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, genl_info_net(info),
  99                     info->snd_portid, GROUP_ID(grp), info->nlhdr, GFP_KERNEL);
 100 }
 101
 102 /**
 103  * DOC: Locking:
 104  *
 105  * All writes e.g. Writes to device state (add/remove datapath, port, set
 106  * operations on vports, etc.), Writes to other state (flow table
 107  * modifications, set miscellaneous datapath parameters, etc.) are protected
 108  * by ovs_lock.
 109  *
 110  * Reads are protected by RCU.
 111  *
 112  * There are a few special cases (mostly stats) that have their own
 113  * synchronization but they nest under all of above and don't interact with
 114  * each other.
 115  *
 116  * The RTNL lock nests inside ovs_mutex.
 117  */
 118
 119 static DEFINE_MUTEX(ovs_mutex);
 120
 121 void ovs_lock(void)
 122 {
 123         mutex_lock(&ovs_mutex);
 124 }
 125
 126 void ovs_unlock(void)
 127 {
 128         mutex_unlock(&ovs_mutex);
 129 }
 130
 131 #ifdef CONFIG_LOCKDEP
 132 int lockdep_ovsl_is_held(void)
 133 {
 134         if (debug_locks)
 135                 return lockdep_is_held(&ovs_mutex);
 136         else
 137                 return 1;
 138 }
 139 EXPORT_SYMBOL_GPL(lockdep_ovsl_is_held);
 140 #endif
 141
 142 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
 143                              const struct sw_flow_key *,
 144                              const struct dp_upcall_info *,
 145                              uint32_t cutlen);
 146 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
 147                                   const struct sw_flow_key *,
 148                                   const struct dp_upcall_info *,
 149                                   uint32_t cutlen);
 150
 151 /* Must be called with rcu_read_lock. */
 152 static struct datapath *get_dp_rcu(struct net *net, int dp_ifindex)
 153 {
 154         struct net_device *dev = dev_get_by_index_rcu(net, dp_ifindex);
 155
 156         if (dev) {
 157                 struct vport *vport = ovs_internal_dev_get_vport(dev);
 158                 if (vport)
 159                         return vport->dp;
 160         }
 161
 162         return NULL;
 163 }
 164
 165 /* The caller must hold either ovs_mutex or rcu_read_lock to keep the
 166  * returned dp pointer valid.
 167  */
 168 static inline struct datapath *get_dp(struct net *net, int dp_ifindex)
 169 {
 170         struct datapath *dp;
 171
 172         WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_ovsl_is_held());
 173         rcu_read_lock();
 174         dp = get_dp_rcu(net, dp_ifindex);
 175         rcu_read_unlock();
 176
 177         return dp;
 178 }
 179
 180 /* Must be called with rcu_read_lock or ovs_mutex. */
 181 const char *ovs_dp_name(const struct datapath *dp)
 182 {
 183         struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
 184         return ovs_vport_name(vport);
 185 }
 186
 187 static int get_dpifindex(const struct datapath *dp)
 188 {
 189         struct vport *local;
 190         int ifindex;
 191
 192         rcu_read_lock();
 193
 194         local = ovs_vport_rcu(dp, OVSP_LOCAL);
 195         if (local)
 196                 ifindex = local->dev->ifindex;
 197         else
 198                 ifindex = 0;
 199
 200         rcu_read_unlock();
 201
 202         return ifindex;
 203 }
 204
 205 static void destroy_dp_rcu(struct rcu_head *rcu)
 206 {
 207         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 208
 209         ovs_flow_tbl_destroy(&dp->table);
 210         free_percpu(dp->stats_percpu);
 211         kfree(dp->ports);
 212         kfree(dp);
 213 }
 214
 215 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
 216                                             u16 port_no)
 217 {
 218         return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
 219 }
 220
 221 /* Called with ovs_mutex or RCU read lock. */
 222 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
 223 {
 224         struct vport *vport;
 225         struct hlist_head *head;
 226
 227         head = vport_hash_bucket(dp, port_no);
 228         hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
 229                 if (vport->port_no == port_no)
 230                         return vport;
 231         }
 232         return NULL;
 233 }
 234
 235 /* Called with ovs_mutex. */
 236 static struct vport *new_vport(const struct vport_parms *parms)
 237 {
 238         struct vport *vport;
 239
 240         vport = ovs_vport_add(parms);
 241         if (!IS_ERR(vport)) {
 242                 struct datapath *dp = parms->dp;
 243                 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
 244
 245                 hlist_add_head_rcu(&vport->dp_hash_node, head);
 246         }
 247         return vport;
 248 }
 249
 250 void ovs_dp_detach_port(struct vport *p)
 251 {
 252         ASSERT_OVSL();
 253
 254         /* First drop references to device. */
 255         hlist_del_rcu(&p->dp_hash_node);
 256
 257         /* Then destroy it. */
 258         ovs_vport_del(p);
 259 }
 260
 261 /* Must be called with rcu_read_lock. */
 262 void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
 263 {
 264         const struct vport *p = OVS_CB(skb)->input_vport;
 265         struct datapath *dp = p->dp;
 266         struct sw_flow *flow;
 267         struct sw_flow_actions *sf_acts;
 268         struct dp_stats_percpu *stats;
 269         u64 *stats_counter;
 270         u32 n_mask_hit;
 271
 272         stats = this_cpu_ptr(dp->stats_percpu);
 273
 274         /* Look up flow. */
 275         flow = ovs_flow_tbl_lookup_stats(&dp->table, key, skb_get_hash(skb),
 276                                          &n_mask_hit);
 277         if (unlikely(!flow)) {
 278                 struct dp_upcall_info upcall;
 279                 int error;
 280
 281                 memset(&upcall, 0, sizeof(upcall));
 282                 upcall.cmd = OVS_PACKET_CMD_MISS;
 283                 upcall.portid = ovs_vport_find_upcall_portid(p, skb);
 284                 upcall.mru = OVS_CB(skb)->mru;
 285                 error = ovs_dp_upcall(dp, skb, key, &upcall, 0);
 286                 if (unlikely(error))
 287                         kfree_skb(skb);
 288                 else
 289                         consume_skb(skb);
 290                 stats_counter = &stats->n_missed;
 291                 goto out;
 292         }
 293
 294         ovs_flow_stats_update(flow, key->tp.flags, skb);
 295         sf_acts = rcu_dereference(flow->sf_acts);
 296         ovs_execute_actions(dp, skb, sf_acts, key);
 297
 298         stats_counter = &stats->n_hit;
 299
 300 out:
 301         /* Update datapath statistics. */
 302         u64_stats_update_begin(&stats->syncp);
 303         (*stats_counter)++;
 304         stats->n_mask_hit += n_mask_hit;
 305         u64_stats_update_end(&stats->syncp);
 306 }
 307
 308 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
 309                   const struct sw_flow_key *key,
 310                   const struct dp_upcall_info *upcall_info,
 311                   uint32_t cutlen)
 312 {
 313         struct dp_stats_percpu *stats;
 314         int err;
 315
 316         if (upcall_info->portid == 0) {
 317                 err = -ENOTCONN;
 318                 goto err;
 319         }
 320
 321         if (!skb_is_gso(skb))
 322                 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
 323         else
 324                 err = queue_gso_packets(dp, skb, key, upcall_info, cutlen);
 325         if (err)
 326                 goto err;
 327
 328         return 0;
 329
 330 err:
 331         stats = this_cpu_ptr(dp->stats_percpu);
 332
 333         u64_stats_update_begin(&stats->syncp);
 334         stats->n_lost++;
 335         u64_stats_update_end(&stats->syncp);
 336
 337         return err;
 338 }
 339
 340 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
 341                              const struct sw_flow_key *key,
 342                              const struct dp_upcall_info *upcall_info,
 343                                  uint32_t cutlen)
 344 {
 345         unsigned short gso_type = skb_shinfo(skb)->gso_type;
 346         struct sw_flow_key later_key;
 347         struct sk_buff *segs, *nskb;
 348         struct ovs_skb_cb ovs_cb;
 349         int err;
 350
 351         ovs_cb = *OVS_CB(skb);
 352         segs = __skb_gso_segment(skb, NETIF_F_SG, false);
 353         *OVS_CB(skb) = ovs_cb;
 354         if (IS_ERR(segs))
 355                 return PTR_ERR(segs);
 356         if (segs == NULL)
 357                 return -EINVAL;
 358
 359         if (gso_type & SKB_GSO_UDP) {
 360                 /* The initial flow key extracted by ovs_flow_key_extract()
 361                  * in this case is for a first fragment, so we need to
 362                  * properly mark later fragments.
 363                  */
 364                 later_key = *key;
 365                 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
 366         }
 367
 368         /* Queue all of the segments. */
 369         skb = segs;
 370         do {
 371                 *OVS_CB(skb) = ovs_cb;
 372                 if (gso_type & SKB_GSO_UDP && skb != segs)
 373                         key = &later_key;
 374
 375                 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
 376                 if (err)
 377                         break;
 378
 379         } while ((skb = skb->next));
 380
 381         /* Free all of the segments. */
 382         skb = segs;
 383         do {
 384                 nskb = skb->next;
 385                 if (err)
 386                         kfree_skb(skb);
 387                 else
 388                         consume_skb(skb);
 389         } while ((skb = nskb));
 390         return err;
 391 }
 392
 393 static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
 394                               unsigned int hdrlen)
 395 {
 396         size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
 397                 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
 398                 + nla_total_size(ovs_key_attr_size()) /* OVS_PACKET_ATTR_KEY */
 399                 + nla_total_size(sizeof(unsigned int)); /* OVS_PACKET_ATTR_LEN */
 400
 401         /* OVS_PACKET_ATTR_USERDATA */
 402         if (upcall_info->userdata)
 403                 size += NLA_ALIGN(upcall_info->userdata->nla_len);
 404
 405         /* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
 406         if (upcall_info->egress_tun_info)
 407                 size += nla_total_size(ovs_tun_key_attr_size());
 408
 409         /* OVS_PACKET_ATTR_ACTIONS */
 410         if (upcall_info->actions_len)
 411                 size += nla_total_size(upcall_info->actions_len);
 412
 413         /* OVS_PACKET_ATTR_MRU */
 414         if (upcall_info->mru)
 415                 size += nla_total_size(sizeof(upcall_info->mru));
 416
 417         return size;
 418 }
 419
 420 static void pad_packet(struct datapath *dp, struct sk_buff *skb)
 421 {
 422         if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
 423                 size_t plen = NLA_ALIGN(skb->len) - skb->len;
 424
 425                 if (plen > 0)
 426                         memset(skb_put(skb, plen), 0, plen);
 427         }
 428 }
 429
 430 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
 431                                   const struct sw_flow_key *key,
 432                                   const struct dp_upcall_info *upcall_info,
 433                                   uint32_t cutlen)
 434 {
 435         struct ovs_header *upcall;
 436         struct sk_buff *nskb = NULL;
 437         struct sk_buff *user_skb = NULL; /* to be queued to userspace */
 438         struct nlattr *nla;
 439         struct genl_info info = {
 440 #ifdef HAVE_GENLMSG_NEW_UNICAST
 441                 .dst_sk = ovs_dp_get_net(dp)->genl_sock,
 442 #endif
 443                 .snd_portid = upcall_info->portid,
 444         };
 445         size_t len;
 446         unsigned int hlen;
 447         int err, dp_ifindex;
 448
 449         dp_ifindex = get_dpifindex(dp);
 450         if (!dp_ifindex)
 451                 return -ENODEV;
 452
 453         if (skb_vlan_tag_present(skb)) {
 454                 nskb = skb_clone(skb, GFP_ATOMIC);
 455                 if (!nskb)
 456                         return -ENOMEM;
 457
 458                 nskb = __vlan_hwaccel_push_inside(nskb);
 459                 if (!nskb)
 460                         return -ENOMEM;
 461
 462                 skb = nskb;
 463         }
 464
 465         if (nla_attr_size(skb->len) > USHRT_MAX) {
 466                 err = -EFBIG;
 467                 goto out;
 468         }
 469
 470         /* Complete checksum if needed */
 471         if (skb->ip_summed == CHECKSUM_PARTIAL &&
 472             (err = skb_checksum_help(skb)))
 473                 goto out;
 474
 475         /* Older versions of OVS user space enforce alignment of the last
 476          * Netlink attribute to NLA_ALIGNTO which would require extensive
 477          * padding logic. Only perform zerocopy if padding is not required.
 478          */
 479         if (dp->user_features & OVS_DP_F_UNALIGNED)
 480                 hlen = skb_zerocopy_headlen(skb);
 481         else
 482                 hlen = skb->len;
 483
 484         len = upcall_msg_size(upcall_info, hlen - cutlen);
 485         user_skb = genlmsg_new_unicast(len, &info, GFP_ATOMIC);
 486         if (!user_skb) {
 487                 err = -ENOMEM;
 488                 goto out;
 489         }
 490
 491         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 492                              0, upcall_info->cmd);
 493         upcall->dp_ifindex = dp_ifindex;
 494
 495         err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
 496         BUG_ON(err);
 497
 498         if (upcall_info->userdata)
 499                 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
 500                           nla_len(upcall_info->userdata),
 501                           nla_data(upcall_info->userdata));
 502
 503
 504         if (upcall_info->egress_tun_info) {
 505                 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_EGRESS_TUN_KEY);
 506                 err = ovs_nla_put_tunnel_info(user_skb,
 507                                               upcall_info->egress_tun_info);
 508                 BUG_ON(err);
 509                 nla_nest_end(user_skb, nla);
 510         }
 511
 512         if (upcall_info->actions_len) {
 513                 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_ACTIONS);
 514                 err = ovs_nla_put_actions(upcall_info->actions,
 515                                           upcall_info->actions_len,
 516                                           user_skb);
 517                 if (!err)
 518                         nla_nest_end(user_skb, nla);
 519                 else
 520                         nla_nest_cancel(user_skb, nla);
 521         }
 522
 523         /* Add OVS_PACKET_ATTR_MRU */
 524         if (upcall_info->mru) {
 525                 if (nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU,
 526                                 upcall_info->mru)) {
 527                         err = -ENOBUFS;
 528                         goto out;
 529                 }
 530                 pad_packet(dp, user_skb);
 531         }
 532
 533         /* Add OVS_PACKET_ATTR_LEN when packet is truncated */
 534         if (cutlen > 0) {
 535                 if (nla_put_u32(user_skb, OVS_PACKET_ATTR_LEN,
 536                                 skb->len)) {
 537                         err = -ENOBUFS;
 538                         goto out;
 539                 }
 540                 pad_packet(dp, user_skb);
 541         }
 542
 543         /* Only reserve room for attribute header, packet data is added
 544          * in skb_zerocopy()
 545          */
 546         if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
 547                 err = -ENOBUFS;
 548                 goto out;
 549         }
 550         nla->nla_len = nla_attr_size(skb->len - cutlen);
 551
 552         err = skb_zerocopy(user_skb, skb, skb->len - cutlen, hlen);
 553         if (err)
 554                 goto out;
 555
 556         /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
 557         pad_packet(dp, user_skb);
 558
 559         ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
 560
 561         err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
 562         user_skb = NULL;
 563 out:
 564         if (err)
 565                 skb_tx_error(skb);
 566         kfree_skb(user_skb);
 567         kfree_skb(nskb);
 568         return err;
 569 }
 570
 571 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 572 {
 573         struct ovs_header *ovs_header = info->userhdr;
 574         struct net *net = sock_net(skb->sk);
 575         struct nlattr **a = info->attrs;
 576         struct sw_flow_actions *acts;
 577         struct sk_buff *packet;
 578         struct sw_flow *flow;
 579         struct sw_flow_actions *sf_acts;
 580         struct datapath *dp;
 581         struct ethhdr *eth;
 582         struct vport *input_vport;
 583         u16 mru = 0;
 584         int len;
 585         int err;
 586         bool log = !a[OVS_PACKET_ATTR_PROBE];
 587
 588         err = -EINVAL;
 589         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 590             !a[OVS_PACKET_ATTR_ACTIONS])
 591                 goto err;
 592
 593         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 594         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 595         err = -ENOMEM;
 596         if (!packet)
 597                 goto err;
 598         skb_reserve(packet, NET_IP_ALIGN);
 599
 600         nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
 601
 602         skb_reset_mac_header(packet);
 603         eth = eth_hdr(packet);
 604
 605         /* Normally, setting the skb 'protocol' field would be handled by a
 606          * call to eth_type_trans(), but it assumes there's a sending
 607          * device, which we may not have.
 608          */
 609         if (eth_proto_is_802_3(eth->h_proto))
 610                 packet->protocol = eth->h_proto;
 611         else
 612                 packet->protocol = htons(ETH_P_802_2);
 613
 614         /* Set packet's mru */
 615         if (a[OVS_PACKET_ATTR_MRU]) {
 616                 mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
 617                 packet->ignore_df = 1;
 618         }
 619         OVS_CB(packet)->mru = mru;
 620
 621         /* Build an sw_flow for sending this packet. */
 622         flow = ovs_flow_alloc();
 623         err = PTR_ERR(flow);
 624         if (IS_ERR(flow))
 625                 goto err_kfree_skb;
 626
 627         err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
 628                                              packet, &flow->key, log);
 629         if (err)
 630                 goto err_flow_free;
 631
 632         err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
 633                                    &flow->key, &acts, log);
 634         if (err)
 635                 goto err_flow_free;
 636
 637         rcu_assign_pointer(flow->sf_acts, acts);
 638         packet->priority = flow->key.phy.priority;
 639         packet->mark = flow->key.phy.skb_mark;
 640
 641         rcu_read_lock();
 642         dp = get_dp_rcu(net, ovs_header->dp_ifindex);
 643         err = -ENODEV;
 644         if (!dp)
 645                 goto err_unlock;
 646
 647         input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
 648         if (!input_vport)
 649                 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
 650
 651         if (!input_vport)
 652                 goto err_unlock;
 653
 654         packet->dev = input_vport->dev;
 655         OVS_CB(packet)->input_vport = input_vport;
 656         sf_acts = rcu_dereference(flow->sf_acts);
 657
 658         local_bh_disable();
 659         err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
 660         local_bh_enable();
 661         rcu_read_unlock();
 662
 663         ovs_flow_free(flow, false);
 664         return err;
 665
 666 err_unlock:
 667         rcu_read_unlock();
 668 err_flow_free:
 669         ovs_flow_free(flow, false);
 670 err_kfree_skb:
 671         kfree_skb(packet);
 672 err:
 673         return err;
 674 }
 675
 676 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 677         [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
 678         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 679         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 680         [OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
 681         [OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
 682 };
 683
 684 static struct genl_ops dp_packet_genl_ops[] = {
 685         { .cmd = OVS_PACKET_CMD_EXECUTE,
 686           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 687           .policy = packet_policy,
 688           .doit = ovs_packet_cmd_execute
 689         }
 690 };
 691
 692 static struct genl_family dp_packet_genl_family = {
 693         .id = GENL_ID_GENERATE,
 694         .hdrsize = sizeof(struct ovs_header),
 695         .name = OVS_PACKET_FAMILY,
 696         .version = OVS_PACKET_VERSION,
 697         .maxattr = OVS_PACKET_ATTR_MAX,
 698         .netnsok = true,
 699         .parallel_ops = true,
 700         .ops = dp_packet_genl_ops,
 701         .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
 702 };
 703
 704 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
 705                          struct ovs_dp_megaflow_stats *mega_stats)
 706 {
 707         int i;
 708
 709         memset(mega_stats, 0, sizeof(*mega_stats));
 710
 711         stats->n_flows = ovs_flow_tbl_count(&dp->table);
 712         mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
 713
 714         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 715
 716         for_each_possible_cpu(i) {
 717                 const struct dp_stats_percpu *percpu_stats;
 718                 struct dp_stats_percpu local_stats;
 719                 unsigned int start;
 720
 721                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 722
 723                 do {
 724                         start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
 725                         local_stats = *percpu_stats;
 726                 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
 727
 728                 stats->n_hit += local_stats.n_hit;
 729                 stats->n_missed += local_stats.n_missed;
 730                 stats->n_lost += local_stats.n_lost;
 731                 mega_stats->n_mask_hit += local_stats.n_mask_hit;
 732         }
 733 }
 734
 735 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
 736 {
 737         return ovs_identifier_is_ufid(sfid) &&
 738                !(ufid_flags & OVS_UFID_F_OMIT_KEY);
 739 }
 740
 741 static bool should_fill_mask(uint32_t ufid_flags)
 742 {
 743         return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
 744 }
 745
 746 static bool should_fill_actions(uint32_t ufid_flags)
 747 {
 748         return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
 749 }
 750
 751 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
 752                                     const struct sw_flow_id *sfid,
 753                                     uint32_t ufid_flags)
 754 {
 755         size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
 756
 757         /* OVS_FLOW_ATTR_UFID */
 758         if (sfid && ovs_identifier_is_ufid(sfid))
 759                 len += nla_total_size(sfid->ufid_len);
 760
 761         /* OVS_FLOW_ATTR_KEY */
 762         if (!sfid || should_fill_key(sfid, ufid_flags))
 763                 len += nla_total_size(ovs_key_attr_size());
 764
 765         /* OVS_FLOW_ATTR_MASK */
 766         if (should_fill_mask(ufid_flags))
 767                 len += nla_total_size(ovs_key_attr_size());
 768
 769         /* OVS_FLOW_ATTR_ACTIONS */
 770         if (should_fill_actions(ufid_flags))
 771                 len += nla_total_size(acts->orig_len);
 772
 773         return len
 774                 + nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
 775                 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
 776                 + nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */
 777 }
 778
 779 /* Called with ovs_mutex or RCU read lock. */
 780 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
 781                                    struct sk_buff *skb)
 782 {
 783         struct ovs_flow_stats stats;
 784         __be16 tcp_flags;
 785         unsigned long used;
 786
 787         ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
 788
 789         if (used &&
 790             nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
 791                               OVS_FLOW_ATTR_PAD))
 792                 return -EMSGSIZE;
 793
 794         if (stats.n_packets &&
 795             nla_put_64bit(skb, OVS_FLOW_ATTR_STATS,
 796                           sizeof(struct ovs_flow_stats), &stats,
 797                           OVS_FLOW_ATTR_PAD))
 798                 return -EMSGSIZE;
 799
 800         if ((u8)ntohs(tcp_flags) &&
 801              nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
 802                 return -EMSGSIZE;
 803
 804         return 0;
 805 }
 806
 807 /* Called with ovs_mutex or RCU read lock. */
 808 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
 809                                      struct sk_buff *skb, int skb_orig_len)
 810 {
 811         struct nlattr *start;
 812         int err;
 813
 814         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 815          * this is the first flow to be dumped into 'skb'.  This is unusual for
 816          * Netlink but individual action lists can be longer than
 817          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 818          * The userspace caller can always fetch the actions separately if it
 819          * really wants them.  (Most userspace callers in fact don't care.)
 820          *
 821          * This can only fail for dump operations because the skb is always
 822          * properly sized for single flows.
 823          */
 824         start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
 825         if (start) {
 826                 const struct sw_flow_actions *sf_acts;
 827
 828                 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
 829                 err = ovs_nla_put_actions(sf_acts->actions,
 830                                           sf_acts->actions_len, skb);
 831
 832                 if (!err)
 833                         nla_nest_end(skb, start);
 834                 else {
 835                         if (skb_orig_len)
 836                                 return err;
 837
 838                         nla_nest_cancel(skb, start);
 839                 }
 840         } else if (skb_orig_len) {
 841                 return -EMSGSIZE;
 842         }
 843
 844         return 0;
 845 }
 846
 847 /* Called with ovs_mutex or RCU read lock. */
 848 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
 849                                   struct sk_buff *skb, u32 portid,
 850                                   u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
 851 {
 852         const int skb_orig_len = skb->len;
 853         struct ovs_header *ovs_header;
 854         int err;
 855
 856         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
 857                                  flags, cmd);
 858         if (!ovs_header)
 859                 return -EMSGSIZE;
 860
 861         ovs_header->dp_ifindex = dp_ifindex;
 862
 863         err = ovs_nla_put_identifier(flow, skb);
 864         if (err)
 865                 goto error;
 866
 867         if (should_fill_key(&flow->id, ufid_flags)) {
 868                 err = ovs_nla_put_masked_key(flow, skb);
 869                 if (err)
 870                         goto error;
 871         }
 872
 873         if (should_fill_mask(ufid_flags)) {
 874                 err = ovs_nla_put_mask(flow, skb);
 875                 if (err)
 876                         goto error;
 877         }
 878
 879         err = ovs_flow_cmd_fill_stats(flow, skb);
 880         if (err)
 881                 goto error;
 882
 883         if (should_fill_actions(ufid_flags)) {
 884                 err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
 885                 if (err)
 886                         goto error;
 887         }
 888
 889         genlmsg_end(skb, ovs_header);
 890         return 0;
 891
 892 error:
 893         genlmsg_cancel(skb, ovs_header);
 894         return err;
 895 }
 896
 897 /* May not be called with RCU read lock. */
 898 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
 899                                                const struct sw_flow_id *sfid,
 900                                                struct genl_info *info,
 901                                                bool always,
 902                                                uint32_t ufid_flags)
 903 {
 904         struct sk_buff *skb;
 905         size_t len;
 906
 907         if (!always && !ovs_must_notify(&dp_flow_genl_family, info,
 908                                         GROUP_ID(&ovs_dp_flow_multicast_group)))
 909                 return NULL;
 910
 911         len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
 912         skb = genlmsg_new_unicast(len, info, GFP_KERNEL);
 913         if (!skb)
 914                 return ERR_PTR(-ENOMEM);
 915
 916         return skb;
 917 }
 918
 919 /* Called with ovs_mutex. */
 920 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
 921                                                int dp_ifindex,
 922                                                struct genl_info *info, u8 cmd,
 923                                                bool always, u32 ufid_flags)
 924 {
 925         struct sk_buff *skb;
 926         int retval;
 927
 928         skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
 929                                       &flow->id, info, always, ufid_flags);
 930         if (IS_ERR_OR_NULL(skb))
 931                 return skb;
 932
 933         retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
 934                                         info->snd_portid, info->snd_seq, 0,
 935                                         cmd, ufid_flags);
 936         BUG_ON(retval < 0);
 937         return skb;
 938 }
 939
 940 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
 941 {
 942         struct net *net = sock_net(skb->sk);
 943         struct nlattr **a = info->attrs;
 944         struct ovs_header *ovs_header = info->userhdr;
 945         struct sw_flow *flow = NULL, *new_flow;
 946         struct sw_flow_mask mask;
 947         struct sk_buff *reply;
 948         struct datapath *dp;
 949         struct sw_flow_key key;
 950         struct sw_flow_actions *acts;
 951         struct sw_flow_match match;
 952         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
 953         int error;
 954         bool log = !a[OVS_FLOW_ATTR_PROBE];
 955
 956         /* Must have key and actions. */
 957         error = -EINVAL;
 958         if (!a[OVS_FLOW_ATTR_KEY]) {
 959                 OVS_NLERR(log, "Flow key attr not present in new flow.");
 960                 goto error;
 961         }
 962         if (!a[OVS_FLOW_ATTR_ACTIONS]) {
 963                 OVS_NLERR(log, "Flow actions attr not present in new flow.");
 964                 goto error;
 965         }
 966
 967         /* Most of the time we need to allocate a new flow, do it before
 968          * locking.
 969          */
 970         new_flow = ovs_flow_alloc();
 971         if (IS_ERR(new_flow)) {
 972                 error = PTR_ERR(new_flow);
 973                 goto error;
 974         }
 975
 976         /* Extract key. */
 977         ovs_match_init(&match, &key, &mask);
 978         error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
 979                                   a[OVS_FLOW_ATTR_MASK], log);
 980         if (error)
 981                 goto err_kfree_flow;
 982
 983         ovs_flow_mask_key(&new_flow->key, &key, true, &mask);
 984
 985         /* Extract flow identifier. */
 986         error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
 987                                        &key, log);
 988         if (error)
 989                 goto err_kfree_flow;
 990
 991         /* Validate actions. */
 992         error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
 993                                      &new_flow->key, &acts, log);
 994         if (error) {
 995                 OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
 996                 goto err_kfree_flow;
 997         }
 998
 999         reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
1000                                         ufid_flags);
1001         if (IS_ERR(reply)) {
1002                 error = PTR_ERR(reply);
1003                 goto err_kfree_acts;
1004         }
1005
1006         ovs_lock();
1007         dp = get_dp(net, ovs_header->dp_ifindex);
1008         if (unlikely(!dp)) {
1009                 error = -ENODEV;
1010                 goto err_unlock_ovs;
1011         }
1012
1013         /* Check if this is a duplicate flow */
1014         if (ovs_identifier_is_ufid(&new_flow->id))
1015                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
1016         if (!flow)
1017                 flow = ovs_flow_tbl_lookup(&dp->table, &key);
1018         if (likely(!flow)) {
1019                 rcu_assign_pointer(new_flow->sf_acts, acts);
1020
1021                 /* Put flow in bucket. */
1022                 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
1023                 if (unlikely(error)) {
1024                         acts = NULL;
1025                         goto err_unlock_ovs;
1026                 }
1027
1028                 if (unlikely(reply)) {
1029                         error = ovs_flow_cmd_fill_info(new_flow,
1030                                                        ovs_header->dp_ifindex,
1031                                                        reply, info->snd_portid,
1032                                                        info->snd_seq, 0,
1033                                                        OVS_FLOW_CMD_NEW,
1034                                                        ufid_flags);
1035                         BUG_ON(error < 0);
1036                 }
1037                 ovs_unlock();
1038         } else {
1039                 struct sw_flow_actions *old_acts;
1040
1041                 /* Bail out if we're not allowed to modify an existing flow.
1042                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1043                  * because Generic Netlink treats the latter as a dump
1044                  * request.  We also accept NLM_F_EXCL in case that bug ever
1045                  * gets fixed.
1046                  */
1047                 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1048                                                          | NLM_F_EXCL))) {
1049                         error = -EEXIST;
1050                         goto err_unlock_ovs;
1051                 }
1052                 /* The flow identifier has to be the same for flow updates.
1053                  * Look for any overlapping flow.
1054                  */
1055                 if (unlikely(!ovs_flow_cmp(flow, &match))) {
1056                         if (ovs_identifier_is_key(&flow->id))
1057                                 flow = ovs_flow_tbl_lookup_exact(&dp->table,
1058                                                                  &match);
1059                         else /* UFID matches but key is different */
1060                                 flow = NULL;
1061                         if (!flow) {
1062                                 error = -ENOENT;
1063                                 goto err_unlock_ovs;
1064                         }
1065                 }
1066                 /* Update actions. */
1067                 old_acts = ovsl_dereference(flow->sf_acts);
1068                 rcu_assign_pointer(flow->sf_acts, acts);
1069
1070                 if (unlikely(reply)) {
1071                         error = ovs_flow_cmd_fill_info(flow,
1072                                                        ovs_header->dp_ifindex,
1073                                                        reply, info->snd_portid,
1074                                                        info->snd_seq, 0,
1075                                                        OVS_FLOW_CMD_NEW,
1076                                                        ufid_flags);
1077                         BUG_ON(error < 0);
1078                 }
1079                 ovs_unlock();
1080
1081                 ovs_nla_free_flow_actions_rcu(old_acts);
1082                 ovs_flow_free(new_flow, false);
1083         }
1084
1085         if (reply)
1086                 ovs_notify(&dp_flow_genl_family, &ovs_dp_flow_multicast_group, reply, info);
1087         return 0;
1088
1089 err_unlock_ovs:
1090         ovs_unlock();
1091         kfree_skb(reply);
1092 err_kfree_acts:
1093         ovs_nla_free_flow_actions(acts);
1094 err_kfree_flow:
1095         ovs_flow_free(new_flow, false);
1096 error:
1097         return error;
1098 }
1099
1100 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1101 static struct sw_flow_actions *get_flow_actions(struct net *net,
1102                                                 const struct nlattr *a,
1103                                                 const struct sw_flow_key *key,
1104                                                 const struct sw_flow_mask *mask,
1105                                                 bool log)
1106 {
1107         struct sw_flow_actions *acts;
1108         struct sw_flow_key masked_key;
1109         int error;
1110
1111         ovs_flow_mask_key(&masked_key, key, true, mask);
1112         error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1113         if (error) {
1114                 OVS_NLERR(log,
1115                           "Actions may not be safe on all matching packets");
1116                 return ERR_PTR(error);
1117         }
1118
1119         return acts;
1120 }
1121
1122 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1123 {
1124         struct net *net = sock_net(skb->sk);
1125         struct nlattr **a = info->attrs;
1126         struct ovs_header *ovs_header = info->userhdr;
1127         struct sw_flow_key key;
1128         struct sw_flow *flow;
1129         struct sw_flow_mask mask;
1130         struct sk_buff *reply = NULL;
1131         struct datapath *dp;
1132         struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1133         struct sw_flow_match match;
1134         struct sw_flow_id sfid;
1135         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1136         int error = 0;
1137         bool log = !a[OVS_FLOW_ATTR_PROBE];
1138         bool ufid_present;
1139
1140         ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1141         if (a[OVS_FLOW_ATTR_KEY]) {
1142                 ovs_match_init(&match, &key, &mask);
1143                 error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1144                                           a[OVS_FLOW_ATTR_MASK], log);
1145         } else if (!ufid_present) {
1146                 OVS_NLERR(log,
1147                           "Flow set message rejected, Key attribute missing.");
1148                 error = -EINVAL;
1149         }
1150         if (error)
1151                 goto error;
1152
1153         /* Validate actions. */
1154         if (a[OVS_FLOW_ATTR_ACTIONS]) {
1155                 if (!a[OVS_FLOW_ATTR_KEY]) {
1156                         OVS_NLERR(log,
1157                                   "Flow key attribute not present in set flow.");
1158                         error = -EINVAL;
1159                         goto error;
1160                 }
1161
1162                 acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], &key,
1163                                         &mask, log);
1164                 if (IS_ERR(acts)) {
1165                         error = PTR_ERR(acts);
1166                         goto error;
1167                 }
1168
1169                 /* Can allocate before locking if have acts. */
1170                 reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1171                                                 ufid_flags);
1172                 if (IS_ERR(reply)) {
1173                         error = PTR_ERR(reply);
1174                         goto err_kfree_acts;
1175                 }
1176         }
1177
1178         ovs_lock();
1179         dp = get_dp(net, ovs_header->dp_ifindex);
1180         if (unlikely(!dp)) {
1181                 error = -ENODEV;
1182                 goto err_unlock_ovs;
1183         }
1184         /* Check that the flow exists. */
1185         if (ufid_present)
1186                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1187         else
1188                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1189         if (unlikely(!flow)) {
1190                 error = -ENOENT;
1191                 goto err_unlock_ovs;
1192         }
1193
1194         /* Update actions, if present. */
1195         if (likely(acts)) {
1196                 old_acts = ovsl_dereference(flow->sf_acts);
1197                 rcu_assign_pointer(flow->sf_acts, acts);
1198
1199                 if (unlikely(reply)) {
1200                         error = ovs_flow_cmd_fill_info(flow,
1201                                                        ovs_header->dp_ifindex,
1202                                                        reply, info->snd_portid,
1203                                                        info->snd_seq, 0,
1204                                                        OVS_FLOW_CMD_NEW,
1205                                                        ufid_flags);
1206                         BUG_ON(error < 0);
1207                 }
1208         } else {
1209                 /* Could not alloc without acts before locking. */
1210                 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1211                                                 info, OVS_FLOW_CMD_NEW, false,
1212                                                 ufid_flags);
1213
1214                 if (unlikely(IS_ERR(reply))) {
1215                         error = PTR_ERR(reply);
1216                         goto err_unlock_ovs;
1217                 }
1218         }
1219
1220         /* Clear stats. */
1221         if (a[OVS_FLOW_ATTR_CLEAR])
1222                 ovs_flow_stats_clear(flow);
1223         ovs_unlock();
1224
1225         if (reply)
1226                 ovs_notify(&dp_flow_genl_family, &ovs_dp_flow_multicast_group, reply, info);
1227         if (old_acts)
1228                 ovs_nla_free_flow_actions_rcu(old_acts);
1229
1230         return 0;
1231
1232 err_unlock_ovs:
1233         ovs_unlock();
1234         kfree_skb(reply);
1235 err_kfree_acts:
1236         ovs_nla_free_flow_actions(acts);
1237 error:
1238         return error;
1239 }
1240
1241 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1242 {
1243         struct nlattr **a = info->attrs;
1244         struct ovs_header *ovs_header = info->userhdr;
1245         struct net *net = sock_net(skb->sk);
1246         struct sw_flow_key key;
1247         struct sk_buff *reply;
1248         struct sw_flow *flow;
1249         struct datapath *dp;
1250         struct sw_flow_match match;
1251         struct sw_flow_id ufid;
1252         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1253         int err = 0;
1254         bool log = !a[OVS_FLOW_ATTR_PROBE];
1255         bool ufid_present;
1256
1257         ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1258         if (a[OVS_FLOW_ATTR_KEY]) {
1259                 ovs_match_init(&match, &key, NULL);
1260                 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1261                                         log);
1262         } else if (!ufid_present) {
1263                 OVS_NLERR(log,
1264                           "Flow get message rejected, Key attribute missing.");
1265                 err = -EINVAL;
1266         }
1267         if (err)
1268                 return err;
1269
1270         ovs_lock();
1271         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1272         if (!dp) {
1273                 err = -ENODEV;
1274                 goto unlock;
1275         }
1276
1277         if (ufid_present)
1278                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1279         else
1280                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1281         if (!flow) {
1282                 err = -ENOENT;
1283                 goto unlock;
1284         }
1285
1286         reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1287                                         OVS_FLOW_CMD_NEW, true, ufid_flags);
1288         if (IS_ERR(reply)) {
1289                 err = PTR_ERR(reply);
1290                 goto unlock;
1291         }
1292
1293         ovs_unlock();
1294         return genlmsg_reply(reply, info);
1295 unlock:
1296         ovs_unlock();
1297         return err;
1298 }
1299
1300 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1301 {
1302         struct nlattr **a = info->attrs;
1303         struct ovs_header *ovs_header = info->userhdr;
1304         struct net *net = sock_net(skb->sk);
1305         struct sw_flow_key key;
1306         struct sk_buff *reply;
1307         struct sw_flow *flow = NULL;
1308         struct datapath *dp;
1309         struct sw_flow_match match;
1310         struct sw_flow_id ufid;
1311         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1312         int err;
1313         bool log = !a[OVS_FLOW_ATTR_PROBE];
1314         bool ufid_present;
1315
1316         ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1317         if (a[OVS_FLOW_ATTR_KEY]) {
1318                 ovs_match_init(&match, &key, NULL);
1319                 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1320                                         NULL, log);
1321                 if (unlikely(err))
1322                         return err;
1323         }
1324
1325         ovs_lock();
1326         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1327         if (unlikely(!dp)) {
1328                 err = -ENODEV;
1329                 goto unlock;
1330         }
1331
1332         if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1333                 err = ovs_flow_tbl_flush(&dp->table);
1334                 goto unlock;
1335         }
1336
1337         if (ufid_present)
1338                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1339         else
1340                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1341         if (unlikely(!flow)) {
1342                 err = -ENOENT;
1343                 goto unlock;
1344         }
1345
1346         ovs_flow_tbl_remove(&dp->table, flow);
1347         ovs_unlock();
1348
1349         reply = ovs_flow_cmd_alloc_info(rcu_dereference_raw(flow->sf_acts),
1350                                         &flow->id, info, false, ufid_flags);
1351
1352         if (likely(reply)) {
1353                 if (likely(!IS_ERR(reply))) {
1354                         rcu_read_lock();        /*To keep RCU checker happy. */
1355                         err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1356                                                      reply, info->snd_portid,
1357                                                      info->snd_seq, 0,
1358                                                      OVS_FLOW_CMD_DEL,
1359                                                      ufid_flags);
1360                         rcu_read_unlock();
1361                         BUG_ON(err < 0);
1362                         ovs_notify(&dp_flow_genl_family, &ovs_dp_flow_multicast_group, reply, info);
1363                 } else {
1364                         genl_set_err(&dp_flow_genl_family, sock_net(skb->sk), 0,
1365                                      GROUP_ID(&ovs_dp_flow_multicast_group), PTR_ERR(reply));
1366
1367                 }
1368         }
1369
1370         ovs_flow_free(flow, true);
1371         return 0;
1372 unlock:
1373         ovs_unlock();
1374         return err;
1375 }
1376
1377 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1378 {
1379         struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1380         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1381         struct table_instance *ti;
1382         struct datapath *dp;
1383         u32 ufid_flags;
1384         int err;
1385
1386         err = genlmsg_parse(cb->nlh, &dp_flow_genl_family, a,
1387                             OVS_FLOW_ATTR_MAX, flow_policy);
1388         if (err)
1389                 return err;
1390         ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1391
1392         rcu_read_lock();
1393         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1394         if (!dp) {
1395                 rcu_read_unlock();
1396                 return -ENODEV;
1397         }
1398
1399         ti = rcu_dereference(dp->table.ti);
1400         for (;;) {
1401                 struct sw_flow *flow;
1402                 u32 bucket, obj;
1403
1404                 bucket = cb->args[0];
1405                 obj = cb->args[1];
1406                 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1407                 if (!flow)
1408                         break;
1409
1410                 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1411                                            NETLINK_CB(cb->skb).portid,
1412                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1413                                            OVS_FLOW_CMD_NEW, ufid_flags) < 0)
1414                         break;
1415
1416                 cb->args[0] = bucket;
1417                 cb->args[1] = obj;
1418         }
1419         rcu_read_unlock();
1420         return skb->len;
1421 }
1422
1423 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1424         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1425         [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1426         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1427         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1428         [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1429         [OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1430         [OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1431 };
1432
1433 static struct genl_ops dp_flow_genl_ops[] = {
1434         { .cmd = OVS_FLOW_CMD_NEW,
1435           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1436           .policy = flow_policy,
1437           .doit = ovs_flow_cmd_new
1438         },
1439         { .cmd = OVS_FLOW_CMD_DEL,
1440           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1441           .policy = flow_policy,
1442           .doit = ovs_flow_cmd_del
1443         },
1444         { .cmd = OVS_FLOW_CMD_GET,
1445           .flags = 0,               /* OK for unprivileged users. */
1446           .policy = flow_policy,
1447           .doit = ovs_flow_cmd_get,
1448           .dumpit = ovs_flow_cmd_dump
1449         },
1450         { .cmd = OVS_FLOW_CMD_SET,
1451           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1452           .policy = flow_policy,
1453           .doit = ovs_flow_cmd_set,
1454         },
1455 };
1456
1457 static struct genl_family dp_flow_genl_family = {
1458         .id = GENL_ID_GENERATE,
1459         .hdrsize = sizeof(struct ovs_header),
1460         .name = OVS_FLOW_FAMILY,
1461         .version = OVS_FLOW_VERSION,
1462         .maxattr = OVS_FLOW_ATTR_MAX,
1463         .netnsok = true,
1464         .parallel_ops = true,
1465         .ops = dp_flow_genl_ops,
1466         .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1467         .mcgrps = &ovs_dp_flow_multicast_group,
1468         .n_mcgrps = 1,
1469 };
1470
1471 static size_t ovs_dp_cmd_msg_size(void)
1472 {
1473         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1474
1475         msgsize += nla_total_size(IFNAMSIZ);
1476         msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1477         msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
1478         msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1479
1480         return msgsize;
1481 }
1482
1483 /* Called with ovs_mutex. */
1484 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1485                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1486 {
1487         struct ovs_header *ovs_header;
1488         struct ovs_dp_stats dp_stats;
1489         struct ovs_dp_megaflow_stats dp_megaflow_stats;
1490         int err;
1491
1492         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1493                                    flags, cmd);
1494         if (!ovs_header)
1495                 goto error;
1496
1497         ovs_header->dp_ifindex = get_dpifindex(dp);
1498
1499         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1500         if (err)
1501                 goto nla_put_failure;
1502
1503         get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1504         if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1505                           &dp_stats, OVS_DP_ATTR_PAD))
1506                 goto nla_put_failure;
1507
1508         if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1509                           sizeof(struct ovs_dp_megaflow_stats),
1510                           &dp_megaflow_stats, OVS_DP_ATTR_PAD))
1511                 goto nla_put_failure;
1512
1513         if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1514                 goto nla_put_failure;
1515
1516         genlmsg_end(skb, ovs_header);
1517         return 0;
1518
1519 nla_put_failure:
1520         genlmsg_cancel(skb, ovs_header);
1521 error:
1522         return -EMSGSIZE;
1523 }
1524
1525 static struct sk_buff *ovs_dp_cmd_alloc_info(struct genl_info *info)
1526 {
1527         return genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info, GFP_KERNEL);
1528 }
1529
1530 /* Called with rcu_read_lock or ovs_mutex. */
1531 static struct datapath *lookup_datapath(struct net *net,
1532                                         const struct ovs_header *ovs_header,
1533                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1534 {
1535         struct datapath *dp;
1536
1537         if (!a[OVS_DP_ATTR_NAME])
1538                 dp = get_dp(net, ovs_header->dp_ifindex);
1539         else {
1540                 struct vport *vport;
1541
1542                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1543                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1544         }
1545         return dp ? dp : ERR_PTR(-ENODEV);
1546 }
1547
1548 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1549 {
1550         struct datapath *dp;
1551
1552         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1553         if (IS_ERR(dp))
1554                 return;
1555
1556         WARN(dp->user_features, "Dropping previously announced user features\n");
1557         dp->user_features = 0;
1558 }
1559
1560 static void ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1561 {
1562         if (a[OVS_DP_ATTR_USER_FEATURES])
1563                 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1564 }
1565
1566 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1567 {
1568         struct nlattr **a = info->attrs;
1569         struct vport_parms parms;
1570         struct sk_buff *reply;
1571         struct datapath *dp;
1572         struct vport *vport;
1573         struct ovs_net *ovs_net;
1574         int err, i;
1575
1576         err = -EINVAL;
1577         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1578                 goto err;
1579
1580         reply = ovs_dp_cmd_alloc_info(info);
1581         if (!reply)
1582                 return -ENOMEM;
1583
1584         err = -ENOMEM;
1585         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1586         if (dp == NULL)
1587                 goto err_free_reply;
1588
1589         ovs_dp_set_net(dp, sock_net(skb->sk));
1590
1591         /* Allocate table. */
1592         err = ovs_flow_tbl_init(&dp->table);
1593         if (err)
1594                 goto err_free_dp;
1595
1596         dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1597         if (!dp->stats_percpu) {
1598                 err = -ENOMEM;
1599                 goto err_destroy_table;
1600         }
1601
1602         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1603                             GFP_KERNEL);
1604         if (!dp->ports) {
1605                 err = -ENOMEM;
1606                 goto err_destroy_percpu;
1607         }
1608
1609         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1610                 INIT_HLIST_HEAD(&dp->ports[i]);
1611
1612         /* Set up our datapath device. */
1613         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1614         parms.type = OVS_VPORT_TYPE_INTERNAL;
1615         parms.options = NULL;
1616         parms.dp = dp;
1617         parms.port_no = OVSP_LOCAL;
1618         parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1619
1620         ovs_dp_change(dp, a);
1621
1622         /* So far only local changes have been made, now need the lock. */
1623         ovs_lock();
1624
1625         vport = new_vport(&parms);
1626         if (IS_ERR(vport)) {
1627                 err = PTR_ERR(vport);
1628                 if (err == -EBUSY)
1629                         err = -EEXIST;
1630
1631                 if (err == -EEXIST) {
1632                         /* An outdated user space instance that does not understand
1633                          * the concept of user_features has attempted to create a new
1634                          * datapath and is likely to reuse it. Drop all user features.
1635                          */
1636                         if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1637                                 ovs_dp_reset_user_features(skb, info);
1638                 }
1639
1640                 goto err_destroy_ports_array;
1641         }
1642
1643         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1644                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1645         BUG_ON(err < 0);
1646
1647         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1648         list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1649
1650         ovs_unlock();
1651
1652         ovs_notify(&dp_datapath_genl_family, &ovs_dp_datapath_multicast_group, reply, info);
1653         return 0;
1654
1655 err_destroy_ports_array:
1656         ovs_unlock();
1657         kfree(dp->ports);
1658 err_destroy_percpu:
1659         free_percpu(dp->stats_percpu);
1660 err_destroy_table:
1661         ovs_flow_tbl_destroy(&dp->table);
1662 err_free_dp:
1663         kfree(dp);
1664 err_free_reply:
1665         kfree_skb(reply);
1666 err:
1667         return err;
1668 }
1669
1670 /* Called with ovs_mutex. */
1671 static void __dp_destroy(struct datapath *dp)
1672 {
1673         int i;
1674
1675         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1676                 struct vport *vport;
1677                 struct hlist_node *n;
1678
1679                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1680                         if (vport->port_no != OVSP_LOCAL)
1681                                 ovs_dp_detach_port(vport);
1682         }
1683
1684         list_del_rcu(&dp->list_node);
1685
1686         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1687          * all ports in datapath are destroyed first before freeing datapath.
1688          */
1689         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1690
1691         /* RCU destroy the flow table */
1692         call_rcu(&dp->rcu, destroy_dp_rcu);
1693 }
1694
1695 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1696 {
1697         struct sk_buff *reply;
1698         struct datapath *dp;
1699         int err;
1700
1701         reply = ovs_dp_cmd_alloc_info(info);
1702         if (!reply)
1703                 return -ENOMEM;
1704
1705         ovs_lock();
1706         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1707         err = PTR_ERR(dp);
1708         if (IS_ERR(dp))
1709                 goto err_unlock_free;
1710
1711         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1712                                    info->snd_seq, 0, OVS_DP_CMD_DEL);
1713         BUG_ON(err < 0);
1714
1715         __dp_destroy(dp);
1716         ovs_unlock();
1717
1718         ovs_notify(&dp_datapath_genl_family, &ovs_dp_datapath_multicast_group, reply, info);
1719         return 0;
1720
1721 err_unlock_free:
1722         ovs_unlock();
1723         kfree_skb(reply);
1724         return err;
1725 }
1726
1727 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1728 {
1729         struct sk_buff *reply;
1730         struct datapath *dp;
1731         int err;
1732
1733         reply = ovs_dp_cmd_alloc_info(info);
1734         if (!reply)
1735                 return -ENOMEM;
1736
1737         ovs_lock();
1738         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1739         err = PTR_ERR(dp);
1740         if (IS_ERR(dp))
1741                 goto err_unlock_free;
1742
1743         ovs_dp_change(dp, info->attrs);
1744
1745         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1746                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1747         BUG_ON(err < 0);
1748
1749         ovs_unlock();
1750
1751         ovs_notify(&dp_datapath_genl_family, &ovs_dp_datapath_multicast_group, reply, info);
1752         return 0;
1753
1754 err_unlock_free:
1755         ovs_unlock();
1756         kfree_skb(reply);
1757         return err;
1758 }
1759
1760 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1761 {
1762         struct sk_buff *reply;
1763         struct datapath *dp;
1764         int err;
1765
1766         reply = ovs_dp_cmd_alloc_info(info);
1767         if (!reply)
1768                 return -ENOMEM;
1769
1770         ovs_lock();
1771         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1772         if (IS_ERR(dp)) {
1773                 err = PTR_ERR(dp);
1774                 goto err_unlock_free;
1775         }
1776         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1777                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1778         BUG_ON(err < 0);
1779         ovs_unlock();
1780
1781         return genlmsg_reply(reply, info);
1782
1783 err_unlock_free:
1784         ovs_unlock();
1785         kfree_skb(reply);
1786         return err;
1787 }
1788
1789 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1790 {
1791         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1792         struct datapath *dp;
1793         int skip = cb->args[0];
1794         int i = 0;
1795
1796         ovs_lock();
1797         list_for_each_entry(dp, &ovs_net->dps, list_node) {
1798                 if (i >= skip &&
1799                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1800                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1801                                          OVS_DP_CMD_NEW) < 0)
1802                         break;
1803                 i++;
1804         }
1805         ovs_unlock();
1806
1807         cb->args[0] = i;
1808
1809         return skb->len;
1810 }
1811
1812 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1813         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1814         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1815         [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1816 };
1817
1818 static struct genl_ops dp_datapath_genl_ops[] = {
1819         { .cmd = OVS_DP_CMD_NEW,
1820           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1821           .policy = datapath_policy,
1822           .doit = ovs_dp_cmd_new
1823         },
1824         { .cmd = OVS_DP_CMD_DEL,
1825           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1826           .policy = datapath_policy,
1827           .doit = ovs_dp_cmd_del
1828         },
1829         { .cmd = OVS_DP_CMD_GET,
1830           .flags = 0,               /* OK for unprivileged users. */
1831           .policy = datapath_policy,
1832           .doit = ovs_dp_cmd_get,
1833           .dumpit = ovs_dp_cmd_dump
1834         },
1835         { .cmd = OVS_DP_CMD_SET,
1836           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1837           .policy = datapath_policy,
1838           .doit = ovs_dp_cmd_set,
1839         },
1840 };
1841
1842 static struct genl_family dp_datapath_genl_family = {
1843         .id = GENL_ID_GENERATE,
1844         .hdrsize = sizeof(struct ovs_header),
1845         .name = OVS_DATAPATH_FAMILY,
1846         .version = OVS_DATAPATH_VERSION,
1847         .maxattr = OVS_DP_ATTR_MAX,
1848         .netnsok = true,
1849         .parallel_ops = true,
1850         .ops = dp_datapath_genl_ops,
1851         .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1852         .mcgrps = &ovs_dp_datapath_multicast_group,
1853         .n_mcgrps = 1,
1854 };
1855
1856 /* Called with ovs_mutex or RCU read lock. */
1857 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1858                                    u32 portid, u32 seq, u32 flags, u8 cmd)
1859 {
1860         struct ovs_header *ovs_header;
1861         struct ovs_vport_stats vport_stats;
1862         int err;
1863
1864         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1865                                  flags, cmd);
1866         if (!ovs_header)
1867                 return -EMSGSIZE;
1868
1869         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1870
1871         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1872             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1873             nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1874                            ovs_vport_name(vport)))
1875                 goto nla_put_failure;
1876
1877         ovs_vport_get_stats(vport, &vport_stats);
1878         if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
1879                           sizeof(struct ovs_vport_stats), &vport_stats,
1880                           OVS_VPORT_ATTR_PAD))
1881                 goto nla_put_failure;
1882
1883         if (ovs_vport_get_upcall_portids(vport, skb))
1884                 goto nla_put_failure;
1885
1886         err = ovs_vport_get_options(vport, skb);
1887         if (err == -EMSGSIZE)
1888                 goto error;
1889
1890         genlmsg_end(skb, ovs_header);
1891         return 0;
1892
1893 nla_put_failure:
1894         err = -EMSGSIZE;
1895 error:
1896         genlmsg_cancel(skb, ovs_header);
1897         return err;
1898 }
1899
1900 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1901 {
1902         return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1903 }
1904
1905 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1906 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1907                                          u32 seq, u8 cmd)
1908 {
1909         struct sk_buff *skb;
1910         int retval;
1911
1912         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1913         if (!skb)
1914                 return ERR_PTR(-ENOMEM);
1915
1916         retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1917         BUG_ON(retval < 0);
1918
1919         return skb;
1920 }
1921
1922 /* Called with ovs_mutex or RCU read lock. */
1923 static struct vport *lookup_vport(struct net *net,
1924                                   const struct ovs_header *ovs_header,
1925                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1926 {
1927         struct datapath *dp;
1928         struct vport *vport;
1929
1930         if (a[OVS_VPORT_ATTR_NAME]) {
1931                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1932                 if (!vport)
1933                         return ERR_PTR(-ENODEV);
1934                 if (ovs_header->dp_ifindex &&
1935                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1936                         return ERR_PTR(-ENODEV);
1937                 return vport;
1938         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1939                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1940
1941                 if (port_no >= DP_MAX_PORTS)
1942                         return ERR_PTR(-EFBIG);
1943
1944                 dp = get_dp(net, ovs_header->dp_ifindex);
1945                 if (!dp)
1946                         return ERR_PTR(-ENODEV);
1947
1948                 vport = ovs_vport_ovsl_rcu(dp, port_no);
1949                 if (!vport)
1950                         return ERR_PTR(-ENODEV);
1951                 return vport;
1952         } else
1953                 return ERR_PTR(-EINVAL);
1954 }
1955
1956 /* Called with ovs_mutex */
1957 static void update_headroom(struct datapath *dp)
1958 {
1959         unsigned dev_headroom, max_headroom = 0;
1960         struct net_device *dev;
1961         struct vport *vport;
1962         int i;
1963
1964         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1965                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1966                         dev = vport->dev;
1967                         dev_headroom = netdev_get_fwd_headroom(dev);
1968                         if (dev_headroom > max_headroom)
1969                                 max_headroom = dev_headroom;
1970                 }
1971         }
1972
1973         dp->max_headroom = max_headroom;
1974         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1975                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node)
1976                         netdev_set_rx_headroom(vport->dev, max_headroom);
1977 }
1978
1979 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1980 {
1981         struct nlattr **a = info->attrs;
1982         struct ovs_header *ovs_header = info->userhdr;
1983         struct vport_parms parms;
1984         struct sk_buff *reply;
1985         struct vport *vport;
1986         struct datapath *dp;
1987         u32 port_no;
1988         int err;
1989
1990         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1991             !a[OVS_VPORT_ATTR_UPCALL_PID])
1992                 return -EINVAL;
1993
1994         port_no = a[OVS_VPORT_ATTR_PORT_NO]
1995                 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
1996         if (port_no >= DP_MAX_PORTS)
1997                 return -EFBIG;
1998
1999         reply = ovs_vport_cmd_alloc_info();
2000         if (!reply)
2001                 return -ENOMEM;
2002
2003         ovs_lock();
2004 restart:
2005         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2006         err = -ENODEV;
2007         if (!dp)
2008                 goto exit_unlock_free;
2009
2010         if (port_no) {
2011                 vport = ovs_vport_ovsl(dp, port_no);
2012                 err = -EBUSY;
2013                 if (vport)
2014                         goto exit_unlock_free;
2015         } else {
2016                 for (port_no = 1; ; port_no++) {
2017                         if (port_no >= DP_MAX_PORTS) {
2018                                 err = -EFBIG;
2019                                 goto exit_unlock_free;
2020                         }
2021                         vport = ovs_vport_ovsl(dp, port_no);
2022                         if (!vport)
2023                                 break;
2024                 }
2025         }
2026
2027         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2028         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2029         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2030         parms.dp = dp;
2031         parms.port_no = port_no;
2032         parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
2033
2034         vport = new_vport(&parms);
2035         err = PTR_ERR(vport);
2036         if (IS_ERR(vport)) {
2037                 if (err == -EAGAIN)
2038                         goto restart;
2039                 goto exit_unlock_free;
2040         }
2041
2042         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2043                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2044         BUG_ON(err < 0);
2045
2046         if (netdev_get_fwd_headroom(vport->dev) > dp->max_headroom)
2047                 update_headroom(dp);
2048         else
2049                 netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2050
2051         ovs_unlock();
2052
2053         ovs_notify(&dp_vport_genl_family, &ovs_dp_vport_multicast_group, reply, info);
2054         return 0;
2055
2056 exit_unlock_free:
2057         ovs_unlock();
2058         kfree_skb(reply);
2059         return err;
2060 }
2061
2062 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2063 {
2064         struct nlattr **a = info->attrs;
2065         struct sk_buff *reply;
2066         struct vport *vport;
2067         int err;
2068
2069         reply = ovs_vport_cmd_alloc_info();
2070         if (!reply)
2071                 return -ENOMEM;
2072
2073         ovs_lock();
2074         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2075         err = PTR_ERR(vport);
2076         if (IS_ERR(vport))
2077                 goto exit_unlock_free;
2078
2079         if (a[OVS_VPORT_ATTR_TYPE] &&
2080             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2081                 err = -EINVAL;
2082                 goto exit_unlock_free;
2083         }
2084
2085         if (a[OVS_VPORT_ATTR_OPTIONS]) {
2086                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2087                 if (err)
2088                         goto exit_unlock_free;
2089         }
2090
2091         if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2092                 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2093
2094                 err = ovs_vport_set_upcall_portids(vport, ids);
2095                 if (err)
2096                         goto exit_unlock_free;
2097         }
2098
2099         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2100                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2101         BUG_ON(err < 0);
2102         ovs_unlock();
2103
2104         ovs_notify(&dp_vport_genl_family, &ovs_dp_vport_multicast_group, reply, info);
2105         return 0;
2106
2107 exit_unlock_free:
2108         ovs_unlock();
2109         kfree_skb(reply);
2110         return err;
2111 }
2112
2113 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2114 {
2115         bool must_update_headroom = false;
2116         struct nlattr **a = info->attrs;
2117         struct sk_buff *reply;
2118         struct datapath *dp;
2119         struct vport *vport;
2120         int err;
2121
2122         reply = ovs_vport_cmd_alloc_info();
2123         if (!reply)
2124                 return -ENOMEM;
2125
2126         ovs_lock();
2127         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2128         err = PTR_ERR(vport);
2129         if (IS_ERR(vport))
2130                 goto exit_unlock_free;
2131
2132         if (vport->port_no == OVSP_LOCAL) {
2133                 err = -EINVAL;
2134                 goto exit_unlock_free;
2135         }
2136
2137         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2138                                       info->snd_seq, 0, OVS_VPORT_CMD_DEL);
2139         BUG_ON(err < 0);
2140
2141         /* the vport deletion may trigger dp headroom update */
2142         dp = vport->dp;
2143         if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2144                 must_update_headroom = true;
2145         netdev_reset_rx_headroom(vport->dev);
2146         ovs_dp_detach_port(vport);
2147
2148         if (must_update_headroom)
2149                 update_headroom(dp);
2150
2151         ovs_unlock();
2152
2153         ovs_notify(&dp_vport_genl_family, &ovs_dp_vport_multicast_group, reply, info);
2154         return 0;
2155
2156 exit_unlock_free:
2157         ovs_unlock();
2158         kfree_skb(reply);
2159         return err;
2160 }
2161
2162 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2163 {
2164         struct nlattr **a = info->attrs;
2165         struct ovs_header *ovs_header = info->userhdr;
2166         struct sk_buff *reply;
2167         struct vport *vport;
2168         int err;
2169
2170         reply = ovs_vport_cmd_alloc_info();
2171         if (!reply)
2172                 return -ENOMEM;
2173
2174         rcu_read_lock();
2175         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2176         err = PTR_ERR(vport);
2177         if (IS_ERR(vport))
2178                 goto exit_unlock_free;
2179         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2180                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2181         BUG_ON(err < 0);
2182         rcu_read_unlock();
2183
2184         return genlmsg_reply(reply, info);
2185
2186 exit_unlock_free:
2187         rcu_read_unlock();
2188         kfree_skb(reply);
2189         return err;
2190 }
2191
2192 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2193 {
2194         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2195         struct datapath *dp;
2196         int bucket = cb->args[0], skip = cb->args[1];
2197         int i, j = 0;
2198
2199         rcu_read_lock();
2200         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2201         if (!dp) {
2202                 rcu_read_unlock();
2203                 return -ENODEV;
2204         }
2205         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2206                 struct vport *vport;
2207
2208                 j = 0;
2209                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2210                         if (j >= skip &&
2211                             ovs_vport_cmd_fill_info(vport, skb,
2212                                                     NETLINK_CB(cb->skb).portid,
2213                                                     cb->nlh->nlmsg_seq,
2214                                                     NLM_F_MULTI,
2215                                                     OVS_VPORT_CMD_NEW) < 0)
2216                                 goto out;
2217
2218                         j++;
2219                 }
2220                 skip = 0;
2221         }
2222 out:
2223         rcu_read_unlock();
2224
2225         cb->args[0] = i;
2226         cb->args[1] = j;
2227
2228         return skb->len;
2229 }
2230
2231 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2232         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2233         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2234         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2235         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2236         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
2237         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2238 };
2239
2240 static struct genl_ops dp_vport_genl_ops[] = {
2241         { .cmd = OVS_VPORT_CMD_NEW,
2242           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2243           .policy = vport_policy,
2244           .doit = ovs_vport_cmd_new
2245         },
2246         { .cmd = OVS_VPORT_CMD_DEL,
2247           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2248           .policy = vport_policy,
2249           .doit = ovs_vport_cmd_del
2250         },
2251         { .cmd = OVS_VPORT_CMD_GET,
2252           .flags = 0,               /* OK for unprivileged users. */
2253           .policy = vport_policy,
2254           .doit = ovs_vport_cmd_get,
2255           .dumpit = ovs_vport_cmd_dump
2256         },
2257         { .cmd = OVS_VPORT_CMD_SET,
2258           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2259           .policy = vport_policy,
2260           .doit = ovs_vport_cmd_set,
2261         },
2262 };
2263
2264 struct genl_family dp_vport_genl_family = {
2265         .id = GENL_ID_GENERATE,
2266         .hdrsize = sizeof(struct ovs_header),
2267         .name = OVS_VPORT_FAMILY,
2268         .version = OVS_VPORT_VERSION,
2269         .maxattr = OVS_VPORT_ATTR_MAX,
2270         .netnsok = true,
2271         .parallel_ops = true,
2272         .ops = dp_vport_genl_ops,
2273         .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2274         .mcgrps = &ovs_dp_vport_multicast_group,
2275         .n_mcgrps = 1,
2276 };
2277
2278 static struct genl_family *dp_genl_families[] = {
2279         &dp_datapath_genl_family,
2280         &dp_vport_genl_family,
2281         &dp_flow_genl_family,
2282         &dp_packet_genl_family,
2283 };
2284
2285 static void dp_unregister_genl(int n_families)
2286 {
2287         int i;
2288
2289         for (i = 0; i < n_families; i++)
2290                 genl_unregister_family(dp_genl_families[i]);
2291 }
2292
2293 static int dp_register_genl(void)
2294 {
2295         int err;
2296         int i;
2297
2298         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2299
2300                 err = genl_register_family(dp_genl_families[i]);
2301                 if (err)
2302                         goto error;
2303         }
2304
2305         return 0;
2306
2307 error:
2308         dp_unregister_genl(i);
2309         return err;
2310 }
2311
2312 static int __net_init ovs_init_net(struct net *net)
2313 {
2314         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2315
2316         INIT_LIST_HEAD(&ovs_net->dps);
2317         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2318         ovs_ct_init(net);
2319         return 0;
2320 }
2321
2322 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2323                                             struct list_head *head)
2324 {
2325         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2326         struct datapath *dp;
2327
2328         list_for_each_entry(dp, &ovs_net->dps, list_node) {
2329                 int i;
2330
2331                 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2332                         struct vport *vport;
2333
2334                         hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2335
2336                                 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2337                                         continue;
2338
2339                                 if (dev_net(vport->dev) == dnet)
2340                                         list_add(&vport->detach_list, head);
2341                         }
2342                 }
2343         }
2344 }
2345
2346 static void __net_exit ovs_exit_net(struct net *dnet)
2347 {
2348         struct datapath *dp, *dp_next;
2349         struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2350         struct vport *vport, *vport_next;
2351         struct net *net;
2352         LIST_HEAD(head);
2353
2354         ovs_ct_exit(dnet);
2355         ovs_lock();
2356         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2357                 __dp_destroy(dp);
2358
2359         rtnl_lock();
2360         for_each_net(net)
2361                 list_vports_from_net(net, dnet, &head);
2362         rtnl_unlock();
2363
2364         /* Detach all vports from given namespace. */
2365         list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2366                 list_del(&vport->detach_list);
2367                 ovs_dp_detach_port(vport);
2368         }
2369
2370         ovs_unlock();
2371
2372         cancel_work_sync(&ovs_net->dp_notify_work);
2373 }
2374
2375 static struct pernet_operations ovs_net_ops = {
2376         .init = ovs_init_net,
2377         .exit = ovs_exit_net,
2378         .id   = &ovs_net_id,
2379         .size = sizeof(struct ovs_net),
2380 };
2381
2382 static int __init dp_init(void)
2383 {
2384         int err;
2385
2386         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2387
2388         pr_info("Open vSwitch switching datapath %s\n", VERSION);
2389
2390         err = compat_init();
2391         if (err)
2392                 goto error;
2393
2394         err = action_fifos_init();
2395         if (err)
2396                 goto error_compat_exit;
2397
2398         err = ovs_internal_dev_rtnl_link_register();
2399         if (err)
2400                 goto error_action_fifos_exit;
2401
2402         err = ovs_flow_init();
2403         if (err)
2404                 goto error_unreg_rtnl_link;
2405
2406         err = ovs_vport_init();
2407         if (err)
2408                 goto error_flow_exit;
2409
2410         err = register_pernet_device(&ovs_net_ops);
2411         if (err)
2412                 goto error_vport_exit;
2413
2414         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2415         if (err)
2416                 goto error_netns_exit;
2417
2418         err = ovs_netdev_init();
2419         if (err)
2420                 goto error_unreg_notifier;
2421
2422         err = dp_register_genl();
2423         if (err < 0)
2424                 goto error_unreg_netdev;
2425
2426         return 0;
2427
2428 error_unreg_netdev:
2429         ovs_netdev_exit();
2430 error_unreg_notifier:
2431         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2432 error_netns_exit:
2433         unregister_pernet_device(&ovs_net_ops);
2434 error_vport_exit:
2435         ovs_vport_exit();
2436 error_flow_exit:
2437         ovs_flow_exit();
2438 error_unreg_rtnl_link:
2439         ovs_internal_dev_rtnl_link_unregister();
2440 error_action_fifos_exit:
2441         action_fifos_exit();
2442 error_compat_exit:
2443         compat_exit();
2444 error:
2445         return err;
2446 }
2447
2448 static void dp_cleanup(void)
2449 {
2450         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2451         ovs_netdev_exit();
2452         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2453         unregister_pernet_device(&ovs_net_ops);
2454         rcu_barrier();
2455         ovs_vport_exit();
2456         ovs_flow_exit();
2457         ovs_internal_dev_rtnl_link_unregister();
2458         action_fifos_exit();
2459         compat_exit();
2460 }
2461
2462 module_init(dp_init);
2463 module_exit(dp_cleanup);
2464
2465 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2466 MODULE_LICENSE("GPL");
2467 MODULE_VERSION(VERSION);