datapath/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/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 <net/genetlink.h>
  52 #include <net/net_namespace.h>
  53 #include <net/netns/generic.h>
  54
  55 #include "datapath.h"
  56 #include "flow.h"
  57 #include "flow_table.h"
  58 #include "flow_netlink.h"
  59 #include "vlan.h"
  60 #include "vport-internal_dev.h"
  61 #include "vport-netdev.h"
  62
  63 int ovs_net_id __read_mostly;
  64
  65 static struct genl_family dp_packet_genl_family;
  66 static struct genl_family dp_flow_genl_family;
  67 static struct genl_family dp_datapath_genl_family;
  68
  69 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
  70         .name = OVS_FLOW_MCGROUP
  71 };
  72
  73 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
  74         .name = OVS_DATAPATH_MCGROUP
  75 };
  76
  77 struct genl_multicast_group ovs_dp_vport_multicast_group = {
  78         .name = OVS_VPORT_MCGROUP
  79 };
  80
  81 /* Check if need to build a reply message.
  82  * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
  83 static bool ovs_must_notify(struct genl_family *family, struct genl_info *info,
  84                             unsigned int group)
  85 {
  86         return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
  87                genl_has_listeners(family, genl_info_net(info)->genl_sock,
  88                                   group);
  89 }
  90
  91 static void ovs_notify(struct genl_family *family, struct genl_multicast_group *grp,
  92                        struct sk_buff *skb, struct genl_info *info)
  93 {
  94         genl_notify(family, skb, genl_info_net(info),
  95                     info->snd_portid, GROUP_ID(grp), 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 #endif
 136
 137 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
 138                              const struct sw_flow_key *,
 139                              const struct dp_upcall_info *);
 140 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
 141                                   const struct sw_flow_key *key,
 142                                   const struct dp_upcall_info *);
 143
 144 /* Must be called with rcu_read_lock. */
 145 static struct datapath *get_dp_rcu(struct net *net, int dp_ifindex)
 146 {
 147         struct net_device *dev = dev_get_by_index_rcu(net, dp_ifindex);
 148
 149         if (dev) {
 150                 struct vport *vport = ovs_internal_dev_get_vport(dev);
 151                 if (vport)
 152                         return vport->dp;
 153         }
 154
 155         return NULL;
 156 }
 157
 158 /* The caller must hold either ovs_mutex or rcu_read_lock to keep the
 159  * returned dp pointer valid. */
 160 static inline struct datapath *get_dp(struct net *net, int dp_ifindex)
 161 {
 162         struct datapath *dp;
 163
 164         WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_ovsl_is_held());
 165         rcu_read_lock();
 166         dp = get_dp_rcu(net, dp_ifindex);
 167         rcu_read_unlock();
 168
 169         return dp;
 170 }
 171
 172 /* Must be called with rcu_read_lock or ovs_mutex. */
 173 const char *ovs_dp_name(const struct datapath *dp)
 174 {
 175         struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
 176         return vport->ops->get_name(vport);
 177 }
 178
 179 static int get_dpifindex(const struct datapath *dp)
 180 {
 181         struct vport *local;
 182         int ifindex;
 183
 184         rcu_read_lock();
 185
 186         local = ovs_vport_rcu(dp, OVSP_LOCAL);
 187         if (local)
 188                 ifindex = netdev_vport_priv(local)->dev->ifindex;
 189         else
 190                 ifindex = 0;
 191
 192         rcu_read_unlock();
 193
 194         return ifindex;
 195 }
 196
 197 static void destroy_dp_rcu(struct rcu_head *rcu)
 198 {
 199         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 200
 201         ovs_flow_tbl_destroy(&dp->table);
 202         free_percpu(dp->stats_percpu);
 203         release_net(ovs_dp_get_net(dp));
 204         kfree(dp->ports);
 205         kfree(dp);
 206 }
 207
 208 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
 209                                             u16 port_no)
 210 {
 211         return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
 212 }
 213
 214 /* Called with ovs_mutex or RCU read lock. */
 215 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
 216 {
 217         struct vport *vport;
 218         struct hlist_head *head;
 219
 220         head = vport_hash_bucket(dp, port_no);
 221         hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
 222                 if (vport->port_no == port_no)
 223                         return vport;
 224         }
 225         return NULL;
 226 }
 227
 228 /* Called with ovs_mutex. */
 229 static struct vport *new_vport(const struct vport_parms *parms)
 230 {
 231         struct vport *vport;
 232
 233         vport = ovs_vport_add(parms);
 234         if (!IS_ERR(vport)) {
 235                 struct datapath *dp = parms->dp;
 236                 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
 237
 238                 hlist_add_head_rcu(&vport->dp_hash_node, head);
 239         }
 240         return vport;
 241 }
 242
 243 void ovs_dp_detach_port(struct vport *p)
 244 {
 245         ASSERT_OVSL();
 246
 247         /* First drop references to device. */
 248         hlist_del_rcu(&p->dp_hash_node);
 249
 250         /* Then destroy it. */
 251         ovs_vport_del(p);
 252 }
 253
 254 /* Must be called with rcu_read_lock. */
 255 void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
 256 {
 257         const struct vport *p = OVS_CB(skb)->input_vport;
 258         struct datapath *dp = p->dp;
 259         struct sw_flow *flow;
 260         struct sw_flow_actions *sf_acts;
 261         struct dp_stats_percpu *stats;
 262         u64 *stats_counter;
 263         u32 n_mask_hit;
 264
 265         stats = this_cpu_ptr(dp->stats_percpu);
 266
 267         /* Look up flow. */
 268         flow = ovs_flow_tbl_lookup_stats(&dp->table, key, skb_get_hash(skb),
 269                                          &n_mask_hit);
 270         if (unlikely(!flow)) {
 271                 struct dp_upcall_info upcall;
 272                 int error;
 273
 274                 upcall.cmd = OVS_PACKET_CMD_MISS;
 275                 upcall.userdata = NULL;
 276                 upcall.portid = ovs_vport_find_upcall_portid(p, skb);
 277                 upcall.egress_tun_info = NULL;
 278
 279                 error = ovs_dp_upcall(dp, skb, key, &upcall);
 280                 if (unlikely(error))
 281                         kfree_skb(skb);
 282                 else
 283                         consume_skb(skb);
 284
 285                 stats_counter = &stats->n_missed;
 286                 goto out;
 287         }
 288
 289         ovs_flow_stats_update(flow, key->tp.flags, skb);
 290
 291         sf_acts = rcu_dereference(flow->sf_acts);
 292         ovs_execute_actions(dp, skb, key, sf_acts);
 293         stats_counter = &stats->n_hit;
 294
 295 out:
 296         /* Update datapath statistics. */
 297         u64_stats_update_begin(&stats->syncp);
 298         (*stats_counter)++;
 299         stats->n_mask_hit += n_mask_hit;
 300         u64_stats_update_end(&stats->syncp);
 301 }
 302
 303 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
 304                   const struct sw_flow_key *key,
 305                   const struct dp_upcall_info *upcall_info)
 306 {
 307         struct dp_stats_percpu *stats;
 308         int err;
 309
 310         if (upcall_info->portid == 0) {
 311                 err = -ENOTCONN;
 312                 goto err;
 313         }
 314
 315         if (!skb_is_gso(skb))
 316                 err = queue_userspace_packet(dp, skb, key, upcall_info);
 317         else
 318                 err = queue_gso_packets(dp, skb, key, upcall_info);
 319         if (err)
 320                 goto err;
 321
 322         return 0;
 323
 324 err:
 325         stats = this_cpu_ptr(dp->stats_percpu);
 326
 327         u64_stats_update_begin(&stats->syncp);
 328         stats->n_lost++;
 329         u64_stats_update_end(&stats->syncp);
 330
 331         return err;
 332 }
 333
 334 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
 335                              const struct sw_flow_key *key,
 336                              const struct dp_upcall_info *upcall_info)
 337 {
 338         unsigned short gso_type = skb_shinfo(skb)->gso_type;
 339         struct sw_flow_key later_key;
 340         struct sk_buff *segs, *nskb;
 341         struct ovs_skb_cb ovs_cb;
 342         int err;
 343
 344         ovs_cb = *OVS_CB(skb);
 345         segs = __skb_gso_segment(skb, NETIF_F_SG, false);
 346         *OVS_CB(skb) = ovs_cb;
 347         if (IS_ERR(segs))
 348                 return PTR_ERR(segs);
 349         if (segs == NULL)
 350                 return -EINVAL;
 351
 352         if (gso_type & SKB_GSO_UDP) {
 353                 /* The initial flow key extracted by ovs_flow_key_extract()
 354                  * in this case is for a first fragment, so we need to
 355                  * properly mark later fragments.
 356                  */
 357                 later_key = *key;
 358                 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
 359         }
 360
 361         /* Queue all of the segments. */
 362         skb = segs;
 363         do {
 364                 *OVS_CB(skb) = ovs_cb;
 365                 if (gso_type & SKB_GSO_UDP && skb != segs)
 366                         key = &later_key;
 367
 368                 err = queue_userspace_packet(dp, skb, key, upcall_info);
 369                 if (err)
 370                         break;
 371
 372         } while ((skb = skb->next));
 373
 374         /* Free all of the segments. */
 375         skb = segs;
 376         do {
 377                 nskb = skb->next;
 378                 if (err)
 379                         kfree_skb(skb);
 380                 else
 381                         consume_skb(skb);
 382         } while ((skb = nskb));
 383         return err;
 384 }
 385
 386 static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
 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(ovs_key_attr_size()); /* OVS_PACKET_ATTR_KEY */
 392
 393         /* OVS_PACKET_ATTR_USERDATA */
 394         if (upcall_info->userdata)
 395                 size += NLA_ALIGN(upcall_info->userdata->nla_len);
 396
 397         /* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
 398         if (upcall_info->egress_tun_info)
 399                 size += nla_total_size(ovs_tun_key_attr_size());
 400
 401         return size;
 402 }
 403
 404 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
 405                                   const struct sw_flow_key *key,
 406                                   const struct dp_upcall_info *upcall_info)
 407 {
 408         struct ovs_header *upcall;
 409         struct sk_buff *nskb = NULL;
 410         struct sk_buff *user_skb = NULL; /* to be queued to userspace */
 411         struct nlattr *nla;
 412         struct genl_info info = {
 413 #ifdef HAVE_GENLMSG_NEW_UNICAST
 414                 .dst_sk = ovs_dp_get_net(dp)->genl_sock,
 415 #endif
 416                 .snd_portid = upcall_info->portid,
 417         };
 418         size_t len;
 419         unsigned int hlen;
 420         int err, dp_ifindex;
 421
 422         dp_ifindex = get_dpifindex(dp);
 423         if (!dp_ifindex)
 424                 return -ENODEV;
 425
 426         if (vlan_tx_tag_present(skb)) {
 427                 nskb = skb_clone(skb, GFP_ATOMIC);
 428                 if (!nskb)
 429                         return -ENOMEM;
 430
 431                 nskb = __vlan_put_tag(nskb, nskb->vlan_proto, vlan_tx_tag_get(nskb));
 432                 if (!nskb)
 433                         return -ENOMEM;
 434
 435                 vlan_set_tci(nskb, 0);
 436
 437                 skb = nskb;
 438         }
 439
 440         if (nla_attr_size(skb->len) > USHRT_MAX) {
 441                 err = -EFBIG;
 442                 goto out;
 443         }
 444
 445         /* Complete checksum if needed */
 446         if (skb->ip_summed == CHECKSUM_PARTIAL &&
 447             (err = skb_checksum_help(skb)))
 448                 goto out;
 449
 450         /* Older versions of OVS user space enforce alignment of the last
 451          * Netlink attribute to NLA_ALIGNTO which would require extensive
 452          * padding logic. Only perform zerocopy if padding is not required.
 453          */
 454         if (dp->user_features & OVS_DP_F_UNALIGNED)
 455                 hlen = skb_zerocopy_headlen(skb);
 456         else
 457                 hlen = skb->len;
 458
 459         len = upcall_msg_size(upcall_info, hlen);
 460         user_skb = genlmsg_new_unicast(len, &info, GFP_ATOMIC);
 461         if (!user_skb) {
 462                 err = -ENOMEM;
 463                 goto out;
 464         }
 465
 466         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 467                              0, upcall_info->cmd);
 468         upcall->dp_ifindex = dp_ifindex;
 469
 470         nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
 471         err = ovs_nla_put_flow(key, key, user_skb);
 472         BUG_ON(err);
 473         nla_nest_end(user_skb, nla);
 474
 475         if (upcall_info->userdata)
 476                 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
 477                           nla_len(upcall_info->userdata),
 478                           nla_data(upcall_info->userdata));
 479
 480         if (upcall_info->egress_tun_info) {
 481                 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_EGRESS_TUN_KEY);
 482                 err = ovs_nla_put_egress_tunnel_key(user_skb,
 483                                                     upcall_info->egress_tun_info);
 484                 BUG_ON(err);
 485                 nla_nest_end(user_skb, nla);
 486         }
 487
 488         /* Only reserve room for attribute header, packet data is added
 489          * in skb_zerocopy() */
 490         if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
 491                 err = -ENOBUFS;
 492                 goto out;
 493         }
 494         nla->nla_len = nla_attr_size(skb->len);
 495
 496         err = skb_zerocopy(user_skb, skb, skb->len, hlen);
 497         if (err)
 498                 goto out;
 499
 500         /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
 501         if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
 502                 size_t plen = NLA_ALIGN(user_skb->len) - user_skb->len;
 503
 504                 if (plen > 0)
 505                         memset(skb_put(user_skb, plen), 0, plen);
 506         }
 507
 508         ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
 509
 510         err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
 511         user_skb = NULL;
 512 out:
 513         if (err)
 514                 skb_tx_error(skb);
 515
 516         kfree_skb(user_skb);
 517         kfree_skb(nskb);
 518         return err;
 519 }
 520
 521 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 522 {
 523         struct ovs_header *ovs_header = info->userhdr;
 524         struct nlattr **a = info->attrs;
 525         struct sw_flow_actions *acts;
 526         struct sk_buff *packet;
 527         struct sw_flow *flow;
 528         struct sw_flow_actions *sf_acts;
 529         struct datapath *dp;
 530         struct ethhdr *eth;
 531         struct vport *input_vport;
 532         int len;
 533         int err;
 534         bool log = !a[OVS_FLOW_ATTR_PROBE];
 535
 536         err = -EINVAL;
 537         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 538             !a[OVS_PACKET_ATTR_ACTIONS])
 539                 goto err;
 540
 541         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 542         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 543         err = -ENOMEM;
 544         if (!packet)
 545                 goto err;
 546         skb_reserve(packet, NET_IP_ALIGN);
 547
 548         nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
 549
 550         skb_reset_mac_header(packet);
 551         eth = eth_hdr(packet);
 552
 553         /* Normally, setting the skb 'protocol' field would be handled by a
 554          * call to eth_type_trans(), but it assumes there's a sending
 555          * device, which we may not have. */
 556         if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
 557                 packet->protocol = eth->h_proto;
 558         else
 559                 packet->protocol = htons(ETH_P_802_2);
 560
 561         /* Build an sw_flow for sending this packet. */
 562         flow = ovs_flow_alloc();
 563         err = PTR_ERR(flow);
 564         if (IS_ERR(flow))
 565                 goto err_kfree_skb;
 566
 567         err = ovs_flow_key_extract_userspace(a[OVS_PACKET_ATTR_KEY], packet,
 568                                              &flow->key, log);
 569         if (err)
 570                 goto err_flow_free;
 571
 572         err = ovs_nla_copy_actions(a[OVS_PACKET_ATTR_ACTIONS],
 573                                    &flow->key, &acts, log);
 574         if (err)
 575                 goto err_flow_free;
 576
 577         rcu_assign_pointer(flow->sf_acts, acts);
 578         OVS_CB(packet)->egress_tun_info = NULL;
 579         packet->priority = flow->key.phy.priority;
 580         packet->mark = flow->key.phy.skb_mark;
 581
 582         rcu_read_lock();
 583         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
 584         err = -ENODEV;
 585         if (!dp)
 586                 goto err_unlock;
 587
 588         input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
 589         if (!input_vport)
 590                 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
 591
 592         if (!input_vport)
 593                 goto err_unlock;
 594
 595         OVS_CB(packet)->input_vport = input_vport;
 596         sf_acts = rcu_dereference(flow->sf_acts);
 597
 598         local_bh_disable();
 599         err = ovs_execute_actions(dp, packet, &flow->key, sf_acts);
 600         local_bh_enable();
 601         rcu_read_unlock();
 602
 603         ovs_flow_free(flow, false);
 604         return err;
 605
 606 err_unlock:
 607         rcu_read_unlock();
 608 err_flow_free:
 609         ovs_flow_free(flow, false);
 610 err_kfree_skb:
 611         kfree_skb(packet);
 612 err:
 613         return err;
 614 }
 615
 616 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 617         [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
 618         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 619         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 620 };
 621
 622 static struct genl_ops dp_packet_genl_ops[] = {
 623         { .cmd = OVS_PACKET_CMD_EXECUTE,
 624           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 625           .policy = packet_policy,
 626           .doit = ovs_packet_cmd_execute
 627         }
 628 };
 629
 630 static struct genl_family dp_packet_genl_family = {
 631         .id = GENL_ID_GENERATE,
 632         .hdrsize = sizeof(struct ovs_header),
 633         .name = OVS_PACKET_FAMILY,
 634         .version = OVS_PACKET_VERSION,
 635         .maxattr = OVS_PACKET_ATTR_MAX,
 636         .netnsok = true,
 637         .parallel_ops = true,
 638         .ops = dp_packet_genl_ops,
 639         .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
 640 };
 641
 642 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
 643                          struct ovs_dp_megaflow_stats *mega_stats)
 644 {
 645         int i;
 646
 647         memset(mega_stats, 0, sizeof(*mega_stats));
 648
 649         stats->n_flows = ovs_flow_tbl_count(&dp->table);
 650         mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
 651
 652         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 653
 654         for_each_possible_cpu(i) {
 655                 const struct dp_stats_percpu *percpu_stats;
 656                 struct dp_stats_percpu local_stats;
 657                 unsigned int start;
 658
 659                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 660
 661                 do {
 662                         start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
 663                         local_stats = *percpu_stats;
 664                 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
 665
 666                 stats->n_hit += local_stats.n_hit;
 667                 stats->n_missed += local_stats.n_missed;
 668                 stats->n_lost += local_stats.n_lost;
 669                 mega_stats->n_mask_hit += local_stats.n_mask_hit;
 670         }
 671 }
 672
 673 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
 674 {
 675         return NLMSG_ALIGN(sizeof(struct ovs_header))
 676                 + nla_total_size(ovs_key_attr_size()) /* OVS_FLOW_ATTR_KEY */
 677                 + nla_total_size(ovs_key_attr_size()) /* OVS_FLOW_ATTR_MASK */
 678                 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
 679                 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
 680                 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
 681                 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
 682 }
 683
 684 /* Called with ovs_mutex or RCU read lock. */
 685 static int ovs_flow_cmd_fill_match(const struct sw_flow *flow,
 686                                    struct sk_buff *skb)
 687 {
 688         struct nlattr *nla;
 689         int err;
 690
 691         /* Fill flow key. */
 692         nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
 693         if (!nla)
 694                 return -EMSGSIZE;
 695
 696         err = ovs_nla_put_flow(&flow->unmasked_key,
 697                                &flow->unmasked_key, skb);
 698         if (err)
 699                 return err;
 700         nla_nest_end(skb, nla);
 701
 702         /* Fill flow mask. */
 703         nla = nla_nest_start(skb, OVS_FLOW_ATTR_MASK);
 704         if (!nla)
 705                 return -EMSGSIZE;
 706
 707         err = ovs_nla_put_flow(&flow->key, &flow->mask->key, skb);
 708         if (err)
 709                 return err;
 710         nla_nest_end(skb, nla);
 711
 712         return 0;
 713 }
 714
 715 /* Called with ovs_mutex or RCU read lock. */
 716 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
 717                                    struct sk_buff *skb)
 718 {
 719         struct ovs_flow_stats stats;
 720         __be16 tcp_flags;
 721         unsigned long used;
 722
 723         ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
 724
 725         if (used &&
 726             nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
 727                 return -EMSGSIZE;
 728
 729         if (stats.n_packets &&
 730             nla_put(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats))
 731                 return -EMSGSIZE;
 732
 733         if ((u8)ntohs(tcp_flags) &&
 734              nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
 735                 return -EMSGSIZE;
 736
 737         return 0;
 738 }
 739
 740 /* Called with ovs_mutex or RCU read lock. */
 741 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
 742                                      struct sk_buff *skb, int skb_orig_len)
 743 {
 744         struct nlattr *start;
 745         int err;
 746
 747         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 748          * this is the first flow to be dumped into 'skb'.  This is unusual for
 749          * Netlink but individual action lists can be longer than
 750          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 751          * The userspace caller can always fetch the actions separately if it
 752          * really wants them.  (Most userspace callers in fact don't care.)
 753          *
 754          * This can only fail for dump operations because the skb is always
 755          * properly sized for single flows.
 756          */
 757         start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
 758         if (start) {
 759                 const struct sw_flow_actions *sf_acts;
 760
 761                 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
 762                 err = ovs_nla_put_actions(sf_acts->actions,
 763                                           sf_acts->actions_len, skb);
 764
 765                 if (!err)
 766                         nla_nest_end(skb, start);
 767                 else {
 768                         if (skb_orig_len)
 769                                 return err;
 770
 771                         nla_nest_cancel(skb, start);
 772                 }
 773         } else if (skb_orig_len) {
 774                 return -EMSGSIZE;
 775         }
 776
 777         return 0;
 778 }
 779
 780 /* Called with ovs_mutex or RCU read lock. */
 781 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
 782                                   struct sk_buff *skb, u32 portid,
 783                                   u32 seq, u32 flags, u8 cmd)
 784 {
 785         const int skb_orig_len = skb->len;
 786         struct ovs_header *ovs_header;
 787         int err;
 788
 789         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
 790         if (!ovs_header)
 791                 return -EMSGSIZE;
 792         ovs_header->dp_ifindex = dp_ifindex;
 793
 794         err = ovs_flow_cmd_fill_match(flow, skb);
 795         if (err)
 796                 goto error;
 797
 798         err = ovs_flow_cmd_fill_stats(flow, skb);
 799         if (err)
 800                 goto error;
 801
 802         err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
 803         if (err)
 804                 goto error;
 805
 806         return genlmsg_end(skb, ovs_header);
 807
 808 error:
 809         genlmsg_cancel(skb, ovs_header);
 810         return err;
 811 }
 812
 813 /* May not be called with RCU read lock. */
 814 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
 815                                                struct genl_info *info,
 816                                                bool always)
 817 {
 818         struct sk_buff *skb;
 819
 820         if (!always && !ovs_must_notify(&dp_flow_genl_family, info,
 821                                         GROUP_ID(&ovs_dp_flow_multicast_group)))
 822                 return NULL;
 823
 824         skb = genlmsg_new_unicast(ovs_flow_cmd_msg_size(acts), info, GFP_KERNEL);
 825
 826         if (!skb)
 827                 return ERR_PTR(-ENOMEM);
 828
 829         return skb;
 830 }
 831
 832 /* Called with ovs_mutex. */
 833 static struct sk_buff *ovs_flow_cmd_build_info(struct datapath *dp,
 834                                                const struct sw_flow *flow,
 835                                                int dp_ifindex,
 836                                                struct genl_info *info, u8 cmd,
 837                                                bool always)
 838 {
 839         struct sk_buff *skb;
 840         int retval;
 841
 842         skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts), info,
 843                                       always);
 844         if (IS_ERR_OR_NULL(skb))
 845                 return skb;
 846
 847         retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
 848                                         info->snd_portid, info->snd_seq, 0,
 849                                         cmd);
 850         BUG_ON(retval < 0);
 851         return skb;
 852 }
 853
 854 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
 855 {
 856         struct nlattr **a = info->attrs;
 857         struct ovs_header *ovs_header = info->userhdr;
 858         struct sw_flow *flow, *new_flow;
 859         struct sw_flow_mask mask;
 860         struct sk_buff *reply;
 861         struct datapath *dp;
 862         struct sw_flow_actions *acts;
 863         struct sw_flow_match match;
 864         int error;
 865         bool log = !a[OVS_FLOW_ATTR_PROBE];
 866
 867         /* Must have key and actions. */
 868         error = -EINVAL;
 869         if (!a[OVS_FLOW_ATTR_KEY]) {
 870                 OVS_NLERR(log, "Flow key attribute not present in new flow.");
 871                 goto error;
 872         }
 873         if (!a[OVS_FLOW_ATTR_ACTIONS]) {
 874                 OVS_NLERR(log,
 875                           "Flow actions attribute not present in new flow.");
 876                 goto error;
 877         }
 878
 879         /* Most of the time we need to allocate a new flow, do it before
 880          * locking. */
 881         new_flow = ovs_flow_alloc();
 882         if (IS_ERR(new_flow)) {
 883                 error = PTR_ERR(new_flow);
 884                 goto error;
 885         }
 886
 887         /* Extract key. */
 888         ovs_match_init(&match, &new_flow->unmasked_key, &mask);
 889         error = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY],
 890                                   a[OVS_FLOW_ATTR_MASK], log);
 891         if (error)
 892                 goto err_kfree_flow;
 893
 894         ovs_flow_mask_key(&new_flow->key, &new_flow->unmasked_key, &mask);
 895
 896         /* Validate actions. */
 897         error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS], &new_flow->key,
 898                                      &acts, log);
 899         if (error) {
 900                 OVS_NLERR(
 901                         log,
 902                         "Flow actions may not be safe on all matching packets."
 903                         );
 904                 goto err_kfree_flow;
 905         }
 906
 907         reply = ovs_flow_cmd_alloc_info(acts, info, false);
 908         if (IS_ERR(reply)) {
 909                 error = PTR_ERR(reply);
 910                 goto err_kfree_acts;
 911         }
 912
 913         ovs_lock();
 914         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 915         if (unlikely(!dp)) {
 916                 error = -ENODEV;
 917                 goto err_unlock_ovs;
 918         }
 919         /* Check if this is a duplicate flow */
 920         flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->unmasked_key);
 921         if (likely(!flow)) {
 922                 rcu_assign_pointer(new_flow->sf_acts, acts);
 923
 924                 /* Put flow in bucket. */
 925                 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
 926                 if (unlikely(error)) {
 927                         acts = NULL;
 928                         goto err_unlock_ovs;
 929                 }
 930
 931                 if (unlikely(reply)) {
 932                         error = ovs_flow_cmd_fill_info(new_flow,
 933                                                        ovs_header->dp_ifindex,
 934                                                        reply, info->snd_portid,
 935                                                        info->snd_seq, 0,
 936                                                        OVS_FLOW_CMD_NEW);
 937                         BUG_ON(error < 0);
 938                 }
 939                 ovs_unlock();
 940         } else {
 941                 struct sw_flow_actions *old_acts;
 942
 943                 /* Bail out if we're not allowed to modify an existing flow.
 944                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
 945                  * because Generic Netlink treats the latter as a dump
 946                  * request.  We also accept NLM_F_EXCL in case that bug ever
 947                  * gets fixed.
 948                  */
 949                 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
 950                                                          | NLM_F_EXCL))) {
 951                         error = -EEXIST;
 952                         goto err_unlock_ovs;
 953                 }
 954                 /* The unmasked key has to be the same for flow updates. */
 955                 if (unlikely(!ovs_flow_cmp_unmasked_key(flow, &match))) {
 956                         /* Look for any overlapping flow. */
 957                         flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
 958                         if (!flow) {
 959                                 error = -ENOENT;
 960                                 goto err_unlock_ovs;
 961                         }
 962                 }
 963                 /* Update actions. */
 964                 old_acts = ovsl_dereference(flow->sf_acts);
 965                 rcu_assign_pointer(flow->sf_acts, acts);
 966
 967                 if (unlikely(reply)) {
 968                         error = ovs_flow_cmd_fill_info(flow,
 969                                                        ovs_header->dp_ifindex,
 970                                                        reply, info->snd_portid,
 971                                                        info->snd_seq, 0,
 972                                                        OVS_FLOW_CMD_NEW);
 973                         BUG_ON(error < 0);
 974                 }
 975                 ovs_unlock();
 976
 977                 ovs_nla_free_flow_actions(old_acts);
 978                 ovs_flow_free(new_flow, false);
 979         }
 980
 981         if (reply)
 982                 ovs_notify(&dp_flow_genl_family, &ovs_dp_flow_multicast_group, reply, info);
 983         return 0;
 984
 985 err_unlock_ovs:
 986         ovs_unlock();
 987         kfree_skb(reply);
 988 err_kfree_acts:
 989         kfree(acts);
 990 err_kfree_flow:
 991         ovs_flow_free(new_flow, false);
 992 error:
 993         return error;
 994 }
 995
 996 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
 997 static struct sw_flow_actions *get_flow_actions(const struct nlattr *a,
 998                                                 const struct sw_flow_key *key,
 999                                                 const struct sw_flow_mask *mask,
1000                                                 bool log)
1001 {
1002         struct sw_flow_actions *acts;
1003         struct sw_flow_key masked_key;
1004         int error;
1005
1006         ovs_flow_mask_key(&masked_key, key, mask);
1007         error = ovs_nla_copy_actions(a, &masked_key, &acts, log);
1008         if (error) {
1009                 OVS_NLERR(log,
1010                           "Actions may not be safe on all matching packets.");
1011                 return ERR_PTR(error);
1012         }
1013
1014         return acts;
1015 }
1016
1017 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1018 {
1019         struct nlattr **a = info->attrs;
1020         struct ovs_header *ovs_header = info->userhdr;
1021         struct sw_flow_key key;
1022         struct sw_flow *flow;
1023         struct sw_flow_mask mask;
1024         struct sk_buff *reply = NULL;
1025         struct datapath *dp;
1026         struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1027         struct sw_flow_match match;
1028         int error;
1029         bool log = !a[OVS_FLOW_ATTR_PROBE];
1030
1031         /* Extract key. */
1032         error = -EINVAL;
1033         if (!a[OVS_FLOW_ATTR_KEY]) {
1034                 OVS_NLERR(log, "Flow key attribute not present in set flow.");
1035                 goto error;
1036         }
1037
1038         ovs_match_init(&match, &key, &mask);
1039         error = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY],
1040                                   a[OVS_FLOW_ATTR_MASK], log);
1041         if (error)
1042                 goto error;
1043
1044         /* Validate actions. */
1045         if (a[OVS_FLOW_ATTR_ACTIONS]) {
1046                 acts = get_flow_actions(a[OVS_FLOW_ATTR_ACTIONS], &key, &mask,
1047                                         log);
1048                 if (IS_ERR(acts)) {
1049                         error = PTR_ERR(acts);
1050                         goto error;
1051                 }
1052
1053                 /* Can allocate before locking if have acts. */
1054                 reply = ovs_flow_cmd_alloc_info(acts, info, false);
1055                 if (IS_ERR(reply)) {
1056                         error = PTR_ERR(reply);
1057                         goto err_kfree_acts;
1058                 }
1059         }
1060
1061         ovs_lock();
1062         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1063         if (unlikely(!dp)) {
1064                 error = -ENODEV;
1065                 goto err_unlock_ovs;
1066         }
1067         /* Check that the flow exists. */
1068         flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1069         if (unlikely(!flow)) {
1070                 error = -ENOENT;
1071                 goto err_unlock_ovs;
1072         }
1073
1074         /* Update actions, if present. */
1075         if (likely(acts)) {
1076                 old_acts = ovsl_dereference(flow->sf_acts);
1077                 rcu_assign_pointer(flow->sf_acts, acts);
1078
1079                 if (unlikely(reply)) {
1080                         error = ovs_flow_cmd_fill_info(flow,
1081                                                        ovs_header->dp_ifindex,
1082                                                        reply, info->snd_portid,
1083                                                        info->snd_seq, 0,
1084                                                        OVS_FLOW_CMD_NEW);
1085                         BUG_ON(error < 0);
1086                 }
1087         } else {
1088                 /* Could not alloc without acts before locking. */
1089                 reply = ovs_flow_cmd_build_info(dp, flow,
1090                                                 ovs_header->dp_ifindex,
1091                                                 info, OVS_FLOW_CMD_NEW, false);
1092                 if (unlikely(IS_ERR(reply))) {
1093                         error = PTR_ERR(reply);
1094                         goto err_unlock_ovs;
1095                 }
1096         }
1097
1098         /* Clear stats. */
1099         if (a[OVS_FLOW_ATTR_CLEAR])
1100                 ovs_flow_stats_clear(flow);
1101         ovs_unlock();
1102
1103         if (reply)
1104                 ovs_notify(&dp_flow_genl_family, &ovs_dp_flow_multicast_group, reply, info);
1105         if (old_acts)
1106                 ovs_nla_free_flow_actions(old_acts);
1107         return 0;
1108
1109 err_unlock_ovs:
1110         ovs_unlock();
1111         kfree_skb(reply);
1112 err_kfree_acts:
1113         kfree(acts);
1114 error:
1115         return error;
1116 }
1117
1118 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1119 {
1120         struct nlattr **a = info->attrs;
1121         struct ovs_header *ovs_header = info->userhdr;
1122         struct sw_flow_key key;
1123         struct sk_buff *reply;
1124         struct sw_flow *flow;
1125         struct datapath *dp;
1126         struct sw_flow_match match;
1127         int err;
1128         bool log = !a[OVS_FLOW_ATTR_PROBE];
1129
1130         if (!a[OVS_FLOW_ATTR_KEY]) {
1131                 OVS_NLERR(log,
1132                           "Flow get message rejected, Key attribute missing.");
1133                 return -EINVAL;
1134         }
1135
1136         ovs_match_init(&match, &key, NULL);
1137         err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL, log);
1138         if (err)
1139                 return err;
1140
1141         ovs_lock();
1142         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1143         if (!dp) {
1144                 err = -ENODEV;
1145                 goto unlock;
1146         }
1147
1148         flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1149         if (!flow) {
1150                 err = -ENOENT;
1151                 goto unlock;
1152         }
1153
1154         reply = ovs_flow_cmd_build_info(dp, flow, ovs_header->dp_ifindex, info,
1155                                         OVS_FLOW_CMD_NEW, true);
1156         if (IS_ERR(reply)) {
1157                 err = PTR_ERR(reply);
1158                 goto unlock;
1159         }
1160
1161         ovs_unlock();
1162         return genlmsg_reply(reply, info);
1163 unlock:
1164         ovs_unlock();
1165         return err;
1166 }
1167
1168 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1169 {
1170         struct nlattr **a = info->attrs;
1171         struct ovs_header *ovs_header = info->userhdr;
1172         struct sw_flow_key key;
1173         struct sk_buff *reply;
1174         struct sw_flow *flow;
1175         struct datapath *dp;
1176         struct sw_flow_match match;
1177         int err;
1178         bool log = !a[OVS_FLOW_ATTR_PROBE];
1179
1180         if (likely(a[OVS_FLOW_ATTR_KEY])) {
1181                 ovs_match_init(&match, &key, NULL);
1182                 err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL,
1183                                         log);
1184                 if (unlikely(err))
1185                         return err;
1186         }
1187
1188         ovs_lock();
1189         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1190         if (unlikely(!dp)) {
1191                 err = -ENODEV;
1192                 goto unlock;
1193         }
1194         if (unlikely(!a[OVS_FLOW_ATTR_KEY])) {
1195                 err = ovs_flow_tbl_flush(&dp->table);
1196                 goto unlock;
1197         }
1198         flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1199         if (unlikely(!flow)) {
1200                 err = -ENOENT;
1201                 goto unlock;
1202         }
1203
1204         ovs_flow_tbl_remove(&dp->table, flow);
1205         ovs_unlock();
1206
1207         reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *)flow->sf_acts,
1208                                         info, false);
1209
1210         if (likely(reply)) {
1211                 if (likely(!IS_ERR(reply))) {
1212                         rcu_read_lock(); /* Keep RCU checker happy. */
1213                         err = ovs_flow_cmd_fill_info(flow,
1214                                                      ovs_header->dp_ifindex,
1215                                                      reply, info->snd_portid,
1216                                                      info->snd_seq, 0,
1217                                                      OVS_FLOW_CMD_DEL);
1218                         rcu_read_unlock();
1219                         BUG_ON(err < 0);
1220                         ovs_notify(&dp_flow_genl_family, &ovs_dp_flow_multicast_group, reply, info);
1221                 } else {
1222                         genl_set_err(&dp_flow_genl_family, sock_net(skb->sk), 0,
1223                                      GROUP_ID(&ovs_dp_flow_multicast_group), PTR_ERR(reply));
1224
1225                 }
1226         }
1227
1228         ovs_flow_free(flow, true);
1229         return 0;
1230 unlock:
1231         ovs_unlock();
1232         return err;
1233 }
1234
1235 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1236 {
1237         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1238         struct table_instance *ti;
1239         struct datapath *dp;
1240
1241         rcu_read_lock();
1242         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1243         if (!dp) {
1244                 rcu_read_unlock();
1245                 return -ENODEV;
1246         }
1247
1248         ti = rcu_dereference(dp->table.ti);
1249         for (;;) {
1250                 struct sw_flow *flow;
1251                 u32 bucket, obj;
1252
1253                 bucket = cb->args[0];
1254                 obj = cb->args[1];
1255                 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1256                 if (!flow)
1257                         break;
1258
1259                 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1260                                            NETLINK_CB(cb->skb).portid,
1261                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1262                                            OVS_FLOW_CMD_NEW) < 0)
1263                         break;
1264
1265                 cb->args[0] = bucket;
1266                 cb->args[1] = obj;
1267         }
1268         rcu_read_unlock();
1269         return skb->len;
1270 }
1271
1272 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1273         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1274         [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1275         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1276         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1277         [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1278 };
1279
1280 static struct genl_ops dp_flow_genl_ops[] = {
1281         { .cmd = OVS_FLOW_CMD_NEW,
1282           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1283           .policy = flow_policy,
1284           .doit = ovs_flow_cmd_new
1285         },
1286         { .cmd = OVS_FLOW_CMD_DEL,
1287           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1288           .policy = flow_policy,
1289           .doit = ovs_flow_cmd_del
1290         },
1291         { .cmd = OVS_FLOW_CMD_GET,
1292           .flags = 0,               /* OK for unprivileged users. */
1293           .policy = flow_policy,
1294           .doit = ovs_flow_cmd_get,
1295           .dumpit = ovs_flow_cmd_dump
1296         },
1297         { .cmd = OVS_FLOW_CMD_SET,
1298           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1299           .policy = flow_policy,
1300           .doit = ovs_flow_cmd_set,
1301         },
1302 };
1303
1304 static struct genl_family dp_flow_genl_family = {
1305         .id = GENL_ID_GENERATE,
1306         .hdrsize = sizeof(struct ovs_header),
1307         .name = OVS_FLOW_FAMILY,
1308         .version = OVS_FLOW_VERSION,
1309         .maxattr = OVS_FLOW_ATTR_MAX,
1310         .netnsok = true,
1311         .parallel_ops = true,
1312         .ops = dp_flow_genl_ops,
1313         .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1314         .mcgrps = &ovs_dp_flow_multicast_group,
1315         .n_mcgrps = 1,
1316 };
1317
1318 static size_t ovs_dp_cmd_msg_size(void)
1319 {
1320         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1321
1322         msgsize += nla_total_size(IFNAMSIZ);
1323         msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1324         msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1325         msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1326
1327         return msgsize;
1328 }
1329
1330 /* Called with ovs_mutex or RCU read lock. */
1331 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1332                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1333 {
1334         struct ovs_header *ovs_header;
1335         struct ovs_dp_stats dp_stats;
1336         struct ovs_dp_megaflow_stats dp_megaflow_stats;
1337         int err;
1338
1339         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1340                                    flags, cmd);
1341         if (!ovs_header)
1342                 goto error;
1343
1344         ovs_header->dp_ifindex = get_dpifindex(dp);
1345
1346         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1347         if (err)
1348                 goto nla_put_failure;
1349
1350         get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1351         if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1352                         &dp_stats))
1353                 goto nla_put_failure;
1354
1355         if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1356                         sizeof(struct ovs_dp_megaflow_stats),
1357                         &dp_megaflow_stats))
1358                 goto nla_put_failure;
1359
1360         if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1361                 goto nla_put_failure;
1362
1363         return genlmsg_end(skb, ovs_header);
1364
1365 nla_put_failure:
1366         genlmsg_cancel(skb, ovs_header);
1367 error:
1368         return -EMSGSIZE;
1369 }
1370
1371 static struct sk_buff *ovs_dp_cmd_alloc_info(struct genl_info *info)
1372 {
1373         return genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info, GFP_KERNEL);
1374 }
1375
1376 /* Called with rcu_read_lock or ovs_mutex. */
1377 static struct datapath *lookup_datapath(struct net *net,
1378                                         const struct ovs_header *ovs_header,
1379                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1380 {
1381         struct datapath *dp;
1382
1383         if (!a[OVS_DP_ATTR_NAME])
1384                 dp = get_dp(net, ovs_header->dp_ifindex);
1385         else {
1386                 struct vport *vport;
1387
1388                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1389                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1390         }
1391         return dp ? dp : ERR_PTR(-ENODEV);
1392 }
1393
1394 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1395 {
1396         struct datapath *dp;
1397
1398         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1399         if (IS_ERR(dp))
1400                 return;
1401
1402         WARN(dp->user_features, "Dropping previously announced user features\n");
1403         dp->user_features = 0;
1404 }
1405
1406 static void ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1407 {
1408         if (a[OVS_DP_ATTR_USER_FEATURES])
1409                 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1410 }
1411
1412 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1413 {
1414         struct nlattr **a = info->attrs;
1415         struct vport_parms parms;
1416         struct sk_buff *reply;
1417         struct datapath *dp;
1418         struct vport *vport;
1419         struct ovs_net *ovs_net;
1420         int err, i;
1421
1422         err = -EINVAL;
1423         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1424                 goto err;
1425
1426         reply = ovs_dp_cmd_alloc_info(info);
1427         if (!reply)
1428                 return -ENOMEM;
1429
1430         err = -ENOMEM;
1431         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1432         if (dp == NULL)
1433                 goto err_free_reply;
1434
1435         ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1436
1437         /* Allocate table. */
1438         err = ovs_flow_tbl_init(&dp->table);
1439         if (err)
1440                 goto err_free_dp;
1441
1442         dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1443         if (!dp->stats_percpu) {
1444                 err = -ENOMEM;
1445                 goto err_destroy_table;
1446         }
1447
1448         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1449                             GFP_KERNEL);
1450         if (!dp->ports) {
1451                 err = -ENOMEM;
1452                 goto err_destroy_percpu;
1453         }
1454
1455         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1456                 INIT_HLIST_HEAD(&dp->ports[i]);
1457
1458         /* Set up our datapath device. */
1459         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1460         parms.type = OVS_VPORT_TYPE_INTERNAL;
1461         parms.options = NULL;
1462         parms.dp = dp;
1463         parms.port_no = OVSP_LOCAL;
1464         parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1465
1466         ovs_dp_change(dp, a);
1467
1468         /* So far only local changes have been made, now need the lock. */
1469         ovs_lock();
1470
1471         vport = new_vport(&parms);
1472         if (IS_ERR(vport)) {
1473                 err = PTR_ERR(vport);
1474                 if (err == -EBUSY)
1475                         err = -EEXIST;
1476
1477                 if (err == -EEXIST) {
1478                         /* An outdated user space instance that does not understand
1479                          * the concept of user_features has attempted to create a new
1480                          * datapath and is likely to reuse it. Drop all user features.
1481                          */
1482                         if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1483                                 ovs_dp_reset_user_features(skb, info);
1484                 }
1485
1486                 goto err_destroy_ports_array;
1487         }
1488
1489         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1490                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1491         BUG_ON(err < 0);
1492
1493         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1494         list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1495
1496         ovs_unlock();
1497
1498         ovs_notify(&dp_datapath_genl_family, &ovs_dp_datapath_multicast_group, reply, info);
1499         return 0;
1500
1501 err_destroy_ports_array:
1502         ovs_unlock();
1503         kfree(dp->ports);
1504 err_destroy_percpu:
1505         free_percpu(dp->stats_percpu);
1506 err_destroy_table:
1507         ovs_flow_tbl_destroy(&dp->table);
1508 err_free_dp:
1509         release_net(ovs_dp_get_net(dp));
1510         kfree(dp);
1511 err_free_reply:
1512         kfree_skb(reply);
1513 err:
1514         return err;
1515 }
1516
1517 /* Called with ovs_mutex. */
1518 static void __dp_destroy(struct datapath *dp)
1519 {
1520         int i;
1521
1522         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1523                 struct vport *vport;
1524                 struct hlist_node *n;
1525
1526                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1527                         if (vport->port_no != OVSP_LOCAL)
1528                                 ovs_dp_detach_port(vport);
1529         }
1530
1531         list_del_rcu(&dp->list_node);
1532
1533         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1534          * all ports in datapath are destroyed first before freeing datapath.
1535          */
1536         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1537
1538         /* RCU destroy the flow table */
1539         call_rcu(&dp->rcu, destroy_dp_rcu);
1540 }
1541
1542 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1543 {
1544         struct sk_buff *reply;
1545         struct datapath *dp;
1546         int err;
1547
1548         reply = ovs_dp_cmd_alloc_info(info);
1549         if (!reply)
1550                 return -ENOMEM;
1551
1552         ovs_lock();
1553         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1554         err = PTR_ERR(dp);
1555         if (IS_ERR(dp))
1556                 goto err_unlock_free;
1557
1558         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1559                                    info->snd_seq, 0, OVS_DP_CMD_DEL);
1560         BUG_ON(err < 0);
1561
1562         __dp_destroy(dp);
1563
1564         ovs_unlock();
1565         ovs_notify(&dp_datapath_genl_family, &ovs_dp_datapath_multicast_group, reply, info);
1566         return 0;
1567
1568 err_unlock_free:
1569         ovs_unlock();
1570         kfree_skb(reply);
1571         return err;
1572 }
1573
1574 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1575 {
1576         struct sk_buff *reply;
1577         struct datapath *dp;
1578         int err;
1579
1580         reply = ovs_dp_cmd_alloc_info(info);
1581         if (!reply)
1582                 return -ENOMEM;
1583
1584         ovs_lock();
1585         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1586         err = PTR_ERR(dp);
1587         if (IS_ERR(dp))
1588                 goto err_unlock_free;
1589
1590         ovs_dp_change(dp, info->attrs);
1591
1592         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1593                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1594         BUG_ON(err < 0);
1595
1596         ovs_unlock();
1597         ovs_notify(&dp_datapath_genl_family, &ovs_dp_datapath_multicast_group, reply, info);
1598         return 0;
1599
1600 err_unlock_free:
1601         ovs_unlock();
1602         kfree_skb(reply);
1603         return err;
1604 }
1605
1606 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1607 {
1608         struct sk_buff *reply;
1609         struct datapath *dp;
1610         int err;
1611
1612         reply = ovs_dp_cmd_alloc_info(info);
1613         if (!reply)
1614                 return -ENOMEM;
1615
1616         rcu_read_lock();
1617         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1618         if (IS_ERR(dp)) {
1619                 err = PTR_ERR(dp);
1620                 goto err_unlock_free;
1621         }
1622         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1623                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1624         BUG_ON(err < 0);
1625         rcu_read_unlock();
1626
1627         return genlmsg_reply(reply, info);
1628
1629 err_unlock_free:
1630         rcu_read_unlock();
1631         kfree_skb(reply);
1632         return err;
1633 }
1634
1635 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1636 {
1637         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1638         struct datapath *dp;
1639         int skip = cb->args[0];
1640         int i = 0;
1641
1642         rcu_read_lock();
1643         list_for_each_entry_rcu(dp, &ovs_net->dps, list_node) {
1644                 if (i >= skip &&
1645                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1646                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1647                                          OVS_DP_CMD_NEW) < 0)
1648                         break;
1649                 i++;
1650         }
1651         rcu_read_unlock();
1652
1653         cb->args[0] = i;
1654
1655         return skb->len;
1656 }
1657
1658 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1659         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1660         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1661         [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1662 };
1663
1664 static struct genl_ops dp_datapath_genl_ops[] = {
1665         { .cmd = OVS_DP_CMD_NEW,
1666           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1667           .policy = datapath_policy,
1668           .doit = ovs_dp_cmd_new
1669         },
1670         { .cmd = OVS_DP_CMD_DEL,
1671           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1672           .policy = datapath_policy,
1673           .doit = ovs_dp_cmd_del
1674         },
1675         { .cmd = OVS_DP_CMD_GET,
1676           .flags = 0,               /* OK for unprivileged users. */
1677           .policy = datapath_policy,
1678           .doit = ovs_dp_cmd_get,
1679           .dumpit = ovs_dp_cmd_dump
1680         },
1681         { .cmd = OVS_DP_CMD_SET,
1682           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1683           .policy = datapath_policy,
1684           .doit = ovs_dp_cmd_set,
1685         },
1686 };
1687
1688 static struct genl_family dp_datapath_genl_family = {
1689         .id = GENL_ID_GENERATE,
1690         .hdrsize = sizeof(struct ovs_header),
1691         .name = OVS_DATAPATH_FAMILY,
1692         .version = OVS_DATAPATH_VERSION,
1693         .maxattr = OVS_DP_ATTR_MAX,
1694         .netnsok = true,
1695         .parallel_ops = true,
1696         .ops = dp_datapath_genl_ops,
1697         .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1698         .mcgrps = &ovs_dp_datapath_multicast_group,
1699         .n_mcgrps = 1,
1700 };
1701
1702 /* Called with ovs_mutex or RCU read lock. */
1703 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1704                                    u32 portid, u32 seq, u32 flags, u8 cmd)
1705 {
1706         struct ovs_header *ovs_header;
1707         struct ovs_vport_stats vport_stats;
1708         int err;
1709
1710         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1711                                  flags, cmd);
1712         if (!ovs_header)
1713                 return -EMSGSIZE;
1714
1715         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1716
1717         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1718             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1719             nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)))
1720                 goto nla_put_failure;
1721
1722         ovs_vport_get_stats(vport, &vport_stats);
1723         if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1724                     &vport_stats))
1725                 goto nla_put_failure;
1726
1727         if (ovs_vport_get_upcall_portids(vport, skb))
1728                 goto nla_put_failure;
1729
1730         err = ovs_vport_get_options(vport, skb);
1731         if (err == -EMSGSIZE)
1732                 goto error;
1733
1734         return genlmsg_end(skb, ovs_header);
1735
1736 nla_put_failure:
1737         err = -EMSGSIZE;
1738 error:
1739         genlmsg_cancel(skb, ovs_header);
1740         return err;
1741 }
1742
1743 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1744 {
1745         return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1746 }
1747
1748 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1749 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1750                                          u32 seq, u8 cmd)
1751 {
1752         struct sk_buff *skb;
1753         int retval;
1754
1755         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1756         if (!skb)
1757                 return ERR_PTR(-ENOMEM);
1758
1759         retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1760         BUG_ON(retval < 0);
1761
1762         return skb;
1763 }
1764
1765 /* Called with ovs_mutex or RCU read lock. */
1766 static struct vport *lookup_vport(struct net *net,
1767                                   const struct ovs_header *ovs_header,
1768                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1769 {
1770         struct datapath *dp;
1771         struct vport *vport;
1772
1773         if (a[OVS_VPORT_ATTR_NAME]) {
1774                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1775                 if (!vport)
1776                         return ERR_PTR(-ENODEV);
1777                 if (ovs_header->dp_ifindex &&
1778                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1779                         return ERR_PTR(-ENODEV);
1780                 return vport;
1781         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1782                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1783
1784                 if (port_no >= DP_MAX_PORTS)
1785                         return ERR_PTR(-EFBIG);
1786
1787                 dp = get_dp(net, ovs_header->dp_ifindex);
1788                 if (!dp)
1789                         return ERR_PTR(-ENODEV);
1790
1791                 vport = ovs_vport_ovsl_rcu(dp, port_no);
1792                 if (!vport)
1793                         return ERR_PTR(-ENODEV);
1794                 return vport;
1795         } else
1796                 return ERR_PTR(-EINVAL);
1797 }
1798
1799 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1800 {
1801         struct nlattr **a = info->attrs;
1802         struct ovs_header *ovs_header = info->userhdr;
1803         struct vport_parms parms;
1804         struct sk_buff *reply;
1805         struct vport *vport;
1806         struct datapath *dp;
1807         u32 port_no;
1808         int err;
1809
1810         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1811             !a[OVS_VPORT_ATTR_UPCALL_PID])
1812                 return -EINVAL;
1813
1814         port_no = a[OVS_VPORT_ATTR_PORT_NO]
1815                 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
1816         if (port_no >= DP_MAX_PORTS)
1817                 return -EFBIG;
1818
1819         reply = ovs_vport_cmd_alloc_info();
1820         if (!reply)
1821                 return -ENOMEM;
1822
1823         ovs_lock();
1824         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1825         err = -ENODEV;
1826         if (!dp)
1827                 goto exit_unlock_free;
1828
1829         if (port_no) {
1830                 vport = ovs_vport_ovsl(dp, port_no);
1831                 err = -EBUSY;
1832                 if (vport)
1833                         goto exit_unlock_free;
1834         } else {
1835                 for (port_no = 1; ; port_no++) {
1836                         if (port_no >= DP_MAX_PORTS) {
1837                                 err = -EFBIG;
1838                                 goto exit_unlock_free;
1839                         }
1840                         vport = ovs_vport_ovsl(dp, port_no);
1841                         if (!vport)
1842                                 break;
1843                 }
1844         }
1845
1846         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1847         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1848         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1849         parms.dp = dp;
1850         parms.port_no = port_no;
1851         parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
1852
1853         vport = new_vport(&parms);
1854         err = PTR_ERR(vport);
1855         if (IS_ERR(vport))
1856                 goto exit_unlock_free;
1857
1858         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1859                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1860         BUG_ON(err < 0);
1861         ovs_unlock();
1862
1863         ovs_notify(&dp_vport_genl_family, &ovs_dp_vport_multicast_group, reply, info);
1864         return 0;
1865
1866 exit_unlock_free:
1867         ovs_unlock();
1868         kfree_skb(reply);
1869         return err;
1870 }
1871
1872 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1873 {
1874         struct nlattr **a = info->attrs;
1875         struct sk_buff *reply;
1876         struct vport *vport;
1877         int err;
1878
1879         reply = ovs_vport_cmd_alloc_info();
1880         if (!reply)
1881                 return -ENOMEM;
1882
1883         ovs_lock();
1884         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1885         err = PTR_ERR(vport);
1886         if (IS_ERR(vport))
1887                 goto exit_unlock_free;
1888
1889         if (a[OVS_VPORT_ATTR_TYPE] &&
1890             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
1891                 err = -EINVAL;
1892                 goto exit_unlock_free;
1893         }
1894
1895         if (a[OVS_VPORT_ATTR_OPTIONS]) {
1896                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1897                 if (err)
1898                         goto exit_unlock_free;
1899         }
1900
1901         if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
1902                 err = ovs_vport_set_upcall_portids(vport,
1903                                                    a[OVS_VPORT_ATTR_UPCALL_PID]);
1904                 if (err)
1905                         goto exit_unlock_free;
1906         }
1907
1908         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1909                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1910         BUG_ON(err < 0);
1911         ovs_unlock();
1912
1913         ovs_notify(&dp_vport_genl_family, &ovs_dp_vport_multicast_group, reply, info);
1914         return 0;
1915
1916 exit_unlock_free:
1917         ovs_unlock();
1918         kfree_skb(reply);
1919         return err;
1920 }
1921
1922 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1923 {
1924         struct nlattr **a = info->attrs;
1925         struct sk_buff *reply;
1926         struct vport *vport;
1927         int err;
1928
1929         reply = ovs_vport_cmd_alloc_info();
1930         if (!reply)
1931                 return -ENOMEM;
1932
1933         ovs_lock();
1934         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1935         err = PTR_ERR(vport);
1936         if (IS_ERR(vport))
1937                 goto exit_unlock_free;
1938
1939         if (vport->port_no == OVSP_LOCAL) {
1940                 err = -EINVAL;
1941                 goto exit_unlock_free;
1942         }
1943
1944         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1945                                       info->snd_seq, 0, OVS_VPORT_CMD_DEL);
1946         BUG_ON(err < 0);
1947         ovs_dp_detach_port(vport);
1948         ovs_unlock();
1949
1950         ovs_notify(&dp_vport_genl_family, &ovs_dp_vport_multicast_group, reply, info);
1951         return 0;
1952
1953 exit_unlock_free:
1954         ovs_unlock();
1955         kfree_skb(reply);
1956         return err;
1957 }
1958
1959 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1960 {
1961         struct nlattr **a = info->attrs;
1962         struct ovs_header *ovs_header = info->userhdr;
1963         struct sk_buff *reply;
1964         struct vport *vport;
1965         int err;
1966
1967         reply = ovs_vport_cmd_alloc_info();
1968         if (!reply)
1969                 return -ENOMEM;
1970
1971         rcu_read_lock();
1972         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1973         err = PTR_ERR(vport);
1974         if (IS_ERR(vport))
1975                 goto exit_unlock_free;
1976         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1977                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1978         BUG_ON(err < 0);
1979         rcu_read_unlock();
1980
1981         return genlmsg_reply(reply, info);
1982
1983 exit_unlock_free:
1984         rcu_read_unlock();
1985         kfree_skb(reply);
1986         return err;
1987 }
1988
1989 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1990 {
1991         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1992         struct datapath *dp;
1993         int bucket = cb->args[0], skip = cb->args[1];
1994         int i, j = 0;
1995
1996         rcu_read_lock();
1997         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1998         if (!dp) {
1999                 rcu_read_unlock();
2000                 return -ENODEV;
2001         }
2002         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2003                 struct vport *vport;
2004
2005                 j = 0;
2006                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2007                         if (j >= skip &&
2008                             ovs_vport_cmd_fill_info(vport, skb,
2009                                                     NETLINK_CB(cb->skb).portid,
2010                                                     cb->nlh->nlmsg_seq,
2011                                                     NLM_F_MULTI,
2012                                                     OVS_VPORT_CMD_NEW) < 0)
2013                                 goto out;
2014
2015                         j++;
2016                 }
2017                 skip = 0;
2018         }
2019 out:
2020         rcu_read_unlock();
2021
2022         cb->args[0] = i;
2023         cb->args[1] = j;
2024
2025         return skb->len;
2026 }
2027
2028 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2029         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2030         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2031         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2032         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2033         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
2034         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2035 };
2036
2037 static struct genl_ops dp_vport_genl_ops[] = {
2038         { .cmd = OVS_VPORT_CMD_NEW,
2039           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2040           .policy = vport_policy,
2041           .doit = ovs_vport_cmd_new
2042         },
2043         { .cmd = OVS_VPORT_CMD_DEL,
2044           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2045           .policy = vport_policy,
2046           .doit = ovs_vport_cmd_del
2047         },
2048         { .cmd = OVS_VPORT_CMD_GET,
2049           .flags = 0,               /* OK for unprivileged users. */
2050           .policy = vport_policy,
2051           .doit = ovs_vport_cmd_get,
2052           .dumpit = ovs_vport_cmd_dump
2053         },
2054         { .cmd = OVS_VPORT_CMD_SET,
2055           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2056           .policy = vport_policy,
2057           .doit = ovs_vport_cmd_set,
2058         },
2059 };
2060
2061 struct genl_family dp_vport_genl_family = {
2062         .id = GENL_ID_GENERATE,
2063         .hdrsize = sizeof(struct ovs_header),
2064         .name = OVS_VPORT_FAMILY,
2065         .version = OVS_VPORT_VERSION,
2066         .maxattr = OVS_VPORT_ATTR_MAX,
2067         .netnsok = true,
2068         .parallel_ops = true,
2069         .ops = dp_vport_genl_ops,
2070         .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2071         .mcgrps = &ovs_dp_vport_multicast_group,
2072         .n_mcgrps = 1,
2073 };
2074
2075 static struct genl_family *dp_genl_families[] = {
2076         &dp_datapath_genl_family,
2077         &dp_vport_genl_family,
2078         &dp_flow_genl_family,
2079         &dp_packet_genl_family,
2080 };
2081
2082 static void dp_unregister_genl(int n_families)
2083 {
2084         int i;
2085
2086         for (i = 0; i < n_families; i++)
2087                 genl_unregister_family(dp_genl_families[i]);
2088 }
2089
2090 static int dp_register_genl(void)
2091 {
2092         int err;
2093         int i;
2094
2095         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2096
2097                 err = genl_register_family(dp_genl_families[i]);
2098                 if (err)
2099                         goto error;
2100         }
2101
2102         return 0;
2103
2104 error:
2105         dp_unregister_genl(i);
2106         return err;
2107 }
2108
2109 static int __net_init ovs_init_net(struct net *net)
2110 {
2111         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2112
2113         INIT_LIST_HEAD(&ovs_net->dps);
2114         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2115         return 0;
2116 }
2117
2118 static void __net_exit ovs_exit_net(struct net *net)
2119 {
2120         struct datapath *dp, *dp_next;
2121         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2122
2123         ovs_lock();
2124         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2125                 __dp_destroy(dp);
2126         ovs_unlock();
2127
2128         cancel_work_sync(&ovs_net->dp_notify_work);
2129 }
2130
2131 static struct pernet_operations ovs_net_ops = {
2132         .init = ovs_init_net,
2133         .exit = ovs_exit_net,
2134         .id   = &ovs_net_id,
2135         .size = sizeof(struct ovs_net),
2136 };
2137
2138 DEFINE_COMPAT_PNET_REG_FUNC(device);
2139
2140 static int __init dp_init(void)
2141 {
2142         int err;
2143
2144         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2145
2146         pr_info("Open vSwitch switching datapath %s, built "__DATE__" "__TIME__"\n",
2147                 VERSION);
2148
2149         err = action_fifos_init();
2150         if (err)
2151                 goto error;
2152
2153         err = ovs_flow_init();
2154         if (err)
2155                 goto error_action_fifos_exit;
2156
2157         err = ovs_vport_init();
2158         if (err)
2159                 goto error_flow_exit;
2160
2161         err = register_pernet_device(&ovs_net_ops);
2162         if (err)
2163                 goto error_vport_exit;
2164
2165         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2166         if (err)
2167                 goto error_netns_exit;
2168
2169         err = dp_register_genl();
2170         if (err < 0)
2171                 goto error_unreg_notifier;
2172
2173         return 0;
2174
2175 error_unreg_notifier:
2176         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2177 error_netns_exit:
2178         unregister_pernet_device(&ovs_net_ops);
2179 error_vport_exit:
2180         ovs_vport_exit();
2181 error_flow_exit:
2182         ovs_flow_exit();
2183 error_action_fifos_exit:
2184         action_fifos_exit();
2185 error:
2186         return err;
2187 }
2188
2189 static void dp_cleanup(void)
2190 {
2191         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2192         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2193         unregister_pernet_device(&ovs_net_ops);
2194         rcu_barrier();
2195         ovs_vport_exit();
2196         ovs_flow_exit();
2197         action_fifos_exit();
2198 }
2199
2200 module_init(dp_init);
2201 module_exit(dp_cleanup);
2202
2203 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2204 MODULE_LICENSE("GPL");
2205 MODULE_VERSION(VERSION);