
 451         else
 452                 hlen = skb->len;
 453
 454         len = upcall_msg_size(upcall_info->userdata, hlen);
 455         user_skb = genlmsg_new_unicast(len, &info, GFP_ATOMIC);
 456         if (!user_skb) {
 457                 err = -ENOMEM;
 458                 goto out;
 459         }
 460
 461         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 462                              0, upcall_info->cmd);
 463         upcall->dp_ifindex = dp_ifindex;
 464
 465         nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
 466         ovs_nla_put_flow(upcall_info->key, upcall_info->key, user_skb);
 467         nla_nest_end(user_skb, nla);
 468
 469         if (upcall_info->userdata)
 470                 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
 471                           nla_len(upcall_info->userdata),
 472                           nla_data(upcall_info->userdata));
 473
 474         /* Only reserve room for attribute header, packet data is added
 475          * in skb_zerocopy() */
 476         if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
 477                 err = -ENOBUFS;
 478                 goto out;
 479         }
 480         nla->nla_len = nla_attr_size(skb->len);
 481
 482         skb_zerocopy(user_skb, skb, skb->len, hlen);
 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 out:
 496         kfree_skb(nskb);
 497         return err;
 498 }
 499
 500 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 501 {
 502         struct ovs_header *ovs_header = info->userhdr;
 503         struct nlattr **a = info->attrs;
 504         struct sw_flow_actions *acts;
 505         struct sk_buff *packet;
 506         struct sw_flow *flow;
 507         struct datapath *dp;
 508         struct ethhdr *eth;
 509         int len;
 510         int err;
 511
 512         err = -EINVAL;
 513         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 514             !a[OVS_PACKET_ATTR_ACTIONS])
 515                 goto err;
 516
 517         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 518         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 519         err = -ENOMEM;
 520         if (!packet)
 521                 goto err;
 522         skb_reserve(packet, NET_IP_ALIGN);
 523
 524         nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
 525
 526         skb_reset_mac_header(packet);
 527         eth = eth_hdr(packet);
 528
 529         /* Normally, setting the skb 'protocol' field would be handled by a
 530          * call to eth_type_trans(), but it assumes there's a sending
 531          * device, which we may not have. */
 532         if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
 533                 packet->protocol = eth->h_proto;
 534         else
 535                 packet->protocol = htons(ETH_P_802_2);
 536
 537         /* Build an sw_flow for sending this packet. */
 538         flow = ovs_flow_alloc();
 539         err = PTR_ERR(flow);
 540         if (IS_ERR(flow))
 541                 goto err_kfree_skb;
 542
 543         err = ovs_flow_extract(packet, -1, &flow->key);
 544         if (err)
 545                 goto err_flow_free;
 546
 547         err = ovs_nla_get_flow_metadata(flow, a[OVS_PACKET_ATTR_KEY]);
 548         if (err)
 549                 goto err_flow_free;
 550         acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_PACKET_ATTR_ACTIONS]));
 551         err = PTR_ERR(acts);
 552         if (IS_ERR(acts))
 553                 goto err_flow_free;
 554
 555         err = ovs_nla_copy_actions(a[OVS_PACKET_ATTR_ACTIONS],
 556                                    &flow->key, 0, &acts);
 557         rcu_assign_pointer(flow->sf_acts, acts);
 558         if (err)
 559                 goto err_flow_free;
 560
 561         OVS_CB(packet)->flow = flow;
 562         OVS_CB(packet)->pkt_key = &flow->key;
 563         packet->priority = flow->key.phy.priority;
 564         packet->mark = flow->key.phy.skb_mark;
 565
 566         rcu_read_lock();
 567         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 568         err = -ENODEV;
 569         if (!dp)
 570                 goto err_unlock;
 571
 572         local_bh_disable();
 573         err = ovs_execute_actions(dp, packet);
 574         local_bh_enable();
 575         rcu_read_unlock();
 576
 577         ovs_flow_free(flow, false);
 578         return err;
 579
 580 err_unlock:
 581         rcu_read_unlock();
 582 err_flow_free:
 583         ovs_flow_free(flow, false);
 584 err_kfree_skb:
 585         kfree_skb(packet);
 586 err:
 587         return err;
 588 }
 589
 590 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 591         [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
 592         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 593         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 594 };
 595
 596 static struct genl_ops dp_packet_genl_ops[] = {
 597         { .cmd = OVS_PACKET_CMD_EXECUTE,
 598           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 599           .policy = packet_policy,
 600           .doit = ovs_packet_cmd_execute
 601         }
 602 };
 603
 604 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats,
 605                          struct ovs_dp_megaflow_stats *mega_stats)
 606 {
 607         int i;
 608
 609         memset(mega_stats, 0, sizeof(*mega_stats));
 610
 611         stats->n_flows = ovs_flow_tbl_count(&dp->table);
 612         mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
 613
 614         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 615
 616         for_each_possible_cpu(i) {
 617                 const struct dp_stats_percpu *percpu_stats;
 618                 struct dp_stats_percpu local_stats;
 619                 unsigned int start;
 620
 621                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 622
 623                 do {
 624                         start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
 625                         local_stats = *percpu_stats;
 626                 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
 627
 628                 stats->n_hit += local_stats.n_hit;
 629                 stats->n_missed += local_stats.n_missed;
 630                 stats->n_lost += local_stats.n_lost;
 631                 mega_stats->n_mask_hit += local_stats.n_mask_hit;
 632         }
 633 }
 634
 635 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
 636         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
 637         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
 638         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
 639 };
 640
 641 static struct genl_family dp_flow_genl_family = {
 642         .id = GENL_ID_GENERATE,
 643         .hdrsize = sizeof(struct ovs_header),
 644         .name = OVS_FLOW_FAMILY,
 645         .version = OVS_FLOW_VERSION,
 646         .maxattr = OVS_FLOW_ATTR_MAX,
 647         .netnsok = true,
 648          SET_PARALLEL_OPS
 649 };
 650
 651 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
 652         .name = OVS_FLOW_MCGROUP
 653 };
 654
 655 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
 656 {
 657         return NLMSG_ALIGN(sizeof(struct ovs_header))
 658                 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
 659                 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_MASK */
 660                 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
 661                 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
 662                 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
 663                 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
 664 }
 665
 666 /* Called with ovs_mutex or RCU read lock. */
 667 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
 668                                   struct sk_buff *skb, u32 portid,
 669                                   u32 seq, u32 flags, u8 cmd)
 670 {
 671         const int skb_orig_len = skb->len;
 672         struct nlattr *start;
 673         struct ovs_flow_stats stats;
 674         __be16 tcp_flags;
 675         unsigned long used;
 676         struct ovs_header *ovs_header;
 677         struct nlattr *nla;
 678         int err;
 679
 680         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
 681         if (!ovs_header)
 682                 return -EMSGSIZE;
 683
 684         ovs_header->dp_ifindex = dp_ifindex;
 685
 686         /* Fill flow key. */
 687         nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
 688         if (!nla)
 689                 goto nla_put_failure;
 690
 691         err = ovs_nla_put_flow(&flow->unmasked_key, &flow->unmasked_key, skb);
 692         if (err)
 693                 goto error;
 694         nla_nest_end(skb, nla);
 695
 696         nla = nla_nest_start(skb, OVS_FLOW_ATTR_MASK);
 697         if (!nla)
 698                 goto nla_put_failure;
 699
 700         err = ovs_nla_put_flow(&flow->key, &flow->mask->key, skb);
 701         if (err)
 702                 goto error;
 703
 704         nla_nest_end(skb, nla);
 705
 706         ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
 707
 708         if (used &&
 709             nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
 710                 goto nla_put_failure;
 711
 712         if (stats.n_packets &&
 713             nla_put(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats))
 714                 goto nla_put_failure;
 715
 716         if ((u8)ntohs(tcp_flags) &&
 717              nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
 718                 goto nla_put_failure;
 719
 720         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 721          * this is the first flow to be dumped into 'skb'.  This is unusual for
 722          * Netlink but individual action lists can be longer than
 723          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 724          * The userspace caller can always fetch the actions separately if it
 725          * really wants them.  (Most userspace callers in fact don't care.)
 726          *
 727          * This can only fail for dump operations because the skb is always
 728          * properly sized for single flows.
 729          */
 730         start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
 731         if (start) {
 732                 const struct sw_flow_actions *sf_acts;
 733
 734                 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
 735                 err = ovs_nla_put_actions(sf_acts->actions,
 736                                           sf_acts->actions_len, skb);
 737
 738                 if (!err)
 739                         nla_nest_end(skb, start);
 740                 else {
 741                         if (skb_orig_len)
 742                                 goto error;
 743
 744                         nla_nest_cancel(skb, start);
 745                 }
 746         } else if (skb_orig_len)
 747                 goto nla_put_failure;
 748
 749         return genlmsg_end(skb, ovs_header);
 750
 751 nla_put_failure:
 752         err = -EMSGSIZE;
 753 error:
 754         genlmsg_cancel(skb, ovs_header);
 755         return err;
 756 }
 757
 758 /* May not be called with RCU read lock. */
 759 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
 760                                                struct genl_info *info,
 761                                                bool always)
 762 {
 763         struct sk_buff *skb;
 764
 765         if (!always && !ovs_must_notify(info, &ovs_dp_flow_multicast_group))
 766                 return NULL;
 767
 768         skb = genlmsg_new_unicast(ovs_flow_cmd_msg_size(acts), info, GFP_KERNEL);
 769
 770         if (!skb)
 771                 return ERR_PTR(-ENOMEM);
 772
 773         return skb;
 774 }
 775
 776 /* Called with ovs_mutex. */
 777 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
 778                                                int dp_ifindex,
 779                                                struct genl_info *info, u8 cmd,
 780                                                bool always)
 781 {
 782         struct sk_buff *skb;
 783         int retval;
 784
 785         skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts), info,
 786                                       always);
 787         if (!skb || IS_ERR(skb))
 788                 return skb;
 789
 790         retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
 791                                         info->snd_portid, info->snd_seq, 0,
 792                                         cmd);
 793         BUG_ON(retval < 0);
 794         return skb;
 795 }
 796
 797 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
 798 {
 799         struct nlattr **a = info->attrs;
 800         struct ovs_header *ovs_header = info->userhdr;
 801         struct sw_flow_key key, masked_key;
 802         struct sw_flow *flow;
 803         struct sw_flow_mask mask;
 804         struct sk_buff *reply;
 805         struct datapath *dp;
 806         struct sw_flow_actions *acts;
 807         struct sw_flow_match match;
 808         int error;
 809
 810         /* Extract key. */
 811         error = -EINVAL;
 812         if (!a[OVS_FLOW_ATTR_KEY])
 813                 goto error;
 814
 815         ovs_match_init(&match, &key, &mask);
 816         error = ovs_nla_get_match(&match,
 817                                   a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
 818         if (error)
 819                 goto error;
 820
 821         /* Validate actions. */
 822         error = -EINVAL;
 823         if (!a[OVS_FLOW_ATTR_ACTIONS])
 824                 goto error;
 825
 826         acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
 827         error = PTR_ERR(acts);
 828         if (IS_ERR(acts))
 829                 goto error;
 830
 831         ovs_flow_mask_key(&masked_key, &key, &mask);
 832         error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS],
 833                                      &masked_key, 0, &acts);
 834         if (error) {
 835                 OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
 836                 goto err_kfree;
 837         }
 838
 839         ovs_lock();
 840         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 841         error = -ENODEV;
 842         if (!dp)
 843                 goto err_unlock_ovs;
 844
 845         /* Check if this is a duplicate flow */
 846         flow = ovs_flow_tbl_lookup(&dp->table, &key);
 847         if (!flow) {
 848                 /* Allocate flow. */
 849                 flow = ovs_flow_alloc();
 850                 if (IS_ERR(flow)) {
 851                         error = PTR_ERR(flow);
 852                         goto err_unlock_ovs;
 853                 }
 854
 855                 flow->key = masked_key;
 856                 flow->unmasked_key = key;
 857                 rcu_assign_pointer(flow->sf_acts, acts);
 858
 859                 /* Put flow in bucket. */
 860                 error = ovs_flow_tbl_insert(&dp->table, flow, &mask);
 861                 if (error) {
 862                         acts = NULL;
 863                         goto err_flow_free;
 864                 }
 865         } else {
 866                 struct sw_flow_actions *old_acts;
 867
 868                 /* Bail out if we're not allowed to modify an existing flow.
 869                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
 870                  * because Generic Netlink treats the latter as a dump
 871                  * request.  We also accept NLM_F_EXCL in case that bug ever
 872                  * gets fixed.
 873                  */
 874                 error = -EEXIST;
 875                 if (info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
 876                         goto err_unlock_ovs;
 877
 878                 /* The unmasked key has to be the same for flow updates. */
 879                 if (!ovs_flow_cmp_unmasked_key(flow, &match))
 880                         goto err_unlock_ovs;
 881
 882                 /* Update actions. */
 883                 old_acts = ovsl_dereference(flow->sf_acts);
 884                 rcu_assign_pointer(flow->sf_acts, acts);
 885                 ovs_nla_free_flow_actions(old_acts);
 886         }
 887
 888         reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
 889                                         info, OVS_FLOW_CMD_NEW, false);
 890         ovs_unlock();
 891
 892         if (reply) {
 893                 if (!IS_ERR(reply))
 894                         ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
 895                 else
 896                         netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
 897                                         ovs_dp_flow_multicast_group.id,
 898                                         PTR_ERR(reply));
 899         }
 900         return 0;
 901
 902 err_flow_free:
 903         ovs_flow_free(flow, false);
 904 err_unlock_ovs:
 905         ovs_unlock();
 906 err_kfree:
 907         kfree(acts);
 908 error:
 909         return error;
 910 }
 911
 912 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
 913 {
 914         struct nlattr **a = info->attrs;
 915         struct ovs_header *ovs_header = info->userhdr;
 916         struct sw_flow_key key, masked_key;
 917         struct sw_flow *flow;
 918         struct sw_flow_mask mask;
 919         struct sk_buff *reply = NULL;
 920         struct datapath *dp;
 921         struct sw_flow_actions *acts = NULL;
 922         struct sw_flow_match match;
 923         int error;
 924
 925         /* Extract key. */
 926         error = -EINVAL;
 927         if (!a[OVS_FLOW_ATTR_KEY])
 928                 goto error;
 929
 930         ovs_match_init(&match, &key, &mask);
 931         error = ovs_nla_get_match(&match,
 932                                   a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
 933         if (error)
 934                 goto error;
 935
 936         /* Validate actions. */
 937         if (a[OVS_FLOW_ATTR_ACTIONS]) {
 938                 acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
 939                 error = PTR_ERR(acts);
 940                 if (IS_ERR(acts))
 941                         goto error;
 942
 943                 ovs_flow_mask_key(&masked_key, &key, &mask);
 944                 error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS],
 945                                              &masked_key, 0, &acts);
 946                 if (error) {
 947                         OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
 948                         goto err_kfree;
 949                 }
 950         }
 951
 952         ovs_lock();
 953         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 954         error = -ENODEV;
 955         if (!dp)
 956                 goto err_unlock_ovs;
 957
 958         /* Check that the flow exists. */
 959         flow = ovs_flow_tbl_lookup(&dp->table, &key);
 960         error = -ENOENT;
 961         if (!flow)
 962                 goto err_unlock_ovs;
 963
 964         /* The unmasked key has to be the same for flow updates. */
 965         error = -EEXIST;
 966         if (!ovs_flow_cmp_unmasked_key(flow, &match))
 967                 goto err_unlock_ovs;
 968
 969         /* Update actions, if present. */
 970         if (acts) {
 971                 struct sw_flow_actions *old_acts;
 972
 973                 old_acts = ovsl_dereference(flow->sf_acts);
 974                 rcu_assign_pointer(flow->sf_acts, acts);
 975                 ovs_nla_free_flow_actions(old_acts);
 976         }
 977
 978         reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
 979                                         info, OVS_FLOW_CMD_NEW, false);
 980         /* Clear stats. */
 981         if (a[OVS_FLOW_ATTR_CLEAR])
 982                 ovs_flow_stats_clear(flow);
 983         ovs_unlock();
 984
 985         if (reply) {
 986                 if (!IS_ERR(reply))
 987                         ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
 988                 else
 989                         netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
 990                                         ovs_dp_flow_multicast_group.id,
 991                                         PTR_ERR(reply));
 992         }
 993         return 0;
 994
 995 err_unlock_ovs:
 996         ovs_unlock();
 997 err_kfree:
 998         kfree(acts);
 999 error:
1000         return error;
1001 }
1002
1003 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1004 {
1005         struct nlattr **a = info->attrs;
1006         struct ovs_header *ovs_header = info->userhdr;
1007         struct sw_flow_key key;
1008         struct sk_buff *reply;
1009         struct sw_flow *flow;
1010         struct datapath *dp;
1011         struct sw_flow_match match;
1012         int err;
1013
1014         if (!a[OVS_FLOW_ATTR_KEY]) {
1015                 OVS_NLERR("Flow get message rejected, Key attribute missing.\n");
1016                 return -EINVAL;
1017         }
1018
1019         ovs_match_init(&match, &key, NULL);
1020         err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
1021         if (err)
1022                 return err;
1023
1024         ovs_lock();
1025         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1026         if (!dp) {
1027                 err = -ENODEV;
1028                 goto unlock;
1029         }
1030
1031         flow = ovs_flow_tbl_lookup(&dp->table, &key);
1032         if (!flow || !ovs_flow_cmp_unmasked_key(flow, &match)) {
1033                 err = -ENOENT;
1034                 goto unlock;
1035         }
1036
1037         reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1038                                         OVS_FLOW_CMD_NEW, true);
1039         if (IS_ERR(reply)) {
1040                 err = PTR_ERR(reply);
1041                 goto unlock;
1042         }
1043
1044         ovs_unlock();
1045         return genlmsg_reply(reply, info);
1046 unlock:
1047         ovs_unlock();
1048         return err;
1049 }
1050
1051 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1052 {
1053         struct nlattr **a = info->attrs;
1054         struct ovs_header *ovs_header = info->userhdr;
1055         struct sw_flow_key key;
1056         struct sk_buff *reply;
1057         struct sw_flow *flow;
1058         struct datapath *dp;
1059         struct sw_flow_match match;
1060         int err;
1061
1062         if (likely(a[OVS_FLOW_ATTR_KEY])) {
1063                 ovs_match_init(&match, &key, NULL);
1064                 err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
1065                 if (unlikely(err))
1066                         return err;
1067         }
1068
1069         ovs_lock();
1070         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1071         if (unlikely(!dp)) {
1072                 err = -ENODEV;
1073                 goto unlock;
1074         }
1075         if (unlikely(!a[OVS_FLOW_ATTR_KEY])) {
1076                 err = ovs_flow_tbl_flush(&dp->table);
1077                 goto unlock;
1078         }
1079         flow = ovs_flow_tbl_lookup(&dp->table, &key);
1080         if (unlikely(!flow || !ovs_flow_cmp_unmasked_key(flow, &match))) {
1081                 err = -ENOENT;
1082                 goto unlock;
1083         }
1084
1085         ovs_flow_tbl_remove(&dp->table, flow);
1086         ovs_unlock();
1087
1088         reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *)flow->sf_acts,
1089                                         info, false);
1090
1091         if (likely(reply)) {
1092                 if (likely(!IS_ERR(reply))) {
1093                         rcu_read_lock(); /* Keep RCU checker happy. */
1094                         err = ovs_flow_cmd_fill_info(flow,
1095                                                      ovs_header->dp_ifindex,
1096                                                      reply, info->snd_portid,
1097                                                      info->snd_seq, 0,
1098                                                      OVS_FLOW_CMD_DEL);
1099                         rcu_read_unlock();
1100                         BUG_ON(err < 0);
1101                         ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
1102                 } else {
1103                         netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
1104                                         ovs_dp_flow_multicast_group.id,
1105                                         PTR_ERR(reply));
1106                 }
1107         }
1108
1109         ovs_flow_free(flow, true);
1110         return 0;
1111 unlock:
1112         ovs_unlock();
1113         return err;
1114 }
1115
1116 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1117 {
1118         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1119         struct table_instance *ti;
1120         struct datapath *dp;
1121
1122         rcu_read_lock();
1123         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1124         if (!dp) {
1125                 rcu_read_unlock();
1126                 return -ENODEV;
1127         }
1128
1129         ti = rcu_dereference(dp->table.ti);
1130         for (;;) {
1131                 struct sw_flow *flow;
1132                 u32 bucket, obj;
1133
1134                 bucket = cb->args[0];
1135                 obj = cb->args[1];
1136                 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1137                 if (!flow)
1138                         break;
1139
1140                 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1141                                            NETLINK_CB(cb->skb).portid,
1142                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1143                                            OVS_FLOW_CMD_NEW) < 0)
1144                         break;
1145
1146                 cb->args[0] = bucket;
1147                 cb->args[1] = obj;
1148         }
1149         rcu_read_unlock();
1150         return skb->len;
1151 }
1152
1153 static struct genl_ops dp_flow_genl_ops[] = {
1154         { .cmd = OVS_FLOW_CMD_NEW,
1155           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1156           .policy = flow_policy,
1157           .doit = ovs_flow_cmd_new
1158         },
1159         { .cmd = OVS_FLOW_CMD_DEL,
1160           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1161           .policy = flow_policy,
1162           .doit = ovs_flow_cmd_del
1163         },
1164         { .cmd = OVS_FLOW_CMD_GET,
1165           .flags = 0,               /* OK for unprivileged users. */
1166           .policy = flow_policy,
1167           .doit = ovs_flow_cmd_get,
1168           .dumpit = ovs_flow_cmd_dump
1169         },
1170         { .cmd = OVS_FLOW_CMD_SET,
1171           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1172           .policy = flow_policy,
1173           .doit = ovs_flow_cmd_set,
1174         },
1175 };
1176
1177 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1178         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1179         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1180         [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1181 };
1182
1183 static struct genl_family dp_datapath_genl_family = {
1184         .id = GENL_ID_GENERATE,
1185         .hdrsize = sizeof(struct ovs_header),
1186         .name = OVS_DATAPATH_FAMILY,
1187         .version = OVS_DATAPATH_VERSION,
1188         .maxattr = OVS_DP_ATTR_MAX,
1189         .netnsok = true,
1190          SET_PARALLEL_OPS
1191 };
1192
1193 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1194         .name = OVS_DATAPATH_MCGROUP
1195 };
1196
1197 static size_t ovs_dp_cmd_msg_size(void)
1198 {
1199         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1200
1201         msgsize += nla_total_size(IFNAMSIZ);
1202         msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1203         msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1204         msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1205
1206         return msgsize;
1207 }
1208
1209 /* Called with ovs_mutex or RCU read lock. */
1210 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1211                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1212 {
1213         struct ovs_header *ovs_header;
1214         struct ovs_dp_stats dp_stats;
1215         struct ovs_dp_megaflow_stats dp_megaflow_stats;
1216         int err;
1217
1218         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1219                                    flags, cmd);
1220         if (!ovs_header)
1221                 goto error;
1222
1223         ovs_header->dp_ifindex = get_dpifindex(dp);
1224
1225         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1226         if (err)
1227                 goto nla_put_failure;
1228
1229         get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1230         if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1231                         &dp_stats))
1232                 goto nla_put_failure;
1233
1234         if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1235                         sizeof(struct ovs_dp_megaflow_stats),
1236                         &dp_megaflow_stats))
1237                 goto nla_put_failure;
1238
1239         if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1240                 goto nla_put_failure;
1241
1242         return genlmsg_end(skb, ovs_header);
1243
1244 nla_put_failure:
1245         genlmsg_cancel(skb, ovs_header);
1246 error:
1247         return -EMSGSIZE;
1248 }
1249
1250 static struct sk_buff *ovs_dp_cmd_alloc_info(struct genl_info *info)
1251 {
1252         return genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info, GFP_KERNEL);
1253 }
1254
1255 /* Called with rcu_read_lock or ovs_mutex. */
1256 static struct datapath *lookup_datapath(struct net *net,
1257                                         struct ovs_header *ovs_header,
1258                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1259 {
1260         struct datapath *dp;
1261
1262         if (!a[OVS_DP_ATTR_NAME])
1263                 dp = get_dp(net, ovs_header->dp_ifindex);
1264         else {
1265                 struct vport *vport;
1266
1267                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1268                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1269         }
1270         return dp ? dp : ERR_PTR(-ENODEV);
1271 }
1272
1273 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1274 {
1275         struct datapath *dp;
1276
1277         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1278         if (IS_ERR(dp))
1279                 return;
1280
1281         WARN(dp->user_features, "Dropping previously announced user features\n");
1282         dp->user_features = 0;
1283 }
1284
1285 static void ovs_dp_change(struct datapath *dp, struct nlattr **a)
1286 {
1287         if (a[OVS_DP_ATTR_USER_FEATURES])
1288                 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1289 }
1290
1291 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1292 {
1293         struct nlattr **a = info->attrs;
1294         struct vport_parms parms;
1295         struct sk_buff *reply;
1296         struct datapath *dp;
1297         struct vport *vport;
1298         struct ovs_net *ovs_net;
1299         int err, i;
1300
1301         err = -EINVAL;
1302         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1303                 goto err;
1304
1305         reply = ovs_dp_cmd_alloc_info(info);
1306         if (!reply)
1307                 return -ENOMEM;
1308
1309         err = -ENOMEM;
1310         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1311         if (dp == NULL)
1312                 goto err_free_reply;
1313
1314         ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1315
1316         /* Allocate table. */
1317         err = ovs_flow_tbl_init(&dp->table);
1318         if (err)
1319                 goto err_free_dp;
1320
1321         dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1322         if (!dp->stats_percpu) {
1323                 err = -ENOMEM;
1324                 goto err_destroy_table;
1325         }
1326
1327         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1328                             GFP_KERNEL);
1329         if (!dp->ports) {
1330                 err = -ENOMEM;
1331                 goto err_destroy_percpu;
1332         }
1333
1334         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1335                 INIT_HLIST_HEAD(&dp->ports[i]);
1336
1337         /* Set up our datapath device. */
1338         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1339         parms.type = OVS_VPORT_TYPE_INTERNAL;
1340         parms.options = NULL;
1341         parms.dp = dp;
1342         parms.port_no = OVSP_LOCAL;
1343         parms.upcall_portid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1344
1345         ovs_dp_change(dp, a);
1346
1347         /* So far only local changes have been made, now need the lock. */
1348         ovs_lock();
1349
1350         vport = new_vport(&parms);
1351         if (IS_ERR(vport)) {
1352                 err = PTR_ERR(vport);
1353                 if (err == -EBUSY)
1354                         err = -EEXIST;
1355
1356                 if (err == -EEXIST) {
1357                         /* An outdated user space instance that does not understand
1358                          * the concept of user_features has attempted to create a new
1359                          * datapath and is likely to reuse it. Drop all user features.
1360                          */
1361                         if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1362                                 ovs_dp_reset_user_features(skb, info);
1363                 }
1364
1365                 goto err_destroy_ports_array;
1366         }
1367
1368         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1369                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1370         BUG_ON(err < 0);
1371
1372         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1373         list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1374
1375         ovs_unlock();
1376
1377         ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1378         return 0;
1379
1380 err_destroy_ports_array:
1381         ovs_unlock();
1382         kfree(dp->ports);
1383 err_destroy_percpu:
1384         free_percpu(dp->stats_percpu);
1385 err_destroy_table:
1386         ovs_flow_tbl_destroy(&dp->table, false);
1387 err_free_dp:
1388         release_net(ovs_dp_get_net(dp));
1389         kfree(dp);
1390 err_free_reply:
1391         kfree_skb(reply);
1392 err:
1393         return err;
1394 }
1395
1396 /* Called with ovs_mutex. */
1397 static void __dp_destroy(struct datapath *dp)
1398 {
1399         int i;
1400
1401         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1402                 struct vport *vport;
1403                 struct hlist_node *n;
1404
1405                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1406                         if (vport->port_no != OVSP_LOCAL)
1407                                 ovs_dp_detach_port(vport);
1408         }
1409
1410         list_del_rcu(&dp->list_node);
1411
1412         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1413          * all ports in datapath are destroyed first before freeing datapath.
1414          */
1415         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1416
1417         /* RCU destroy the flow table */
1418         ovs_flow_tbl_destroy(&dp->table, true);
1419
1420         call_rcu(&dp->rcu, destroy_dp_rcu);
1421 }
1422
1423 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1424 {
1425         struct sk_buff *reply;
1426         struct datapath *dp;
1427         int err;
1428
1429         reply = ovs_dp_cmd_alloc_info(info);
1430         if (!reply)
1431                 return -ENOMEM;
1432
1433         ovs_lock();
1434         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1435         err = PTR_ERR(dp);
1436         if (IS_ERR(dp))
1437                 goto err_unlock_free;
1438
1439         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1440                                    info->snd_seq, 0, OVS_DP_CMD_DEL);
1441         BUG_ON(err < 0);
1442
1443         __dp_destroy(dp);
1444
1445         ovs_unlock();
1446         ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1447         return 0;
1448
1449 err_unlock_free:
1450         ovs_unlock();
1451         kfree_skb(reply);
1452         return err;
1453 }
1454
1455 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1456 {
1457         struct sk_buff *reply;
1458         struct datapath *dp;
1459         int err;
1460
1461         reply = ovs_dp_cmd_alloc_info(info);
1462         if (!reply)
1463                 return -ENOMEM;
1464
1465         ovs_lock();
1466         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1467         err = PTR_ERR(dp);
1468         if (IS_ERR(dp))
1469                 goto err_unlock_free;
1470
1471         ovs_dp_change(dp, info->attrs);
1472
1473         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1474                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1475         BUG_ON(err < 0);
1476
1477         ovs_unlock();
1478         ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1479         return 0;
1480
1481 err_unlock_free:
1482         ovs_unlock();
1483         kfree_skb(reply);
1484         return err;
1485 }
1486
1487 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1488 {
1489         struct sk_buff *reply;
1490         struct datapath *dp;
1491         int err;
1492
1493         reply = ovs_dp_cmd_alloc_info(info);
1494         if (!reply)
1495                 return -ENOMEM;
1496
1497         rcu_read_lock();
1498         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1499         if (IS_ERR(dp)) {
1500                 err = PTR_ERR(dp);
1501                 goto err_unlock_free;
1502         }
1503         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1504                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1505         BUG_ON(err < 0);
1506         rcu_read_unlock();
1507
1508         return genlmsg_reply(reply, info);
1509
1510 err_unlock_free:
1511         rcu_read_unlock();
1512         kfree_skb(reply);
1513         return err;
1514 }
1515
1516 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1517 {
1518         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1519         struct datapath *dp;
1520         int skip = cb->args[0];
1521         int i = 0;
1522
1523         rcu_read_lock();
1524         list_for_each_entry_rcu(dp, &ovs_net->dps, list_node) {
1525                 if (i >= skip &&
1526                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1527                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1528                                          OVS_DP_CMD_NEW) < 0)
1529                         break;
1530                 i++;
1531         }
1532         rcu_read_unlock();
1533
1534         cb->args[0] = i;
1535
1536         return skb->len;
1537 }
1538
1539 static struct genl_ops dp_datapath_genl_ops[] = {
1540         { .cmd = OVS_DP_CMD_NEW,
1541           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1542           .policy = datapath_policy,
1543           .doit = ovs_dp_cmd_new
1544         },
1545         { .cmd = OVS_DP_CMD_DEL,
1546           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1547           .policy = datapath_policy,
1548           .doit = ovs_dp_cmd_del
1549         },
1550         { .cmd = OVS_DP_CMD_GET,
1551           .flags = 0,               /* OK for unprivileged users. */
1552           .policy = datapath_policy,
1553           .doit = ovs_dp_cmd_get,
1554           .dumpit = ovs_dp_cmd_dump
1555         },
1556         { .cmd = OVS_DP_CMD_SET,
1557           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1558           .policy = datapath_policy,
1559           .doit = ovs_dp_cmd_set,
1560         },
1561 };
1562
1563 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1564         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1565         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1566         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1567         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1568         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1569         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1570 };
1571
1572 static struct genl_family dp_vport_genl_family = {
1573         .id = GENL_ID_GENERATE,
1574         .hdrsize = sizeof(struct ovs_header),
1575         .name = OVS_VPORT_FAMILY,
1576         .version = OVS_VPORT_VERSION,
1577         .maxattr = OVS_VPORT_ATTR_MAX,
1578         .netnsok = true,
1579          SET_PARALLEL_OPS
1580 };
1581
1582 struct genl_multicast_group ovs_dp_vport_multicast_group = {
1583         .name = OVS_VPORT_MCGROUP
1584 };
1585
1586 /* Called with ovs_mutex or RCU read lock. */
1587 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1588                                    u32 portid, u32 seq, u32 flags, u8 cmd)
1589 {
1590         struct ovs_header *ovs_header;
1591         struct ovs_vport_stats vport_stats;
1592         int err;
1593
1594         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1595                                  flags, cmd);
1596         if (!ovs_header)
1597                 return -EMSGSIZE;
1598
1599         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1600
1601         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1602             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1603             nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1604             nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_portid))
1605                 goto nla_put_failure;
1606
1607         ovs_vport_get_stats(vport, &vport_stats);
1608         if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1609                     &vport_stats))
1610                 goto nla_put_failure;
1611
1612         err = ovs_vport_get_options(vport, skb);
1613         if (err == -EMSGSIZE)
1614                 goto error;
1615
1616         return genlmsg_end(skb, ovs_header);
1617
1618 nla_put_failure:
1619         err = -EMSGSIZE;
1620 error:
1621         genlmsg_cancel(skb, ovs_header);
1622         return err;
1623 }
1624
1625 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1626 {
1627         return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1628 }
1629
1630 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1631 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1632                                          u32 seq, u8 cmd)
1633 {
1634         struct sk_buff *skb;
1635         int retval;
1636
1637         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1638         if (!skb)
1639                 return ERR_PTR(-ENOMEM);
1640
1641         retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1642         BUG_ON(retval < 0);
1643
1644         return skb;
1645 }
1646
1647 /* Called with ovs_mutex or RCU read lock. */
1648 static struct vport *lookup_vport(struct net *net,
1649                                   struct ovs_header *ovs_header,
1650                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1651 {
1652         struct datapath *dp;
1653         struct vport *vport;
1654
1655         if (a[OVS_VPORT_ATTR_NAME]) {
1656                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1657                 if (!vport)
1658                         return ERR_PTR(-ENODEV);
1659                 if (ovs_header->dp_ifindex &&
1660                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1661                         return ERR_PTR(-ENODEV);
1662                 return vport;
1663         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1664                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1665
1666                 if (port_no >= DP_MAX_PORTS)
1667                         return ERR_PTR(-EFBIG);
1668
1669                 dp = get_dp(net, ovs_header->dp_ifindex);
1670                 if (!dp)
1671                         return ERR_PTR(-ENODEV);
1672
1673                 vport = ovs_vport_ovsl_rcu(dp, port_no);
1674                 if (!vport)
1675                         return ERR_PTR(-ENODEV);
1676                 return vport;
1677         } else
1678                 return ERR_PTR(-EINVAL);
1679 }
1680
1681 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1682 {
1683         struct nlattr **a = info->attrs;
1684         struct ovs_header *ovs_header = info->userhdr;
1685         struct vport_parms parms;
1686         struct sk_buff *reply;
1687         struct vport *vport;
1688         struct datapath *dp;
1689         u32 port_no;
1690         int err;
1691
1692         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1693             !a[OVS_VPORT_ATTR_UPCALL_PID])
1694                 return -EINVAL;
1695
1696         port_no = a[OVS_VPORT_ATTR_PORT_NO]
1697                 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
1698         if (port_no >= DP_MAX_PORTS)
1699                 return -EFBIG;
1700
1701         reply = ovs_vport_cmd_alloc_info();
1702         if (!reply)
1703                 return -ENOMEM;
1704
1705         ovs_lock();
1706         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1707         err = -ENODEV;
1708         if (!dp)
1709                 goto exit_unlock_free;
1710
1711         if (port_no) {
1712                 vport = ovs_vport_ovsl(dp, port_no);
1713                 err = -EBUSY;
1714                 if (vport)
1715                         goto exit_unlock_free;
1716         } else {
1717                 for (port_no = 1; ; port_no++) {
1718                         if (port_no >= DP_MAX_PORTS) {
1719                                 err = -EFBIG;
1720                                 goto exit_unlock_free;
1721                         }
1722                         vport = ovs_vport_ovsl(dp, port_no);
1723                         if (!vport)
1724                                 break;
1725                 }
1726         }
1727
1728         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1729         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1730         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1731         parms.dp = dp;
1732         parms.port_no = port_no;
1733         parms.upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1734
1735         vport = new_vport(&parms);
1736         err = PTR_ERR(vport);
1737         if (IS_ERR(vport))
1738                 goto exit_unlock_free;
1739
1740         err = 0;
1741         if (a[OVS_VPORT_ATTR_STATS])
1742                 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1743
1744         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1745                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1746         BUG_ON(err < 0);
1747         ovs_unlock();
1748
1749         ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
1750         return 0;
1751
1752 exit_unlock_free:
1753         ovs_unlock();
1754         kfree_skb(reply);
1755         return err;
1756 }
1757
1758 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1759 {
1760         struct nlattr **a = info->attrs;
1761         struct sk_buff *reply;
1762         struct vport *vport;
1763         int err;
1764
1765         reply = ovs_vport_cmd_alloc_info();
1766         if (!reply)
1767                 return -ENOMEM;
1768
1769         ovs_lock();
1770         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1771         err = PTR_ERR(vport);
1772         if (IS_ERR(vport))
1773                 goto exit_unlock_free;
1774
1775         if (a[OVS_VPORT_ATTR_TYPE] &&
1776             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
1777                 err = -EINVAL;
1778                 goto exit_unlock_free;
1779         }
1780
1781         if (a[OVS_VPORT_ATTR_OPTIONS]) {
1782                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1783                 if (err)
1784                         goto exit_unlock_free;
1785         }
1786
1787         if (a[OVS_VPORT_ATTR_STATS])
1788                 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1789
1790         if (a[OVS_VPORT_ATTR_UPCALL_PID])
1791                 vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1792
1793         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1794                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1795         BUG_ON(err < 0);
1796         ovs_unlock();
1797
1798         ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
1799         return 0;
1800
1801 exit_unlock_free:
1802         ovs_unlock();
1803         kfree_skb(reply);
1804         return err;
1805 }
1806
1807 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1808 {
1809         struct nlattr **a = info->attrs;
1810         struct sk_buff *reply;
1811         struct vport *vport;
1812         int err;
1813
1814         reply = ovs_vport_cmd_alloc_info();
1815         if (!reply)
1816                 return -ENOMEM;
1817
1818         ovs_lock();
1819         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1820         err = PTR_ERR(vport);
1821         if (IS_ERR(vport))
1822                 goto exit_unlock_free;
1823
1824         if (vport->port_no == OVSP_LOCAL) {
1825                 err = -EINVAL;
1826                 goto exit_unlock_free;
1827         }
1828
1829         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1830                                       info->snd_seq, 0, OVS_VPORT_CMD_DEL);
1831         BUG_ON(err < 0);
1832         ovs_dp_detach_port(vport);
1833         ovs_unlock();
1834
1835         ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
1836         return 0;
1837
1838 exit_unlock_free:
1839         ovs_unlock();
1840         kfree_skb(reply);
1841         return err;
1842 }
1843
1844 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1845 {
1846         struct nlattr **a = info->attrs;
1847         struct ovs_header *ovs_header = info->userhdr;
1848         struct sk_buff *reply;
1849         struct vport *vport;
1850         int err;
1851
1852         reply = ovs_vport_cmd_alloc_info();
1853         if (!reply)
1854                 return -ENOMEM;
1855
1856         rcu_read_lock();
1857         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1858         err = PTR_ERR(vport);
1859         if (IS_ERR(vport))
1860                 goto exit_unlock_free;
1861         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1862                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1863         BUG_ON(err < 0);
1864         rcu_read_unlock();
1865
1866         return genlmsg_reply(reply, info);
1867
1868 exit_unlock_free:
1869         rcu_read_unlock();
1870         kfree_skb(reply);
1871         return err;
1872 }
1873
1874 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1875 {
1876         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1877         struct datapath *dp;
1878         int bucket = cb->args[0], skip = cb->args[1];
1879         int i, j = 0;
1880
1881         rcu_read_lock();
1882         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1883         if (!dp) {
1884                 rcu_read_unlock();
1885                 return -ENODEV;
1886         }
1887         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
1888                 struct vport *vport;
1889
1890                 j = 0;
1891                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1892                         if (j >= skip &&
1893                             ovs_vport_cmd_fill_info(vport, skb,
1894                                                     NETLINK_CB(cb->skb).portid,
1895                                                     cb->nlh->nlmsg_seq,
1896                                                     NLM_F_MULTI,
1897                                                     OVS_VPORT_CMD_NEW) < 0)
1898                                 goto out;
1899
1900                         j++;
1901                 }
1902                 skip = 0;
1903         }
1904 out:
1905         rcu_read_unlock();
1906
1907         cb->args[0] = i;
1908         cb->args[1] = j;
1909
1910         return skb->len;
1911 }
1912
1913 static struct genl_ops dp_vport_genl_ops[] = {
1914         { .cmd = OVS_VPORT_CMD_NEW,
1915           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1916           .policy = vport_policy,
1917           .doit = ovs_vport_cmd_new
1918         },
1919         { .cmd = OVS_VPORT_CMD_DEL,
1920           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1921           .policy = vport_policy,
1922           .doit = ovs_vport_cmd_del
1923         },
1924         { .cmd = OVS_VPORT_CMD_GET,
1925           .flags = 0,               /* OK for unprivileged users. */
1926           .policy = vport_policy,
1927           .doit = ovs_vport_cmd_get,
1928           .dumpit = ovs_vport_cmd_dump
1929         },
1930         { .cmd = OVS_VPORT_CMD_SET,
1931           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1932           .policy = vport_policy,
1933           .doit = ovs_vport_cmd_set,
1934         },
1935 };
1936
1937 struct genl_family_and_ops {
1938         struct genl_family *family;
1939         struct genl_ops *ops;
1940         int n_ops;
1941         struct genl_multicast_group *group;
1942 };
1943
1944 static const struct genl_family_and_ops dp_genl_families[] = {
1945         { &dp_datapath_genl_family,
1946           dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1947           &ovs_dp_datapath_multicast_group },
1948         { &dp_vport_genl_family,
1949           dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1950           &ovs_dp_vport_multicast_group },
1951         { &dp_flow_genl_family,
1952           dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1953           &ovs_dp_flow_multicast_group },
1954         { &dp_packet_genl_family,
1955           dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1956           NULL },
1957 };
1958
1959 static void dp_unregister_genl(int n_families)
1960 {
1961         int i;
1962
1963         for (i = 0; i < n_families; i++)
1964                 genl_unregister_family(dp_genl_families[i].family);
1965 }
1966
1967 static int dp_register_genl(void)
1968 {
1969         int n_registered;
1970         int err;
1971         int i;
1972
1973         n_registered = 0;
1974         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
1975                 const struct genl_family_and_ops *f = &dp_genl_families[i];
1976
1977                 err = genl_register_family_with_ops(f->family, f->ops,
1978                                                     f->n_ops);
1979                 if (err)
1980                         goto error;
1981                 n_registered++;
1982
1983                 if (f->group) {
1984                         err = genl_register_mc_group(f->family, f->group);
1985                         if (err)
1986                                 goto error;
1987                 }
1988         }
1989
1990         return 0;
1991
1992 error:
1993         dp_unregister_genl(n_registered);
1994         return err;
1995 }
1996
1997 static int __net_init ovs_init_net(struct net *net)
1998 {
1999         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2000
2001         INIT_LIST_HEAD(&ovs_net->dps);
2002         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2003         return 0;
2004 }
2005
2006 static void __net_exit ovs_exit_net(struct net *net)
2007 {
2008         struct datapath *dp, *dp_next;
2009         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2010
2011         ovs_lock();
2012         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2013                 __dp_destroy(dp);
2014         ovs_unlock();
2015
2016         cancel_work_sync(&ovs_net->dp_notify_work);
2017 }
2018
2019 static struct pernet_operations ovs_net_ops = {
2020         .init = ovs_init_net,
2021         .exit = ovs_exit_net,
2022         .id   = &ovs_net_id,
2023         .size = sizeof(struct ovs_net),
2024 };
2025
2026 DEFINE_COMPAT_PNET_REG_FUNC(device);
2027
2028 static int __init dp_init(void)
2029 {
2030         int err;
2031
2032         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2033
2034         pr_info("Open vSwitch switching datapath %s, built "__DATE__" "__TIME__"\n",
2035                 VERSION);
2036
2037         err = ovs_flow_init();
2038         if (err)
2039                 goto error;
2040
2041         err = ovs_vport_init();
2042         if (err)
2043                 goto error_flow_exit;
2044
2045         err = register_pernet_device(&ovs_net_ops);
2046         if (err)
2047                 goto error_vport_exit;
2048
2049         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2050         if (err)
2051                 goto error_netns_exit;
2052
2053         err = dp_register_genl();
2054         if (err < 0)
2055                 goto error_unreg_notifier;
2056
2057         return 0;
2058
2059 error_unreg_notifier:
2060         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2061 error_netns_exit:
2062         unregister_pernet_device(&ovs_net_ops);
2063 error_vport_exit:
2064         ovs_vport_exit();
2065 error_flow_exit:
2066         ovs_flow_exit();
2067 error:
2068         return err;
2069 }
2070
2071 static void dp_cleanup(void)
2072 {
2073         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2074         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2075         unregister_pernet_device(&ovs_net_ops);
2076         rcu_barrier();
2077         ovs_vport_exit();
2078         ovs_flow_exit();
2079 }
2080
2081 module_init(dp_init);
2082 module_exit(dp_cleanup);
2083
2084 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2085 MODULE_LICENSE("GPL");
2086 MODULE_VERSION(VERSION);