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                 BUG_ON(err);
 496                 nla_nest_end(user_skb, nla);
 497         }
 498
 499         if (upcall_info->actions_len) {
 500                 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_ACTIONS);
 501                 err = ovs_nla_put_actions(upcall_info->actions,
 502                                           upcall_info->actions_len,
 503                                           user_skb);
 504                 if (!err)
 505                         nla_nest_end(user_skb, nla);
 506                 else
 507                         nla_nest_cancel(user_skb, nla);
 508         }
 509
 510         /* Add OVS_PACKET_ATTR_MRU */
 511         if (upcall_info->mru) {
 512                 if (nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU,
 513                                 upcall_info->mru)) {
 514                         err = -ENOBUFS;
 515                         goto out;
 516                 }
 517                 pad_packet(dp, user_skb);
 518         }
 519
 520         /* Only reserve room for attribute header, packet data is added
 521          * in skb_zerocopy() */
 522         if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
 523                 err = -ENOBUFS;
 524                 goto out;
 525         }
 526         nla->nla_len = nla_attr_size(skb->len);
 527
 528         err = skb_zerocopy(user_skb, skb, skb->len, hlen);
 529         if (err)
 530                 goto out;
 531
 532         /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
 533         pad_packet(dp, user_skb);
 534
 535         ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
 536
 537         err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
 538         user_skb = NULL;
 539 out:
 540         if (err)
 541                 skb_tx_error(skb);
 542         kfree_skb(user_skb);
 543         kfree_skb(nskb);
 544         return err;
 545 }
 546
 547 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 548 {
 549         struct ovs_header *ovs_header = info->userhdr;
 550         struct net *net = sock_net(skb->sk);
 551         struct nlattr **a = info->attrs;
 552         struct sw_flow_actions *acts;
 553         struct sk_buff *packet;
 554         struct sw_flow *flow;
 555         struct sw_flow_actions *sf_acts;
 556         struct datapath *dp;
 557         struct ethhdr *eth;
 558         struct vport *input_vport;
 559         u16 mru = 0;
 560         int len;
 561         int err;
 562         bool log = !a[OVS_PACKET_ATTR_PROBE];
 563
 564         err = -EINVAL;
 565         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 566             !a[OVS_PACKET_ATTR_ACTIONS])
 567                 goto err;
 568
 569         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 570         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 571         err = -ENOMEM;
 572         if (!packet)
 573                 goto err;
 574         skb_reserve(packet, NET_IP_ALIGN);
 575
 576         nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
 577
 578         skb_reset_mac_header(packet);
 579         eth = eth_hdr(packet);
 580
 581         /* Normally, setting the skb 'protocol' field would be handled by a
 582          * call to eth_type_trans(), but it assumes there's a sending
 583          * device, which we may not have. */
 584         if (eth_proto_is_802_3(eth->h_proto))
 585                 packet->protocol = eth->h_proto;
 586         else
 587                 packet->protocol = htons(ETH_P_802_2);
 588
 589         /* Set packet's mru */
 590         if (a[OVS_PACKET_ATTR_MRU]) {
 591                 mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
 592                 packet->ignore_df = 1;
 593         }
 594         OVS_CB(packet)->mru = mru;
 595
 596         /* Build an sw_flow for sending this packet. */
 597         flow = ovs_flow_alloc();
 598         err = PTR_ERR(flow);
 599         if (IS_ERR(flow))
 600                 goto err_kfree_skb;
 601
 602         err = ovs_flow_key_extract_userspace(a[OVS_PACKET_ATTR_KEY], packet,
 603                                              &flow->key, log);
 604         if (err)
 605                 goto err_flow_free;
 606
 607         err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
 608                                    &flow->key, &acts, log);
 609         if (err)
 610                 goto err_flow_free;
 611
 612         rcu_assign_pointer(flow->sf_acts, acts);
 613         OVS_CB(packet)->egress_tun_info = NULL;
 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(&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, &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, 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(&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 sw_flow_key key;
1212         struct sk_buff *reply;
1213         struct sw_flow *flow;
1214         struct datapath *dp;
1215         struct sw_flow_match match;
1216         struct sw_flow_id ufid;
1217         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1218         int err = 0;
1219         bool log = !a[OVS_FLOW_ATTR_PROBE];
1220         bool ufid_present;
1221
1222         ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1223         if (a[OVS_FLOW_ATTR_KEY]) {
1224                 ovs_match_init(&match, &key, NULL);
1225                 err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL,
1226                                         log);
1227         } else if (!ufid_present) {
1228                 OVS_NLERR(log,
1229                           "Flow get message rejected, Key attribute missing.");
1230                 err = -EINVAL;
1231         }
1232         if (err)
1233                 return err;
1234
1235         ovs_lock();
1236         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1237         if (!dp) {
1238                 err = -ENODEV;
1239                 goto unlock;
1240         }
1241
1242         if (ufid_present)
1243                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1244         else
1245                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1246         if (!flow) {
1247                 err = -ENOENT;
1248                 goto unlock;
1249         }
1250
1251         reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1252                                         OVS_FLOW_CMD_NEW, true, ufid_flags);
1253         if (IS_ERR(reply)) {
1254                 err = PTR_ERR(reply);
1255                 goto unlock;
1256         }
1257
1258         ovs_unlock();
1259         return genlmsg_reply(reply, info);
1260 unlock:
1261         ovs_unlock();
1262         return err;
1263 }
1264
1265 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1266 {
1267         struct nlattr **a = info->attrs;
1268         struct ovs_header *ovs_header = info->userhdr;
1269         struct sw_flow_key key;
1270         struct sk_buff *reply;
1271         struct sw_flow *flow = NULL;
1272         struct datapath *dp;
1273         struct sw_flow_match match;
1274         struct sw_flow_id ufid;
1275         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1276         int err;
1277         bool log = !a[OVS_FLOW_ATTR_PROBE];
1278         bool ufid_present;
1279
1280         ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1281         if (a[OVS_FLOW_ATTR_KEY]) {
1282                 ovs_match_init(&match, &key, NULL);
1283                 err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL,
1284                                         log);
1285                 if (unlikely(err))
1286                         return err;
1287         }
1288
1289         ovs_lock();
1290         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1291         if (unlikely(!dp)) {
1292                 err = -ENODEV;
1293                 goto unlock;
1294         }
1295
1296         if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1297                 err = ovs_flow_tbl_flush(&dp->table);
1298                 goto unlock;
1299         }
1300
1301         if (ufid_present)
1302                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1303         else
1304                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1305         if (unlikely(!flow)) {
1306                 err = -ENOENT;
1307                 goto unlock;
1308         }
1309
1310         ovs_flow_tbl_remove(&dp->table, flow);
1311         ovs_unlock();
1312
1313         reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1314                                         &flow->id, info, false, ufid_flags);
1315         if (likely(reply)) {
1316                 if (likely(!IS_ERR(reply))) {
1317                         rcu_read_lock();        /*To keep RCU checker happy. */
1318                         err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1319                                                      reply, info->snd_portid,
1320                                                      info->snd_seq, 0,
1321                                                      OVS_FLOW_CMD_DEL,
1322                                                      ufid_flags);
1323                         rcu_read_unlock();
1324                         BUG_ON(err < 0);
1325
1326                         ovs_notify(&dp_flow_genl_family, reply, info);
1327                 } else {
1328                         netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
1329                 }
1330         }
1331
1332         ovs_flow_free(flow, true);
1333         return 0;
1334 unlock:
1335         ovs_unlock();
1336         return err;
1337 }
1338
1339 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1340 {
1341         struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1342         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1343         struct table_instance *ti;
1344         struct datapath *dp;
1345         u32 ufid_flags;
1346         int err;
1347
1348         err = genlmsg_parse(cb->nlh, &dp_flow_genl_family, a,
1349                             OVS_FLOW_ATTR_MAX, flow_policy);
1350         if (err)
1351                 return err;
1352         ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1353
1354         rcu_read_lock();
1355         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1356         if (!dp) {
1357                 rcu_read_unlock();
1358                 return -ENODEV;
1359         }
1360
1361         ti = rcu_dereference(dp->table.ti);
1362         for (;;) {
1363                 struct sw_flow *flow;
1364                 u32 bucket, obj;
1365
1366                 bucket = cb->args[0];
1367                 obj = cb->args[1];
1368                 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1369                 if (!flow)
1370                         break;
1371
1372                 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1373                                            NETLINK_CB(cb->skb).portid,
1374                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1375                                            OVS_FLOW_CMD_NEW, ufid_flags) < 0)
1376                         break;
1377
1378                 cb->args[0] = bucket;
1379                 cb->args[1] = obj;
1380         }
1381         rcu_read_unlock();
1382         return skb->len;
1383 }
1384
1385 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1386         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1387         [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1388         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1389         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1390         [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1391         [OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1392         [OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1393 };
1394
1395 static const struct genl_ops dp_flow_genl_ops[] = {
1396         { .cmd = OVS_FLOW_CMD_NEW,
1397           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1398           .policy = flow_policy,
1399           .doit = ovs_flow_cmd_new
1400         },
1401         { .cmd = OVS_FLOW_CMD_DEL,
1402           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1403           .policy = flow_policy,
1404           .doit = ovs_flow_cmd_del
1405         },
1406         { .cmd = OVS_FLOW_CMD_GET,
1407           .flags = 0,               /* OK for unprivileged users. */
1408           .policy = flow_policy,
1409           .doit = ovs_flow_cmd_get,
1410           .dumpit = ovs_flow_cmd_dump
1411         },
1412         { .cmd = OVS_FLOW_CMD_SET,
1413           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1414           .policy = flow_policy,
1415           .doit = ovs_flow_cmd_set,
1416         },
1417 };
1418
1419 static struct genl_family dp_flow_genl_family = {
1420         .id = GENL_ID_GENERATE,
1421         .hdrsize = sizeof(struct ovs_header),
1422         .name = OVS_FLOW_FAMILY,
1423         .version = OVS_FLOW_VERSION,
1424         .maxattr = OVS_FLOW_ATTR_MAX,
1425         .netnsok = true,
1426         .parallel_ops = true,
1427         .ops = dp_flow_genl_ops,
1428         .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1429         .mcgrps = &ovs_dp_flow_multicast_group,
1430         .n_mcgrps = 1,
1431 };
1432
1433 static size_t ovs_dp_cmd_msg_size(void)
1434 {
1435         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1436
1437         msgsize += nla_total_size(IFNAMSIZ);
1438         msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1439         msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1440         msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1441
1442         return msgsize;
1443 }
1444
1445 /* Called with ovs_mutex. */
1446 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1447                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1448 {
1449         struct ovs_header *ovs_header;
1450         struct ovs_dp_stats dp_stats;
1451         struct ovs_dp_megaflow_stats dp_megaflow_stats;
1452         int err;
1453
1454         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1455                                    flags, cmd);
1456         if (!ovs_header)
1457                 goto error;
1458
1459         ovs_header->dp_ifindex = get_dpifindex(dp);
1460
1461         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1462         if (err)
1463                 goto nla_put_failure;
1464
1465         get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1466         if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1467                         &dp_stats))
1468                 goto nla_put_failure;
1469
1470         if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1471                         sizeof(struct ovs_dp_megaflow_stats),
1472                         &dp_megaflow_stats))
1473                 goto nla_put_failure;
1474
1475         if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1476                 goto nla_put_failure;
1477
1478         genlmsg_end(skb, ovs_header);
1479         return 0;
1480
1481 nla_put_failure:
1482         genlmsg_cancel(skb, ovs_header);
1483 error:
1484         return -EMSGSIZE;
1485 }
1486
1487 static struct sk_buff *ovs_dp_cmd_alloc_info(struct genl_info *info)
1488 {
1489         return genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info, GFP_KERNEL);
1490 }
1491
1492 /* Called with rcu_read_lock or ovs_mutex. */
1493 static struct datapath *lookup_datapath(struct net *net,
1494                                         const struct ovs_header *ovs_header,
1495                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1496 {
1497         struct datapath *dp;
1498
1499         if (!a[OVS_DP_ATTR_NAME])
1500                 dp = get_dp(net, ovs_header->dp_ifindex);
1501         else {
1502                 struct vport *vport;
1503
1504                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1505                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1506         }
1507         return dp ? dp : ERR_PTR(-ENODEV);
1508 }
1509
1510 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1511 {
1512         struct datapath *dp;
1513
1514         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1515         if (IS_ERR(dp))
1516                 return;
1517
1518         WARN(dp->user_features, "Dropping previously announced user features\n");
1519         dp->user_features = 0;
1520 }
1521
1522 static void ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1523 {
1524         if (a[OVS_DP_ATTR_USER_FEATURES])
1525                 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1526 }
1527
1528 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1529 {
1530         struct nlattr **a = info->attrs;
1531         struct vport_parms parms;
1532         struct sk_buff *reply;
1533         struct datapath *dp;
1534         struct vport *vport;
1535         struct ovs_net *ovs_net;
1536         int err, i;
1537
1538         err = -EINVAL;
1539         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1540                 goto err;
1541
1542         reply = ovs_dp_cmd_alloc_info(info);
1543         if (!reply)
1544                 return -ENOMEM;
1545
1546         err = -ENOMEM;
1547         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1548         if (dp == NULL)
1549                 goto err_free_reply;
1550
1551         ovs_dp_set_net(dp, sock_net(skb->sk));
1552
1553         /* Allocate table. */
1554         err = ovs_flow_tbl_init(&dp->table);
1555         if (err)
1556                 goto err_free_dp;
1557
1558         dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1559         if (!dp->stats_percpu) {
1560                 err = -ENOMEM;
1561                 goto err_destroy_table;
1562         }
1563
1564         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1565                             GFP_KERNEL);
1566         if (!dp->ports) {
1567                 err = -ENOMEM;
1568                 goto err_destroy_percpu;
1569         }
1570
1571         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1572                 INIT_HLIST_HEAD(&dp->ports[i]);
1573
1574         /* Set up our datapath device. */
1575         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1576         parms.type = OVS_VPORT_TYPE_INTERNAL;
1577         parms.options = NULL;
1578         parms.dp = dp;
1579         parms.port_no = OVSP_LOCAL;
1580         parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1581
1582         ovs_dp_change(dp, a);
1583
1584         /* So far only local changes have been made, now need the lock. */
1585         ovs_lock();
1586
1587         vport = new_vport(&parms);
1588         if (IS_ERR(vport)) {
1589                 err = PTR_ERR(vport);
1590                 if (err == -EBUSY)
1591                         err = -EEXIST;
1592
1593                 if (err == -EEXIST) {
1594                         /* An outdated user space instance that does not understand
1595                          * the concept of user_features has attempted to create a new
1596                          * datapath and is likely to reuse it. Drop all user features.
1597                          */
1598                         if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1599                                 ovs_dp_reset_user_features(skb, info);
1600                 }
1601
1602                 goto err_destroy_ports_array;
1603         }
1604
1605         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1606                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1607         BUG_ON(err < 0);
1608
1609         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1610         list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1611
1612         ovs_unlock();
1613
1614         ovs_notify(&dp_datapath_genl_family, reply, info);
1615         return 0;
1616
1617 err_destroy_ports_array:
1618         ovs_unlock();
1619         kfree(dp->ports);
1620 err_destroy_percpu:
1621         free_percpu(dp->stats_percpu);
1622 err_destroy_table:
1623         ovs_flow_tbl_destroy(&dp->table);
1624 err_free_dp:
1625         kfree(dp);
1626 err_free_reply:
1627         kfree_skb(reply);
1628 err:
1629         return err;
1630 }
1631
1632 /* Called with ovs_mutex. */
1633 static void __dp_destroy(struct datapath *dp)
1634 {
1635         int i;
1636
1637         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1638                 struct vport *vport;
1639                 struct hlist_node *n;
1640
1641                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1642                         if (vport->port_no != OVSP_LOCAL)
1643                                 ovs_dp_detach_port(vport);
1644         }
1645
1646         list_del_rcu(&dp->list_node);
1647
1648         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1649          * all ports in datapath are destroyed first before freeing datapath.
1650          */
1651         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1652
1653         /* RCU destroy the flow table */
1654         call_rcu(&dp->rcu, destroy_dp_rcu);
1655 }
1656
1657 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1658 {
1659         struct sk_buff *reply;
1660         struct datapath *dp;
1661         int err;
1662
1663         reply = ovs_dp_cmd_alloc_info(info);
1664         if (!reply)
1665                 return -ENOMEM;
1666
1667         ovs_lock();
1668         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1669         err = PTR_ERR(dp);
1670         if (IS_ERR(dp))
1671                 goto err_unlock_free;
1672
1673         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1674                                    info->snd_seq, 0, OVS_DP_CMD_DEL);
1675         BUG_ON(err < 0);
1676
1677         __dp_destroy(dp);
1678         ovs_unlock();
1679
1680         ovs_notify(&dp_datapath_genl_family, reply, info);
1681
1682         return 0;
1683
1684 err_unlock_free:
1685         ovs_unlock();
1686         kfree_skb(reply);
1687         return err;
1688 }
1689
1690 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1691 {
1692         struct sk_buff *reply;
1693         struct datapath *dp;
1694         int err;
1695
1696         reply = ovs_dp_cmd_alloc_info(info);
1697         if (!reply)
1698                 return -ENOMEM;
1699
1700         ovs_lock();
1701         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1702         err = PTR_ERR(dp);
1703         if (IS_ERR(dp))
1704                 goto err_unlock_free;
1705
1706         ovs_dp_change(dp, info->attrs);
1707
1708         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1709                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1710         BUG_ON(err < 0);
1711
1712         ovs_unlock();
1713         ovs_notify(&dp_datapath_genl_family, reply, info);
1714
1715         return 0;
1716
1717 err_unlock_free:
1718         ovs_unlock();
1719         kfree_skb(reply);
1720         return err;
1721 }
1722
1723 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1724 {
1725         struct sk_buff *reply;
1726         struct datapath *dp;
1727         int err;
1728
1729         reply = ovs_dp_cmd_alloc_info(info);
1730         if (!reply)
1731                 return -ENOMEM;
1732
1733         ovs_lock();
1734         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1735         if (IS_ERR(dp)) {
1736                 err = PTR_ERR(dp);
1737                 goto err_unlock_free;
1738         }
1739         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1740                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1741         BUG_ON(err < 0);
1742         ovs_unlock();
1743
1744         return genlmsg_reply(reply, info);
1745
1746 err_unlock_free:
1747         ovs_unlock();
1748         kfree_skb(reply);
1749         return err;
1750 }
1751
1752 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1753 {
1754         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1755         struct datapath *dp;
1756         int skip = cb->args[0];
1757         int i = 0;
1758
1759         ovs_lock();
1760         list_for_each_entry(dp, &ovs_net->dps, list_node) {
1761                 if (i >= skip &&
1762                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1763                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1764                                          OVS_DP_CMD_NEW) < 0)
1765                         break;
1766                 i++;
1767         }
1768         ovs_unlock();
1769
1770         cb->args[0] = i;
1771
1772         return skb->len;
1773 }
1774
1775 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1776         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1777         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1778         [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1779 };
1780
1781 static const struct genl_ops dp_datapath_genl_ops[] = {
1782         { .cmd = OVS_DP_CMD_NEW,
1783           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1784           .policy = datapath_policy,
1785           .doit = ovs_dp_cmd_new
1786         },
1787         { .cmd = OVS_DP_CMD_DEL,
1788           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1789           .policy = datapath_policy,
1790           .doit = ovs_dp_cmd_del
1791         },
1792         { .cmd = OVS_DP_CMD_GET,
1793           .flags = 0,               /* OK for unprivileged users. */
1794           .policy = datapath_policy,
1795           .doit = ovs_dp_cmd_get,
1796           .dumpit = ovs_dp_cmd_dump
1797         },
1798         { .cmd = OVS_DP_CMD_SET,
1799           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1800           .policy = datapath_policy,
1801           .doit = ovs_dp_cmd_set,
1802         },
1803 };
1804
1805 static struct genl_family dp_datapath_genl_family = {
1806         .id = GENL_ID_GENERATE,
1807         .hdrsize = sizeof(struct ovs_header),
1808         .name = OVS_DATAPATH_FAMILY,
1809         .version = OVS_DATAPATH_VERSION,
1810         .maxattr = OVS_DP_ATTR_MAX,
1811         .netnsok = true,
1812         .parallel_ops = true,
1813         .ops = dp_datapath_genl_ops,
1814         .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1815         .mcgrps = &ovs_dp_datapath_multicast_group,
1816         .n_mcgrps = 1,
1817 };
1818
1819 /* Called with ovs_mutex or RCU read lock. */
1820 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1821                                    u32 portid, u32 seq, u32 flags, u8 cmd)
1822 {
1823         struct ovs_header *ovs_header;
1824         struct ovs_vport_stats vport_stats;
1825         int err;
1826
1827         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1828                                  flags, cmd);
1829         if (!ovs_header)
1830                 return -EMSGSIZE;
1831
1832         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1833
1834         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1835             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1836             nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1837                            ovs_vport_name(vport)))
1838                 goto nla_put_failure;
1839
1840         ovs_vport_get_stats(vport, &vport_stats);
1841         if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1842                     &vport_stats))
1843                 goto nla_put_failure;
1844
1845         if (ovs_vport_get_upcall_portids(vport, skb))
1846                 goto nla_put_failure;
1847
1848         err = ovs_vport_get_options(vport, skb);
1849         if (err == -EMSGSIZE)
1850                 goto error;
1851
1852         genlmsg_end(skb, ovs_header);
1853         return 0;
1854
1855 nla_put_failure:
1856         err = -EMSGSIZE;
1857 error:
1858         genlmsg_cancel(skb, ovs_header);
1859         return err;
1860 }
1861
1862 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1863 {
1864         return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1865 }
1866
1867 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1868 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1869                                          u32 seq, u8 cmd)
1870 {
1871         struct sk_buff *skb;
1872         int retval;
1873
1874         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1875         if (!skb)
1876                 return ERR_PTR(-ENOMEM);
1877
1878         retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1879         BUG_ON(retval < 0);
1880
1881         return skb;
1882 }
1883
1884 /* Called with ovs_mutex or RCU read lock. */
1885 static struct vport *lookup_vport(struct net *net,
1886                                   const struct ovs_header *ovs_header,
1887                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1888 {
1889         struct datapath *dp;
1890         struct vport *vport;
1891
1892         if (a[OVS_VPORT_ATTR_NAME]) {
1893                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1894                 if (!vport)
1895                         return ERR_PTR(-ENODEV);
1896                 if (ovs_header->dp_ifindex &&
1897                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1898                         return ERR_PTR(-ENODEV);
1899                 return vport;
1900         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1901                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1902
1903                 if (port_no >= DP_MAX_PORTS)
1904                         return ERR_PTR(-EFBIG);
1905
1906                 dp = get_dp(net, ovs_header->dp_ifindex);
1907                 if (!dp)
1908                         return ERR_PTR(-ENODEV);
1909
1910                 vport = ovs_vport_ovsl_rcu(dp, port_no);
1911                 if (!vport)
1912                         return ERR_PTR(-ENODEV);
1913                 return vport;
1914         } else
1915                 return ERR_PTR(-EINVAL);
1916 }
1917
1918 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1919 {
1920         struct nlattr **a = info->attrs;
1921         struct ovs_header *ovs_header = info->userhdr;
1922         struct vport_parms parms;
1923         struct sk_buff *reply;
1924         struct vport *vport;
1925         struct datapath *dp;
1926         u32 port_no;
1927         int err;
1928
1929         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1930             !a[OVS_VPORT_ATTR_UPCALL_PID])
1931                 return -EINVAL;
1932
1933         port_no = a[OVS_VPORT_ATTR_PORT_NO]
1934                 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
1935         if (port_no >= DP_MAX_PORTS)
1936                 return -EFBIG;
1937
1938         reply = ovs_vport_cmd_alloc_info();
1939         if (!reply)
1940                 return -ENOMEM;
1941
1942         ovs_lock();
1943 restart:
1944         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1945         err = -ENODEV;
1946         if (!dp)
1947                 goto exit_unlock_free;
1948
1949         if (port_no) {
1950                 vport = ovs_vport_ovsl(dp, port_no);
1951                 err = -EBUSY;
1952                 if (vport)
1953                         goto exit_unlock_free;
1954         } else {
1955                 for (port_no = 1; ; port_no++) {
1956                         if (port_no >= DP_MAX_PORTS) {
1957                                 err = -EFBIG;
1958                                 goto exit_unlock_free;
1959                         }
1960                         vport = ovs_vport_ovsl(dp, port_no);
1961                         if (!vport)
1962                                 break;
1963                 }
1964         }
1965
1966         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1967         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1968         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1969         parms.dp = dp;
1970         parms.port_no = port_no;
1971         parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
1972
1973         vport = new_vport(&parms);
1974         err = PTR_ERR(vport);
1975         if (IS_ERR(vport)) {
1976                 if (err == -EAGAIN)
1977                         goto restart;
1978                 goto exit_unlock_free;
1979         }
1980
1981         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1982                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1983         BUG_ON(err < 0);
1984         ovs_unlock();
1985
1986         ovs_notify(&dp_vport_genl_family, reply, info);
1987         return 0;
1988
1989 exit_unlock_free:
1990         ovs_unlock();
1991         kfree_skb(reply);
1992         return err;
1993 }
1994
1995 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1996 {
1997         struct nlattr **a = info->attrs;
1998         struct sk_buff *reply;
1999         struct vport *vport;
2000         int err;
2001
2002         reply = ovs_vport_cmd_alloc_info();
2003         if (!reply)
2004                 return -ENOMEM;
2005
2006         ovs_lock();
2007         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2008         err = PTR_ERR(vport);
2009         if (IS_ERR(vport))
2010                 goto exit_unlock_free;
2011
2012         if (a[OVS_VPORT_ATTR_TYPE] &&
2013             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2014                 err = -EINVAL;
2015                 goto exit_unlock_free;
2016         }
2017
2018         if (a[OVS_VPORT_ATTR_OPTIONS]) {
2019                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2020                 if (err)
2021                         goto exit_unlock_free;
2022         }
2023
2024
2025         if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2026                 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2027
2028                 err = ovs_vport_set_upcall_portids(vport, ids);
2029                 if (err)
2030                         goto exit_unlock_free;
2031         }
2032
2033         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2034                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2035         BUG_ON(err < 0);
2036
2037         ovs_unlock();
2038         ovs_notify(&dp_vport_genl_family, reply, info);
2039         return 0;
2040
2041 exit_unlock_free:
2042         ovs_unlock();
2043         kfree_skb(reply);
2044         return err;
2045 }
2046
2047 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2048 {
2049         struct nlattr **a = info->attrs;
2050         struct sk_buff *reply;
2051         struct vport *vport;
2052         int err;
2053
2054         reply = ovs_vport_cmd_alloc_info();
2055         if (!reply)
2056                 return -ENOMEM;
2057
2058         ovs_lock();
2059         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2060         err = PTR_ERR(vport);
2061         if (IS_ERR(vport))
2062                 goto exit_unlock_free;
2063
2064         if (vport->port_no == OVSP_LOCAL) {
2065                 err = -EINVAL;
2066                 goto exit_unlock_free;
2067         }
2068
2069         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2070                                       info->snd_seq, 0, OVS_VPORT_CMD_DEL);
2071         BUG_ON(err < 0);
2072         ovs_dp_detach_port(vport);
2073         ovs_unlock();
2074
2075         ovs_notify(&dp_vport_genl_family, reply, info);
2076         return 0;
2077
2078 exit_unlock_free:
2079         ovs_unlock();
2080         kfree_skb(reply);
2081         return err;
2082 }
2083
2084 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2085 {
2086         struct nlattr **a = info->attrs;
2087         struct ovs_header *ovs_header = info->userhdr;
2088         struct sk_buff *reply;
2089         struct vport *vport;
2090         int err;
2091
2092         reply = ovs_vport_cmd_alloc_info();
2093         if (!reply)
2094                 return -ENOMEM;
2095
2096         rcu_read_lock();
2097         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2098         err = PTR_ERR(vport);
2099         if (IS_ERR(vport))
2100                 goto exit_unlock_free;
2101         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2102                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2103         BUG_ON(err < 0);
2104         rcu_read_unlock();
2105
2106         return genlmsg_reply(reply, info);
2107
2108 exit_unlock_free:
2109         rcu_read_unlock();
2110         kfree_skb(reply);
2111         return err;
2112 }
2113
2114 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2115 {
2116         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2117         struct datapath *dp;
2118         int bucket = cb->args[0], skip = cb->args[1];
2119         int i, j = 0;
2120
2121         rcu_read_lock();
2122         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2123         if (!dp) {
2124                 rcu_read_unlock();
2125                 return -ENODEV;
2126         }
2127         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2128                 struct vport *vport;
2129
2130                 j = 0;
2131                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2132                         if (j >= skip &&
2133                             ovs_vport_cmd_fill_info(vport, skb,
2134                                                     NETLINK_CB(cb->skb).portid,
2135                                                     cb->nlh->nlmsg_seq,
2136                                                     NLM_F_MULTI,
2137                                                     OVS_VPORT_CMD_NEW) < 0)
2138                                 goto out;
2139
2140                         j++;
2141                 }
2142                 skip = 0;
2143         }
2144 out:
2145         rcu_read_unlock();
2146
2147         cb->args[0] = i;
2148         cb->args[1] = j;
2149
2150         return skb->len;
2151 }
2152
2153 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2154         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2155         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2156         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2157         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2158         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
2159         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2160 };
2161
2162 static const struct genl_ops dp_vport_genl_ops[] = {
2163         { .cmd = OVS_VPORT_CMD_NEW,
2164           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2165           .policy = vport_policy,
2166           .doit = ovs_vport_cmd_new
2167         },
2168         { .cmd = OVS_VPORT_CMD_DEL,
2169           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2170           .policy = vport_policy,
2171           .doit = ovs_vport_cmd_del
2172         },
2173         { .cmd = OVS_VPORT_CMD_GET,
2174           .flags = 0,               /* OK for unprivileged users. */
2175           .policy = vport_policy,
2176           .doit = ovs_vport_cmd_get,
2177           .dumpit = ovs_vport_cmd_dump
2178         },
2179         { .cmd = OVS_VPORT_CMD_SET,
2180           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2181           .policy = vport_policy,
2182           .doit = ovs_vport_cmd_set,
2183         },
2184 };
2185
2186 struct genl_family dp_vport_genl_family = {
2187         .id = GENL_ID_GENERATE,
2188         .hdrsize = sizeof(struct ovs_header),
2189         .name = OVS_VPORT_FAMILY,
2190         .version = OVS_VPORT_VERSION,
2191         .maxattr = OVS_VPORT_ATTR_MAX,
2192         .netnsok = true,
2193         .parallel_ops = true,
2194         .ops = dp_vport_genl_ops,
2195         .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2196         .mcgrps = &ovs_dp_vport_multicast_group,
2197         .n_mcgrps = 1,
2198 };
2199
2200 static struct genl_family * const dp_genl_families[] = {
2201         &dp_datapath_genl_family,
2202         &dp_vport_genl_family,
2203         &dp_flow_genl_family,
2204         &dp_packet_genl_family,
2205 };
2206
2207 static void dp_unregister_genl(int n_families)
2208 {
2209         int i;
2210
2211         for (i = 0; i < n_families; i++)
2212                 genl_unregister_family(dp_genl_families[i]);
2213 }
2214
2215 static int dp_register_genl(void)
2216 {
2217         int err;
2218         int i;
2219
2220         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2221
2222                 err = genl_register_family(dp_genl_families[i]);
2223                 if (err)
2224                         goto error;
2225         }
2226
2227         return 0;
2228
2229 error:
2230         dp_unregister_genl(i);
2231         return err;
2232 }
2233
2234 static int __net_init ovs_init_net(struct net *net)
2235 {
2236         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2237
2238         INIT_LIST_HEAD(&ovs_net->dps);
2239         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2240         return 0;
2241 }
2242
2243 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2244                                             struct list_head *head)
2245 {
2246         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2247         struct datapath *dp;
2248
2249         list_for_each_entry(dp, &ovs_net->dps, list_node) {
2250                 int i;
2251
2252                 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2253                         struct vport *vport;
2254
2255                         hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2256                                 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2257                                         continue;
2258
2259                                 if (dev_net(vport->dev) == dnet)
2260                                         list_add(&vport->detach_list, head);
2261                         }
2262                 }
2263         }
2264 }
2265
2266 static void __net_exit ovs_exit_net(struct net *dnet)
2267 {
2268         struct datapath *dp, *dp_next;
2269         struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2270         struct vport *vport, *vport_next;
2271         struct net *net;
2272         LIST_HEAD(head);
2273
2274         ovs_lock();
2275         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2276                 __dp_destroy(dp);
2277
2278         rtnl_lock();
2279         for_each_net(net)
2280                 list_vports_from_net(net, dnet, &head);
2281         rtnl_unlock();
2282
2283         /* Detach all vports from given namespace. */
2284         list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2285                 list_del(&vport->detach_list);
2286                 ovs_dp_detach_port(vport);
2287         }
2288
2289         ovs_unlock();
2290
2291         cancel_work_sync(&ovs_net->dp_notify_work);
2292 }
2293
2294 static struct pernet_operations ovs_net_ops = {
2295         .init = ovs_init_net,
2296         .exit = ovs_exit_net,
2297         .id   = &ovs_net_id,
2298         .size = sizeof(struct ovs_net),
2299 };
2300
2301 static int __init dp_init(void)
2302 {
2303         int err;
2304
2305         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2306
2307         pr_info("Open vSwitch switching datapath\n");
2308
2309         err = action_fifos_init();
2310         if (err)
2311                 goto error;
2312
2313         err = ovs_internal_dev_rtnl_link_register();
2314         if (err)
2315                 goto error_action_fifos_exit;
2316
2317         err = ovs_flow_init();
2318         if (err)
2319                 goto error_unreg_rtnl_link;
2320
2321         err = ovs_vport_init();
2322         if (err)
2323                 goto error_flow_exit;
2324
2325         err = register_pernet_device(&ovs_net_ops);
2326         if (err)
2327                 goto error_vport_exit;
2328
2329         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2330         if (err)
2331                 goto error_netns_exit;
2332
2333         err = ovs_netdev_init();
2334         if (err)
2335                 goto error_unreg_notifier;
2336
2337         err = dp_register_genl();
2338         if (err < 0)
2339                 goto error_unreg_netdev;
2340
2341         return 0;
2342
2343 error_unreg_netdev:
2344         ovs_netdev_exit();
2345 error_unreg_notifier:
2346         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2347 error_netns_exit:
2348         unregister_pernet_device(&ovs_net_ops);
2349 error_vport_exit:
2350         ovs_vport_exit();
2351 error_flow_exit:
2352         ovs_flow_exit();
2353 error_unreg_rtnl_link:
2354         ovs_internal_dev_rtnl_link_unregister();
2355 error_action_fifos_exit:
2356         action_fifos_exit();
2357 error:
2358         return err;
2359 }
2360
2361 static void dp_cleanup(void)
2362 {
2363         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2364         ovs_netdev_exit();
2365         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2366         unregister_pernet_device(&ovs_net_ops);
2367         rcu_barrier();
2368         ovs_vport_exit();
2369         ovs_flow_exit();
2370         ovs_internal_dev_rtnl_link_unregister();
2371         action_fifos_exit();
2372 }
2373
2374 module_init(dp_init);
2375 module_exit(dp_cleanup);
2376
2377 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2378 MODULE_LICENSE("GPL");