datapath/datapath.c

   1 /*
   2  * Copyright (c) 2007-2015 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 "conntrack.h"
  57 #include "flow.h"
  58 #include "flow_table.h"
  59 #include "flow_netlink.h"
  60 #include "gso.h"
  61 #include "vport-internal_dev.h"
  62 #include "vport-netdev.h"
  63
  64 int ovs_net_id __read_mostly;
  65 EXPORT_SYMBOL_GPL(ovs_net_id);
  66
  67 static struct genl_family dp_packet_genl_family;
  68 static struct genl_family dp_flow_genl_family;
  69 static struct genl_family dp_datapath_genl_family;
  70
  71 static const struct nla_policy flow_policy[];
  72
  73 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
  74         .name = OVS_FLOW_MCGROUP
  75 };
  76
  77 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
  78         .name = OVS_DATAPATH_MCGROUP
  79 };
  80
  81 struct genl_multicast_group ovs_dp_vport_multicast_group = {
  82         .name = OVS_VPORT_MCGROUP
  83 };
  84
  85 /* Check if need to build a reply message.
  86  * OVS userspace sets the NLM_F_ECHO flag if it needs the reply.
  87  */
  88 static bool ovs_must_notify(struct genl_family *family, struct genl_info *info,
  89                             unsigned int group)
  90 {
  91         return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
  92                genl_has_listeners(family, genl_info_net(info), group);
  93 }
  94
  95 static void ovs_notify(struct genl_family *family, struct genl_multicast_group *grp,
  96                        struct sk_buff *skb, struct genl_info *info)
  97 {
  98         genl_notify(family, skb, info, GROUP_ID(grp), GFP_KERNEL);
  99 }
 100
 101 /**
 102  * DOC: Locking:
 103  *
 104  * All writes e.g. Writes to device state (add/remove datapath, port, set
 105  * operations on vports, etc.), Writes to other state (flow table
 106  * modifications, set miscellaneous datapath parameters, etc.) are protected
 107  * by ovs_lock.
 108  *
 109  * Reads are protected by RCU.
 110  *
 111  * There are a few special cases (mostly stats) that have their own
 112  * synchronization but they nest under all of above and don't interact with
 113  * each other.
 114  *
 115  * The RTNL lock nests inside ovs_mutex.
 116  */
 117
 118 static DEFINE_MUTEX(ovs_mutex);
 119
 120 void ovs_lock(void)
 121 {
 122         mutex_lock(&ovs_mutex);
 123 }
 124
 125 void ovs_unlock(void)
 126 {
 127         mutex_unlock(&ovs_mutex);
 128 }
 129
 130 #ifdef CONFIG_LOCKDEP
 131 int lockdep_ovsl_is_held(void)
 132 {
 133         if (debug_locks)
 134                 return lockdep_is_held(&ovs_mutex);
 135         else
 136                 return 1;
 137 }
 138 EXPORT_SYMBOL_GPL(lockdep_ovsl_is_held);
 139 #endif
 140
 141 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
 142                              const struct sw_flow_key *,
 143                              const struct dp_upcall_info *,
 144                              uint32_t cutlen);
 145 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
 146                                   const struct sw_flow_key *,
 147                                   const struct dp_upcall_info *,
 148                                   uint32_t cutlen);
 149
 150 /* Must be called with rcu_read_lock. */
 151 static struct datapath *get_dp_rcu(struct net *net, int dp_ifindex)
 152 {
 153         struct net_device *dev = dev_get_by_index_rcu(net, dp_ifindex);
 154
 155         if (dev) {
 156                 struct vport *vport = ovs_internal_dev_get_vport(dev);
 157                 if (vport)
 158                         return vport->dp;
 159         }
 160
 161         return NULL;
 162 }
 163
 164 /* The caller must hold either ovs_mutex or rcu_read_lock to keep the
 165  * returned dp pointer valid.
 166  */
 167 static inline struct datapath *get_dp(struct net *net, int dp_ifindex)
 168 {
 169         struct datapath *dp;
 170
 171         WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_ovsl_is_held());
 172         rcu_read_lock();
 173         dp = get_dp_rcu(net, dp_ifindex);
 174         rcu_read_unlock();
 175
 176         return dp;
 177 }
 178
 179 /* Must be called with rcu_read_lock or ovs_mutex. */
 180 const char *ovs_dp_name(const struct datapath *dp)
 181 {
 182         struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
 183         return ovs_vport_name(vport);
 184 }
 185
 186 static int get_dpifindex(const struct datapath *dp)
 187 {
 188         struct vport *local;
 189         int ifindex;
 190
 191         rcu_read_lock();
 192
 193         local = ovs_vport_rcu(dp, OVSP_LOCAL);
 194         if (local)
 195                 ifindex = local->dev->ifindex;
 196         else
 197                 ifindex = 0;
 198
 199         rcu_read_unlock();
 200
 201         return ifindex;
 202 }
 203
 204 static void destroy_dp_rcu(struct rcu_head *rcu)
 205 {
 206         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 207
 208         ovs_flow_tbl_destroy(&dp->table);
 209         free_percpu(dp->stats_percpu);
 210         kfree(dp->ports);
 211         kfree(dp);
 212 }
 213
 214 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
 215                                             u16 port_no)
 216 {
 217         return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
 218 }
 219
 220 /* Called with ovs_mutex or RCU read lock. */
 221 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
 222 {
 223         struct vport *vport;
 224         struct hlist_head *head;
 225
 226         head = vport_hash_bucket(dp, port_no);
 227         hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
 228                 if (vport->port_no == port_no)
 229                         return vport;
 230         }
 231         return NULL;
 232 }
 233
 234 /* Called with ovs_mutex. */
 235 static struct vport *new_vport(const struct vport_parms *parms)
 236 {
 237         struct vport *vport;
 238
 239         vport = ovs_vport_add(parms);
 240         if (!IS_ERR(vport)) {
 241                 struct datapath *dp = parms->dp;
 242                 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
 243
 244                 hlist_add_head_rcu(&vport->dp_hash_node, head);
 245         }
 246         return vport;
 247 }
 248
 249 void ovs_dp_detach_port(struct vport *p)
 250 {
 251         ASSERT_OVSL();
 252
 253         /* First drop references to device. */
 254         hlist_del_rcu(&p->dp_hash_node);
 255
 256         /* Then destroy it. */
 257         ovs_vport_del(p);
 258 }
 259
 260 /* Must be called with rcu_read_lock. */
 261 void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
 262 {
 263         const struct vport *p = OVS_CB(skb)->input_vport;
 264         struct datapath *dp = p->dp;
 265         struct sw_flow *flow;
 266         struct sw_flow_actions *sf_acts;
 267         struct dp_stats_percpu *stats;
 268         u64 *stats_counter;
 269         u32 n_mask_hit;
 270
 271         stats = this_cpu_ptr(dp->stats_percpu);
 272
 273         /* Look up flow. */
 274         flow = ovs_flow_tbl_lookup_stats(&dp->table, key, skb_get_hash(skb),
 275                                          &n_mask_hit);
 276         if (unlikely(!flow)) {
 277                 struct dp_upcall_info upcall;
 278                 int error;
 279
 280                 memset(&upcall, 0, sizeof(upcall));
 281                 upcall.cmd = OVS_PACKET_CMD_MISS;
 282                 upcall.portid = ovs_vport_find_upcall_portid(p, skb);
 283                 upcall.mru = OVS_CB(skb)->mru;
 284                 error = ovs_dp_upcall(dp, skb, key, &upcall, 0);
 285                 if (unlikely(error))
 286                         kfree_skb(skb);
 287                 else
 288                         consume_skb(skb);
 289                 stats_counter = &stats->n_missed;
 290                 goto out;
 291         }
 292
 293         ovs_flow_stats_update(flow, key->tp.flags, skb);
 294         sf_acts = rcu_dereference(flow->sf_acts);
 295         ovs_execute_actions(dp, skb, sf_acts, key);
 296
 297         stats_counter = &stats->n_hit;
 298
 299 out:
 300         /* Update datapath statistics. */
 301         u64_stats_update_begin(&stats->syncp);
 302         (*stats_counter)++;
 303         stats->n_mask_hit += n_mask_hit;
 304         u64_stats_update_end(&stats->syncp);
 305 }
 306
 307 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
 308                   const struct sw_flow_key *key,
 309                   const struct dp_upcall_info *upcall_info,
 310                   uint32_t cutlen)
 311 {
 312         struct dp_stats_percpu *stats;
 313         int err;
 314
 315         if (upcall_info->portid == 0) {
 316                 err = -ENOTCONN;
 317                 goto err;
 318         }
 319
 320         if (!skb_is_gso(skb))
 321                 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
 322         else
 323                 err = queue_gso_packets(dp, skb, key, upcall_info, cutlen);
 324         if (err)
 325                 goto err;
 326
 327         return 0;
 328
 329 err:
 330         stats = this_cpu_ptr(dp->stats_percpu);
 331
 332         u64_stats_update_begin(&stats->syncp);
 333         stats->n_lost++;
 334         u64_stats_update_end(&stats->syncp);
 335
 336         return err;
 337 }
 338
 339 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
 340                              const struct sw_flow_key *key,
 341                              const struct dp_upcall_info *upcall_info,
 342                                  uint32_t cutlen)
 343 {
 344         unsigned short gso_type = skb_shinfo(skb)->gso_type;
 345         struct sw_flow_key later_key;
 346         struct sk_buff *segs, *nskb;
 347         struct ovs_skb_cb ovs_cb;
 348         int err;
 349
 350         ovs_cb = *OVS_CB(skb);
 351         segs = __skb_gso_segment(skb, NETIF_F_SG, false);
 352         *OVS_CB(skb) = ovs_cb;
 353         if (IS_ERR(segs))
 354                 return PTR_ERR(segs);
 355         if (segs == NULL)
 356                 return -EINVAL;
 357
 358         if (gso_type & SKB_GSO_UDP) {
 359                 /* The initial flow key extracted by ovs_flow_key_extract()
 360                  * in this case is for a first fragment, so we need to
 361                  * properly mark later fragments.
 362                  */
 363                 later_key = *key;
 364                 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
 365         }
 366
 367         /* Queue all of the segments. */
 368         skb = segs;
 369         do {
 370                 *OVS_CB(skb) = ovs_cb;
 371                 if (gso_type & SKB_GSO_UDP && skb != segs)
 372                         key = &later_key;
 373
 374                 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
 375                 if (err)
 376                         break;
 377
 378         } while ((skb = skb->next));
 379
 380         /* Free all of the segments. */
 381         skb = segs;
 382         do {
 383                 nskb = skb->next;
 384                 if (err)
 385                         kfree_skb(skb);
 386                 else
 387                         consume_skb(skb);
 388         } while ((skb = nskb));
 389         return err;
 390 }
 391
 392 static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
 393                               unsigned int hdrlen)
 394 {
 395         size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
 396                 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
 397                 + nla_total_size(ovs_key_attr_size()) /* OVS_PACKET_ATTR_KEY */
 398                 + nla_total_size(sizeof(unsigned int)); /* OVS_PACKET_ATTR_LEN */
 399
 400         /* OVS_PACKET_ATTR_USERDATA */
 401         if (upcall_info->userdata)
 402                 size += NLA_ALIGN(upcall_info->userdata->nla_len);
 403
 404         /* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
 405         if (upcall_info->egress_tun_info)
 406                 size += nla_total_size(ovs_tun_key_attr_size());
 407
 408         /* OVS_PACKET_ATTR_ACTIONS */
 409         if (upcall_info->actions_len)
 410                 size += nla_total_size(upcall_info->actions_len);
 411
 412         /* OVS_PACKET_ATTR_MRU */
 413         if (upcall_info->mru)
 414                 size += nla_total_size(sizeof(upcall_info->mru));
 415
 416         return size;
 417 }
 418
 419 static void pad_packet(struct datapath *dp, struct sk_buff *skb)
 420 {
 421         if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
 422                 size_t plen = NLA_ALIGN(skb->len) - skb->len;
 423
 424                 if (plen > 0)
 425                         memset(skb_put(skb, plen), 0, plen);
 426         }
 427 }
 428
 429 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
 430                                   const struct sw_flow_key *key,
 431                                   const struct dp_upcall_info *upcall_info,
 432                                   uint32_t cutlen)
 433 {
 434         struct ovs_header *upcall;
 435         struct sk_buff *nskb = NULL;
 436         struct sk_buff *user_skb = NULL; /* to be queued to userspace */
 437         struct nlattr *nla;
 438         size_t len;
 439         unsigned int hlen;
 440         int err, dp_ifindex;
 441
 442         dp_ifindex = get_dpifindex(dp);
 443         if (!dp_ifindex)
 444                 return -ENODEV;
 445
 446         if (skb_vlan_tag_present(skb)) {
 447                 nskb = skb_clone(skb, GFP_ATOMIC);
 448                 if (!nskb)
 449                         return -ENOMEM;
 450
 451                 nskb = __vlan_hwaccel_push_inside(nskb);
 452                 if (!nskb)
 453                         return -ENOMEM;
 454
 455                 skb = nskb;
 456         }
 457
 458         if (nla_attr_size(skb->len) > USHRT_MAX) {
 459                 err = -EFBIG;
 460                 goto out;
 461         }
 462
 463         /* Complete checksum if needed */
 464         if (skb->ip_summed == CHECKSUM_PARTIAL &&
 465             (err = skb_checksum_help(skb)))
 466                 goto out;
 467
 468         /* Older versions of OVS user space enforce alignment of the last
 469          * Netlink attribute to NLA_ALIGNTO which would require extensive
 470          * padding logic. Only perform zerocopy if padding is not required.
 471          */
 472         if (dp->user_features & OVS_DP_F_UNALIGNED)
 473                 hlen = skb_zerocopy_headlen(skb);
 474         else
 475                 hlen = skb->len;
 476
 477         len = upcall_msg_size(upcall_info, hlen - cutlen);
 478         user_skb = genlmsg_new(len, GFP_ATOMIC);
 479         if (!user_skb) {
 480                 err = -ENOMEM;
 481                 goto out;
 482         }
 483
 484         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 485                              0, upcall_info->cmd);
 486         upcall->dp_ifindex = dp_ifindex;
 487
 488         err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
 489         BUG_ON(err);
 490
 491         if (upcall_info->userdata)
 492                 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
 493                           nla_len(upcall_info->userdata),
 494                           nla_data(upcall_info->userdata));
 495
 496
 497         if (upcall_info->egress_tun_info) {
 498                 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_EGRESS_TUN_KEY);
 499                 err = ovs_nla_put_tunnel_info(user_skb,
 500                                               upcall_info->egress_tun_info);
 501                 BUG_ON(err);
 502                 nla_nest_end(user_skb, nla);
 503         }
 504
 505         if (upcall_info->actions_len) {
 506                 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_ACTIONS);
 507                 err = ovs_nla_put_actions(upcall_info->actions,
 508                                           upcall_info->actions_len,
 509                                           user_skb);
 510                 if (!err)
 511                         nla_nest_end(user_skb, nla);
 512                 else
 513                         nla_nest_cancel(user_skb, nla);
 514         }
 515
 516         /* Add OVS_PACKET_ATTR_MRU */
 517         if (upcall_info->mru) {
 518                 if (nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU,
 519                                 upcall_info->mru)) {
 520                         err = -ENOBUFS;
 521                         goto out;
 522                 }
 523                 pad_packet(dp, user_skb);
 524         }
 525
 526         /* Add OVS_PACKET_ATTR_LEN when packet is truncated */
 527         if (cutlen > 0) {
 528                 if (nla_put_u32(user_skb, OVS_PACKET_ATTR_LEN,
 529                                 skb->len)) {
 530                         err = -ENOBUFS;
 531                         goto out;
 532                 }
 533                 pad_packet(dp, user_skb);
 534         }
 535
 536         /* Only reserve room for attribute header, packet data is added
 537          * in skb_zerocopy()
 538          */
 539         if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
 540                 err = -ENOBUFS;
 541                 goto out;
 542         }
 543         nla->nla_len = nla_attr_size(skb->len - cutlen);
 544
 545         err = skb_zerocopy(user_skb, skb, skb->len - cutlen, hlen);
 546         if (err)
 547                 goto out;
 548
 549         /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
 550         pad_packet(dp, user_skb);
 551
 552         ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
 553
 554         err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
 555         user_skb = NULL;
 556 out:
 557         if (err)
 558                 skb_tx_error(skb);
 559         kfree_skb(user_skb);
 560         kfree_skb(nskb);
 561         return err;
 562 }
 563
 564 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 565 {
 566         struct ovs_header *ovs_header = info->userhdr;
 567         struct net *net = sock_net(skb->sk);
 568         struct nlattr **a = info->attrs;
 569         struct sw_flow_actions *acts;
 570         struct sk_buff *packet;
 571         struct sw_flow *flow;
 572         struct sw_flow_actions *sf_acts;
 573         struct datapath *dp;
 574         struct ethhdr *eth;
 575         struct vport *input_vport;
 576         u16 mru = 0;
 577         int len;
 578         int err;
 579         bool log = !a[OVS_PACKET_ATTR_PROBE];
 580
 581         err = -EINVAL;
 582         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 583             !a[OVS_PACKET_ATTR_ACTIONS])
 584                 goto err;
 585
 586         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 587         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 588         err = -ENOMEM;
 589         if (!packet)
 590                 goto err;
 591         skb_reserve(packet, NET_IP_ALIGN);
 592
 593         nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
 594
 595         skb_reset_mac_header(packet);
 596         eth = eth_hdr(packet);
 597
 598         /* Normally, setting the skb 'protocol' field would be handled by a
 599          * call to eth_type_trans(), but it assumes there's a sending
 600          * device, which we may not have.
 601          */
 602         if (eth_proto_is_802_3(eth->h_proto))
 603                 packet->protocol = eth->h_proto;
 604         else
 605                 packet->protocol = htons(ETH_P_802_2);
 606
 607         /* Set packet's mru */
 608         if (a[OVS_PACKET_ATTR_MRU]) {
 609                 mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
 610                 packet->ignore_df = 1;
 611         }
 612         OVS_CB(packet)->mru = mru;
 613
 614         /* Build an sw_flow for sending this packet. */
 615         flow = ovs_flow_alloc();
 616         err = PTR_ERR(flow);
 617         if (IS_ERR(flow))
 618                 goto err_kfree_skb;
 619
 620         err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
 621                                              packet, &flow->key, log);
 622         if (err)
 623                 goto err_flow_free;
 624
 625         err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
 626                                    &flow->key, &acts, log);
 627         if (err)
 628                 goto err_flow_free;
 629
 630         rcu_assign_pointer(flow->sf_acts, acts);
 631         packet->priority = flow->key.phy.priority;
 632         packet->mark = flow->key.phy.skb_mark;
 633
 634         rcu_read_lock();
 635         dp = get_dp_rcu(net, ovs_header->dp_ifindex);
 636         err = -ENODEV;
 637         if (!dp)
 638                 goto err_unlock;
 639
 640         input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
 641         if (!input_vport)
 642                 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
 643
 644         if (!input_vport)
 645                 goto err_unlock;
 646
 647         packet->dev = input_vport->dev;
 648         OVS_CB(packet)->input_vport = input_vport;
 649         sf_acts = rcu_dereference(flow->sf_acts);
 650
 651         local_bh_disable();
 652         err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
 653         local_bh_enable();
 654         rcu_read_unlock();
 655
 656         ovs_flow_free(flow, false);
 657         return err;
 658
 659 err_unlock:
 660         rcu_read_unlock();
 661 err_flow_free:
 662         ovs_flow_free(flow, false);
 663 err_kfree_skb:
 664         kfree_skb(packet);
 665 err:
 666         return err;
 667 }
 668
 669 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 670         [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
 671         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 672         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 673         [OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
 674         [OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
 675 };
 676
 677 static struct genl_ops dp_packet_genl_ops[] = {
 678         { .cmd = OVS_PACKET_CMD_EXECUTE,
 679           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 680           .policy = packet_policy,
 681           .doit = ovs_packet_cmd_execute
 682         }
 683 };
 684
 685 static struct genl_family dp_packet_genl_family = {
 686         .id = GENL_ID_GENERATE,
 687         .hdrsize = sizeof(struct ovs_header),
 688         .name = OVS_PACKET_FAMILY,
 689         .version = OVS_PACKET_VERSION,
 690         .maxattr = OVS_PACKET_ATTR_MAX,
 691         .netnsok = true,
 692         .parallel_ops = true,
 693         .ops = dp_packet_genl_ops,
 694         .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
 695 };
 696
 697 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
 698                          struct ovs_dp_megaflow_stats *mega_stats)
 699 {
 700         int i;
 701
 702         memset(mega_stats, 0, sizeof(*mega_stats));
 703
 704         stats->n_flows = ovs_flow_tbl_count(&dp->table);
 705         mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
 706
 707         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 708
 709         for_each_possible_cpu(i) {
 710                 const struct dp_stats_percpu *percpu_stats;
 711                 struct dp_stats_percpu local_stats;
 712                 unsigned int start;
 713
 714                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 715
 716                 do {
 717                         start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
 718                         local_stats = *percpu_stats;
 719                 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
 720
 721                 stats->n_hit += local_stats.n_hit;
 722                 stats->n_missed += local_stats.n_missed;
 723                 stats->n_lost += local_stats.n_lost;
 724                 mega_stats->n_mask_hit += local_stats.n_mask_hit;
 725         }
 726 }
 727
 728 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
 729 {
 730         return ovs_identifier_is_ufid(sfid) &&
 731                !(ufid_flags & OVS_UFID_F_OMIT_KEY);
 732 }
 733
 734 static bool should_fill_mask(uint32_t ufid_flags)
 735 {
 736         return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
 737 }
 738
 739 static bool should_fill_actions(uint32_t ufid_flags)
 740 {
 741         return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
 742 }
 743
 744 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
 745                                     const struct sw_flow_id *sfid,
 746                                     uint32_t ufid_flags)
 747 {
 748         size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
 749
 750         /* OVS_FLOW_ATTR_UFID */
 751         if (sfid && ovs_identifier_is_ufid(sfid))
 752                 len += nla_total_size(sfid->ufid_len);
 753
 754         /* OVS_FLOW_ATTR_KEY */
 755         if (!sfid || should_fill_key(sfid, ufid_flags))
 756                 len += nla_total_size(ovs_key_attr_size());
 757
 758         /* OVS_FLOW_ATTR_MASK */
 759         if (should_fill_mask(ufid_flags))
 760                 len += nla_total_size(ovs_key_attr_size());
 761
 762         /* OVS_FLOW_ATTR_ACTIONS */
 763         if (should_fill_actions(ufid_flags))
 764                 len += nla_total_size(acts->orig_len);
 765
 766         return len
 767                 + nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
 768                 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
 769                 + nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */
 770 }
 771
 772 /* Called with ovs_mutex or RCU read lock. */
 773 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
 774                                    struct sk_buff *skb)
 775 {
 776         struct ovs_flow_stats stats;
 777         __be16 tcp_flags;
 778         unsigned long used;
 779
 780         ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
 781
 782         if (used &&
 783             nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
 784                               OVS_FLOW_ATTR_PAD))
 785                 return -EMSGSIZE;
 786
 787         if (stats.n_packets &&
 788             nla_put_64bit(skb, OVS_FLOW_ATTR_STATS,
 789                           sizeof(struct ovs_flow_stats), &stats,
 790                           OVS_FLOW_ATTR_PAD))
 791                 return -EMSGSIZE;
 792
 793         if ((u8)ntohs(tcp_flags) &&
 794              nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
 795                 return -EMSGSIZE;
 796
 797         return 0;
 798 }
 799
 800 /* Called with ovs_mutex or RCU read lock. */
 801 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
 802                                      struct sk_buff *skb, int skb_orig_len)
 803 {
 804         struct nlattr *start;
 805         int err;
 806
 807         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 808          * this is the first flow to be dumped into 'skb'.  This is unusual for
 809          * Netlink but individual action lists can be longer than
 810          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 811          * The userspace caller can always fetch the actions separately if it
 812          * really wants them.  (Most userspace callers in fact don't care.)
 813          *
 814          * This can only fail for dump operations because the skb is always
 815          * properly sized for single flows.
 816          */
 817         start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
 818         if (start) {
 819                 const struct sw_flow_actions *sf_acts;
 820
 821                 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
 822                 err = ovs_nla_put_actions(sf_acts->actions,
 823                                           sf_acts->actions_len, skb);
 824
 825                 if (!err)
 826                         nla_nest_end(skb, start);
 827                 else {
 828                         if (skb_orig_len)
 829                                 return err;
 830
 831                         nla_nest_cancel(skb, start);
 832                 }
 833         } else if (skb_orig_len) {
 834                 return -EMSGSIZE;
 835         }
 836
 837         return 0;
 838 }
 839
 840 /* Called with ovs_mutex or RCU read lock. */
 841 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
 842                                   struct sk_buff *skb, u32 portid,
 843                                   u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
 844 {
 845         const int skb_orig_len = skb->len;
 846         struct ovs_header *ovs_header;
 847         int err;
 848
 849         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
 850                                  flags, cmd);
 851         if (!ovs_header)
 852                 return -EMSGSIZE;
 853
 854         ovs_header->dp_ifindex = dp_ifindex;
 855
 856         err = ovs_nla_put_identifier(flow, skb);
 857         if (err)
 858                 goto error;
 859
 860         if (should_fill_key(&flow->id, ufid_flags)) {
 861                 err = ovs_nla_put_masked_key(flow, skb);
 862                 if (err)
 863                         goto error;
 864         }
 865
 866         if (should_fill_mask(ufid_flags)) {
 867                 err = ovs_nla_put_mask(flow, skb);
 868                 if (err)
 869                         goto error;
 870         }
 871
 872         err = ovs_flow_cmd_fill_stats(flow, skb);
 873         if (err)
 874                 goto error;
 875
 876         if (should_fill_actions(ufid_flags)) {
 877                 err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
 878                 if (err)
 879                         goto error;
 880         }
 881
 882         genlmsg_end(skb, ovs_header);
 883         return 0;
 884
 885 error:
 886         genlmsg_cancel(skb, ovs_header);
 887         return err;
 888 }
 889
 890 /* May not be called with RCU read lock. */
 891 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
 892                                                const struct sw_flow_id *sfid,
 893                                                struct genl_info *info,
 894                                                bool always,
 895                                                uint32_t ufid_flags)
 896 {
 897         struct sk_buff *skb;
 898         size_t len;
 899
 900         if (!always && !ovs_must_notify(&dp_flow_genl_family, info,
 901                                         GROUP_ID(&ovs_dp_flow_multicast_group)))
 902                 return NULL;
 903
 904         len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
 905         skb = genlmsg_new(len, GFP_KERNEL);
 906         if (!skb)
 907                 return ERR_PTR(-ENOMEM);
 908
 909         return skb;
 910 }
 911
 912 /* Called with ovs_mutex. */
 913 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
 914                                                int dp_ifindex,
 915                                                struct genl_info *info, u8 cmd,
 916                                                bool always, u32 ufid_flags)
 917 {
 918         struct sk_buff *skb;
 919         int retval;
 920
 921         skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
 922                                       &flow->id, info, always, ufid_flags);
 923         if (IS_ERR_OR_NULL(skb))
 924                 return skb;
 925
 926         retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
 927                                         info->snd_portid, info->snd_seq, 0,
 928                                         cmd, ufid_flags);
 929         BUG_ON(retval < 0);
 930         return skb;
 931 }
 932
 933 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
 934 {
 935         struct net *net = sock_net(skb->sk);
 936         struct nlattr **a = info->attrs;
 937         struct ovs_header *ovs_header = info->userhdr;
 938         struct sw_flow *flow = NULL, *new_flow;
 939         struct sw_flow_mask mask;
 940         struct sk_buff *reply;
 941         struct datapath *dp;
 942         struct sw_flow_key key;
 943         struct sw_flow_actions *acts;
 944         struct sw_flow_match match;
 945         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
 946         int error;
 947         bool log = !a[OVS_FLOW_ATTR_PROBE];
 948
 949         /* Must have key and actions. */
 950         error = -EINVAL;
 951         if (!a[OVS_FLOW_ATTR_KEY]) {
 952                 OVS_NLERR(log, "Flow key attr not present in new flow.");
 953                 goto error;
 954         }
 955         if (!a[OVS_FLOW_ATTR_ACTIONS]) {
 956                 OVS_NLERR(log, "Flow actions attr not present in new flow.");
 957                 goto error;
 958         }
 959
 960         /* Most of the time we need to allocate a new flow, do it before
 961          * locking.
 962          */
 963         new_flow = ovs_flow_alloc();
 964         if (IS_ERR(new_flow)) {
 965                 error = PTR_ERR(new_flow);
 966                 goto error;
 967         }
 968
 969         /* Extract key. */
 970         ovs_match_init(&match, &key, &mask);
 971         error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
 972                                   a[OVS_FLOW_ATTR_MASK], log);
 973         if (error)
 974                 goto err_kfree_flow;
 975
 976         ovs_flow_mask_key(&new_flow->key, &key, true, &mask);
 977
 978         /* Extract flow identifier. */
 979         error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
 980                                        &key, log);
 981         if (error)
 982                 goto err_kfree_flow;
 983
 984         /* Validate actions. */
 985         error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
 986                                      &new_flow->key, &acts, log);
 987         if (error) {
 988                 OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
 989                 goto err_kfree_flow;
 990         }
 991
 992         reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
 993                                         ufid_flags);
 994         if (IS_ERR(reply)) {
 995                 error = PTR_ERR(reply);
 996                 goto err_kfree_acts;
 997         }
 998
 999         ovs_lock();
1000         dp = get_dp(net, ovs_header->dp_ifindex);
1001         if (unlikely(!dp)) {
1002                 error = -ENODEV;
1003                 goto err_unlock_ovs;
1004         }
1005
1006         /* Check if this is a duplicate flow */
1007         if (ovs_identifier_is_ufid(&new_flow->id))
1008                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
1009         if (!flow)
1010                 flow = ovs_flow_tbl_lookup(&dp->table, &key);
1011         if (likely(!flow)) {
1012                 rcu_assign_pointer(new_flow->sf_acts, acts);
1013
1014                 /* Put flow in bucket. */
1015                 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
1016                 if (unlikely(error)) {
1017                         acts = NULL;
1018                         goto err_unlock_ovs;
1019                 }
1020
1021                 if (unlikely(reply)) {
1022                         error = ovs_flow_cmd_fill_info(new_flow,
1023                                                        ovs_header->dp_ifindex,
1024                                                        reply, info->snd_portid,
1025                                                        info->snd_seq, 0,
1026                                                        OVS_FLOW_CMD_NEW,
1027                                                        ufid_flags);
1028                         BUG_ON(error < 0);
1029                 }
1030                 ovs_unlock();
1031         } else {
1032                 struct sw_flow_actions *old_acts;
1033
1034                 /* Bail out if we're not allowed to modify an existing flow.
1035                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1036                  * because Generic Netlink treats the latter as a dump
1037                  * request.  We also accept NLM_F_EXCL in case that bug ever
1038                  * gets fixed.
1039                  */
1040                 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1041                                                          | NLM_F_EXCL))) {
1042                         error = -EEXIST;
1043                         goto err_unlock_ovs;
1044                 }
1045                 /* The flow identifier has to be the same for flow updates.
1046                  * Look for any overlapping flow.
1047                  */
1048                 if (unlikely(!ovs_flow_cmp(flow, &match))) {
1049                         if (ovs_identifier_is_key(&flow->id))
1050                                 flow = ovs_flow_tbl_lookup_exact(&dp->table,
1051                                                                  &match);
1052                         else /* UFID matches but key is different */
1053                                 flow = NULL;
1054                         if (!flow) {
1055                                 error = -ENOENT;
1056                                 goto err_unlock_ovs;
1057                         }
1058                 }
1059                 /* Update actions. */
1060                 old_acts = ovsl_dereference(flow->sf_acts);
1061                 rcu_assign_pointer(flow->sf_acts, acts);
1062
1063                 if (unlikely(reply)) {
1064                         error = ovs_flow_cmd_fill_info(flow,
1065                                                        ovs_header->dp_ifindex,
1066                                                        reply, info->snd_portid,
1067                                                        info->snd_seq, 0,
1068                                                        OVS_FLOW_CMD_NEW,
1069                                                        ufid_flags);
1070                         BUG_ON(error < 0);
1071                 }
1072                 ovs_unlock();
1073
1074                 ovs_nla_free_flow_actions_rcu(old_acts);
1075                 ovs_flow_free(new_flow, false);
1076         }
1077
1078         if (reply)
1079                 ovs_notify(&dp_flow_genl_family, &ovs_dp_flow_multicast_group, reply, info);
1080         return 0;
1081
1082 err_unlock_ovs:
1083         ovs_unlock();
1084         kfree_skb(reply);
1085 err_kfree_acts:
1086         ovs_nla_free_flow_actions(acts);
1087 err_kfree_flow:
1088         ovs_flow_free(new_flow, false);
1089 error:
1090         return error;
1091 }
1092
1093 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1094 static struct sw_flow_actions *get_flow_actions(struct net *net,
1095                                                 const struct nlattr *a,
1096                                                 const struct sw_flow_key *key,
1097                                                 const struct sw_flow_mask *mask,
1098                                                 bool log)
1099 {
1100         struct sw_flow_actions *acts;
1101         struct sw_flow_key masked_key;
1102         int error;
1103
1104         ovs_flow_mask_key(&masked_key, key, true, mask);
1105         error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1106         if (error) {
1107                 OVS_NLERR(log,
1108                           "Actions may not be safe on all matching packets");
1109                 return ERR_PTR(error);
1110         }
1111
1112         return acts;
1113 }
1114
1115 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1116 {
1117         struct net *net = sock_net(skb->sk);
1118         struct nlattr **a = info->attrs;
1119         struct ovs_header *ovs_header = info->userhdr;
1120         struct sw_flow_key key;
1121         struct sw_flow *flow;
1122         struct sw_flow_mask mask;
1123         struct sk_buff *reply = NULL;
1124         struct datapath *dp;
1125         struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1126         struct sw_flow_match match;
1127         struct sw_flow_id sfid;
1128         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1129         int error = 0;
1130         bool log = !a[OVS_FLOW_ATTR_PROBE];
1131         bool ufid_present;
1132
1133         ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1134         if (a[OVS_FLOW_ATTR_KEY]) {
1135                 ovs_match_init(&match, &key, &mask);
1136                 error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1137                                           a[OVS_FLOW_ATTR_MASK], log);
1138         } else if (!ufid_present) {
1139                 OVS_NLERR(log,
1140                           "Flow set message rejected, Key attribute missing.");
1141                 error = -EINVAL;
1142         }
1143         if (error)
1144                 goto error;
1145
1146         /* Validate actions. */
1147         if (a[OVS_FLOW_ATTR_ACTIONS]) {
1148                 if (!a[OVS_FLOW_ATTR_KEY]) {
1149                         OVS_NLERR(log,
1150                                   "Flow key attribute not present in set flow.");
1151                         error = -EINVAL;
1152                         goto error;
1153                 }
1154
1155                 acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], &key,
1156                                         &mask, log);
1157                 if (IS_ERR(acts)) {
1158                         error = PTR_ERR(acts);
1159                         goto error;
1160                 }
1161
1162                 /* Can allocate before locking if have acts. */
1163                 reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1164                                                 ufid_flags);
1165                 if (IS_ERR(reply)) {
1166                         error = PTR_ERR(reply);
1167                         goto err_kfree_acts;
1168                 }
1169         }
1170
1171         ovs_lock();
1172         dp = get_dp(net, ovs_header->dp_ifindex);
1173         if (unlikely(!dp)) {
1174                 error = -ENODEV;
1175                 goto err_unlock_ovs;
1176         }
1177         /* Check that the flow exists. */
1178         if (ufid_present)
1179                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1180         else
1181                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1182         if (unlikely(!flow)) {
1183                 error = -ENOENT;
1184                 goto err_unlock_ovs;
1185         }
1186
1187         /* Update actions, if present. */
1188         if (likely(acts)) {
1189                 old_acts = ovsl_dereference(flow->sf_acts);
1190                 rcu_assign_pointer(flow->sf_acts, acts);
1191
1192                 if (unlikely(reply)) {
1193                         error = ovs_flow_cmd_fill_info(flow,
1194                                                        ovs_header->dp_ifindex,
1195                                                        reply, info->snd_portid,
1196                                                        info->snd_seq, 0,
1197                                                        OVS_FLOW_CMD_NEW,
1198                                                        ufid_flags);
1199                         BUG_ON(error < 0);
1200                 }
1201         } else {
1202                 /* Could not alloc without acts before locking. */
1203                 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1204                                                 info, OVS_FLOW_CMD_NEW, false,
1205                                                 ufid_flags);
1206
1207                 if (unlikely(IS_ERR(reply))) {
1208                         error = PTR_ERR(reply);
1209                         goto err_unlock_ovs;
1210                 }
1211         }
1212
1213         /* Clear stats. */
1214         if (a[OVS_FLOW_ATTR_CLEAR])
1215                 ovs_flow_stats_clear(flow);
1216         ovs_unlock();
1217
1218         if (reply)
1219                 ovs_notify(&dp_flow_genl_family, &ovs_dp_flow_multicast_group, reply, info);
1220         if (old_acts)
1221                 ovs_nla_free_flow_actions_rcu(old_acts);
1222
1223         return 0;
1224
1225 err_unlock_ovs:
1226         ovs_unlock();
1227         kfree_skb(reply);
1228 err_kfree_acts:
1229         ovs_nla_free_flow_actions(acts);
1230 error:
1231         return error;
1232 }
1233
1234 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1235 {
1236         struct nlattr **a = info->attrs;
1237         struct ovs_header *ovs_header = info->userhdr;
1238         struct net *net = sock_net(skb->sk);
1239         struct sw_flow_key key;
1240         struct sk_buff *reply;
1241         struct sw_flow *flow;
1242         struct datapath *dp;
1243         struct sw_flow_match match;
1244         struct sw_flow_id ufid;
1245         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1246         int err = 0;
1247         bool log = !a[OVS_FLOW_ATTR_PROBE];
1248         bool ufid_present;
1249
1250         ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1251         if (a[OVS_FLOW_ATTR_KEY]) {
1252                 ovs_match_init(&match, &key, NULL);
1253                 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1254                                         log);
1255         } else if (!ufid_present) {
1256                 OVS_NLERR(log,
1257                           "Flow get message rejected, Key attribute missing.");
1258                 err = -EINVAL;
1259         }
1260         if (err)
1261                 return err;
1262
1263         ovs_lock();
1264         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1265         if (!dp) {
1266                 err = -ENODEV;
1267                 goto unlock;
1268         }
1269
1270         if (ufid_present)
1271                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1272         else
1273                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1274         if (!flow) {
1275                 err = -ENOENT;
1276                 goto unlock;
1277         }
1278
1279         reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1280                                         OVS_FLOW_CMD_NEW, true, ufid_flags);
1281         if (IS_ERR(reply)) {
1282                 err = PTR_ERR(reply);
1283                 goto unlock;
1284         }
1285
1286         ovs_unlock();
1287         return genlmsg_reply(reply, info);
1288 unlock:
1289         ovs_unlock();
1290         return err;
1291 }
1292
1293 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1294 {
1295         struct nlattr **a = info->attrs;
1296         struct ovs_header *ovs_header = info->userhdr;
1297         struct net *net = sock_net(skb->sk);
1298         struct sw_flow_key key;
1299         struct sk_buff *reply;
1300         struct sw_flow *flow = NULL;
1301         struct datapath *dp;
1302         struct sw_flow_match match;
1303         struct sw_flow_id ufid;
1304         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1305         int err;
1306         bool log = !a[OVS_FLOW_ATTR_PROBE];
1307         bool ufid_present;
1308
1309         ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1310         if (a[OVS_FLOW_ATTR_KEY]) {
1311                 ovs_match_init(&match, &key, NULL);
1312                 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1313                                         NULL, log);
1314                 if (unlikely(err))
1315                         return err;
1316         }
1317
1318         ovs_lock();
1319         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1320         if (unlikely(!dp)) {
1321                 err = -ENODEV;
1322                 goto unlock;
1323         }
1324
1325         if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1326                 err = ovs_flow_tbl_flush(&dp->table);
1327                 goto unlock;
1328         }
1329
1330         if (ufid_present)
1331                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1332         else
1333                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1334         if (unlikely(!flow)) {
1335                 err = -ENOENT;
1336                 goto unlock;
1337         }
1338
1339         ovs_flow_tbl_remove(&dp->table, flow);
1340         ovs_unlock();
1341
1342         reply = ovs_flow_cmd_alloc_info(rcu_dereference_raw(flow->sf_acts),
1343                                         &flow->id, info, false, ufid_flags);
1344
1345         if (likely(reply)) {
1346                 if (likely(!IS_ERR(reply))) {
1347                         rcu_read_lock();        /*To keep RCU checker happy. */
1348                         err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1349                                                      reply, info->snd_portid,
1350                                                      info->snd_seq, 0,
1351                                                      OVS_FLOW_CMD_DEL,
1352                                                      ufid_flags);
1353                         rcu_read_unlock();
1354                         BUG_ON(err < 0);
1355                         ovs_notify(&dp_flow_genl_family, &ovs_dp_flow_multicast_group, reply, info);
1356                 } else {
1357                         genl_set_err(&dp_flow_genl_family, sock_net(skb->sk), 0,
1358                                      GROUP_ID(&ovs_dp_flow_multicast_group), PTR_ERR(reply));
1359
1360                 }
1361         }
1362
1363         ovs_flow_free(flow, true);
1364         return 0;
1365 unlock:
1366         ovs_unlock();
1367         return err;
1368 }
1369
1370 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1371 {
1372         struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1373         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1374         struct table_instance *ti;
1375         struct datapath *dp;
1376         u32 ufid_flags;
1377         int err;
1378
1379         err = genlmsg_parse(cb->nlh, &dp_flow_genl_family, a,
1380                             OVS_FLOW_ATTR_MAX, flow_policy);
1381         if (err)
1382                 return err;
1383         ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1384
1385         rcu_read_lock();
1386         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1387         if (!dp) {
1388                 rcu_read_unlock();
1389                 return -ENODEV;
1390         }
1391
1392         ti = rcu_dereference(dp->table.ti);
1393         for (;;) {
1394                 struct sw_flow *flow;
1395                 u32 bucket, obj;
1396
1397                 bucket = cb->args[0];
1398                 obj = cb->args[1];
1399                 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1400                 if (!flow)
1401                         break;
1402
1403                 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1404                                            NETLINK_CB(cb->skb).portid,
1405                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1406                                            OVS_FLOW_CMD_NEW, ufid_flags) < 0)
1407                         break;
1408
1409                 cb->args[0] = bucket;
1410                 cb->args[1] = obj;
1411         }
1412         rcu_read_unlock();
1413         return skb->len;
1414 }
1415
1416 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1417         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1418         [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1419         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1420         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1421         [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1422         [OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1423         [OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1424 };
1425
1426 static struct genl_ops dp_flow_genl_ops[] = {
1427         { .cmd = OVS_FLOW_CMD_NEW,
1428           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1429           .policy = flow_policy,
1430           .doit = ovs_flow_cmd_new
1431         },
1432         { .cmd = OVS_FLOW_CMD_DEL,
1433           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1434           .policy = flow_policy,
1435           .doit = ovs_flow_cmd_del
1436         },
1437         { .cmd = OVS_FLOW_CMD_GET,
1438           .flags = 0,               /* OK for unprivileged users. */
1439           .policy = flow_policy,
1440           .doit = ovs_flow_cmd_get,
1441           .dumpit = ovs_flow_cmd_dump
1442         },
1443         { .cmd = OVS_FLOW_CMD_SET,
1444           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1445           .policy = flow_policy,
1446           .doit = ovs_flow_cmd_set,
1447         },
1448 };
1449
1450 static struct genl_family dp_flow_genl_family = {
1451         .id = GENL_ID_GENERATE,
1452         .hdrsize = sizeof(struct ovs_header),
1453         .name = OVS_FLOW_FAMILY,
1454         .version = OVS_FLOW_VERSION,
1455         .maxattr = OVS_FLOW_ATTR_MAX,
1456         .netnsok = true,
1457         .parallel_ops = true,
1458         .ops = dp_flow_genl_ops,
1459         .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1460         .mcgrps = &ovs_dp_flow_multicast_group,
1461         .n_mcgrps = 1,
1462 };
1463
1464 static size_t ovs_dp_cmd_msg_size(void)
1465 {
1466         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1467
1468         msgsize += nla_total_size(IFNAMSIZ);
1469         msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1470         msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
1471         msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1472
1473         return msgsize;
1474 }
1475
1476 /* Called with ovs_mutex. */
1477 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1478                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1479 {
1480         struct ovs_header *ovs_header;
1481         struct ovs_dp_stats dp_stats;
1482         struct ovs_dp_megaflow_stats dp_megaflow_stats;
1483         int err;
1484
1485         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1486                                    flags, cmd);
1487         if (!ovs_header)
1488                 goto error;
1489
1490         ovs_header->dp_ifindex = get_dpifindex(dp);
1491
1492         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1493         if (err)
1494                 goto nla_put_failure;
1495
1496         get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1497         if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1498                           &dp_stats, OVS_DP_ATTR_PAD))
1499                 goto nla_put_failure;
1500
1501         if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1502                           sizeof(struct ovs_dp_megaflow_stats),
1503                           &dp_megaflow_stats, OVS_DP_ATTR_PAD))
1504                 goto nla_put_failure;
1505
1506         if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1507                 goto nla_put_failure;
1508
1509         genlmsg_end(skb, ovs_header);
1510         return 0;
1511
1512 nla_put_failure:
1513         genlmsg_cancel(skb, ovs_header);
1514 error:
1515         return -EMSGSIZE;
1516 }
1517
1518 static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1519 {
1520         return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1521 }
1522
1523 /* Called with rcu_read_lock or ovs_mutex. */
1524 static struct datapath *lookup_datapath(struct net *net,
1525                                         const struct ovs_header *ovs_header,
1526                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1527 {
1528         struct datapath *dp;
1529
1530         if (!a[OVS_DP_ATTR_NAME])
1531                 dp = get_dp(net, ovs_header->dp_ifindex);
1532         else {
1533                 struct vport *vport;
1534
1535                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1536                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1537         }
1538         return dp ? dp : ERR_PTR(-ENODEV);
1539 }
1540
1541 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1542 {
1543         struct datapath *dp;
1544
1545         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1546         if (IS_ERR(dp))
1547                 return;
1548
1549         WARN(dp->user_features, "Dropping previously announced user features\n");
1550         dp->user_features = 0;
1551 }
1552
1553 static void ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1554 {
1555         if (a[OVS_DP_ATTR_USER_FEATURES])
1556                 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1557 }
1558
1559 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1560 {
1561         struct nlattr **a = info->attrs;
1562         struct vport_parms parms;
1563         struct sk_buff *reply;
1564         struct datapath *dp;
1565         struct vport *vport;
1566         struct ovs_net *ovs_net;
1567         int err, i;
1568
1569         err = -EINVAL;
1570         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1571                 goto err;
1572
1573         reply = ovs_dp_cmd_alloc_info();
1574         if (!reply)
1575                 return -ENOMEM;
1576
1577         err = -ENOMEM;
1578         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1579         if (dp == NULL)
1580                 goto err_free_reply;
1581
1582         ovs_dp_set_net(dp, sock_net(skb->sk));
1583
1584         /* Allocate table. */
1585         err = ovs_flow_tbl_init(&dp->table);
1586         if (err)
1587                 goto err_free_dp;
1588
1589         dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1590         if (!dp->stats_percpu) {
1591                 err = -ENOMEM;
1592                 goto err_destroy_table;
1593         }
1594
1595         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1596                             GFP_KERNEL);
1597         if (!dp->ports) {
1598                 err = -ENOMEM;
1599                 goto err_destroy_percpu;
1600         }
1601
1602         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1603                 INIT_HLIST_HEAD(&dp->ports[i]);
1604
1605         /* Set up our datapath device. */
1606         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1607         parms.type = OVS_VPORT_TYPE_INTERNAL;
1608         parms.options = NULL;
1609         parms.dp = dp;
1610         parms.port_no = OVSP_LOCAL;
1611         parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1612
1613         ovs_dp_change(dp, a);
1614
1615         /* So far only local changes have been made, now need the lock. */
1616         ovs_lock();
1617
1618         vport = new_vport(&parms);
1619         if (IS_ERR(vport)) {
1620                 err = PTR_ERR(vport);
1621                 if (err == -EBUSY)
1622                         err = -EEXIST;
1623
1624                 if (err == -EEXIST) {
1625                         /* An outdated user space instance that does not understand
1626                          * the concept of user_features has attempted to create a new
1627                          * datapath and is likely to reuse it. Drop all user features.
1628                          */
1629                         if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1630                                 ovs_dp_reset_user_features(skb, info);
1631                 }
1632
1633                 goto err_destroy_ports_array;
1634         }
1635
1636         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1637                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1638         BUG_ON(err < 0);
1639
1640         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1641         list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1642
1643         ovs_unlock();
1644
1645         ovs_notify(&dp_datapath_genl_family, &ovs_dp_datapath_multicast_group, reply, info);
1646         return 0;
1647
1648 err_destroy_ports_array:
1649         ovs_unlock();
1650         kfree(dp->ports);
1651 err_destroy_percpu:
1652         free_percpu(dp->stats_percpu);
1653 err_destroy_table:
1654         ovs_flow_tbl_destroy(&dp->table);
1655 err_free_dp:
1656         kfree(dp);
1657 err_free_reply:
1658         kfree_skb(reply);
1659 err:
1660         return err;
1661 }
1662
1663 /* Called with ovs_mutex. */
1664 static void __dp_destroy(struct datapath *dp)
1665 {
1666         int i;
1667
1668         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1669                 struct vport *vport;
1670                 struct hlist_node *n;
1671
1672                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1673                         if (vport->port_no != OVSP_LOCAL)
1674                                 ovs_dp_detach_port(vport);
1675         }
1676
1677         list_del_rcu(&dp->list_node);
1678
1679         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1680          * all ports in datapath are destroyed first before freeing datapath.
1681          */
1682         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1683
1684         /* RCU destroy the flow table */
1685         call_rcu(&dp->rcu, destroy_dp_rcu);
1686 }
1687
1688 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1689 {
1690         struct sk_buff *reply;
1691         struct datapath *dp;
1692         int err;
1693
1694         reply = ovs_dp_cmd_alloc_info();
1695         if (!reply)
1696                 return -ENOMEM;
1697
1698         ovs_lock();
1699         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1700         err = PTR_ERR(dp);
1701         if (IS_ERR(dp))
1702                 goto err_unlock_free;
1703
1704         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1705                                    info->snd_seq, 0, OVS_DP_CMD_DEL);
1706         BUG_ON(err < 0);
1707
1708         __dp_destroy(dp);
1709         ovs_unlock();
1710
1711         ovs_notify(&dp_datapath_genl_family, &ovs_dp_datapath_multicast_group, reply, info);
1712         return 0;
1713
1714 err_unlock_free:
1715         ovs_unlock();
1716         kfree_skb(reply);
1717         return err;
1718 }
1719
1720 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1721 {
1722         struct sk_buff *reply;
1723         struct datapath *dp;
1724         int err;
1725
1726         reply = ovs_dp_cmd_alloc_info();
1727         if (!reply)
1728                 return -ENOMEM;
1729
1730         ovs_lock();
1731         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1732         err = PTR_ERR(dp);
1733         if (IS_ERR(dp))
1734                 goto err_unlock_free;
1735
1736         ovs_dp_change(dp, info->attrs);
1737
1738         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1739                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1740         BUG_ON(err < 0);
1741
1742         ovs_unlock();
1743
1744         ovs_notify(&dp_datapath_genl_family, &ovs_dp_datapath_multicast_group, reply, info);
1745         return 0;
1746
1747 err_unlock_free:
1748         ovs_unlock();
1749         kfree_skb(reply);
1750         return err;
1751 }
1752
1753 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1754 {
1755         struct sk_buff *reply;
1756         struct datapath *dp;
1757         int err;
1758
1759         reply = ovs_dp_cmd_alloc_info();
1760         if (!reply)
1761                 return -ENOMEM;
1762
1763         ovs_lock();
1764         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1765         if (IS_ERR(dp)) {
1766                 err = PTR_ERR(dp);
1767                 goto err_unlock_free;
1768         }
1769         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1770                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1771         BUG_ON(err < 0);
1772         ovs_unlock();
1773
1774         return genlmsg_reply(reply, info);
1775
1776 err_unlock_free:
1777         ovs_unlock();
1778         kfree_skb(reply);
1779         return err;
1780 }
1781
1782 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1783 {
1784         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1785         struct datapath *dp;
1786         int skip = cb->args[0];
1787         int i = 0;
1788
1789         ovs_lock();
1790         list_for_each_entry(dp, &ovs_net->dps, list_node) {
1791                 if (i >= skip &&
1792                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1793                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1794                                          OVS_DP_CMD_NEW) < 0)
1795                         break;
1796                 i++;
1797         }
1798         ovs_unlock();
1799
1800         cb->args[0] = i;
1801
1802         return skb->len;
1803 }
1804
1805 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1806         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1807         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1808         [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1809 };
1810
1811 static struct genl_ops dp_datapath_genl_ops[] = {
1812         { .cmd = OVS_DP_CMD_NEW,
1813           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1814           .policy = datapath_policy,
1815           .doit = ovs_dp_cmd_new
1816         },
1817         { .cmd = OVS_DP_CMD_DEL,
1818           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1819           .policy = datapath_policy,
1820           .doit = ovs_dp_cmd_del
1821         },
1822         { .cmd = OVS_DP_CMD_GET,
1823           .flags = 0,               /* OK for unprivileged users. */
1824           .policy = datapath_policy,
1825           .doit = ovs_dp_cmd_get,
1826           .dumpit = ovs_dp_cmd_dump
1827         },
1828         { .cmd = OVS_DP_CMD_SET,
1829           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1830           .policy = datapath_policy,
1831           .doit = ovs_dp_cmd_set,
1832         },
1833 };
1834
1835 static struct genl_family dp_datapath_genl_family = {
1836         .id = GENL_ID_GENERATE,
1837         .hdrsize = sizeof(struct ovs_header),
1838         .name = OVS_DATAPATH_FAMILY,
1839         .version = OVS_DATAPATH_VERSION,
1840         .maxattr = OVS_DP_ATTR_MAX,
1841         .netnsok = true,
1842         .parallel_ops = true,
1843         .ops = dp_datapath_genl_ops,
1844         .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1845         .mcgrps = &ovs_dp_datapath_multicast_group,
1846         .n_mcgrps = 1,
1847 };
1848
1849 /* Called with ovs_mutex or RCU read lock. */
1850 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1851                                    u32 portid, u32 seq, u32 flags, u8 cmd)
1852 {
1853         struct ovs_header *ovs_header;
1854         struct ovs_vport_stats vport_stats;
1855         int err;
1856
1857         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1858                                  flags, cmd);
1859         if (!ovs_header)
1860                 return -EMSGSIZE;
1861
1862         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1863
1864         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1865             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1866             nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1867                            ovs_vport_name(vport)))
1868                 goto nla_put_failure;
1869
1870         ovs_vport_get_stats(vport, &vport_stats);
1871         if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
1872                           sizeof(struct ovs_vport_stats), &vport_stats,
1873                           OVS_VPORT_ATTR_PAD))
1874                 goto nla_put_failure;
1875
1876         if (ovs_vport_get_upcall_portids(vport, skb))
1877                 goto nla_put_failure;
1878
1879         err = ovs_vport_get_options(vport, skb);
1880         if (err == -EMSGSIZE)
1881                 goto error;
1882
1883         genlmsg_end(skb, ovs_header);
1884         return 0;
1885
1886 nla_put_failure:
1887         err = -EMSGSIZE;
1888 error:
1889         genlmsg_cancel(skb, ovs_header);
1890         return err;
1891 }
1892
1893 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1894 {
1895         return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1896 }
1897
1898 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1899 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1900                                          u32 seq, u8 cmd)
1901 {
1902         struct sk_buff *skb;
1903         int retval;
1904
1905         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1906         if (!skb)
1907                 return ERR_PTR(-ENOMEM);
1908
1909         retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1910         BUG_ON(retval < 0);
1911
1912         return skb;
1913 }
1914
1915 /* Called with ovs_mutex or RCU read lock. */
1916 static struct vport *lookup_vport(struct net *net,
1917                                   const struct ovs_header *ovs_header,
1918                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1919 {
1920         struct datapath *dp;
1921         struct vport *vport;
1922
1923         if (a[OVS_VPORT_ATTR_NAME]) {
1924                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1925                 if (!vport)
1926                         return ERR_PTR(-ENODEV);
1927                 if (ovs_header->dp_ifindex &&
1928                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1929                         return ERR_PTR(-ENODEV);
1930                 return vport;
1931         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1932                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1933
1934                 if (port_no >= DP_MAX_PORTS)
1935                         return ERR_PTR(-EFBIG);
1936
1937                 dp = get_dp(net, ovs_header->dp_ifindex);
1938                 if (!dp)
1939                         return ERR_PTR(-ENODEV);
1940
1941                 vport = ovs_vport_ovsl_rcu(dp, port_no);
1942                 if (!vport)
1943                         return ERR_PTR(-ENODEV);
1944                 return vport;
1945         } else
1946                 return ERR_PTR(-EINVAL);
1947 }
1948
1949 /* Called with ovs_mutex */
1950 static void update_headroom(struct datapath *dp)
1951 {
1952         unsigned dev_headroom, max_headroom = 0;
1953         struct net_device *dev;
1954         struct vport *vport;
1955         int i;
1956
1957         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1958                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1959                         dev = vport->dev;
1960                         dev_headroom = netdev_get_fwd_headroom(dev);
1961                         if (dev_headroom > max_headroom)
1962                                 max_headroom = dev_headroom;
1963                 }
1964         }
1965
1966         dp->max_headroom = max_headroom;
1967         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1968                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node)
1969                         netdev_set_rx_headroom(vport->dev, max_headroom);
1970 }
1971
1972 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1973 {
1974         struct nlattr **a = info->attrs;
1975         struct ovs_header *ovs_header = info->userhdr;
1976         struct vport_parms parms;
1977         struct sk_buff *reply;
1978         struct vport *vport;
1979         struct datapath *dp;
1980         u32 port_no;
1981         int err;
1982
1983         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1984             !a[OVS_VPORT_ATTR_UPCALL_PID])
1985                 return -EINVAL;
1986
1987         port_no = a[OVS_VPORT_ATTR_PORT_NO]
1988                 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
1989         if (port_no >= DP_MAX_PORTS)
1990                 return -EFBIG;
1991
1992         reply = ovs_vport_cmd_alloc_info();
1993         if (!reply)
1994                 return -ENOMEM;
1995
1996         ovs_lock();
1997 restart:
1998         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1999         err = -ENODEV;
2000         if (!dp)
2001                 goto exit_unlock_free;
2002
2003         if (port_no) {
2004                 vport = ovs_vport_ovsl(dp, port_no);
2005                 err = -EBUSY;
2006                 if (vport)
2007                         goto exit_unlock_free;
2008         } else {
2009                 for (port_no = 1; ; port_no++) {
2010                         if (port_no >= DP_MAX_PORTS) {
2011                                 err = -EFBIG;
2012                                 goto exit_unlock_free;
2013                         }
2014                         vport = ovs_vport_ovsl(dp, port_no);
2015                         if (!vport)
2016                                 break;
2017                 }
2018         }
2019
2020         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2021         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2022         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2023         parms.dp = dp;
2024         parms.port_no = port_no;
2025         parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
2026
2027         vport = new_vport(&parms);
2028         err = PTR_ERR(vport);
2029         if (IS_ERR(vport)) {
2030                 if (err == -EAGAIN)
2031                         goto restart;
2032                 goto exit_unlock_free;
2033         }
2034
2035         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2036                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2037         BUG_ON(err < 0);
2038
2039         if (netdev_get_fwd_headroom(vport->dev) > dp->max_headroom)
2040                 update_headroom(dp);
2041         else
2042                 netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2043
2044         ovs_unlock();
2045
2046         ovs_notify(&dp_vport_genl_family, &ovs_dp_vport_multicast_group, reply, info);
2047         return 0;
2048
2049 exit_unlock_free:
2050         ovs_unlock();
2051         kfree_skb(reply);
2052         return err;
2053 }
2054
2055 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2056 {
2057         struct nlattr **a = info->attrs;
2058         struct sk_buff *reply;
2059         struct vport *vport;
2060         int err;
2061
2062         reply = ovs_vport_cmd_alloc_info();
2063         if (!reply)
2064                 return -ENOMEM;
2065
2066         ovs_lock();
2067         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2068         err = PTR_ERR(vport);
2069         if (IS_ERR(vport))
2070                 goto exit_unlock_free;
2071
2072         if (a[OVS_VPORT_ATTR_TYPE] &&
2073             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2074                 err = -EINVAL;
2075                 goto exit_unlock_free;
2076         }
2077
2078         if (a[OVS_VPORT_ATTR_OPTIONS]) {
2079                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2080                 if (err)
2081                         goto exit_unlock_free;
2082         }
2083
2084         if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2085                 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2086
2087                 err = ovs_vport_set_upcall_portids(vport, ids);
2088                 if (err)
2089                         goto exit_unlock_free;
2090         }
2091
2092         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2093                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2094         BUG_ON(err < 0);
2095         ovs_unlock();
2096
2097         ovs_notify(&dp_vport_genl_family, &ovs_dp_vport_multicast_group, reply, info);
2098         return 0;
2099
2100 exit_unlock_free:
2101         ovs_unlock();
2102         kfree_skb(reply);
2103         return err;
2104 }
2105
2106 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2107 {
2108         bool must_update_headroom = false;
2109         struct nlattr **a = info->attrs;
2110         struct sk_buff *reply;
2111         struct datapath *dp;
2112         struct vport *vport;
2113         int err;
2114
2115         reply = ovs_vport_cmd_alloc_info();
2116         if (!reply)
2117                 return -ENOMEM;
2118
2119         ovs_lock();
2120         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2121         err = PTR_ERR(vport);
2122         if (IS_ERR(vport))
2123                 goto exit_unlock_free;
2124
2125         if (vport->port_no == OVSP_LOCAL) {
2126                 err = -EINVAL;
2127                 goto exit_unlock_free;
2128         }
2129
2130         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2131                                       info->snd_seq, 0, OVS_VPORT_CMD_DEL);
2132         BUG_ON(err < 0);
2133
2134         /* the vport deletion may trigger dp headroom update */
2135         dp = vport->dp;
2136         if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2137                 must_update_headroom = true;
2138         netdev_reset_rx_headroom(vport->dev);
2139         ovs_dp_detach_port(vport);
2140
2141         if (must_update_headroom)
2142                 update_headroom(dp);
2143
2144         ovs_unlock();
2145
2146         ovs_notify(&dp_vport_genl_family, &ovs_dp_vport_multicast_group, reply, info);
2147         return 0;
2148
2149 exit_unlock_free:
2150         ovs_unlock();
2151         kfree_skb(reply);
2152         return err;
2153 }
2154
2155 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2156 {
2157         struct nlattr **a = info->attrs;
2158         struct ovs_header *ovs_header = info->userhdr;
2159         struct sk_buff *reply;
2160         struct vport *vport;
2161         int err;
2162
2163         reply = ovs_vport_cmd_alloc_info();
2164         if (!reply)
2165                 return -ENOMEM;
2166
2167         rcu_read_lock();
2168         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2169         err = PTR_ERR(vport);
2170         if (IS_ERR(vport))
2171                 goto exit_unlock_free;
2172         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2173                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2174         BUG_ON(err < 0);
2175         rcu_read_unlock();
2176
2177         return genlmsg_reply(reply, info);
2178
2179 exit_unlock_free:
2180         rcu_read_unlock();
2181         kfree_skb(reply);
2182         return err;
2183 }
2184
2185 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2186 {
2187         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2188         struct datapath *dp;
2189         int bucket = cb->args[0], skip = cb->args[1];
2190         int i, j = 0;
2191
2192         rcu_read_lock();
2193         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2194         if (!dp) {
2195                 rcu_read_unlock();
2196                 return -ENODEV;
2197         }
2198         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2199                 struct vport *vport;
2200
2201                 j = 0;
2202                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2203                         if (j >= skip &&
2204                             ovs_vport_cmd_fill_info(vport, skb,
2205                                                     NETLINK_CB(cb->skb).portid,
2206                                                     cb->nlh->nlmsg_seq,
2207                                                     NLM_F_MULTI,
2208                                                     OVS_VPORT_CMD_NEW) < 0)
2209                                 goto out;
2210
2211                         j++;
2212                 }
2213                 skip = 0;
2214         }
2215 out:
2216         rcu_read_unlock();
2217
2218         cb->args[0] = i;
2219         cb->args[1] = j;
2220
2221         return skb->len;
2222 }
2223
2224 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2225         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2226         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2227         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2228         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2229         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
2230         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2231 };
2232
2233 static struct genl_ops dp_vport_genl_ops[] = {
2234         { .cmd = OVS_VPORT_CMD_NEW,
2235           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2236           .policy = vport_policy,
2237           .doit = ovs_vport_cmd_new
2238         },
2239         { .cmd = OVS_VPORT_CMD_DEL,
2240           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2241           .policy = vport_policy,
2242           .doit = ovs_vport_cmd_del
2243         },
2244         { .cmd = OVS_VPORT_CMD_GET,
2245           .flags = 0,               /* OK for unprivileged users. */
2246           .policy = vport_policy,
2247           .doit = ovs_vport_cmd_get,
2248           .dumpit = ovs_vport_cmd_dump
2249         },
2250         { .cmd = OVS_VPORT_CMD_SET,
2251           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2252           .policy = vport_policy,
2253           .doit = ovs_vport_cmd_set,
2254         },
2255 };
2256
2257 struct genl_family dp_vport_genl_family = {
2258         .id = GENL_ID_GENERATE,
2259         .hdrsize = sizeof(struct ovs_header),
2260         .name = OVS_VPORT_FAMILY,
2261         .version = OVS_VPORT_VERSION,
2262         .maxattr = OVS_VPORT_ATTR_MAX,
2263         .netnsok = true,
2264         .parallel_ops = true,
2265         .ops = dp_vport_genl_ops,
2266         .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2267         .mcgrps = &ovs_dp_vport_multicast_group,
2268         .n_mcgrps = 1,
2269 };
2270
2271 static struct genl_family *dp_genl_families[] = {
2272         &dp_datapath_genl_family,
2273         &dp_vport_genl_family,
2274         &dp_flow_genl_family,
2275         &dp_packet_genl_family,
2276 };
2277
2278 static void dp_unregister_genl(int n_families)
2279 {
2280         int i;
2281
2282         for (i = 0; i < n_families; i++)
2283                 genl_unregister_family(dp_genl_families[i]);
2284 }
2285
2286 static int dp_register_genl(void)
2287 {
2288         int err;
2289         int i;
2290
2291         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2292
2293                 err = genl_register_family(dp_genl_families[i]);
2294                 if (err)
2295                         goto error;
2296         }
2297
2298         return 0;
2299
2300 error:
2301         dp_unregister_genl(i);
2302         return err;
2303 }
2304
2305 static int __net_init ovs_init_net(struct net *net)
2306 {
2307         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2308
2309         INIT_LIST_HEAD(&ovs_net->dps);
2310         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2311         ovs_ct_init(net);
2312         return 0;
2313 }
2314
2315 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2316                                             struct list_head *head)
2317 {
2318         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2319         struct datapath *dp;
2320
2321         list_for_each_entry(dp, &ovs_net->dps, list_node) {
2322                 int i;
2323
2324                 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2325                         struct vport *vport;
2326
2327                         hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2328
2329                                 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2330                                         continue;
2331
2332                                 if (dev_net(vport->dev) == dnet)
2333                                         list_add(&vport->detach_list, head);
2334                         }
2335                 }
2336         }
2337 }
2338
2339 static void __net_exit ovs_exit_net(struct net *dnet)
2340 {
2341         struct datapath *dp, *dp_next;
2342         struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2343         struct vport *vport, *vport_next;
2344         struct net *net;
2345         LIST_HEAD(head);
2346
2347         ovs_ct_exit(dnet);
2348         ovs_lock();
2349         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2350                 __dp_destroy(dp);
2351
2352         rtnl_lock();
2353         for_each_net(net)
2354                 list_vports_from_net(net, dnet, &head);
2355         rtnl_unlock();
2356
2357         /* Detach all vports from given namespace. */
2358         list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2359                 list_del(&vport->detach_list);
2360                 ovs_dp_detach_port(vport);
2361         }
2362
2363         ovs_unlock();
2364
2365         cancel_work_sync(&ovs_net->dp_notify_work);
2366 }
2367
2368 static struct pernet_operations ovs_net_ops = {
2369         .init = ovs_init_net,
2370         .exit = ovs_exit_net,
2371         .id   = &ovs_net_id,
2372         .size = sizeof(struct ovs_net),
2373 };
2374
2375 static int __init dp_init(void)
2376 {
2377         int err;
2378
2379         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2380
2381         pr_info("Open vSwitch switching datapath %s\n", VERSION);
2382
2383         err = compat_init();
2384         if (err)
2385                 goto error;
2386
2387         err = action_fifos_init();
2388         if (err)
2389                 goto error_compat_exit;
2390
2391         err = ovs_internal_dev_rtnl_link_register();
2392         if (err)
2393                 goto error_action_fifos_exit;
2394
2395         err = ovs_flow_init();
2396         if (err)
2397                 goto error_unreg_rtnl_link;
2398
2399         err = ovs_vport_init();
2400         if (err)
2401                 goto error_flow_exit;
2402
2403         err = register_pernet_device(&ovs_net_ops);
2404         if (err)
2405                 goto error_vport_exit;
2406
2407         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2408         if (err)
2409                 goto error_netns_exit;
2410
2411         err = ovs_netdev_init();
2412         if (err)
2413                 goto error_unreg_notifier;
2414
2415         err = dp_register_genl();
2416         if (err < 0)
2417                 goto error_unreg_netdev;
2418
2419         return 0;
2420
2421 error_unreg_netdev:
2422         ovs_netdev_exit();
2423 error_unreg_notifier:
2424         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2425 error_netns_exit:
2426         unregister_pernet_device(&ovs_net_ops);
2427 error_vport_exit:
2428         ovs_vport_exit();
2429 error_flow_exit:
2430         ovs_flow_exit();
2431 error_unreg_rtnl_link:
2432         ovs_internal_dev_rtnl_link_unregister();
2433 error_action_fifos_exit:
2434         action_fifos_exit();
2435 error_compat_exit:
2436         compat_exit();
2437 error:
2438         return err;
2439 }
2440
2441 static void dp_cleanup(void)
2442 {
2443         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2444         ovs_netdev_exit();
2445         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2446         unregister_pernet_device(&ovs_net_ops);
2447         rcu_barrier();
2448         ovs_vport_exit();
2449         ovs_flow_exit();
2450         ovs_internal_dev_rtnl_link_unregister();
2451         action_fifos_exit();
2452         compat_exit();
2453 }
2454
2455 module_init(dp_init);
2456 module_exit(dp_cleanup);
2457
2458 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2459 MODULE_LICENSE("GPL");
2460 MODULE_VERSION(VERSION);