kernel/net/openvswitch/datapath.c

   1 /*
   2  * Copyright (c) 2007-2014 Nicira, Inc.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of version 2 of the GNU General Public
   6  * License as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program; if not, write to the Free Software
  15  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
  16  * 02110-1301, USA
  17  */
  18
  19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  20
  21 #include <linux/init.h>
  22 #include <linux/module.h>
  23 #include <linux/if_arp.h>
  24 #include <linux/if_vlan.h>
  25 #include <linux/in.h>
  26 #include <linux/ip.h>
  27 #include <linux/jhash.h>
  28 #include <linux/delay.h>
  29 #include <linux/time.h>
  30 #include <linux/etherdevice.h>
  31 #include <linux/genetlink.h>
  32 #include <linux/kernel.h>
  33 #include <linux/kthread.h>
  34 #include <linux/mutex.h>
  35 #include <linux/percpu.h>
  36 #include <linux/rcupdate.h>
  37 #include <linux/tcp.h>
  38 #include <linux/udp.h>
  39 #include <linux/ethtool.h>
  40 #include <linux/wait.h>
  41 #include <asm/div64.h>
  42 #include <linux/highmem.h>
  43 #include <linux/netfilter_bridge.h>
  44 #include <linux/netfilter_ipv4.h>
  45 #include <linux/inetdevice.h>
  46 #include <linux/list.h>
  47 #include <linux/openvswitch.h>
  48 #include <linux/rculist.h>
  49 #include <linux/dmi.h>
  50 #include <net/genetlink.h>
  51 #include <net/net_namespace.h>
  52 #include <net/netns/generic.h>
  53
  54 #include "datapath.h"
  55 #include "flow.h"
  56 #include "flow_table.h"
  57 #include "flow_netlink.h"
  58 #include "vport-internal_dev.h"
  59 #include "vport-netdev.h"
  60
  61 int ovs_net_id __read_mostly;
  62 EXPORT_SYMBOL_GPL(ovs_net_id);
  63
  64 static struct genl_family dp_packet_genl_family;
  65 static struct genl_family dp_flow_genl_family;
  66 static struct genl_family dp_datapath_genl_family;
  67
  68 static const struct nla_policy flow_policy[];
  69
  70 static const struct genl_multicast_group ovs_dp_flow_multicast_group = {
  71         .name = OVS_FLOW_MCGROUP,
  72 };
  73
  74 static const struct genl_multicast_group ovs_dp_datapath_multicast_group = {
  75         .name = OVS_DATAPATH_MCGROUP,
  76 };
  77
  78 static const struct genl_multicast_group ovs_dp_vport_multicast_group = {
  79         .name = OVS_VPORT_MCGROUP,
  80 };
  81
  82 /* Check if need to build a reply message.
  83  * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
  84 static bool ovs_must_notify(struct genl_family *family, struct genl_info *info,
  85                             unsigned int group)
  86 {
  87         return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
  88                genl_has_listeners(family, genl_info_net(info), group);
  89 }
  90
  91 static void ovs_notify(struct genl_family *family,
  92                        struct sk_buff *skb, struct genl_info *info)
  93 {
  94         genl_notify(family, skb, info, 0, GFP_KERNEL);
  95 }
  96
  97 /**
  98  * DOC: Locking:
  99  *
 100  * All writes e.g. Writes to device state (add/remove datapath, port, set
 101  * operations on vports, etc.), Writes to other state (flow table
 102  * modifications, set miscellaneous datapath parameters, etc.) are protected
 103  * by ovs_lock.
 104  *
 105  * Reads are protected by RCU.
 106  *
 107  * There are a few special cases (mostly stats) that have their own
 108  * synchronization but they nest under all of above and don't interact with
 109  * each other.
 110  *
 111  * The RTNL lock nests inside ovs_mutex.
 112  */
 113
 114 static DEFINE_MUTEX(ovs_mutex);
 115
 116 void ovs_lock(void)
 117 {
 118         mutex_lock(&ovs_mutex);
 119 }
 120
 121 void ovs_unlock(void)
 122 {
 123         mutex_unlock(&ovs_mutex);
 124 }
 125
 126 #ifdef CONFIG_LOCKDEP
 127 int lockdep_ovsl_is_held(void)
 128 {
 129         if (debug_locks)
 130                 return lockdep_is_held(&ovs_mutex);
 131         else
 132                 return 1;
 133 }
 134 EXPORT_SYMBOL_GPL(lockdep_ovsl_is_held);
 135 #endif
 136
 137 static struct vport *new_vport(const struct vport_parms *);
 138 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
 139                              const struct sw_flow_key *,
 140                              const struct dp_upcall_info *);
 141 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
 142                                   const struct sw_flow_key *,
 143                                   const struct dp_upcall_info *);
 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);
 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 {
 305         struct dp_stats_percpu *stats;
 306         int err;
 307
 308         if (upcall_info->portid == 0) {
 309                 err = -ENOTCONN;
 310                 goto err;
 311         }
 312
 313         if (!skb_is_gso(skb))
 314                 err = queue_userspace_packet(dp, skb, key, upcall_info);
 315         else
 316                 err = queue_gso_packets(dp, skb, key, upcall_info);
 317         if (err)
 318                 goto err;
 319
 320         return 0;
 321
 322 err:
 323         stats = this_cpu_ptr(dp->stats_percpu);
 324
 325         u64_stats_update_begin(&stats->syncp);
 326         stats->n_lost++;
 327         u64_stats_update_end(&stats->syncp);
 328
 329         return err;
 330 }
 331
 332 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
 333                              const struct sw_flow_key *key,
 334                              const struct dp_upcall_info *upcall_info)
 335 {
 336         unsigned short gso_type = skb_shinfo(skb)->gso_type;
 337         struct sw_flow_key later_key;
 338         struct sk_buff *segs, *nskb;
 339         int err;
 340
 341         BUILD_BUG_ON(sizeof(*OVS_CB(skb)) > SKB_SGO_CB_OFFSET);
 342         segs = __skb_gso_segment(skb, NETIF_F_SG, false);
 343         if (IS_ERR(segs))
 344                 return PTR_ERR(segs);
 345         if (segs == NULL)
 346                 return -EINVAL;
 347
 348         if (gso_type & SKB_GSO_UDP) {
 349                 /* The initial flow key extracted by ovs_flow_key_extract()
 350                  * in this case is for a first fragment, so we need to
 351                  * properly mark later fragments.
 352                  */
 353                 later_key = *key;
 354                 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
 355         }
 356
 357         /* Queue all of the segments. */
 358         skb = segs;
 359         do {
 360                 if (gso_type & SKB_GSO_UDP && skb != segs)
 361                         key = &later_key;
 362
 363                 err = queue_userspace_packet(dp, skb, key, upcall_info);
 364                 if (err)
 365                         break;
 366
 367         } while ((skb = skb->next));
 368
 369         /* Free all of the segments. */
 370         skb = segs;
 371         do {
 372                 nskb = skb->next;
 373                 if (err)
 374                         kfree_skb(skb);
 375                 else
 376                         consume_skb(skb);
 377         } while ((skb = nskb));
 378         return err;
 379 }
 380
 381 static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
 382                               unsigned int hdrlen)
 383 {
 384         size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
 385                 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
 386                 + nla_total_size(ovs_key_attr_size()); /* OVS_PACKET_ATTR_KEY */
 387
 388         /* OVS_PACKET_ATTR_USERDATA */
 389         if (upcall_info->userdata)
 390                 size += NLA_ALIGN(upcall_info->userdata->nla_len);
 391
 392         /* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
 393         if (upcall_info->egress_tun_info)
 394                 size += nla_total_size(ovs_tun_key_attr_size());
 395
 396         /* OVS_PACKET_ATTR_ACTIONS */
 397         if (upcall_info->actions_len)
 398                 size += nla_total_size(upcall_info->actions_len);
 399
 400         /* OVS_PACKET_ATTR_MRU */
 401         if (upcall_info->mru)
 402                 size += nla_total_size(sizeof(upcall_info->mru));
 403
 404         return size;
 405 }
 406
 407 static void pad_packet(struct datapath *dp, struct sk_buff *skb)
 408 {
 409         if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
 410                 size_t plen = NLA_ALIGN(skb->len) - skb->len;
 411
 412                 if (plen > 0)
 413                         memset(skb_put(skb, plen), 0, plen);
 414         }
 415 }
 416
 417 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
 418                                   const struct sw_flow_key *key,
 419                                   const struct dp_upcall_info *upcall_info)
 420 {
 421         struct ovs_header *upcall;
 422         struct sk_buff *nskb = NULL;
 423         struct sk_buff *user_skb = NULL; /* to be queued to userspace */
 424         struct nlattr *nla;
 425         struct genl_info info = {
 426                 .dst_sk = ovs_dp_get_net(dp)->genl_sock,
 427                 .snd_portid = upcall_info->portid,
 428         };
 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);
 469         user_skb = genlmsg_new_unicast(len, &info, 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         /* Only reserve room for attribute header, packet data is added
 517          * in skb_zerocopy() */
 518         if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
 519                 err = -ENOBUFS;
 520                 goto out;
 521         }
 522         nla->nla_len = nla_attr_size(skb->len);
 523
 524         err = skb_zerocopy(user_skb, skb, skb->len, hlen);
 525         if (err)
 526                 goto out;
 527
 528         /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
 529         pad_packet(dp, user_skb);
 530
 531         ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
 532
 533         err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
 534         user_skb = NULL;
 535 out:
 536         if (err)
 537                 skb_tx_error(skb);
 538         kfree_skb(user_skb);
 539         kfree_skb(nskb);
 540         return err;
 541 }
 542
 543 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 544 {
 545         struct ovs_header *ovs_header = info->userhdr;
 546         struct net *net = sock_net(skb->sk);
 547         struct nlattr **a = info->attrs;
 548         struct sw_flow_actions *acts;
 549         struct sk_buff *packet;
 550         struct sw_flow *flow;
 551         struct sw_flow_actions *sf_acts;
 552         struct datapath *dp;
 553         struct ethhdr *eth;
 554         struct vport *input_vport;
 555         u16 mru = 0;
 556         int len;
 557         int err;
 558         bool log = !a[OVS_PACKET_ATTR_PROBE];
 559
 560         err = -EINVAL;
 561         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 562             !a[OVS_PACKET_ATTR_ACTIONS])
 563                 goto err;
 564
 565         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 566         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 567         err = -ENOMEM;
 568         if (!packet)
 569                 goto err;
 570         skb_reserve(packet, NET_IP_ALIGN);
 571
 572         nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
 573
 574         skb_reset_mac_header(packet);
 575         eth = eth_hdr(packet);
 576
 577         /* Normally, setting the skb 'protocol' field would be handled by a
 578          * call to eth_type_trans(), but it assumes there's a sending
 579          * device, which we may not have. */
 580         if (eth_proto_is_802_3(eth->h_proto))
 581                 packet->protocol = eth->h_proto;
 582         else
 583                 packet->protocol = htons(ETH_P_802_2);
 584
 585         /* Set packet's mru */
 586         if (a[OVS_PACKET_ATTR_MRU]) {
 587                 mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
 588                 packet->ignore_df = 1;
 589         }
 590         OVS_CB(packet)->mru = mru;
 591
 592         /* Build an sw_flow for sending this packet. */
 593         flow = ovs_flow_alloc();
 594         err = PTR_ERR(flow);
 595         if (IS_ERR(flow))
 596                 goto err_kfree_skb;
 597
 598         err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
 599                                              packet, &flow->key, log);
 600         if (err)
 601                 goto err_flow_free;
 602
 603         err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
 604                                    &flow->key, &acts, log);
 605         if (err)
 606                 goto err_flow_free;
 607
 608         rcu_assign_pointer(flow->sf_acts, acts);
 609         packet->priority = flow->key.phy.priority;
 610         packet->mark = flow->key.phy.skb_mark;
 611
 612         rcu_read_lock();
 613         dp = get_dp_rcu(net, ovs_header->dp_ifindex);
 614         err = -ENODEV;
 615         if (!dp)
 616                 goto err_unlock;
 617
 618         input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
 619         if (!input_vport)
 620                 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
 621
 622         if (!input_vport)
 623                 goto err_unlock;
 624
 625         packet->dev = input_vport->dev;
 626         OVS_CB(packet)->input_vport = input_vport;
 627         sf_acts = rcu_dereference(flow->sf_acts);
 628
 629         local_bh_disable();
 630         err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
 631         local_bh_enable();
 632         rcu_read_unlock();
 633
 634         ovs_flow_free(flow, false);
 635         return err;
 636
 637 err_unlock:
 638         rcu_read_unlock();
 639 err_flow_free:
 640         ovs_flow_free(flow, false);
 641 err_kfree_skb:
 642         kfree_skb(packet);
 643 err:
 644         return err;
 645 }
 646
 647 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 648         [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
 649         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 650         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 651         [OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
 652         [OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
 653 };
 654
 655 static const struct genl_ops dp_packet_genl_ops[] = {
 656         { .cmd = OVS_PACKET_CMD_EXECUTE,
 657           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 658           .policy = packet_policy,
 659           .doit = ovs_packet_cmd_execute
 660         }
 661 };
 662
 663 static struct genl_family dp_packet_genl_family = {
 664         .id = GENL_ID_GENERATE,
 665         .hdrsize = sizeof(struct ovs_header),
 666         .name = OVS_PACKET_FAMILY,
 667         .version = OVS_PACKET_VERSION,
 668         .maxattr = OVS_PACKET_ATTR_MAX,
 669         .netnsok = true,
 670         .parallel_ops = true,
 671         .ops = dp_packet_genl_ops,
 672         .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
 673 };
 674
 675 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
 676                          struct ovs_dp_megaflow_stats *mega_stats)
 677 {
 678         int i;
 679
 680         memset(mega_stats, 0, sizeof(*mega_stats));
 681
 682         stats->n_flows = ovs_flow_tbl_count(&dp->table);
 683         mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
 684
 685         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 686
 687         for_each_possible_cpu(i) {
 688                 const struct dp_stats_percpu *percpu_stats;
 689                 struct dp_stats_percpu local_stats;
 690                 unsigned int start;
 691
 692                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 693
 694                 do {
 695                         start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
 696                         local_stats = *percpu_stats;
 697                 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
 698
 699                 stats->n_hit += local_stats.n_hit;
 700                 stats->n_missed += local_stats.n_missed;
 701                 stats->n_lost += local_stats.n_lost;
 702                 mega_stats->n_mask_hit += local_stats.n_mask_hit;
 703         }
 704 }
 705
 706 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
 707 {
 708         return ovs_identifier_is_ufid(sfid) &&
 709                !(ufid_flags & OVS_UFID_F_OMIT_KEY);
 710 }
 711
 712 static bool should_fill_mask(uint32_t ufid_flags)
 713 {
 714         return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
 715 }
 716
 717 static bool should_fill_actions(uint32_t ufid_flags)
 718 {
 719         return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
 720 }
 721
 722 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
 723                                     const struct sw_flow_id *sfid,
 724                                     uint32_t ufid_flags)
 725 {
 726         size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
 727
 728         /* OVS_FLOW_ATTR_UFID */
 729         if (sfid && ovs_identifier_is_ufid(sfid))
 730                 len += nla_total_size(sfid->ufid_len);
 731
 732         /* OVS_FLOW_ATTR_KEY */
 733         if (!sfid || should_fill_key(sfid, ufid_flags))
 734                 len += nla_total_size(ovs_key_attr_size());
 735
 736         /* OVS_FLOW_ATTR_MASK */
 737         if (should_fill_mask(ufid_flags))
 738                 len += nla_total_size(ovs_key_attr_size());
 739
 740         /* OVS_FLOW_ATTR_ACTIONS */
 741         if (should_fill_actions(ufid_flags))
 742                 len += nla_total_size(acts->orig_len);
 743
 744         return len
 745                 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
 746                 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
 747                 + nla_total_size(8); /* OVS_FLOW_ATTR_USED */
 748 }
 749
 750 /* Called with ovs_mutex or RCU read lock. */
 751 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
 752                                    struct sk_buff *skb)
 753 {
 754         struct ovs_flow_stats stats;
 755         __be16 tcp_flags;
 756         unsigned long used;
 757
 758         ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
 759
 760         if (used &&
 761             nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
 762                 return -EMSGSIZE;
 763
 764         if (stats.n_packets &&
 765             nla_put(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats))
 766                 return -EMSGSIZE;
 767
 768         if ((u8)ntohs(tcp_flags) &&
 769              nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
 770                 return -EMSGSIZE;
 771
 772         return 0;
 773 }
 774
 775 /* Called with ovs_mutex or RCU read lock. */
 776 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
 777                                      struct sk_buff *skb, int skb_orig_len)
 778 {
 779         struct nlattr *start;
 780         int err;
 781
 782         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 783          * this is the first flow to be dumped into 'skb'.  This is unusual for
 784          * Netlink but individual action lists can be longer than
 785          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 786          * The userspace caller can always fetch the actions separately if it
 787          * really wants them.  (Most userspace callers in fact don't care.)
 788          *
 789          * This can only fail for dump operations because the skb is always
 790          * properly sized for single flows.
 791          */
 792         start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
 793         if (start) {
 794                 const struct sw_flow_actions *sf_acts;
 795
 796                 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
 797                 err = ovs_nla_put_actions(sf_acts->actions,
 798                                           sf_acts->actions_len, skb);
 799
 800                 if (!err)
 801                         nla_nest_end(skb, start);
 802                 else {
 803                         if (skb_orig_len)
 804                                 return err;
 805
 806                         nla_nest_cancel(skb, start);
 807                 }
 808         } else if (skb_orig_len) {
 809                 return -EMSGSIZE;
 810         }
 811
 812         return 0;
 813 }
 814
 815 /* Called with ovs_mutex or RCU read lock. */
 816 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
 817                                   struct sk_buff *skb, u32 portid,
 818                                   u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
 819 {
 820         const int skb_orig_len = skb->len;
 821         struct ovs_header *ovs_header;
 822         int err;
 823
 824         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
 825                                  flags, cmd);
 826         if (!ovs_header)
 827                 return -EMSGSIZE;
 828
 829         ovs_header->dp_ifindex = dp_ifindex;
 830
 831         err = ovs_nla_put_identifier(flow, skb);
 832         if (err)
 833                 goto error;
 834
 835         if (should_fill_key(&flow->id, ufid_flags)) {
 836                 err = ovs_nla_put_masked_key(flow, skb);
 837                 if (err)
 838                         goto error;
 839         }
 840
 841         if (should_fill_mask(ufid_flags)) {
 842                 err = ovs_nla_put_mask(flow, skb);
 843                 if (err)
 844                         goto error;
 845         }
 846
 847         err = ovs_flow_cmd_fill_stats(flow, skb);
 848         if (err)
 849                 goto error;
 850
 851         if (should_fill_actions(ufid_flags)) {
 852                 err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
 853                 if (err)
 854                         goto error;
 855         }
 856
 857         genlmsg_end(skb, ovs_header);
 858         return 0;
 859
 860 error:
 861         genlmsg_cancel(skb, ovs_header);
 862         return err;
 863 }
 864
 865 /* May not be called with RCU read lock. */
 866 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
 867                                                const struct sw_flow_id *sfid,
 868                                                struct genl_info *info,
 869                                                bool always,
 870                                                uint32_t ufid_flags)
 871 {
 872         struct sk_buff *skb;
 873         size_t len;
 874
 875         if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
 876                 return NULL;
 877
 878         len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
 879         skb = genlmsg_new_unicast(len, info, GFP_KERNEL);
 880         if (!skb)
 881                 return ERR_PTR(-ENOMEM);
 882
 883         return skb;
 884 }
 885
 886 /* Called with ovs_mutex. */
 887 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
 888                                                int dp_ifindex,
 889                                                struct genl_info *info, u8 cmd,
 890                                                bool always, u32 ufid_flags)
 891 {
 892         struct sk_buff *skb;
 893         int retval;
 894
 895         skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
 896                                       &flow->id, info, always, ufid_flags);
 897         if (IS_ERR_OR_NULL(skb))
 898                 return skb;
 899
 900         retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
 901                                         info->snd_portid, info->snd_seq, 0,
 902                                         cmd, ufid_flags);
 903         BUG_ON(retval < 0);
 904         return skb;
 905 }
 906
 907 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
 908 {
 909         struct net *net = sock_net(skb->sk);
 910         struct nlattr **a = info->attrs;
 911         struct ovs_header *ovs_header = info->userhdr;
 912         struct sw_flow *flow = NULL, *new_flow;
 913         struct sw_flow_mask mask;
 914         struct sk_buff *reply;
 915         struct datapath *dp;
 916         struct sw_flow_key key;
 917         struct sw_flow_actions *acts;
 918         struct sw_flow_match match;
 919         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
 920         int error;
 921         bool log = !a[OVS_FLOW_ATTR_PROBE];
 922
 923         /* Must have key and actions. */
 924         error = -EINVAL;
 925         if (!a[OVS_FLOW_ATTR_KEY]) {
 926                 OVS_NLERR(log, "Flow key attr not present in new flow.");
 927                 goto error;
 928         }
 929         if (!a[OVS_FLOW_ATTR_ACTIONS]) {
 930                 OVS_NLERR(log, "Flow actions attr not present in new flow.");
 931                 goto error;
 932         }
 933
 934         /* Most of the time we need to allocate a new flow, do it before
 935          * locking.
 936          */
 937         new_flow = ovs_flow_alloc();
 938         if (IS_ERR(new_flow)) {
 939                 error = PTR_ERR(new_flow);
 940                 goto error;
 941         }
 942
 943         /* Extract key. */
 944         ovs_match_init(&match, &key, &mask);
 945         error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
 946                                   a[OVS_FLOW_ATTR_MASK], log);
 947         if (error)
 948                 goto err_kfree_flow;
 949
 950         ovs_flow_mask_key(&new_flow->key, &key, true, &mask);
 951
 952         /* Extract flow identifier. */
 953         error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
 954                                        &key, log);
 955         if (error)
 956                 goto err_kfree_flow;
 957
 958         /* Validate actions. */
 959         error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
 960                                      &new_flow->key, &acts, log);
 961         if (error) {
 962                 OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
 963                 goto err_kfree_flow;
 964         }
 965
 966         reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
 967                                         ufid_flags);
 968         if (IS_ERR(reply)) {
 969                 error = PTR_ERR(reply);
 970                 goto err_kfree_acts;
 971         }
 972
 973         ovs_lock();
 974         dp = get_dp(net, ovs_header->dp_ifindex);
 975         if (unlikely(!dp)) {
 976                 error = -ENODEV;
 977                 goto err_unlock_ovs;
 978         }
 979
 980         /* Check if this is a duplicate flow */
 981         if (ovs_identifier_is_ufid(&new_flow->id))
 982                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
 983         if (!flow)
 984                 flow = ovs_flow_tbl_lookup(&dp->table, &key);
 985         if (likely(!flow)) {
 986                 rcu_assign_pointer(new_flow->sf_acts, acts);
 987
 988                 /* Put flow in bucket. */
 989                 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
 990                 if (unlikely(error)) {
 991                         acts = NULL;
 992                         goto err_unlock_ovs;
 993                 }
 994
 995                 if (unlikely(reply)) {
 996                         error = ovs_flow_cmd_fill_info(new_flow,
 997                                                        ovs_header->dp_ifindex,
 998                                                        reply, info->snd_portid,
 999                                                        info->snd_seq, 0,
1000                                                        OVS_FLOW_CMD_NEW,
1001                                                        ufid_flags);
1002                         BUG_ON(error < 0);
1003                 }
1004                 ovs_unlock();
1005         } else {
1006                 struct sw_flow_actions *old_acts;
1007
1008                 /* Bail out if we're not allowed to modify an existing flow.
1009                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1010                  * because Generic Netlink treats the latter as a dump
1011                  * request.  We also accept NLM_F_EXCL in case that bug ever
1012                  * gets fixed.
1013                  */
1014                 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1015                                                          | NLM_F_EXCL))) {
1016                         error = -EEXIST;
1017                         goto err_unlock_ovs;
1018                 }
1019                 /* The flow identifier has to be the same for flow updates.
1020                  * Look for any overlapping flow.
1021                  */
1022                 if (unlikely(!ovs_flow_cmp(flow, &match))) {
1023                         if (ovs_identifier_is_key(&flow->id))
1024                                 flow = ovs_flow_tbl_lookup_exact(&dp->table,
1025                                                                  &match);
1026                         else /* UFID matches but key is different */
1027                                 flow = NULL;
1028                         if (!flow) {
1029                                 error = -ENOENT;
1030                                 goto err_unlock_ovs;
1031                         }
1032                 }
1033                 /* Update actions. */
1034                 old_acts = ovsl_dereference(flow->sf_acts);
1035                 rcu_assign_pointer(flow->sf_acts, acts);
1036
1037                 if (unlikely(reply)) {
1038                         error = ovs_flow_cmd_fill_info(flow,
1039                                                        ovs_header->dp_ifindex,
1040                                                        reply, info->snd_portid,
1041                                                        info->snd_seq, 0,
1042                                                        OVS_FLOW_CMD_NEW,
1043                                                        ufid_flags);
1044                         BUG_ON(error < 0);
1045                 }
1046                 ovs_unlock();
1047
1048                 ovs_nla_free_flow_actions_rcu(old_acts);
1049                 ovs_flow_free(new_flow, false);
1050         }
1051
1052         if (reply)
1053                 ovs_notify(&dp_flow_genl_family, reply, info);
1054         return 0;
1055
1056 err_unlock_ovs:
1057         ovs_unlock();
1058         kfree_skb(reply);
1059 err_kfree_acts:
1060         ovs_nla_free_flow_actions(acts);
1061 err_kfree_flow:
1062         ovs_flow_free(new_flow, false);
1063 error:
1064         return error;
1065 }
1066
1067 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1068 static struct sw_flow_actions *get_flow_actions(struct net *net,
1069                                                 const struct nlattr *a,
1070                                                 const struct sw_flow_key *key,
1071                                                 const struct sw_flow_mask *mask,
1072                                                 bool log)
1073 {
1074         struct sw_flow_actions *acts;
1075         struct sw_flow_key masked_key;
1076         int error;
1077
1078         ovs_flow_mask_key(&masked_key, key, true, mask);
1079         error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1080         if (error) {
1081                 OVS_NLERR(log,
1082                           "Actions may not be safe on all matching packets");
1083                 return ERR_PTR(error);
1084         }
1085
1086         return acts;
1087 }
1088
1089 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1090 {
1091         struct net *net = sock_net(skb->sk);
1092         struct nlattr **a = info->attrs;
1093         struct ovs_header *ovs_header = info->userhdr;
1094         struct sw_flow_key key;
1095         struct sw_flow *flow;
1096         struct sw_flow_mask mask;
1097         struct sk_buff *reply = NULL;
1098         struct datapath *dp;
1099         struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1100         struct sw_flow_match match;
1101         struct sw_flow_id sfid;
1102         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1103         int error;
1104         bool log = !a[OVS_FLOW_ATTR_PROBE];
1105         bool ufid_present;
1106
1107         /* Extract key. */
1108         error = -EINVAL;
1109         if (!a[OVS_FLOW_ATTR_KEY]) {
1110                 OVS_NLERR(log, "Flow key attribute not present in set flow.");
1111                 goto error;
1112         }
1113
1114         ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1115         ovs_match_init(&match, &key, &mask);
1116         error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1117                                   a[OVS_FLOW_ATTR_MASK], log);
1118         if (error)
1119                 goto error;
1120
1121         /* Validate actions. */
1122         if (a[OVS_FLOW_ATTR_ACTIONS]) {
1123                 acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], &key,
1124                                         &mask, log);
1125                 if (IS_ERR(acts)) {
1126                         error = PTR_ERR(acts);
1127                         goto error;
1128                 }
1129
1130                 /* Can allocate before locking if have acts. */
1131                 reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1132                                                 ufid_flags);
1133                 if (IS_ERR(reply)) {
1134                         error = PTR_ERR(reply);
1135                         goto err_kfree_acts;
1136                 }
1137         }
1138
1139         ovs_lock();
1140         dp = get_dp(net, ovs_header->dp_ifindex);
1141         if (unlikely(!dp)) {
1142                 error = -ENODEV;
1143                 goto err_unlock_ovs;
1144         }
1145         /* Check that the flow exists. */
1146         if (ufid_present)
1147                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1148         else
1149                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1150         if (unlikely(!flow)) {
1151                 error = -ENOENT;
1152                 goto err_unlock_ovs;
1153         }
1154
1155         /* Update actions, if present. */
1156         if (likely(acts)) {
1157                 old_acts = ovsl_dereference(flow->sf_acts);
1158                 rcu_assign_pointer(flow->sf_acts, acts);
1159
1160                 if (unlikely(reply)) {
1161                         error = ovs_flow_cmd_fill_info(flow,
1162                                                        ovs_header->dp_ifindex,
1163                                                        reply, info->snd_portid,
1164                                                        info->snd_seq, 0,
1165                                                        OVS_FLOW_CMD_NEW,
1166                                                        ufid_flags);
1167                         BUG_ON(error < 0);
1168                 }
1169         } else {
1170                 /* Could not alloc without acts before locking. */
1171                 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1172                                                 info, OVS_FLOW_CMD_NEW, false,
1173                                                 ufid_flags);
1174
1175                 if (IS_ERR(reply)) {
1176                         error = PTR_ERR(reply);
1177                         goto err_unlock_ovs;
1178                 }
1179         }
1180
1181         /* Clear stats. */
1182         if (a[OVS_FLOW_ATTR_CLEAR])
1183                 ovs_flow_stats_clear(flow);
1184         ovs_unlock();
1185
1186         if (reply)
1187                 ovs_notify(&dp_flow_genl_family, reply, info);
1188         if (old_acts)
1189                 ovs_nla_free_flow_actions_rcu(old_acts);
1190
1191         return 0;
1192
1193 err_unlock_ovs:
1194         ovs_unlock();
1195         kfree_skb(reply);
1196 err_kfree_acts:
1197         ovs_nla_free_flow_actions(acts);
1198 error:
1199         return error;
1200 }
1201
1202 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1203 {
1204         struct nlattr **a = info->attrs;
1205         struct ovs_header *ovs_header = info->userhdr;
1206         struct net *net = sock_net(skb->sk);
1207         struct sw_flow_key key;
1208         struct sk_buff *reply;
1209         struct sw_flow *flow;
1210         struct datapath *dp;
1211         struct sw_flow_match match;
1212         struct sw_flow_id ufid;
1213         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1214         int err = 0;
1215         bool log = !a[OVS_FLOW_ATTR_PROBE];
1216         bool ufid_present;
1217
1218         ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1219         if (a[OVS_FLOW_ATTR_KEY]) {
1220                 ovs_match_init(&match, &key, NULL);
1221                 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1222                                         log);
1223         } else if (!ufid_present) {
1224                 OVS_NLERR(log,
1225                           "Flow get message rejected, Key attribute missing.");
1226                 err = -EINVAL;
1227         }
1228         if (err)
1229                 return err;
1230
1231         ovs_lock();
1232         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1233         if (!dp) {
1234                 err = -ENODEV;
1235                 goto unlock;
1236         }
1237
1238         if (ufid_present)
1239                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1240         else
1241                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1242         if (!flow) {
1243                 err = -ENOENT;
1244                 goto unlock;
1245         }
1246
1247         reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1248                                         OVS_FLOW_CMD_NEW, true, ufid_flags);
1249         if (IS_ERR(reply)) {
1250                 err = PTR_ERR(reply);
1251                 goto unlock;
1252         }
1253
1254         ovs_unlock();
1255         return genlmsg_reply(reply, info);
1256 unlock:
1257         ovs_unlock();
1258         return err;
1259 }
1260
1261 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1262 {
1263         struct nlattr **a = info->attrs;
1264         struct ovs_header *ovs_header = info->userhdr;
1265         struct net *net = sock_net(skb->sk);
1266         struct sw_flow_key key;
1267         struct sk_buff *reply;
1268         struct sw_flow *flow = NULL;
1269         struct datapath *dp;
1270         struct sw_flow_match match;
1271         struct sw_flow_id ufid;
1272         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1273         int err;
1274         bool log = !a[OVS_FLOW_ATTR_PROBE];
1275         bool ufid_present;
1276
1277         ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1278         if (a[OVS_FLOW_ATTR_KEY]) {
1279                 ovs_match_init(&match, &key, NULL);
1280                 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1281                                         NULL, log);
1282                 if (unlikely(err))
1283                         return err;
1284         }
1285
1286         ovs_lock();
1287         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1288         if (unlikely(!dp)) {
1289                 err = -ENODEV;
1290                 goto unlock;
1291         }
1292
1293         if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1294                 err = ovs_flow_tbl_flush(&dp->table);
1295                 goto unlock;
1296         }
1297
1298         if (ufid_present)
1299                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1300         else
1301                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1302         if (unlikely(!flow)) {
1303                 err = -ENOENT;
1304                 goto unlock;
1305         }
1306
1307         ovs_flow_tbl_remove(&dp->table, flow);
1308         ovs_unlock();
1309
1310         reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1311                                         &flow->id, info, false, ufid_flags);
1312         if (likely(reply)) {
1313                 if (likely(!IS_ERR(reply))) {
1314                         rcu_read_lock();        /*To keep RCU checker happy. */
1315                         err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1316                                                      reply, info->snd_portid,
1317                                                      info->snd_seq, 0,
1318                                                      OVS_FLOW_CMD_DEL,
1319                                                      ufid_flags);
1320                         rcu_read_unlock();
1321                         BUG_ON(err < 0);
1322
1323                         ovs_notify(&dp_flow_genl_family, reply, info);
1324                 } else {
1325                         netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
1326                 }
1327         }
1328
1329         ovs_flow_free(flow, true);
1330         return 0;
1331 unlock:
1332         ovs_unlock();
1333         return err;
1334 }
1335
1336 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1337 {
1338         struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1339         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1340         struct table_instance *ti;
1341         struct datapath *dp;
1342         u32 ufid_flags;
1343         int err;
1344
1345         err = genlmsg_parse(cb->nlh, &dp_flow_genl_family, a,
1346                             OVS_FLOW_ATTR_MAX, flow_policy);
1347         if (err)
1348                 return err;
1349         ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1350
1351         rcu_read_lock();
1352         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1353         if (!dp) {
1354                 rcu_read_unlock();
1355                 return -ENODEV;
1356         }
1357
1358         ti = rcu_dereference(dp->table.ti);
1359         for (;;) {
1360                 struct sw_flow *flow;
1361                 u32 bucket, obj;
1362
1363                 bucket = cb->args[0];
1364                 obj = cb->args[1];
1365                 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1366                 if (!flow)
1367                         break;
1368
1369                 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1370                                            NETLINK_CB(cb->skb).portid,
1371                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1372                                            OVS_FLOW_CMD_NEW, ufid_flags) < 0)
1373                         break;
1374
1375                 cb->args[0] = bucket;
1376                 cb->args[1] = obj;
1377         }
1378         rcu_read_unlock();
1379         return skb->len;
1380 }
1381
1382 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1383         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1384         [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1385         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1386         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1387         [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1388         [OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1389         [OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1390 };
1391
1392 static const struct genl_ops dp_flow_genl_ops[] = {
1393         { .cmd = OVS_FLOW_CMD_NEW,
1394           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1395           .policy = flow_policy,
1396           .doit = ovs_flow_cmd_new
1397         },
1398         { .cmd = OVS_FLOW_CMD_DEL,
1399           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1400           .policy = flow_policy,
1401           .doit = ovs_flow_cmd_del
1402         },
1403         { .cmd = OVS_FLOW_CMD_GET,
1404           .flags = 0,               /* OK for unprivileged users. */
1405           .policy = flow_policy,
1406           .doit = ovs_flow_cmd_get,
1407           .dumpit = ovs_flow_cmd_dump
1408         },
1409         { .cmd = OVS_FLOW_CMD_SET,
1410           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1411           .policy = flow_policy,
1412           .doit = ovs_flow_cmd_set,
1413         },
1414 };
1415
1416 static struct genl_family dp_flow_genl_family = {
1417         .id = GENL_ID_GENERATE,
1418         .hdrsize = sizeof(struct ovs_header),
1419         .name = OVS_FLOW_FAMILY,
1420         .version = OVS_FLOW_VERSION,
1421         .maxattr = OVS_FLOW_ATTR_MAX,
1422         .netnsok = true,
1423         .parallel_ops = true,
1424         .ops = dp_flow_genl_ops,
1425         .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1426         .mcgrps = &ovs_dp_flow_multicast_group,
1427         .n_mcgrps = 1,
1428 };
1429
1430 static size_t ovs_dp_cmd_msg_size(void)
1431 {
1432         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1433
1434         msgsize += nla_total_size(IFNAMSIZ);
1435         msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1436         msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1437         msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1438
1439         return msgsize;
1440 }
1441
1442 /* Called with ovs_mutex. */
1443 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1444                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1445 {
1446         struct ovs_header *ovs_header;
1447         struct ovs_dp_stats dp_stats;
1448         struct ovs_dp_megaflow_stats dp_megaflow_stats;
1449         int err;
1450
1451         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1452                                    flags, cmd);
1453         if (!ovs_header)
1454                 goto error;
1455
1456         ovs_header->dp_ifindex = get_dpifindex(dp);
1457
1458         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1459         if (err)
1460                 goto nla_put_failure;
1461
1462         get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1463         if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1464                         &dp_stats))
1465                 goto nla_put_failure;
1466
1467         if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1468                         sizeof(struct ovs_dp_megaflow_stats),
1469                         &dp_megaflow_stats))
1470                 goto nla_put_failure;
1471
1472         if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1473                 goto nla_put_failure;
1474
1475         genlmsg_end(skb, ovs_header);
1476         return 0;
1477
1478 nla_put_failure:
1479         genlmsg_cancel(skb, ovs_header);
1480 error:
1481         return -EMSGSIZE;
1482 }
1483
1484 static struct sk_buff *ovs_dp_cmd_alloc_info(struct genl_info *info)
1485 {
1486         return genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info, GFP_KERNEL);
1487 }
1488
1489 /* Called with rcu_read_lock or ovs_mutex. */
1490 static struct datapath *lookup_datapath(struct net *net,
1491                                         const struct ovs_header *ovs_header,
1492                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1493 {
1494         struct datapath *dp;
1495
1496         if (!a[OVS_DP_ATTR_NAME])
1497                 dp = get_dp(net, ovs_header->dp_ifindex);
1498         else {
1499                 struct vport *vport;
1500
1501                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1502                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1503         }
1504         return dp ? dp : ERR_PTR(-ENODEV);
1505 }
1506
1507 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1508 {
1509         struct datapath *dp;
1510
1511         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1512         if (IS_ERR(dp))
1513                 return;
1514
1515         WARN(dp->user_features, "Dropping previously announced user features\n");
1516         dp->user_features = 0;
1517 }
1518
1519 static void ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1520 {
1521         if (a[OVS_DP_ATTR_USER_FEATURES])
1522                 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1523 }
1524
1525 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1526 {
1527         struct nlattr **a = info->attrs;
1528         struct vport_parms parms;
1529         struct sk_buff *reply;
1530         struct datapath *dp;
1531         struct vport *vport;
1532         struct ovs_net *ovs_net;
1533         int err, i;
1534
1535         err = -EINVAL;
1536         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1537                 goto err;
1538
1539         reply = ovs_dp_cmd_alloc_info(info);
1540         if (!reply)
1541                 return -ENOMEM;
1542
1543         err = -ENOMEM;
1544         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1545         if (dp == NULL)
1546                 goto err_free_reply;
1547
1548         ovs_dp_set_net(dp, sock_net(skb->sk));
1549
1550         /* Allocate table. */
1551         err = ovs_flow_tbl_init(&dp->table);
1552         if (err)
1553                 goto err_free_dp;
1554
1555         dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1556         if (!dp->stats_percpu) {
1557                 err = -ENOMEM;
1558                 goto err_destroy_table;
1559         }
1560
1561         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1562                             GFP_KERNEL);
1563         if (!dp->ports) {
1564                 err = -ENOMEM;
1565                 goto err_destroy_percpu;
1566         }
1567
1568         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1569                 INIT_HLIST_HEAD(&dp->ports[i]);
1570
1571         /* Set up our datapath device. */
1572         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1573         parms.type = OVS_VPORT_TYPE_INTERNAL;
1574         parms.options = NULL;
1575         parms.dp = dp;
1576         parms.port_no = OVSP_LOCAL;
1577         parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1578
1579         ovs_dp_change(dp, a);
1580
1581         /* So far only local changes have been made, now need the lock. */
1582         ovs_lock();
1583
1584         vport = new_vport(&parms);
1585         if (IS_ERR(vport)) {
1586                 err = PTR_ERR(vport);
1587                 if (err == -EBUSY)
1588                         err = -EEXIST;
1589
1590                 if (err == -EEXIST) {
1591                         /* An outdated user space instance that does not understand
1592                          * the concept of user_features has attempted to create a new
1593                          * datapath and is likely to reuse it. Drop all user features.
1594                          */
1595                         if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1596                                 ovs_dp_reset_user_features(skb, info);
1597                 }
1598
1599                 goto err_destroy_ports_array;
1600         }
1601
1602         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1603                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1604         BUG_ON(err < 0);
1605
1606         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1607         list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1608
1609         ovs_unlock();
1610
1611         ovs_notify(&dp_datapath_genl_family, reply, info);
1612         return 0;
1613
1614 err_destroy_ports_array:
1615         ovs_unlock();
1616         kfree(dp->ports);
1617 err_destroy_percpu:
1618         free_percpu(dp->stats_percpu);
1619 err_destroy_table:
1620         ovs_flow_tbl_destroy(&dp->table);
1621 err_free_dp:
1622         kfree(dp);
1623 err_free_reply:
1624         kfree_skb(reply);
1625 err:
1626         return err;
1627 }
1628
1629 /* Called with ovs_mutex. */
1630 static void __dp_destroy(struct datapath *dp)
1631 {
1632         int i;
1633
1634         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1635                 struct vport *vport;
1636                 struct hlist_node *n;
1637
1638                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1639                         if (vport->port_no != OVSP_LOCAL)
1640                                 ovs_dp_detach_port(vport);
1641         }
1642
1643         list_del_rcu(&dp->list_node);
1644
1645         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1646          * all ports in datapath are destroyed first before freeing datapath.
1647          */
1648         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1649
1650         /* RCU destroy the flow table */
1651         call_rcu(&dp->rcu, destroy_dp_rcu);
1652 }
1653
1654 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1655 {
1656         struct sk_buff *reply;
1657         struct datapath *dp;
1658         int err;
1659
1660         reply = ovs_dp_cmd_alloc_info(info);
1661         if (!reply)
1662                 return -ENOMEM;
1663
1664         ovs_lock();
1665         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1666         err = PTR_ERR(dp);
1667         if (IS_ERR(dp))
1668                 goto err_unlock_free;
1669
1670         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1671                                    info->snd_seq, 0, OVS_DP_CMD_DEL);
1672         BUG_ON(err < 0);
1673
1674         __dp_destroy(dp);
1675         ovs_unlock();
1676
1677         ovs_notify(&dp_datapath_genl_family, reply, info);
1678
1679         return 0;
1680
1681 err_unlock_free:
1682         ovs_unlock();
1683         kfree_skb(reply);
1684         return err;
1685 }
1686
1687 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1688 {
1689         struct sk_buff *reply;
1690         struct datapath *dp;
1691         int err;
1692
1693         reply = ovs_dp_cmd_alloc_info(info);
1694         if (!reply)
1695                 return -ENOMEM;
1696
1697         ovs_lock();
1698         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1699         err = PTR_ERR(dp);
1700         if (IS_ERR(dp))
1701                 goto err_unlock_free;
1702
1703         ovs_dp_change(dp, info->attrs);
1704
1705         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1706                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1707         BUG_ON(err < 0);
1708
1709         ovs_unlock();
1710         ovs_notify(&dp_datapath_genl_family, reply, info);
1711
1712         return 0;
1713
1714 err_unlock_free:
1715         ovs_unlock();
1716         kfree_skb(reply);
1717         return err;
1718 }
1719
1720 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1721 {
1722         struct sk_buff *reply;
1723         struct datapath *dp;
1724         int err;
1725
1726         reply = ovs_dp_cmd_alloc_info(info);
1727         if (!reply)
1728                 return -ENOMEM;
1729
1730         ovs_lock();
1731         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1732         if (IS_ERR(dp)) {
1733                 err = PTR_ERR(dp);
1734                 goto err_unlock_free;
1735         }
1736         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1737                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1738         BUG_ON(err < 0);
1739         ovs_unlock();
1740
1741         return genlmsg_reply(reply, info);
1742
1743 err_unlock_free:
1744         ovs_unlock();
1745         kfree_skb(reply);
1746         return err;
1747 }
1748
1749 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1750 {
1751         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1752         struct datapath *dp;
1753         int skip = cb->args[0];
1754         int i = 0;
1755
1756         ovs_lock();
1757         list_for_each_entry(dp, &ovs_net->dps, list_node) {
1758                 if (i >= skip &&
1759                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1760                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1761                                          OVS_DP_CMD_NEW) < 0)
1762                         break;
1763                 i++;
1764         }
1765         ovs_unlock();
1766
1767         cb->args[0] = i;
1768
1769         return skb->len;
1770 }
1771
1772 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1773         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1774         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1775         [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1776 };
1777
1778 static const struct genl_ops dp_datapath_genl_ops[] = {
1779         { .cmd = OVS_DP_CMD_NEW,
1780           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1781           .policy = datapath_policy,
1782           .doit = ovs_dp_cmd_new
1783         },
1784         { .cmd = OVS_DP_CMD_DEL,
1785           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1786           .policy = datapath_policy,
1787           .doit = ovs_dp_cmd_del
1788         },
1789         { .cmd = OVS_DP_CMD_GET,
1790           .flags = 0,               /* OK for unprivileged users. */
1791           .policy = datapath_policy,
1792           .doit = ovs_dp_cmd_get,
1793           .dumpit = ovs_dp_cmd_dump
1794         },
1795         { .cmd = OVS_DP_CMD_SET,
1796           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1797           .policy = datapath_policy,
1798           .doit = ovs_dp_cmd_set,
1799         },
1800 };
1801
1802 static struct genl_family dp_datapath_genl_family = {
1803         .id = GENL_ID_GENERATE,
1804         .hdrsize = sizeof(struct ovs_header),
1805         .name = OVS_DATAPATH_FAMILY,
1806         .version = OVS_DATAPATH_VERSION,
1807         .maxattr = OVS_DP_ATTR_MAX,
1808         .netnsok = true,
1809         .parallel_ops = true,
1810         .ops = dp_datapath_genl_ops,
1811         .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1812         .mcgrps = &ovs_dp_datapath_multicast_group,
1813         .n_mcgrps = 1,
1814 };
1815
1816 /* Called with ovs_mutex or RCU read lock. */
1817 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1818                                    u32 portid, u32 seq, u32 flags, u8 cmd)
1819 {
1820         struct ovs_header *ovs_header;
1821         struct ovs_vport_stats vport_stats;
1822         int err;
1823
1824         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1825                                  flags, cmd);
1826         if (!ovs_header)
1827                 return -EMSGSIZE;
1828
1829         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1830
1831         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1832             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1833             nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1834                            ovs_vport_name(vport)))
1835                 goto nla_put_failure;
1836
1837         ovs_vport_get_stats(vport, &vport_stats);
1838         if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1839                     &vport_stats))
1840                 goto nla_put_failure;
1841
1842         if (ovs_vport_get_upcall_portids(vport, skb))
1843                 goto nla_put_failure;
1844
1845         err = ovs_vport_get_options(vport, skb);
1846         if (err == -EMSGSIZE)
1847                 goto error;
1848
1849         genlmsg_end(skb, ovs_header);
1850         return 0;
1851
1852 nla_put_failure:
1853         err = -EMSGSIZE;
1854 error:
1855         genlmsg_cancel(skb, ovs_header);
1856         return err;
1857 }
1858
1859 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1860 {
1861         return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1862 }
1863
1864 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1865 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1866                                          u32 seq, u8 cmd)
1867 {
1868         struct sk_buff *skb;
1869         int retval;
1870
1871         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1872         if (!skb)
1873                 return ERR_PTR(-ENOMEM);
1874
1875         retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1876         BUG_ON(retval < 0);
1877
1878         return skb;
1879 }
1880
1881 /* Called with ovs_mutex or RCU read lock. */
1882 static struct vport *lookup_vport(struct net *net,
1883                                   const struct ovs_header *ovs_header,
1884                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1885 {
1886         struct datapath *dp;
1887         struct vport *vport;
1888
1889         if (a[OVS_VPORT_ATTR_NAME]) {
1890                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1891                 if (!vport)
1892                         return ERR_PTR(-ENODEV);
1893                 if (ovs_header->dp_ifindex &&
1894                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1895                         return ERR_PTR(-ENODEV);
1896                 return vport;
1897         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1898                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1899
1900                 if (port_no >= DP_MAX_PORTS)
1901                         return ERR_PTR(-EFBIG);
1902
1903                 dp = get_dp(net, ovs_header->dp_ifindex);
1904                 if (!dp)
1905                         return ERR_PTR(-ENODEV);
1906
1907                 vport = ovs_vport_ovsl_rcu(dp, port_no);
1908                 if (!vport)
1909                         return ERR_PTR(-ENODEV);
1910                 return vport;
1911         } else
1912                 return ERR_PTR(-EINVAL);
1913 }
1914
1915 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1916 {
1917         struct nlattr **a = info->attrs;
1918         struct ovs_header *ovs_header = info->userhdr;
1919         struct vport_parms parms;
1920         struct sk_buff *reply;
1921         struct vport *vport;
1922         struct datapath *dp;
1923         u32 port_no;
1924         int err;
1925
1926         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1927             !a[OVS_VPORT_ATTR_UPCALL_PID])
1928                 return -EINVAL;
1929
1930         port_no = a[OVS_VPORT_ATTR_PORT_NO]
1931                 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
1932         if (port_no >= DP_MAX_PORTS)
1933                 return -EFBIG;
1934
1935         reply = ovs_vport_cmd_alloc_info();
1936         if (!reply)
1937                 return -ENOMEM;
1938
1939         ovs_lock();
1940 restart:
1941         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1942         err = -ENODEV;
1943         if (!dp)
1944                 goto exit_unlock_free;
1945
1946         if (port_no) {
1947                 vport = ovs_vport_ovsl(dp, port_no);
1948                 err = -EBUSY;
1949                 if (vport)
1950                         goto exit_unlock_free;
1951         } else {
1952                 for (port_no = 1; ; port_no++) {
1953                         if (port_no >= DP_MAX_PORTS) {
1954                                 err = -EFBIG;
1955                                 goto exit_unlock_free;
1956                         }
1957                         vport = ovs_vport_ovsl(dp, port_no);
1958                         if (!vport)
1959                                 break;
1960                 }
1961         }
1962
1963         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1964         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1965         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1966         parms.dp = dp;
1967         parms.port_no = port_no;
1968         parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
1969
1970         vport = new_vport(&parms);
1971         err = PTR_ERR(vport);
1972         if (IS_ERR(vport)) {
1973                 if (err == -EAGAIN)
1974                         goto restart;
1975                 goto exit_unlock_free;
1976         }
1977
1978         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1979                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1980         BUG_ON(err < 0);
1981         ovs_unlock();
1982
1983         ovs_notify(&dp_vport_genl_family, reply, info);
1984         return 0;
1985
1986 exit_unlock_free:
1987         ovs_unlock();
1988         kfree_skb(reply);
1989         return err;
1990 }
1991
1992 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1993 {
1994         struct nlattr **a = info->attrs;
1995         struct sk_buff *reply;
1996         struct vport *vport;
1997         int err;
1998
1999         reply = ovs_vport_cmd_alloc_info();
2000         if (!reply)
2001                 return -ENOMEM;
2002
2003         ovs_lock();
2004         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2005         err = PTR_ERR(vport);
2006         if (IS_ERR(vport))
2007                 goto exit_unlock_free;
2008
2009         if (a[OVS_VPORT_ATTR_TYPE] &&
2010             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2011                 err = -EINVAL;
2012                 goto exit_unlock_free;
2013         }
2014
2015         if (a[OVS_VPORT_ATTR_OPTIONS]) {
2016                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2017                 if (err)
2018                         goto exit_unlock_free;
2019         }
2020
2021
2022         if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2023                 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2024
2025                 err = ovs_vport_set_upcall_portids(vport, ids);
2026                 if (err)
2027                         goto exit_unlock_free;
2028         }
2029
2030         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2031                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2032         BUG_ON(err < 0);
2033
2034         ovs_unlock();
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_del(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 (vport->port_no == OVSP_LOCAL) {
2062                 err = -EINVAL;
2063                 goto exit_unlock_free;
2064         }
2065
2066         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2067                                       info->snd_seq, 0, OVS_VPORT_CMD_DEL);
2068         BUG_ON(err < 0);
2069         ovs_dp_detach_port(vport);
2070         ovs_unlock();
2071
2072         ovs_notify(&dp_vport_genl_family, reply, info);
2073         return 0;
2074
2075 exit_unlock_free:
2076         ovs_unlock();
2077         kfree_skb(reply);
2078         return err;
2079 }
2080
2081 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2082 {
2083         struct nlattr **a = info->attrs;
2084         struct ovs_header *ovs_header = info->userhdr;
2085         struct sk_buff *reply;
2086         struct vport *vport;
2087         int err;
2088
2089         reply = ovs_vport_cmd_alloc_info();
2090         if (!reply)
2091                 return -ENOMEM;
2092
2093         rcu_read_lock();
2094         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2095         err = PTR_ERR(vport);
2096         if (IS_ERR(vport))
2097                 goto exit_unlock_free;
2098         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2099                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2100         BUG_ON(err < 0);
2101         rcu_read_unlock();
2102
2103         return genlmsg_reply(reply, info);
2104
2105 exit_unlock_free:
2106         rcu_read_unlock();
2107         kfree_skb(reply);
2108         return err;
2109 }
2110
2111 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2112 {
2113         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2114         struct datapath *dp;
2115         int bucket = cb->args[0], skip = cb->args[1];
2116         int i, j = 0;
2117
2118         rcu_read_lock();
2119         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2120         if (!dp) {
2121                 rcu_read_unlock();
2122                 return -ENODEV;
2123         }
2124         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2125                 struct vport *vport;
2126
2127                 j = 0;
2128                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2129                         if (j >= skip &&
2130                             ovs_vport_cmd_fill_info(vport, skb,
2131                                                     NETLINK_CB(cb->skb).portid,
2132                                                     cb->nlh->nlmsg_seq,
2133                                                     NLM_F_MULTI,
2134                                                     OVS_VPORT_CMD_NEW) < 0)
2135                                 goto out;
2136
2137                         j++;
2138                 }
2139                 skip = 0;
2140         }
2141 out:
2142         rcu_read_unlock();
2143
2144         cb->args[0] = i;
2145         cb->args[1] = j;
2146
2147         return skb->len;
2148 }
2149
2150 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2151         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2152         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2153         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2154         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2155         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
2156         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2157 };
2158
2159 static const struct genl_ops dp_vport_genl_ops[] = {
2160         { .cmd = OVS_VPORT_CMD_NEW,
2161           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2162           .policy = vport_policy,
2163           .doit = ovs_vport_cmd_new
2164         },
2165         { .cmd = OVS_VPORT_CMD_DEL,
2166           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2167           .policy = vport_policy,
2168           .doit = ovs_vport_cmd_del
2169         },
2170         { .cmd = OVS_VPORT_CMD_GET,
2171           .flags = 0,               /* OK for unprivileged users. */
2172           .policy = vport_policy,
2173           .doit = ovs_vport_cmd_get,
2174           .dumpit = ovs_vport_cmd_dump
2175         },
2176         { .cmd = OVS_VPORT_CMD_SET,
2177           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2178           .policy = vport_policy,
2179           .doit = ovs_vport_cmd_set,
2180         },
2181 };
2182
2183 struct genl_family dp_vport_genl_family = {
2184         .id = GENL_ID_GENERATE,
2185         .hdrsize = sizeof(struct ovs_header),
2186         .name = OVS_VPORT_FAMILY,
2187         .version = OVS_VPORT_VERSION,
2188         .maxattr = OVS_VPORT_ATTR_MAX,
2189         .netnsok = true,
2190         .parallel_ops = true,
2191         .ops = dp_vport_genl_ops,
2192         .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2193         .mcgrps = &ovs_dp_vport_multicast_group,
2194         .n_mcgrps = 1,
2195 };
2196
2197 static struct genl_family * const dp_genl_families[] = {
2198         &dp_datapath_genl_family,
2199         &dp_vport_genl_family,
2200         &dp_flow_genl_family,
2201         &dp_packet_genl_family,
2202 };
2203
2204 static void dp_unregister_genl(int n_families)
2205 {
2206         int i;
2207
2208         for (i = 0; i < n_families; i++)
2209                 genl_unregister_family(dp_genl_families[i]);
2210 }
2211
2212 static int dp_register_genl(void)
2213 {
2214         int err;
2215         int i;
2216
2217         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2218
2219                 err = genl_register_family(dp_genl_families[i]);
2220                 if (err)
2221                         goto error;
2222         }
2223
2224         return 0;
2225
2226 error:
2227         dp_unregister_genl(i);
2228         return err;
2229 }
2230
2231 static int __net_init ovs_init_net(struct net *net)
2232 {
2233         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2234
2235         INIT_LIST_HEAD(&ovs_net->dps);
2236         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2237         ovs_ct_init(net);
2238         return 0;
2239 }
2240
2241 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2242                                             struct list_head *head)
2243 {
2244         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2245         struct datapath *dp;
2246
2247         list_for_each_entry(dp, &ovs_net->dps, list_node) {
2248                 int i;
2249
2250                 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2251                         struct vport *vport;
2252
2253                         hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2254                                 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2255                                         continue;
2256
2257                                 if (dev_net(vport->dev) == dnet)
2258                                         list_add(&vport->detach_list, head);
2259                         }
2260                 }
2261         }
2262 }
2263
2264 static void __net_exit ovs_exit_net(struct net *dnet)
2265 {
2266         struct datapath *dp, *dp_next;
2267         struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2268         struct vport *vport, *vport_next;
2269         struct net *net;
2270         LIST_HEAD(head);
2271
2272         ovs_ct_exit(dnet);
2273         ovs_lock();
2274         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2275                 __dp_destroy(dp);
2276
2277         rtnl_lock();
2278         for_each_net(net)
2279                 list_vports_from_net(net, dnet, &head);
2280         rtnl_unlock();
2281
2282         /* Detach all vports from given namespace. */
2283         list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2284                 list_del(&vport->detach_list);
2285                 ovs_dp_detach_port(vport);
2286         }
2287
2288         ovs_unlock();
2289
2290         cancel_work_sync(&ovs_net->dp_notify_work);
2291 }
2292
2293 static struct pernet_operations ovs_net_ops = {
2294         .init = ovs_init_net,
2295         .exit = ovs_exit_net,
2296         .id   = &ovs_net_id,
2297         .size = sizeof(struct ovs_net),
2298 };
2299
2300 static int __init dp_init(void)
2301 {
2302         int err;
2303
2304         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2305
2306         pr_info("Open vSwitch switching datapath\n");
2307
2308         err = action_fifos_init();
2309         if (err)
2310                 goto error;
2311
2312         err = ovs_internal_dev_rtnl_link_register();
2313         if (err)
2314                 goto error_action_fifos_exit;
2315
2316         err = ovs_flow_init();
2317         if (err)
2318                 goto error_unreg_rtnl_link;
2319
2320         err = ovs_vport_init();
2321         if (err)
2322                 goto error_flow_exit;
2323
2324         err = register_pernet_device(&ovs_net_ops);
2325         if (err)
2326                 goto error_vport_exit;
2327
2328         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2329         if (err)
2330                 goto error_netns_exit;
2331
2332         err = ovs_netdev_init();
2333         if (err)
2334                 goto error_unreg_notifier;
2335
2336         err = dp_register_genl();
2337         if (err < 0)
2338                 goto error_unreg_netdev;
2339
2340         return 0;
2341
2342 error_unreg_netdev:
2343         ovs_netdev_exit();
2344 error_unreg_notifier:
2345         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2346 error_netns_exit:
2347         unregister_pernet_device(&ovs_net_ops);
2348 error_vport_exit:
2349         ovs_vport_exit();
2350 error_flow_exit:
2351         ovs_flow_exit();
2352 error_unreg_rtnl_link:
2353         ovs_internal_dev_rtnl_link_unregister();
2354 error_action_fifos_exit:
2355         action_fifos_exit();
2356 error:
2357         return err;
2358 }
2359
2360 static void dp_cleanup(void)
2361 {
2362         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2363         ovs_netdev_exit();
2364         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2365         unregister_pernet_device(&ovs_net_ops);
2366         rcu_barrier();
2367         ovs_vport_exit();
2368         ovs_flow_exit();
2369         ovs_internal_dev_rtnl_link_unregister();
2370         action_fifos_exit();
2371 }
2372
2373 module_init(dp_init);
2374 module_exit(dp_cleanup);
2375
2376 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2377 MODULE_LICENSE("GPL");