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