net/openvswitch/datapath.c

   1 /*
   2  * Copyright (c) 2007-2014 Nicira, Inc.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of version 2 of the GNU General Public
   6  * License as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program; if not, write to the Free Software
  15  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
  16  * 02110-1301, USA
  17  */
  18
  19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  20
  21 #include <linux/init.h>
  22 #include <linux/module.h>
  23 #include <linux/if_arp.h>
  24 #include <linux/if_vlan.h>
  25 #include <linux/in.h>
  26 #include <linux/ip.h>
  27 #include <linux/jhash.h>
  28 #include <linux/delay.h>
  29 #include <linux/time.h>
  30 #include <linux/etherdevice.h>
  31 #include <linux/genetlink.h>
  32 #include <linux/kernel.h>
  33 #include <linux/kthread.h>
  34 #include <linux/mutex.h>
  35 #include <linux/percpu.h>
  36 #include <linux/rcupdate.h>
  37 #include <linux/tcp.h>
  38 #include <linux/udp.h>
  39 #include <linux/ethtool.h>
  40 #include <linux/wait.h>
  41 #include <asm/div64.h>
  42 #include <linux/highmem.h>
  43 #include <linux/netfilter_bridge.h>
  44 #include <linux/netfilter_ipv4.h>
  45 #include <linux/inetdevice.h>
  46 #include <linux/list.h>
  47 #include <linux/openvswitch.h>
  48 #include <linux/rculist.h>
  49 #include <linux/dmi.h>
  50 #include <net/genetlink.h>
  51 #include <net/net_namespace.h>
  52 #include <net/netns/generic.h>
  53
  54 #include "datapath.h"
  55 #include "flow.h"
  56 #include "flow_table.h"
  57 #include "flow_netlink.h"
  58 #include "vport-internal_dev.h"
  59 #include "vport-netdev.h"
  60
  61 int ovs_net_id __read_mostly;
  62 EXPORT_SYMBOL_GPL(ovs_net_id);
  63
  64 static struct genl_family dp_packet_genl_family;
  65 static struct genl_family dp_flow_genl_family;
  66 static struct genl_family dp_datapath_genl_family;
  67
  68 static const struct nla_policy flow_policy[];
  69
  70 static const struct genl_multicast_group ovs_dp_flow_multicast_group = {
  71         .name = OVS_FLOW_MCGROUP,
  72 };
  73
  74 static const struct genl_multicast_group ovs_dp_datapath_multicast_group = {
  75         .name = OVS_DATAPATH_MCGROUP,
  76 };
  77
  78 static const struct genl_multicast_group ovs_dp_vport_multicast_group = {
  79         .name = OVS_VPORT_MCGROUP,
  80 };
  81
  82 /* Check if need to build a reply message.
  83  * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
  84 static bool ovs_must_notify(struct genl_family *family, struct genl_info *info,
  85                             unsigned int group)
  86 {
  87         return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
  88                genl_has_listeners(family, genl_info_net(info), group);
  89 }
  90
  91 static void ovs_notify(struct genl_family *family,
  92                        struct sk_buff *skb, struct genl_info *info)
  93 {
  94         genl_notify(family, skb, genl_info_net(info), info->snd_portid,
  95                     0, info->nlhdr, GFP_KERNEL);
  96 }
  97
  98 /**
  99  * DOC: Locking:
 100  *
 101  * All writes e.g. Writes to device state (add/remove datapath, port, set
 102  * operations on vports, etc.), Writes to other state (flow table
 103  * modifications, set miscellaneous datapath parameters, etc.) are protected
 104  * by ovs_lock.
 105  *
 106  * Reads are protected by RCU.
 107  *
 108  * There are a few special cases (mostly stats) that have their own
 109  * synchronization but they nest under all of above and don't interact with
 110  * each other.
 111  *
 112  * The RTNL lock nests inside ovs_mutex.
 113  */
 114
 115 static DEFINE_MUTEX(ovs_mutex);
 116
 117 void ovs_lock(void)
 118 {
 119         mutex_lock(&ovs_mutex);
 120 }
 121
 122 void ovs_unlock(void)
 123 {
 124         mutex_unlock(&ovs_mutex);
 125 }
 126
 127 #ifdef CONFIG_LOCKDEP
 128 int lockdep_ovsl_is_held(void)
 129 {
 130         if (debug_locks)
 131                 return lockdep_is_held(&ovs_mutex);
 132         else
 133                 return 1;
 134 }
 135 EXPORT_SYMBOL_GPL(lockdep_ovsl_is_held);
 136 #endif
 137
 138 static struct vport *new_vport(const struct vport_parms *);
 139 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
 140                              const struct sw_flow_key *,
 141                              const struct dp_upcall_info *);
 142 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
 143                                   const struct sw_flow_key *,
 144                                   const struct dp_upcall_info *);
 145
 146 /* Must be called with rcu_read_lock. */
 147 static struct datapath *get_dp_rcu(struct net *net, int dp_ifindex)
 148 {
 149         struct net_device *dev = dev_get_by_index_rcu(net, dp_ifindex);
 150
 151         if (dev) {
 152                 struct vport *vport = ovs_internal_dev_get_vport(dev);
 153                 if (vport)
 154                         return vport->dp;
 155         }
 156
 157         return NULL;
 158 }
 159
 160 /* The caller must hold either ovs_mutex or rcu_read_lock to keep the
 161  * returned dp pointer valid.
 162  */
 163 static inline struct datapath *get_dp(struct net *net, int dp_ifindex)
 164 {
 165         struct datapath *dp;
 166
 167         WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_ovsl_is_held());
 168         rcu_read_lock();
 169         dp = get_dp_rcu(net, dp_ifindex);
 170         rcu_read_unlock();
 171
 172         return dp;
 173 }
 174
 175 /* Must be called with rcu_read_lock or ovs_mutex. */
 176 const char *ovs_dp_name(const struct datapath *dp)
 177 {
 178         struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
 179         return ovs_vport_name(vport);
 180 }
 181
 182 static int get_dpifindex(const struct datapath *dp)
 183 {
 184         struct vport *local;
 185         int ifindex;
 186
 187         rcu_read_lock();
 188
 189         local = ovs_vport_rcu(dp, OVSP_LOCAL);
 190         if (local)
 191                 ifindex = local->dev->ifindex;
 192         else
 193                 ifindex = 0;
 194
 195         rcu_read_unlock();
 196
 197         return ifindex;
 198 }
 199
 200 static void destroy_dp_rcu(struct rcu_head *rcu)
 201 {
 202         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 203
 204         ovs_flow_tbl_destroy(&dp->table);
 205         free_percpu(dp->stats_percpu);
 206         kfree(dp->ports);
 207         kfree(dp);
 208 }
 209
 210 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
 211                                             u16 port_no)
 212 {
 213         return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
 214 }
 215
 216 /* Called with ovs_mutex or RCU read lock. */
 217 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
 218 {
 219         struct vport *vport;
 220         struct hlist_head *head;
 221
 222         head = vport_hash_bucket(dp, port_no);
 223         hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
 224                 if (vport->port_no == port_no)
 225                         return vport;
 226         }
 227         return NULL;
 228 }
 229
 230 /* Called with ovs_mutex. */
 231 static struct vport *new_vport(const struct vport_parms *parms)
 232 {
 233         struct vport *vport;
 234
 235         vport = ovs_vport_add(parms);
 236         if (!IS_ERR(vport)) {
 237                 struct datapath *dp = parms->dp;
 238                 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
 239
 240                 hlist_add_head_rcu(&vport->dp_hash_node, head);
 241         }
 242         return vport;
 243 }
 244
 245 void ovs_dp_detach_port(struct vport *p)
 246 {
 247         ASSERT_OVSL();
 248
 249         /* First drop references to device. */
 250         hlist_del_rcu(&p->dp_hash_node);
 251
 252         /* Then destroy it. */
 253         ovs_vport_del(p);
 254 }
 255
 256 /* Must be called with rcu_read_lock. */
 257 void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
 258 {
 259         const struct vport *p = OVS_CB(skb)->input_vport;
 260         struct datapath *dp = p->dp;
 261         struct sw_flow *flow;
 262         struct sw_flow_actions *sf_acts;
 263         struct dp_stats_percpu *stats;
 264         u64 *stats_counter;
 265         u32 n_mask_hit;
 266
 267         stats = this_cpu_ptr(dp->stats_percpu);
 268
 269         /* Look up flow. */
 270         flow = ovs_flow_tbl_lookup_stats(&dp->table, key, &n_mask_hit);
 271         if (unlikely(!flow)) {
 272                 struct dp_upcall_info upcall;
 273                 int error;
 274
 275                 memset(&upcall, 0, sizeof(upcall));
 276                 upcall.cmd = OVS_PACKET_CMD_MISS;
 277                 upcall.portid = ovs_vport_find_upcall_portid(p, skb);
 278                 upcall.mru = OVS_CB(skb)->mru;
 279                 error = ovs_dp_upcall(dp, skb, key, &upcall);
 280                 if (unlikely(error))
 281                         kfree_skb(skb);
 282                 else
 283                         consume_skb(skb);
 284                 stats_counter = &stats->n_missed;
 285                 goto out;
 286         }
 287
 288         ovs_flow_stats_update(flow, key->tp.flags, skb);
 289         sf_acts = rcu_dereference(flow->sf_acts);
 290         ovs_execute_actions(dp, skb, sf_acts, key);
 291
 292         stats_counter = &stats->n_hit;
 293
 294 out:
 295         /* Update datapath statistics. */
 296         u64_stats_update_begin(&stats->syncp);
 297         (*stats_counter)++;
 298         stats->n_mask_hit += n_mask_hit;
 299         u64_stats_update_end(&stats->syncp);
 300 }
 301
 302 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
 303                   const struct sw_flow_key *key,
 304                   const struct dp_upcall_info *upcall_info)
 305 {
 306         struct dp_stats_percpu *stats;
 307         int err;
 308
 309         if (upcall_info->portid == 0) {
 310                 err = -ENOTCONN;
 311                 goto err;
 312         }
 313
 314         if (!skb_is_gso(skb))
 315                 err = queue_userspace_packet(dp, skb, key, upcall_info);
 316         else
 317                 err = queue_gso_packets(dp, skb, key, upcall_info);
 318         if (err)
 319                 goto err;
 320
 321         return 0;
 322
 323 err:
 324         stats = this_cpu_ptr(dp->stats_percpu);
 325
 326         u64_stats_update_begin(&stats->syncp);
 327         stats->n_lost++;
 328         u64_stats_update_end(&stats->syncp);
 329
 330         return err;
 331 }
 332
 333 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
 334                              const struct sw_flow_key *key,
 335                              const struct dp_upcall_info *upcall_info)
 336 {
 337         unsigned short gso_type = skb_shinfo(skb)->gso_type;
 338         struct sw_flow_key later_key;
 339         struct sk_buff *segs, *nskb;
 340         struct ovs_skb_cb ovs_cb;
 341         int err;
 342
 343         ovs_cb = *OVS_CB(skb);
 344         segs = __skb_gso_segment(skb, NETIF_F_SG, false);
 345         *OVS_CB(skb) = ovs_cb;
 346         if (IS_ERR(segs))
 347                 return PTR_ERR(segs);
 348         if (segs == NULL)
 349                 return -EINVAL;
 350
 351         if (gso_type & SKB_GSO_UDP) {
 352                 /* The initial flow key extracted by ovs_flow_key_extract()
 353                  * in this case is for a first fragment, so we need to
 354                  * properly mark later fragments.
 355                  */
 356                 later_key = *key;
 357                 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
 358         }
 359
 360         /* Queue all of the segments. */
 361         skb = segs;
 362         do {
 363                 *OVS_CB(skb) = ovs_cb;
 364                 if (gso_type & SKB_GSO_UDP && skb != segs)
 365                         key = &later_key;
 366
 367                 err = queue_userspace_packet(dp, skb, key, upcall_info);
 368                 if (err)
 369                         break;
 370
 371         } while ((skb = skb->next));
 372
 373         /* Free all of the segments. */
 374         skb = segs;
 375         do {
 376                 nskb = skb->next;
 377                 if (err)
 378                         kfree_skb(skb);
 379                 else
 380                         consume_skb(skb);
 381         } while ((skb = nskb));
 382         return err;
 383 }
 384
 385 static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
 386                               unsigned int hdrlen)
 387 {
 388         size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
 389                 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
 390                 + nla_total_size(ovs_key_attr_size()); /* OVS_PACKET_ATTR_KEY */
 391
 392         /* OVS_PACKET_ATTR_USERDATA */
 393         if (upcall_info->userdata)
 394                 size += NLA_ALIGN(upcall_info->userdata->nla_len);
 395
 396         /* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
 397         if (upcall_info->egress_tun_info)
 398                 size += nla_total_size(ovs_tun_key_attr_size());
 399
 400         /* OVS_PACKET_ATTR_ACTIONS */
 401         if (upcall_info->actions_len)
 402                 size += nla_total_size(upcall_info->actions_len);
 403
 404         /* OVS_PACKET_ATTR_MRU */
 405         if (upcall_info->mru)
 406                 size += nla_total_size(sizeof(upcall_info->mru));
 407
 408         return size;
 409 }
 410
 411 static void pad_packet(struct datapath *dp, struct sk_buff *skb)
 412 {
 413         if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
 414                 size_t plen = NLA_ALIGN(skb->len) - skb->len;
 415
 416                 if (plen > 0)
 417                         memset(skb_put(skb, plen), 0, plen);
 418         }
 419 }
 420
 421 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
 422                                   const struct sw_flow_key *key,
 423                                   const struct dp_upcall_info *upcall_info)
 424 {
 425         struct ovs_header *upcall;
 426         struct sk_buff *nskb = NULL;
 427         struct sk_buff *user_skb = NULL; /* to be queued to userspace */
 428         struct nlattr *nla;
 429         struct genl_info info = {
 430                 .dst_sk = ovs_dp_get_net(dp)->genl_sock,
 431                 .snd_portid = upcall_info->portid,
 432         };
 433         size_t len;
 434         unsigned int hlen;
 435         int err, dp_ifindex;
 436
 437         dp_ifindex = get_dpifindex(dp);
 438         if (!dp_ifindex)
 439                 return -ENODEV;
 440
 441         if (skb_vlan_tag_present(skb)) {
 442                 nskb = skb_clone(skb, GFP_ATOMIC);
 443                 if (!nskb)
 444                         return -ENOMEM;
 445
 446                 nskb = __vlan_hwaccel_push_inside(nskb);
 447                 if (!nskb)
 448                         return -ENOMEM;
 449
 450                 skb = nskb;
 451         }
 452
 453         if (nla_attr_size(skb->len) > USHRT_MAX) {
 454                 err = -EFBIG;
 455                 goto out;
 456         }
 457
 458         /* Complete checksum if needed */
 459         if (skb->ip_summed == CHECKSUM_PARTIAL &&
 460             (err = skb_checksum_help(skb)))
 461                 goto out;
 462
 463         /* Older versions of OVS user space enforce alignment of the last
 464          * Netlink attribute to NLA_ALIGNTO which would require extensive
 465          * padding logic. Only perform zerocopy if padding is not required.
 466          */
 467         if (dp->user_features & OVS_DP_F_UNALIGNED)
 468                 hlen = skb_zerocopy_headlen(skb);
 469         else
 470                 hlen = skb->len;
 471
 472         len = upcall_msg_size(upcall_info, hlen);
 473         user_skb = genlmsg_new_unicast(len, &info, GFP_ATOMIC);
 474         if (!user_skb) {
 475                 err = -ENOMEM;
 476                 goto out;
 477         }
 478
 479         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 480                              0, upcall_info->cmd);
 481         upcall->dp_ifindex = dp_ifindex;
 482
 483         err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
 484         BUG_ON(err);
 485
 486         if (upcall_info->userdata)
 487                 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
 488                           nla_len(upcall_info->userdata),
 489                           nla_data(upcall_info->userdata));
 490
 491         if (upcall_info->egress_tun_info) {
 492                 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_EGRESS_TUN_KEY);
 493                 err = ovs_nla_put_egress_tunnel_key(user_skb,
 494                                                     upcall_info->egress_tun_info,
 495                                                     upcall_info->egress_tun_opts);
 496                 BUG_ON(err);
 497                 nla_nest_end(user_skb, nla);
 498         }
 499
 500         if (upcall_info->actions_len) {
 501                 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_ACTIONS);
 502                 err = ovs_nla_put_actions(upcall_info->actions,
 503                                           upcall_info->actions_len,
 504                                           user_skb);
 505                 if (!err)
 506                         nla_nest_end(user_skb, nla);
 507                 else
 508                         nla_nest_cancel(user_skb, nla);
 509         }
 510
 511         /* Add OVS_PACKET_ATTR_MRU */
 512         if (upcall_info->mru) {
 513                 if (nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU,
 514                                 upcall_info->mru)) {
 515                         err = -ENOBUFS;
 516                         goto out;
 517                 }
 518                 pad_packet(dp, user_skb);
 519         }
 520
 521         /* Only reserve room for attribute header, packet data is added
 522          * in skb_zerocopy() */
 523         if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
 524                 err = -ENOBUFS;
 525                 goto out;
 526         }
 527         nla->nla_len = nla_attr_size(skb->len);
 528
 529         err = skb_zerocopy(user_skb, skb, skb->len, hlen);
 530         if (err)
 531                 goto out;
 532
 533         /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
 534         pad_packet(dp, user_skb);
 535
 536         ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
 537
 538         err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
 539         user_skb = NULL;
 540 out:
 541         if (err)
 542                 skb_tx_error(skb);
 543         kfree_skb(user_skb);
 544         kfree_skb(nskb);
 545         return err;
 546 }
 547
 548 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 549 {
 550         struct ovs_header *ovs_header = info->userhdr;
 551         struct net *net = sock_net(skb->sk);
 552         struct nlattr **a = info->attrs;
 553         struct sw_flow_actions *acts;
 554         struct sk_buff *packet;
 555         struct sw_flow *flow;
 556         struct sw_flow_actions *sf_acts;
 557         struct datapath *dp;
 558         struct ethhdr *eth;
 559         struct vport *input_vport;
 560         u16 mru = 0;
 561         int len;
 562         int err;
 563         bool log = !a[OVS_PACKET_ATTR_PROBE];
 564
 565         err = -EINVAL;
 566         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 567             !a[OVS_PACKET_ATTR_ACTIONS])
 568                 goto err;
 569
 570         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 571         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 572         err = -ENOMEM;
 573         if (!packet)
 574                 goto err;
 575         skb_reserve(packet, NET_IP_ALIGN);
 576
 577         nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
 578
 579         skb_reset_mac_header(packet);
 580         eth = eth_hdr(packet);
 581
 582         /* Normally, setting the skb 'protocol' field would be handled by a
 583          * call to eth_type_trans(), but it assumes there's a sending
 584          * device, which we may not have. */
 585         if (eth_proto_is_802_3(eth->h_proto))
 586                 packet->protocol = eth->h_proto;
 587         else
 588                 packet->protocol = htons(ETH_P_802_2);
 589
 590         /* Set packet's mru */
 591         if (a[OVS_PACKET_ATTR_MRU]) {
 592                 mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
 593                 packet->ignore_df = 1;
 594         }
 595         OVS_CB(packet)->mru = mru;
 596
 597         /* Build an sw_flow for sending this packet. */
 598         flow = ovs_flow_alloc();
 599         err = PTR_ERR(flow);
 600         if (IS_ERR(flow))
 601                 goto err_kfree_skb;
 602
 603         err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
 604                                              packet, &flow->key, log);
 605         if (err)
 606                 goto err_flow_free;
 607
 608         err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
 609                                    &flow->key, &acts, log);
 610         if (err)
 611                 goto err_flow_free;
 612
 613         rcu_assign_pointer(flow->sf_acts, acts);
 614         packet->priority = flow->key.phy.priority;
 615         packet->mark = flow->key.phy.skb_mark;
 616
 617         rcu_read_lock();
 618         dp = get_dp_rcu(net, ovs_header->dp_ifindex);
 619         err = -ENODEV;
 620         if (!dp)
 621                 goto err_unlock;
 622
 623         input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
 624         if (!input_vport)
 625                 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
 626
 627         if (!input_vport)
 628                 goto err_unlock;
 629
 630         packet->dev = input_vport->dev;
 631         OVS_CB(packet)->input_vport = input_vport;
 632         sf_acts = rcu_dereference(flow->sf_acts);
 633
 634         local_bh_disable();
 635         err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
 636         local_bh_enable();
 637         rcu_read_unlock();
 638
 639         ovs_flow_free(flow, false);
 640         return err;
 641
 642 err_unlock:
 643         rcu_read_unlock();
 644 err_flow_free:
 645         ovs_flow_free(flow, false);
 646 err_kfree_skb:
 647         kfree_skb(packet);
 648 err:
 649         return err;
 650 }
 651
 652 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 653         [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
 654         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 655         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 656         [OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
 657         [OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
 658 };
 659
 660 static const struct genl_ops dp_packet_genl_ops[] = {
 661         { .cmd = OVS_PACKET_CMD_EXECUTE,
 662           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 663           .policy = packet_policy,
 664           .doit = ovs_packet_cmd_execute
 665         }
 666 };
 667
 668 static struct genl_family dp_packet_genl_family = {
 669         .id = GENL_ID_GENERATE,
 670         .hdrsize = sizeof(struct ovs_header),
 671         .name = OVS_PACKET_FAMILY,
 672         .version = OVS_PACKET_VERSION,
 673         .maxattr = OVS_PACKET_ATTR_MAX,
 674         .netnsok = true,
 675         .parallel_ops = true,
 676         .ops = dp_packet_genl_ops,
 677         .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
 678 };
 679
 680 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
 681                          struct ovs_dp_megaflow_stats *mega_stats)
 682 {
 683         int i;
 684
 685         memset(mega_stats, 0, sizeof(*mega_stats));
 686
 687         stats->n_flows = ovs_flow_tbl_count(&dp->table);
 688         mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
 689
 690         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 691
 692         for_each_possible_cpu(i) {
 693                 const struct dp_stats_percpu *percpu_stats;
 694                 struct dp_stats_percpu local_stats;
 695                 unsigned int start;
 696
 697                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 698
 699                 do {
 700                         start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
 701                         local_stats = *percpu_stats;
 702                 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
 703
 704                 stats->n_hit += local_stats.n_hit;
 705                 stats->n_missed += local_stats.n_missed;
 706                 stats->n_lost += local_stats.n_lost;
 707                 mega_stats->n_mask_hit += local_stats.n_mask_hit;
 708         }
 709 }
 710
 711 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
 712 {
 713         return ovs_identifier_is_ufid(sfid) &&
 714                !(ufid_flags & OVS_UFID_F_OMIT_KEY);
 715 }
 716
 717 static bool should_fill_mask(uint32_t ufid_flags)
 718 {
 719         return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
 720 }
 721
 722 static bool should_fill_actions(uint32_t ufid_flags)
 723 {
 724         return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
 725 }
 726
 727 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
 728                                     const struct sw_flow_id *sfid,
 729                                     uint32_t ufid_flags)
 730 {
 731         size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
 732
 733         /* OVS_FLOW_ATTR_UFID */
 734         if (sfid && ovs_identifier_is_ufid(sfid))
 735                 len += nla_total_size(sfid->ufid_len);
 736
 737         /* OVS_FLOW_ATTR_KEY */
 738         if (!sfid || should_fill_key(sfid, ufid_flags))
 739                 len += nla_total_size(ovs_key_attr_size());
 740
 741         /* OVS_FLOW_ATTR_MASK */
 742         if (should_fill_mask(ufid_flags))
 743                 len += nla_total_size(ovs_key_attr_size());
 744
 745         /* OVS_FLOW_ATTR_ACTIONS */
 746         if (should_fill_actions(ufid_flags))
 747                 len += nla_total_size(acts->orig_len);
 748
 749         return len
 750                 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
 751                 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
 752                 + nla_total_size(8); /* OVS_FLOW_ATTR_USED */
 753 }
 754
 755 /* Called with ovs_mutex or RCU read lock. */
 756 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
 757                                    struct sk_buff *skb)
 758 {
 759         struct ovs_flow_stats stats;
 760         __be16 tcp_flags;
 761         unsigned long used;
 762
 763         ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
 764
 765         if (used &&
 766             nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
 767                 return -EMSGSIZE;
 768
 769         if (stats.n_packets &&
 770             nla_put(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats))
 771                 return -EMSGSIZE;
 772
 773         if ((u8)ntohs(tcp_flags) &&
 774              nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
 775                 return -EMSGSIZE;
 776
 777         return 0;
 778 }
 779
 780 /* Called with ovs_mutex or RCU read lock. */
 781 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
 782                                      struct sk_buff *skb, int skb_orig_len)
 783 {
 784         struct nlattr *start;
 785         int err;
 786
 787         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 788          * this is the first flow to be dumped into 'skb'.  This is unusual for
 789          * Netlink but individual action lists can be longer than
 790          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 791          * The userspace caller can always fetch the actions separately if it
 792          * really wants them.  (Most userspace callers in fact don't care.)
 793          *
 794          * This can only fail for dump operations because the skb is always
 795          * properly sized for single flows.
 796          */
 797         start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
 798         if (start) {
 799                 const struct sw_flow_actions *sf_acts;
 800
 801                 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
 802                 err = ovs_nla_put_actions(sf_acts->actions,
 803                                           sf_acts->actions_len, skb);
 804
 805                 if (!err)
 806                         nla_nest_end(skb, start);
 807                 else {
 808                         if (skb_orig_len)
 809                                 return err;
 810
 811                         nla_nest_cancel(skb, start);
 812                 }
 813         } else if (skb_orig_len) {
 814                 return -EMSGSIZE;
 815         }
 816
 817         return 0;
 818 }
 819
 820 /* Called with ovs_mutex or RCU read lock. */
 821 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
 822                                   struct sk_buff *skb, u32 portid,
 823                                   u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
 824 {
 825         const int skb_orig_len = skb->len;
 826         struct ovs_header *ovs_header;
 827         int err;
 828
 829         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
 830                                  flags, cmd);
 831         if (!ovs_header)
 832                 return -EMSGSIZE;
 833
 834         ovs_header->dp_ifindex = dp_ifindex;
 835
 836         err = ovs_nla_put_identifier(flow, skb);
 837         if (err)
 838                 goto error;
 839
 840         if (should_fill_key(&flow->id, ufid_flags)) {
 841                 err = ovs_nla_put_masked_key(flow, skb);
 842                 if (err)
 843                         goto error;
 844         }
 845
 846         if (should_fill_mask(ufid_flags)) {
 847                 err = ovs_nla_put_mask(flow, skb);
 848                 if (err)
 849                         goto error;
 850         }
 851
 852         err = ovs_flow_cmd_fill_stats(flow, skb);
 853         if (err)
 854                 goto error;
 855
 856         if (should_fill_actions(ufid_flags)) {
 857                 err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
 858                 if (err)
 859                         goto error;
 860         }
 861
 862         genlmsg_end(skb, ovs_header);
 863         return 0;
 864
 865 error:
 866         genlmsg_cancel(skb, ovs_header);
 867         return err;
 868 }
 869
 870 /* May not be called with RCU read lock. */
 871 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
 872                                                const struct sw_flow_id *sfid,
 873                                                struct genl_info *info,
 874                                                bool always,
 875                                                uint32_t ufid_flags)
 876 {
 877         struct sk_buff *skb;
 878         size_t len;
 879
 880         if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
 881                 return NULL;
 882
 883         len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
 884         skb = genlmsg_new_unicast(len, info, GFP_KERNEL);
 885         if (!skb)
 886                 return ERR_PTR(-ENOMEM);
 887
 888         return skb;
 889 }
 890
 891 /* Called with ovs_mutex. */
 892 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
 893                                                int dp_ifindex,
 894                                                struct genl_info *info, u8 cmd,
 895                                                bool always, u32 ufid_flags)
 896 {
 897         struct sk_buff *skb;
 898         int retval;
 899
 900         skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
 901                                       &flow->id, info, always, ufid_flags);
 902         if (IS_ERR_OR_NULL(skb))
 903                 return skb;
 904
 905         retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
 906                                         info->snd_portid, info->snd_seq, 0,
 907                                         cmd, ufid_flags);
 908         BUG_ON(retval < 0);
 909         return skb;
 910 }
 911
 912 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
 913 {
 914         struct net *net = sock_net(skb->sk);
 915         struct nlattr **a = info->attrs;
 916         struct ovs_header *ovs_header = info->userhdr;
 917         struct sw_flow *flow = NULL, *new_flow;
 918         struct sw_flow_mask mask;
 919         struct sk_buff *reply;
 920         struct datapath *dp;
 921         struct sw_flow_key key;
 922         struct sw_flow_actions *acts;
 923         struct sw_flow_match match;
 924         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
 925         int error;
 926         bool log = !a[OVS_FLOW_ATTR_PROBE];
 927
 928         /* Must have key and actions. */
 929         error = -EINVAL;
 930         if (!a[OVS_FLOW_ATTR_KEY]) {
 931                 OVS_NLERR(log, "Flow key attr not present in new flow.");
 932                 goto error;
 933         }
 934         if (!a[OVS_FLOW_ATTR_ACTIONS]) {
 935                 OVS_NLERR(log, "Flow actions attr not present in new flow.");
 936                 goto error;
 937         }
 938
 939         /* Most of the time we need to allocate a new flow, do it before
 940          * locking.
 941          */
 942         new_flow = ovs_flow_alloc();
 943         if (IS_ERR(new_flow)) {
 944                 error = PTR_ERR(new_flow);
 945                 goto error;
 946         }
 947
 948         /* Extract key. */
 949         ovs_match_init(&match, &key, &mask);
 950         error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
 951                                   a[OVS_FLOW_ATTR_MASK], log);
 952         if (error)
 953                 goto err_kfree_flow;
 954
 955         ovs_flow_mask_key(&new_flow->key, &key, true, &mask);
 956
 957         /* Extract flow identifier. */
 958         error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
 959                                        &key, log);
 960         if (error)
 961                 goto err_kfree_flow;
 962
 963         /* Validate actions. */
 964         error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
 965                                      &new_flow->key, &acts, log);
 966         if (error) {
 967                 OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
 968                 goto err_kfree_flow;
 969         }
 970
 971         reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
 972                                         ufid_flags);
 973         if (IS_ERR(reply)) {
 974                 error = PTR_ERR(reply);
 975                 goto err_kfree_acts;
 976         }
 977
 978         ovs_lock();
 979         dp = get_dp(net, ovs_header->dp_ifindex);
 980         if (unlikely(!dp)) {
 981                 error = -ENODEV;
 982                 goto err_unlock_ovs;
 983         }
 984
 985         /* Check if this is a duplicate flow */
 986         if (ovs_identifier_is_ufid(&new_flow->id))
 987                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
 988         if (!flow)
 989                 flow = ovs_flow_tbl_lookup(&dp->table, &key);
 990         if (likely(!flow)) {
 991                 rcu_assign_pointer(new_flow->sf_acts, acts);
 992
 993                 /* Put flow in bucket. */
 994                 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
 995                 if (unlikely(error)) {
 996                         acts = NULL;
 997                         goto err_unlock_ovs;
 998                 }
 999
1000                 if (unlikely(reply)) {
1001                         error = ovs_flow_cmd_fill_info(new_flow,
1002                                                        ovs_header->dp_ifindex,
1003                                                        reply, info->snd_portid,
1004                                                        info->snd_seq, 0,
1005                                                        OVS_FLOW_CMD_NEW,
1006                                                        ufid_flags);
1007                         BUG_ON(error < 0);
1008                 }
1009                 ovs_unlock();
1010         } else {
1011                 struct sw_flow_actions *old_acts;
1012
1013                 /* Bail out if we're not allowed to modify an existing flow.
1014                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1015                  * because Generic Netlink treats the latter as a dump
1016                  * request.  We also accept NLM_F_EXCL in case that bug ever
1017                  * gets fixed.
1018                  */
1019                 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1020                                                          | NLM_F_EXCL))) {
1021                         error = -EEXIST;
1022                         goto err_unlock_ovs;
1023                 }
1024                 /* The flow identifier has to be the same for flow updates.
1025                  * Look for any overlapping flow.
1026                  */
1027                 if (unlikely(!ovs_flow_cmp(flow, &match))) {
1028                         if (ovs_identifier_is_key(&flow->id))
1029                                 flow = ovs_flow_tbl_lookup_exact(&dp->table,
1030                                                                  &match);
1031                         else /* UFID matches but key is different */
1032                                 flow = NULL;
1033                         if (!flow) {
1034                                 error = -ENOENT;
1035                                 goto err_unlock_ovs;
1036                         }
1037                 }
1038                 /* Update actions. */
1039                 old_acts = ovsl_dereference(flow->sf_acts);
1040                 rcu_assign_pointer(flow->sf_acts, acts);
1041
1042                 if (unlikely(reply)) {
1043                         error = ovs_flow_cmd_fill_info(flow,
1044                                                        ovs_header->dp_ifindex,
1045                                                        reply, info->snd_portid,
1046                                                        info->snd_seq, 0,
1047                                                        OVS_FLOW_CMD_NEW,
1048                                                        ufid_flags);
1049                         BUG_ON(error < 0);
1050                 }
1051                 ovs_unlock();
1052
1053                 ovs_nla_free_flow_actions_rcu(old_acts);
1054                 ovs_flow_free(new_flow, false);
1055         }
1056
1057         if (reply)
1058                 ovs_notify(&dp_flow_genl_family, reply, info);
1059         return 0;
1060
1061 err_unlock_ovs:
1062         ovs_unlock();
1063         kfree_skb(reply);
1064 err_kfree_acts:
1065         ovs_nla_free_flow_actions(acts);
1066 err_kfree_flow:
1067         ovs_flow_free(new_flow, false);
1068 error:
1069         return error;
1070 }
1071
1072 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1073 static struct sw_flow_actions *get_flow_actions(struct net *net,
1074                                                 const struct nlattr *a,
1075                                                 const struct sw_flow_key *key,
1076                                                 const struct sw_flow_mask *mask,
1077                                                 bool log)
1078 {
1079         struct sw_flow_actions *acts;
1080         struct sw_flow_key masked_key;
1081         int error;
1082
1083         ovs_flow_mask_key(&masked_key, key, true, mask);
1084         error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1085         if (error) {
1086                 OVS_NLERR(log,
1087                           "Actions may not be safe on all matching packets");
1088                 return ERR_PTR(error);
1089         }
1090
1091         return acts;
1092 }
1093
1094 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1095 {
1096         struct net *net = sock_net(skb->sk);
1097         struct nlattr **a = info->attrs;
1098         struct ovs_header *ovs_header = info->userhdr;
1099         struct sw_flow_key key;
1100         struct sw_flow *flow;
1101         struct sw_flow_mask mask;
1102         struct sk_buff *reply = NULL;
1103         struct datapath *dp;
1104         struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1105         struct sw_flow_match match;
1106         struct sw_flow_id sfid;
1107         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1108         int error;
1109         bool log = !a[OVS_FLOW_ATTR_PROBE];
1110         bool ufid_present;
1111
1112         /* Extract key. */
1113         error = -EINVAL;
1114         if (!a[OVS_FLOW_ATTR_KEY]) {
1115                 OVS_NLERR(log, "Flow key attribute not present in set flow.");
1116                 goto error;
1117         }
1118
1119         ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1120         ovs_match_init(&match, &key, &mask);
1121         error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1122                                   a[OVS_FLOW_ATTR_MASK], log);
1123         if (error)
1124                 goto error;
1125
1126         /* Validate actions. */
1127         if (a[OVS_FLOW_ATTR_ACTIONS]) {
1128                 acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], &key,
1129                                         &mask, log);
1130                 if (IS_ERR(acts)) {
1131                         error = PTR_ERR(acts);
1132                         goto error;
1133                 }
1134
1135                 /* Can allocate before locking if have acts. */
1136                 reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1137                                                 ufid_flags);
1138                 if (IS_ERR(reply)) {
1139                         error = PTR_ERR(reply);
1140                         goto err_kfree_acts;
1141                 }
1142         }
1143
1144         ovs_lock();
1145         dp = get_dp(net, ovs_header->dp_ifindex);
1146         if (unlikely(!dp)) {
1147                 error = -ENODEV;
1148                 goto err_unlock_ovs;
1149         }
1150         /* Check that the flow exists. */
1151         if (ufid_present)
1152                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1153         else
1154                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1155         if (unlikely(!flow)) {
1156                 error = -ENOENT;
1157                 goto err_unlock_ovs;
1158         }
1159
1160         /* Update actions, if present. */
1161         if (likely(acts)) {
1162                 old_acts = ovsl_dereference(flow->sf_acts);
1163                 rcu_assign_pointer(flow->sf_acts, acts);
1164
1165                 if (unlikely(reply)) {
1166                         error = ovs_flow_cmd_fill_info(flow,
1167                                                        ovs_header->dp_ifindex,
1168                                                        reply, info->snd_portid,
1169                                                        info->snd_seq, 0,
1170                                                        OVS_FLOW_CMD_NEW,
1171                                                        ufid_flags);
1172                         BUG_ON(error < 0);
1173                 }
1174         } else {
1175                 /* Could not alloc without acts before locking. */
1176                 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1177                                                 info, OVS_FLOW_CMD_NEW, false,
1178                                                 ufid_flags);
1179
1180                 if (unlikely(IS_ERR(reply))) {
1181                         error = PTR_ERR(reply);
1182                         goto err_unlock_ovs;
1183                 }
1184         }
1185
1186         /* Clear stats. */
1187         if (a[OVS_FLOW_ATTR_CLEAR])
1188                 ovs_flow_stats_clear(flow);
1189         ovs_unlock();
1190
1191         if (reply)
1192                 ovs_notify(&dp_flow_genl_family, reply, info);
1193         if (old_acts)
1194                 ovs_nla_free_flow_actions_rcu(old_acts);
1195
1196         return 0;
1197
1198 err_unlock_ovs:
1199         ovs_unlock();
1200         kfree_skb(reply);
1201 err_kfree_acts:
1202         ovs_nla_free_flow_actions(acts);
1203 error:
1204         return error;
1205 }
1206
1207 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1208 {
1209         struct nlattr **a = info->attrs;
1210         struct ovs_header *ovs_header = info->userhdr;
1211         struct net *net = sock_net(skb->sk);
1212         struct sw_flow_key key;
1213         struct sk_buff *reply;
1214         struct sw_flow *flow;
1215         struct datapath *dp;
1216         struct sw_flow_match match;
1217         struct sw_flow_id ufid;
1218         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1219         int err = 0;
1220         bool log = !a[OVS_FLOW_ATTR_PROBE];
1221         bool ufid_present;
1222
1223         ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1224         if (a[OVS_FLOW_ATTR_KEY]) {
1225                 ovs_match_init(&match, &key, NULL);
1226                 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1227                                         log);
1228         } else if (!ufid_present) {
1229                 OVS_NLERR(log,
1230                           "Flow get message rejected, Key attribute missing.");
1231                 err = -EINVAL;
1232         }
1233         if (err)
1234                 return err;
1235
1236         ovs_lock();
1237         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1238         if (!dp) {
1239                 err = -ENODEV;
1240                 goto unlock;
1241         }
1242
1243         if (ufid_present)
1244                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1245         else
1246                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1247         if (!flow) {
1248                 err = -ENOENT;
1249                 goto unlock;
1250         }
1251
1252         reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1253                                         OVS_FLOW_CMD_NEW, true, ufid_flags);
1254         if (IS_ERR(reply)) {
1255                 err = PTR_ERR(reply);
1256                 goto unlock;
1257         }
1258
1259         ovs_unlock();
1260         return genlmsg_reply(reply, info);
1261 unlock:
1262         ovs_unlock();
1263         return err;
1264 }
1265
1266 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1267 {
1268         struct nlattr **a = info->attrs;
1269         struct ovs_header *ovs_header = info->userhdr;
1270         struct net *net = sock_net(skb->sk);
1271         struct sw_flow_key key;
1272         struct sk_buff *reply;
1273         struct sw_flow *flow = NULL;
1274         struct datapath *dp;
1275         struct sw_flow_match match;
1276         struct sw_flow_id ufid;
1277         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1278         int err;
1279         bool log = !a[OVS_FLOW_ATTR_PROBE];
1280         bool ufid_present;
1281
1282         ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1283         if (a[OVS_FLOW_ATTR_KEY]) {
1284                 ovs_match_init(&match, &key, NULL);
1285                 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1286                                         NULL, log);
1287                 if (unlikely(err))
1288                         return err;
1289         }
1290
1291         ovs_lock();
1292         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1293         if (unlikely(!dp)) {
1294                 err = -ENODEV;
1295                 goto unlock;
1296         }
1297
1298         if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1299                 err = ovs_flow_tbl_flush(&dp->table);
1300                 goto unlock;
1301         }
1302
1303         if (ufid_present)
1304                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1305         else
1306                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1307         if (unlikely(!flow)) {
1308                 err = -ENOENT;
1309                 goto unlock;
1310         }
1311
1312         ovs_flow_tbl_remove(&dp->table, flow);
1313         ovs_unlock();
1314
1315         reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1316                                         &flow->id, info, false, ufid_flags);
1317         if (likely(reply)) {
1318                 if (likely(!IS_ERR(reply))) {
1319                         rcu_read_lock();        /*To keep RCU checker happy. */
1320                         err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1321                                                      reply, info->snd_portid,
1322                                                      info->snd_seq, 0,
1323                                                      OVS_FLOW_CMD_DEL,
1324                                                      ufid_flags);
1325                         rcu_read_unlock();
1326                         BUG_ON(err < 0);
1327
1328                         ovs_notify(&dp_flow_genl_family, reply, info);
1329                 } else {
1330                         netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
1331                 }
1332         }
1333
1334         ovs_flow_free(flow, true);
1335         return 0;
1336 unlock:
1337         ovs_unlock();
1338         return err;
1339 }
1340
1341 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1342 {
1343         struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1344         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1345         struct table_instance *ti;
1346         struct datapath *dp;
1347         u32 ufid_flags;
1348         int err;
1349
1350         err = genlmsg_parse(cb->nlh, &dp_flow_genl_family, a,
1351                             OVS_FLOW_ATTR_MAX, flow_policy);
1352         if (err)
1353                 return err;
1354         ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1355
1356         rcu_read_lock();
1357         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1358         if (!dp) {
1359                 rcu_read_unlock();
1360                 return -ENODEV;
1361         }
1362
1363         ti = rcu_dereference(dp->table.ti);
1364         for (;;) {
1365                 struct sw_flow *flow;
1366                 u32 bucket, obj;
1367
1368                 bucket = cb->args[0];
1369                 obj = cb->args[1];
1370                 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1371                 if (!flow)
1372                         break;
1373
1374                 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1375                                            NETLINK_CB(cb->skb).portid,
1376                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1377                                            OVS_FLOW_CMD_NEW, ufid_flags) < 0)
1378                         break;
1379
1380                 cb->args[0] = bucket;
1381                 cb->args[1] = obj;
1382         }
1383         rcu_read_unlock();
1384         return skb->len;
1385 }
1386
1387 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1388         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1389         [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1390         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1391         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1392         [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1393         [OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1394         [OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1395 };
1396
1397 static const struct genl_ops dp_flow_genl_ops[] = {
1398         { .cmd = OVS_FLOW_CMD_NEW,
1399           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1400           .policy = flow_policy,
1401           .doit = ovs_flow_cmd_new
1402         },
1403         { .cmd = OVS_FLOW_CMD_DEL,
1404           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1405           .policy = flow_policy,
1406           .doit = ovs_flow_cmd_del
1407         },
1408         { .cmd = OVS_FLOW_CMD_GET,
1409           .flags = 0,               /* OK for unprivileged users. */
1410           .policy = flow_policy,
1411           .doit = ovs_flow_cmd_get,
1412           .dumpit = ovs_flow_cmd_dump
1413         },
1414         { .cmd = OVS_FLOW_CMD_SET,
1415           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1416           .policy = flow_policy,
1417           .doit = ovs_flow_cmd_set,
1418         },
1419 };
1420
1421 static struct genl_family dp_flow_genl_family = {
1422         .id = GENL_ID_GENERATE,
1423         .hdrsize = sizeof(struct ovs_header),
1424         .name = OVS_FLOW_FAMILY,
1425         .version = OVS_FLOW_VERSION,
1426         .maxattr = OVS_FLOW_ATTR_MAX,
1427         .netnsok = true,
1428         .parallel_ops = true,
1429         .ops = dp_flow_genl_ops,
1430         .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1431         .mcgrps = &ovs_dp_flow_multicast_group,
1432         .n_mcgrps = 1,
1433 };
1434
1435 static size_t ovs_dp_cmd_msg_size(void)
1436 {
1437         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1438
1439         msgsize += nla_total_size(IFNAMSIZ);
1440         msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1441         msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1442         msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1443
1444         return msgsize;
1445 }
1446
1447 /* Called with ovs_mutex. */
1448 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1449                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1450 {
1451         struct ovs_header *ovs_header;
1452         struct ovs_dp_stats dp_stats;
1453         struct ovs_dp_megaflow_stats dp_megaflow_stats;
1454         int err;
1455
1456         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1457                                    flags, cmd);
1458         if (!ovs_header)
1459                 goto error;
1460
1461         ovs_header->dp_ifindex = get_dpifindex(dp);
1462
1463         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1464         if (err)
1465                 goto nla_put_failure;
1466
1467         get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1468         if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1469                         &dp_stats))
1470                 goto nla_put_failure;
1471
1472         if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1473                         sizeof(struct ovs_dp_megaflow_stats),
1474                         &dp_megaflow_stats))
1475                 goto nla_put_failure;
1476
1477         if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1478                 goto nla_put_failure;
1479
1480         genlmsg_end(skb, ovs_header);
1481         return 0;
1482
1483 nla_put_failure:
1484         genlmsg_cancel(skb, ovs_header);
1485 error:
1486         return -EMSGSIZE;
1487 }
1488
1489 static struct sk_buff *ovs_dp_cmd_alloc_info(struct genl_info *info)
1490 {
1491         return genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info, GFP_KERNEL);
1492 }
1493
1494 /* Called with rcu_read_lock or ovs_mutex. */
1495 static struct datapath *lookup_datapath(struct net *net,
1496                                         const struct ovs_header *ovs_header,
1497                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1498 {
1499         struct datapath *dp;
1500
1501         if (!a[OVS_DP_ATTR_NAME])
1502                 dp = get_dp(net, ovs_header->dp_ifindex);
1503         else {
1504                 struct vport *vport;
1505
1506                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1507                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1508         }
1509         return dp ? dp : ERR_PTR(-ENODEV);
1510 }
1511
1512 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1513 {
1514         struct datapath *dp;
1515
1516         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1517         if (IS_ERR(dp))
1518                 return;
1519
1520         WARN(dp->user_features, "Dropping previously announced user features\n");
1521         dp->user_features = 0;
1522 }
1523
1524 static void ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1525 {
1526         if (a[OVS_DP_ATTR_USER_FEATURES])
1527                 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1528 }
1529
1530 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1531 {
1532         struct nlattr **a = info->attrs;
1533         struct vport_parms parms;
1534         struct sk_buff *reply;
1535         struct datapath *dp;
1536         struct vport *vport;
1537         struct ovs_net *ovs_net;
1538         int err, i;
1539
1540         err = -EINVAL;
1541         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1542                 goto err;
1543
1544         reply = ovs_dp_cmd_alloc_info(info);
1545         if (!reply)
1546                 return -ENOMEM;
1547
1548         err = -ENOMEM;
1549         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1550         if (dp == NULL)
1551                 goto err_free_reply;
1552
1553         ovs_dp_set_net(dp, sock_net(skb->sk));
1554
1555         /* Allocate table. */
1556         err = ovs_flow_tbl_init(&dp->table);
1557         if (err)
1558                 goto err_free_dp;
1559
1560         dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1561         if (!dp->stats_percpu) {
1562                 err = -ENOMEM;
1563                 goto err_destroy_table;
1564         }
1565
1566         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1567                             GFP_KERNEL);
1568         if (!dp->ports) {
1569                 err = -ENOMEM;
1570                 goto err_destroy_percpu;
1571         }
1572
1573         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1574                 INIT_HLIST_HEAD(&dp->ports[i]);
1575
1576         /* Set up our datapath device. */
1577         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1578         parms.type = OVS_VPORT_TYPE_INTERNAL;
1579         parms.options = NULL;
1580         parms.dp = dp;
1581         parms.port_no = OVSP_LOCAL;
1582         parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1583
1584         ovs_dp_change(dp, a);
1585
1586         /* So far only local changes have been made, now need the lock. */
1587         ovs_lock();
1588
1589         vport = new_vport(&parms);
1590         if (IS_ERR(vport)) {
1591                 err = PTR_ERR(vport);
1592                 if (err == -EBUSY)
1593                         err = -EEXIST;
1594
1595                 if (err == -EEXIST) {
1596                         /* An outdated user space instance that does not understand
1597                          * the concept of user_features has attempted to create a new
1598                          * datapath and is likely to reuse it. Drop all user features.
1599                          */
1600                         if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1601                                 ovs_dp_reset_user_features(skb, info);
1602                 }
1603
1604                 goto err_destroy_ports_array;
1605         }
1606
1607         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1608                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1609         BUG_ON(err < 0);
1610
1611         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1612         list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1613
1614         ovs_unlock();
1615
1616         ovs_notify(&dp_datapath_genl_family, reply, info);
1617         return 0;
1618
1619 err_destroy_ports_array:
1620         ovs_unlock();
1621         kfree(dp->ports);
1622 err_destroy_percpu:
1623         free_percpu(dp->stats_percpu);
1624 err_destroy_table:
1625         ovs_flow_tbl_destroy(&dp->table);
1626 err_free_dp:
1627         kfree(dp);
1628 err_free_reply:
1629         kfree_skb(reply);
1630 err:
1631         return err;
1632 }
1633
1634 /* Called with ovs_mutex. */
1635 static void __dp_destroy(struct datapath *dp)
1636 {
1637         int i;
1638
1639         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1640                 struct vport *vport;
1641                 struct hlist_node *n;
1642
1643                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1644                         if (vport->port_no != OVSP_LOCAL)
1645                                 ovs_dp_detach_port(vport);
1646         }
1647
1648         list_del_rcu(&dp->list_node);
1649
1650         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1651          * all ports in datapath are destroyed first before freeing datapath.
1652          */
1653         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1654
1655         /* RCU destroy the flow table */
1656         call_rcu(&dp->rcu, destroy_dp_rcu);
1657 }
1658
1659 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1660 {
1661         struct sk_buff *reply;
1662         struct datapath *dp;
1663         int err;
1664
1665         reply = ovs_dp_cmd_alloc_info(info);
1666         if (!reply)
1667                 return -ENOMEM;
1668
1669         ovs_lock();
1670         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1671         err = PTR_ERR(dp);
1672         if (IS_ERR(dp))
1673                 goto err_unlock_free;
1674
1675         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1676                                    info->snd_seq, 0, OVS_DP_CMD_DEL);
1677         BUG_ON(err < 0);
1678
1679         __dp_destroy(dp);
1680         ovs_unlock();
1681
1682         ovs_notify(&dp_datapath_genl_family, reply, info);
1683
1684         return 0;
1685
1686 err_unlock_free:
1687         ovs_unlock();
1688         kfree_skb(reply);
1689         return err;
1690 }
1691
1692 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1693 {
1694         struct sk_buff *reply;
1695         struct datapath *dp;
1696         int err;
1697
1698         reply = ovs_dp_cmd_alloc_info(info);
1699         if (!reply)
1700                 return -ENOMEM;
1701
1702         ovs_lock();
1703         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1704         err = PTR_ERR(dp);
1705         if (IS_ERR(dp))
1706                 goto err_unlock_free;
1707
1708         ovs_dp_change(dp, info->attrs);
1709
1710         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1711                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1712         BUG_ON(err < 0);
1713
1714         ovs_unlock();
1715         ovs_notify(&dp_datapath_genl_family, reply, info);
1716
1717         return 0;
1718
1719 err_unlock_free:
1720         ovs_unlock();
1721         kfree_skb(reply);
1722         return err;
1723 }
1724
1725 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1726 {
1727         struct sk_buff *reply;
1728         struct datapath *dp;
1729         int err;
1730
1731         reply = ovs_dp_cmd_alloc_info(info);
1732         if (!reply)
1733                 return -ENOMEM;
1734
1735         ovs_lock();
1736         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1737         if (IS_ERR(dp)) {
1738                 err = PTR_ERR(dp);
1739                 goto err_unlock_free;
1740         }
1741         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1742                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1743         BUG_ON(err < 0);
1744         ovs_unlock();
1745
1746         return genlmsg_reply(reply, info);
1747
1748 err_unlock_free:
1749         ovs_unlock();
1750         kfree_skb(reply);
1751         return err;
1752 }
1753
1754 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1755 {
1756         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1757         struct datapath *dp;
1758         int skip = cb->args[0];
1759         int i = 0;
1760
1761         ovs_lock();
1762         list_for_each_entry(dp, &ovs_net->dps, list_node) {
1763                 if (i >= skip &&
1764                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1765                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1766                                          OVS_DP_CMD_NEW) < 0)
1767                         break;
1768                 i++;
1769         }
1770         ovs_unlock();
1771
1772         cb->args[0] = i;
1773
1774         return skb->len;
1775 }
1776
1777 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1778         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1779         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1780         [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1781 };
1782
1783 static const struct genl_ops dp_datapath_genl_ops[] = {
1784         { .cmd = OVS_DP_CMD_NEW,
1785           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1786           .policy = datapath_policy,
1787           .doit = ovs_dp_cmd_new
1788         },
1789         { .cmd = OVS_DP_CMD_DEL,
1790           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1791           .policy = datapath_policy,
1792           .doit = ovs_dp_cmd_del
1793         },
1794         { .cmd = OVS_DP_CMD_GET,
1795           .flags = 0,               /* OK for unprivileged users. */
1796           .policy = datapath_policy,
1797           .doit = ovs_dp_cmd_get,
1798           .dumpit = ovs_dp_cmd_dump
1799         },
1800         { .cmd = OVS_DP_CMD_SET,
1801           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1802           .policy = datapath_policy,
1803           .doit = ovs_dp_cmd_set,
1804         },
1805 };
1806
1807 static struct genl_family dp_datapath_genl_family = {
1808         .id = GENL_ID_GENERATE,
1809         .hdrsize = sizeof(struct ovs_header),
1810         .name = OVS_DATAPATH_FAMILY,
1811         .version = OVS_DATAPATH_VERSION,
1812         .maxattr = OVS_DP_ATTR_MAX,
1813         .netnsok = true,
1814         .parallel_ops = true,
1815         .ops = dp_datapath_genl_ops,
1816         .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1817         .mcgrps = &ovs_dp_datapath_multicast_group,
1818         .n_mcgrps = 1,
1819 };
1820
1821 /* Called with ovs_mutex or RCU read lock. */
1822 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1823                                    u32 portid, u32 seq, u32 flags, u8 cmd)
1824 {
1825         struct ovs_header *ovs_header;
1826         struct ovs_vport_stats vport_stats;
1827         int err;
1828
1829         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1830                                  flags, cmd);
1831         if (!ovs_header)
1832                 return -EMSGSIZE;
1833
1834         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1835
1836         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1837             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1838             nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1839                            ovs_vport_name(vport)))
1840                 goto nla_put_failure;
1841
1842         ovs_vport_get_stats(vport, &vport_stats);
1843         if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1844                     &vport_stats))
1845                 goto nla_put_failure;
1846
1847         if (ovs_vport_get_upcall_portids(vport, skb))
1848                 goto nla_put_failure;
1849
1850         err = ovs_vport_get_options(vport, skb);
1851         if (err == -EMSGSIZE)
1852                 goto error;
1853
1854         genlmsg_end(skb, ovs_header);
1855         return 0;
1856
1857 nla_put_failure:
1858         err = -EMSGSIZE;
1859 error:
1860         genlmsg_cancel(skb, ovs_header);
1861         return err;
1862 }
1863
1864 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1865 {
1866         return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1867 }
1868
1869 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1870 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1871                                          u32 seq, u8 cmd)
1872 {
1873         struct sk_buff *skb;
1874         int retval;
1875
1876         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1877         if (!skb)
1878                 return ERR_PTR(-ENOMEM);
1879
1880         retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1881         BUG_ON(retval < 0);
1882
1883         return skb;
1884 }
1885
1886 /* Called with ovs_mutex or RCU read lock. */
1887 static struct vport *lookup_vport(struct net *net,
1888                                   const struct ovs_header *ovs_header,
1889                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1890 {
1891         struct datapath *dp;
1892         struct vport *vport;
1893
1894         if (a[OVS_VPORT_ATTR_NAME]) {
1895                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1896                 if (!vport)
1897                         return ERR_PTR(-ENODEV);
1898                 if (ovs_header->dp_ifindex &&
1899                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1900                         return ERR_PTR(-ENODEV);
1901                 return vport;
1902         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1903                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1904
1905                 if (port_no >= DP_MAX_PORTS)
1906                         return ERR_PTR(-EFBIG);
1907
1908                 dp = get_dp(net, ovs_header->dp_ifindex);
1909                 if (!dp)
1910                         return ERR_PTR(-ENODEV);
1911
1912                 vport = ovs_vport_ovsl_rcu(dp, port_no);
1913                 if (!vport)
1914                         return ERR_PTR(-ENODEV);
1915                 return vport;
1916         } else
1917                 return ERR_PTR(-EINVAL);
1918 }
1919
1920 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1921 {
1922         struct nlattr **a = info->attrs;
1923         struct ovs_header *ovs_header = info->userhdr;
1924         struct vport_parms parms;
1925         struct sk_buff *reply;
1926         struct vport *vport;
1927         struct datapath *dp;
1928         u32 port_no;
1929         int err;
1930
1931         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1932             !a[OVS_VPORT_ATTR_UPCALL_PID])
1933                 return -EINVAL;
1934
1935         port_no = a[OVS_VPORT_ATTR_PORT_NO]
1936                 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
1937         if (port_no >= DP_MAX_PORTS)
1938                 return -EFBIG;
1939
1940         reply = ovs_vport_cmd_alloc_info();
1941         if (!reply)
1942                 return -ENOMEM;
1943
1944         ovs_lock();
1945 restart:
1946         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1947         err = -ENODEV;
1948         if (!dp)
1949                 goto exit_unlock_free;
1950
1951         if (port_no) {
1952                 vport = ovs_vport_ovsl(dp, port_no);
1953                 err = -EBUSY;
1954                 if (vport)
1955                         goto exit_unlock_free;
1956         } else {
1957                 for (port_no = 1; ; port_no++) {
1958                         if (port_no >= DP_MAX_PORTS) {
1959                                 err = -EFBIG;
1960                                 goto exit_unlock_free;
1961                         }
1962                         vport = ovs_vport_ovsl(dp, port_no);
1963                         if (!vport)
1964                                 break;
1965                 }
1966         }
1967
1968         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1969         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1970         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1971         parms.dp = dp;
1972         parms.port_no = port_no;
1973         parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
1974
1975         vport = new_vport(&parms);
1976         err = PTR_ERR(vport);
1977         if (IS_ERR(vport)) {
1978                 if (err == -EAGAIN)
1979                         goto restart;
1980                 goto exit_unlock_free;
1981         }
1982
1983         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1984                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1985         BUG_ON(err < 0);
1986         ovs_unlock();
1987
1988         ovs_notify(&dp_vport_genl_family, reply, info);
1989         return 0;
1990
1991 exit_unlock_free:
1992         ovs_unlock();
1993         kfree_skb(reply);
1994         return err;
1995 }
1996
1997 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1998 {
1999         struct nlattr **a = info->attrs;
2000         struct sk_buff *reply;
2001         struct vport *vport;
2002         int err;
2003
2004         reply = ovs_vport_cmd_alloc_info();
2005         if (!reply)
2006                 return -ENOMEM;
2007
2008         ovs_lock();
2009         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2010         err = PTR_ERR(vport);
2011         if (IS_ERR(vport))
2012                 goto exit_unlock_free;
2013
2014         if (a[OVS_VPORT_ATTR_TYPE] &&
2015             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2016                 err = -EINVAL;
2017                 goto exit_unlock_free;
2018         }
2019
2020         if (a[OVS_VPORT_ATTR_OPTIONS]) {
2021                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2022                 if (err)
2023                         goto exit_unlock_free;
2024         }
2025
2026
2027         if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2028                 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2029
2030                 err = ovs_vport_set_upcall_portids(vport, ids);
2031                 if (err)
2032                         goto exit_unlock_free;
2033         }
2034
2035         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2036                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2037         BUG_ON(err < 0);
2038
2039         ovs_unlock();
2040         ovs_notify(&dp_vport_genl_family, reply, info);
2041         return 0;
2042
2043 exit_unlock_free:
2044         ovs_unlock();
2045         kfree_skb(reply);
2046         return err;
2047 }
2048
2049 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2050 {
2051         struct nlattr **a = info->attrs;
2052         struct sk_buff *reply;
2053         struct vport *vport;
2054         int err;
2055
2056         reply = ovs_vport_cmd_alloc_info();
2057         if (!reply)
2058                 return -ENOMEM;
2059
2060         ovs_lock();
2061         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2062         err = PTR_ERR(vport);
2063         if (IS_ERR(vport))
2064                 goto exit_unlock_free;
2065
2066         if (vport->port_no == OVSP_LOCAL) {
2067                 err = -EINVAL;
2068                 goto exit_unlock_free;
2069         }
2070
2071         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2072                                       info->snd_seq, 0, OVS_VPORT_CMD_DEL);
2073         BUG_ON(err < 0);
2074         ovs_dp_detach_port(vport);
2075         ovs_unlock();
2076
2077         ovs_notify(&dp_vport_genl_family, reply, info);
2078         return 0;
2079
2080 exit_unlock_free:
2081         ovs_unlock();
2082         kfree_skb(reply);
2083         return err;
2084 }
2085
2086 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2087 {
2088         struct nlattr **a = info->attrs;
2089         struct ovs_header *ovs_header = info->userhdr;
2090         struct sk_buff *reply;
2091         struct vport *vport;
2092         int err;
2093
2094         reply = ovs_vport_cmd_alloc_info();
2095         if (!reply)
2096                 return -ENOMEM;
2097
2098         rcu_read_lock();
2099         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2100         err = PTR_ERR(vport);
2101         if (IS_ERR(vport))
2102                 goto exit_unlock_free;
2103         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2104                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2105         BUG_ON(err < 0);
2106         rcu_read_unlock();
2107
2108         return genlmsg_reply(reply, info);
2109
2110 exit_unlock_free:
2111         rcu_read_unlock();
2112         kfree_skb(reply);
2113         return err;
2114 }
2115
2116 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2117 {
2118         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2119         struct datapath *dp;
2120         int bucket = cb->args[0], skip = cb->args[1];
2121         int i, j = 0;
2122
2123         rcu_read_lock();
2124         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2125         if (!dp) {
2126                 rcu_read_unlock();
2127                 return -ENODEV;
2128         }
2129         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2130                 struct vport *vport;
2131
2132                 j = 0;
2133                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2134                         if (j >= skip &&
2135                             ovs_vport_cmd_fill_info(vport, skb,
2136                                                     NETLINK_CB(cb->skb).portid,
2137                                                     cb->nlh->nlmsg_seq,
2138                                                     NLM_F_MULTI,
2139                                                     OVS_VPORT_CMD_NEW) < 0)
2140                                 goto out;
2141
2142                         j++;
2143                 }
2144                 skip = 0;
2145         }
2146 out:
2147         rcu_read_unlock();
2148
2149         cb->args[0] = i;
2150         cb->args[1] = j;
2151
2152         return skb->len;
2153 }
2154
2155 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2156         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2157         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2158         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2159         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2160         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
2161         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2162 };
2163
2164 static const struct genl_ops dp_vport_genl_ops[] = {
2165         { .cmd = OVS_VPORT_CMD_NEW,
2166           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2167           .policy = vport_policy,
2168           .doit = ovs_vport_cmd_new
2169         },
2170         { .cmd = OVS_VPORT_CMD_DEL,
2171           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2172           .policy = vport_policy,
2173           .doit = ovs_vport_cmd_del
2174         },
2175         { .cmd = OVS_VPORT_CMD_GET,
2176           .flags = 0,               /* OK for unprivileged users. */
2177           .policy = vport_policy,
2178           .doit = ovs_vport_cmd_get,
2179           .dumpit = ovs_vport_cmd_dump
2180         },
2181         { .cmd = OVS_VPORT_CMD_SET,
2182           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2183           .policy = vport_policy,
2184           .doit = ovs_vport_cmd_set,
2185         },
2186 };
2187
2188 struct genl_family dp_vport_genl_family = {
2189         .id = GENL_ID_GENERATE,
2190         .hdrsize = sizeof(struct ovs_header),
2191         .name = OVS_VPORT_FAMILY,
2192         .version = OVS_VPORT_VERSION,
2193         .maxattr = OVS_VPORT_ATTR_MAX,
2194         .netnsok = true,
2195         .parallel_ops = true,
2196         .ops = dp_vport_genl_ops,
2197         .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2198         .mcgrps = &ovs_dp_vport_multicast_group,
2199         .n_mcgrps = 1,
2200 };
2201
2202 static struct genl_family * const dp_genl_families[] = {
2203         &dp_datapath_genl_family,
2204         &dp_vport_genl_family,
2205         &dp_flow_genl_family,
2206         &dp_packet_genl_family,
2207 };
2208
2209 static void dp_unregister_genl(int n_families)
2210 {
2211         int i;
2212
2213         for (i = 0; i < n_families; i++)
2214                 genl_unregister_family(dp_genl_families[i]);
2215 }
2216
2217 static int dp_register_genl(void)
2218 {
2219         int err;
2220         int i;
2221
2222         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2223
2224                 err = genl_register_family(dp_genl_families[i]);
2225                 if (err)
2226                         goto error;
2227         }
2228
2229         return 0;
2230
2231 error:
2232         dp_unregister_genl(i);
2233         return err;
2234 }
2235
2236 static int __net_init ovs_init_net(struct net *net)
2237 {
2238         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2239
2240         INIT_LIST_HEAD(&ovs_net->dps);
2241         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2242         ovs_ct_init(net);
2243         return 0;
2244 }
2245
2246 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2247                                             struct list_head *head)
2248 {
2249         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2250         struct datapath *dp;
2251
2252         list_for_each_entry(dp, &ovs_net->dps, list_node) {
2253                 int i;
2254
2255                 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2256                         struct vport *vport;
2257
2258                         hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2259                                 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2260                                         continue;
2261
2262                                 if (dev_net(vport->dev) == dnet)
2263                                         list_add(&vport->detach_list, head);
2264                         }
2265                 }
2266         }
2267 }
2268
2269 static void __net_exit ovs_exit_net(struct net *dnet)
2270 {
2271         struct datapath *dp, *dp_next;
2272         struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2273         struct vport *vport, *vport_next;
2274         struct net *net;
2275         LIST_HEAD(head);
2276
2277         ovs_ct_exit(dnet);
2278         ovs_lock();
2279         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2280                 __dp_destroy(dp);
2281
2282         rtnl_lock();
2283         for_each_net(net)
2284                 list_vports_from_net(net, dnet, &head);
2285         rtnl_unlock();
2286
2287         /* Detach all vports from given namespace. */
2288         list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2289                 list_del(&vport->detach_list);
2290                 ovs_dp_detach_port(vport);
2291         }
2292
2293         ovs_unlock();
2294
2295         cancel_work_sync(&ovs_net->dp_notify_work);
2296 }
2297
2298 static struct pernet_operations ovs_net_ops = {
2299         .init = ovs_init_net,
2300         .exit = ovs_exit_net,
2301         .id   = &ovs_net_id,
2302         .size = sizeof(struct ovs_net),
2303 };
2304
2305 static int __init dp_init(void)
2306 {
2307         int err;
2308
2309         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2310
2311         pr_info("Open vSwitch switching datapath\n");
2312
2313         err = action_fifos_init();
2314         if (err)
2315                 goto error;
2316
2317         err = ovs_internal_dev_rtnl_link_register();
2318         if (err)
2319                 goto error_action_fifos_exit;
2320
2321         err = ovs_flow_init();
2322         if (err)
2323                 goto error_unreg_rtnl_link;
2324
2325         err = ovs_vport_init();
2326         if (err)
2327                 goto error_flow_exit;
2328
2329         err = register_pernet_device(&ovs_net_ops);
2330         if (err)
2331                 goto error_vport_exit;
2332
2333         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2334         if (err)
2335                 goto error_netns_exit;
2336
2337         err = ovs_netdev_init();
2338         if (err)
2339                 goto error_unreg_notifier;
2340
2341         err = dp_register_genl();
2342         if (err < 0)
2343                 goto error_unreg_netdev;
2344
2345         return 0;
2346
2347 error_unreg_netdev:
2348         ovs_netdev_exit();
2349 error_unreg_notifier:
2350         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2351 error_netns_exit:
2352         unregister_pernet_device(&ovs_net_ops);
2353 error_vport_exit:
2354         ovs_vport_exit();
2355 error_flow_exit:
2356         ovs_flow_exit();
2357 error_unreg_rtnl_link:
2358         ovs_internal_dev_rtnl_link_unregister();
2359 error_action_fifos_exit:
2360         action_fifos_exit();
2361 error:
2362         return err;
2363 }
2364
2365 static void dp_cleanup(void)
2366 {
2367         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2368         ovs_netdev_exit();
2369         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2370         unregister_pernet_device(&ovs_net_ops);
2371         rcu_barrier();
2372         ovs_vport_exit();
2373         ovs_flow_exit();
2374         ovs_internal_dev_rtnl_link_unregister();
2375         action_fifos_exit();
2376 }
2377
2378 module_init(dp_init);
2379 module_exit(dp_cleanup);
2380
2381 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2382 MODULE_LICENSE("GPL");