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         size_t len;
 426         unsigned int hlen;
 427         int err, dp_ifindex;
 428
 429         dp_ifindex = get_dpifindex(dp);
 430         if (!dp_ifindex)
 431                 return -ENODEV;
 432
 433         if (skb_vlan_tag_present(skb)) {
 434                 nskb = skb_clone(skb, GFP_ATOMIC);
 435                 if (!nskb)
 436                         return -ENOMEM;
 437
 438                 nskb = __vlan_hwaccel_push_inside(nskb);
 439                 if (!nskb)
 440                         return -ENOMEM;
 441
 442                 skb = nskb;
 443         }
 444
 445         if (nla_attr_size(skb->len) > USHRT_MAX) {
 446                 err = -EFBIG;
 447                 goto out;
 448         }
 449
 450         /* Complete checksum if needed */
 451         if (skb->ip_summed == CHECKSUM_PARTIAL &&
 452             (err = skb_checksum_help(skb)))
 453                 goto out;
 454
 455         /* Older versions of OVS user space enforce alignment of the last
 456          * Netlink attribute to NLA_ALIGNTO which would require extensive
 457          * padding logic. Only perform zerocopy if padding is not required.
 458          */
 459         if (dp->user_features & OVS_DP_F_UNALIGNED)
 460                 hlen = skb_zerocopy_headlen(skb);
 461         else
 462                 hlen = skb->len;
 463
 464         len = upcall_msg_size(upcall_info, hlen);
 465         user_skb = genlmsg_new(len, GFP_ATOMIC);
 466         if (!user_skb) {
 467                 err = -ENOMEM;
 468                 goto out;
 469         }
 470
 471         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 472                              0, upcall_info->cmd);
 473         upcall->dp_ifindex = dp_ifindex;
 474
 475         err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
 476         BUG_ON(err);
 477
 478         if (upcall_info->userdata)
 479                 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
 480                           nla_len(upcall_info->userdata),
 481                           nla_data(upcall_info->userdata));
 482
 483         if (upcall_info->egress_tun_info) {
 484                 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_EGRESS_TUN_KEY);
 485                 err = ovs_nla_put_tunnel_info(user_skb,
 486                                               upcall_info->egress_tun_info);
 487                 BUG_ON(err);
 488                 nla_nest_end(user_skb, nla);
 489         }
 490
 491         if (upcall_info->actions_len) {
 492                 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_ACTIONS);
 493                 err = ovs_nla_put_actions(upcall_info->actions,
 494                                           upcall_info->actions_len,
 495                                           user_skb);
 496                 if (!err)
 497                         nla_nest_end(user_skb, nla);
 498                 else
 499                         nla_nest_cancel(user_skb, nla);
 500         }
 501
 502         /* Add OVS_PACKET_ATTR_MRU */
 503         if (upcall_info->mru) {
 504                 if (nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU,
 505                                 upcall_info->mru)) {
 506                         err = -ENOBUFS;
 507                         goto out;
 508                 }
 509                 pad_packet(dp, user_skb);
 510         }
 511
 512         /* Only reserve room for attribute header, packet data is added
 513          * in skb_zerocopy() */
 514         if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
 515                 err = -ENOBUFS;
 516                 goto out;
 517         }
 518         nla->nla_len = nla_attr_size(skb->len);
 519
 520         err = skb_zerocopy(user_skb, skb, skb->len, hlen);
 521         if (err)
 522                 goto out;
 523
 524         /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
 525         pad_packet(dp, user_skb);
 526
 527         ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
 528
 529         err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
 530         user_skb = NULL;
 531 out:
 532         if (err)
 533                 skb_tx_error(skb);
 534         kfree_skb(user_skb);
 535         kfree_skb(nskb);
 536         return err;
 537 }
 538
 539 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 540 {
 541         struct ovs_header *ovs_header = info->userhdr;
 542         struct net *net = sock_net(skb->sk);
 543         struct nlattr **a = info->attrs;
 544         struct sw_flow_actions *acts;
 545         struct sk_buff *packet;
 546         struct sw_flow *flow;
 547         struct sw_flow_actions *sf_acts;
 548         struct datapath *dp;
 549         struct ethhdr *eth;
 550         struct vport *input_vport;
 551         u16 mru = 0;
 552         int len;
 553         int err;
 554         bool log = !a[OVS_PACKET_ATTR_PROBE];
 555
 556         err = -EINVAL;
 557         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 558             !a[OVS_PACKET_ATTR_ACTIONS])
 559                 goto err;
 560
 561         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 562         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 563         err = -ENOMEM;
 564         if (!packet)
 565                 goto err;
 566         skb_reserve(packet, NET_IP_ALIGN);
 567
 568         nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
 569
 570         skb_reset_mac_header(packet);
 571         eth = eth_hdr(packet);
 572
 573         /* Normally, setting the skb 'protocol' field would be handled by a
 574          * call to eth_type_trans(), but it assumes there's a sending
 575          * device, which we may not have. */
 576         if (eth_proto_is_802_3(eth->h_proto))
 577                 packet->protocol = eth->h_proto;
 578         else
 579                 packet->protocol = htons(ETH_P_802_2);
 580
 581         /* Set packet's mru */
 582         if (a[OVS_PACKET_ATTR_MRU]) {
 583                 mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
 584                 packet->ignore_df = 1;
 585         }
 586         OVS_CB(packet)->mru = mru;
 587
 588         /* Build an sw_flow for sending this packet. */
 589         flow = ovs_flow_alloc();
 590         err = PTR_ERR(flow);
 591         if (IS_ERR(flow))
 592                 goto err_kfree_skb;
 593
 594         err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
 595                                              packet, &flow->key, log);
 596         if (err)
 597                 goto err_flow_free;
 598
 599         err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
 600                                    &flow->key, &acts, log);
 601         if (err)
 602                 goto err_flow_free;
 603
 604         rcu_assign_pointer(flow->sf_acts, acts);
 605         packet->priority = flow->key.phy.priority;
 606         packet->mark = flow->key.phy.skb_mark;
 607
 608         rcu_read_lock();
 609         dp = get_dp_rcu(net, ovs_header->dp_ifindex);
 610         err = -ENODEV;
 611         if (!dp)
 612                 goto err_unlock;
 613
 614         input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
 615         if (!input_vport)
 616                 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
 617
 618         if (!input_vport)
 619                 goto err_unlock;
 620
 621         packet->dev = input_vport->dev;
 622         OVS_CB(packet)->input_vport = input_vport;
 623         sf_acts = rcu_dereference(flow->sf_acts);
 624
 625         local_bh_disable();
 626         err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
 627         local_bh_enable();
 628         rcu_read_unlock();
 629
 630         ovs_flow_free(flow, false);
 631         return err;
 632
 633 err_unlock:
 634         rcu_read_unlock();
 635 err_flow_free:
 636         ovs_flow_free(flow, false);
 637 err_kfree_skb:
 638         kfree_skb(packet);
 639 err:
 640         return err;
 641 }
 642
 643 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 644         [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
 645         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 646         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 647         [OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
 648         [OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
 649 };
 650
 651 static const struct genl_ops dp_packet_genl_ops[] = {
 652         { .cmd = OVS_PACKET_CMD_EXECUTE,
 653           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 654           .policy = packet_policy,
 655           .doit = ovs_packet_cmd_execute
 656         }
 657 };
 658
 659 static struct genl_family dp_packet_genl_family = {
 660         .id = GENL_ID_GENERATE,
 661         .hdrsize = sizeof(struct ovs_header),
 662         .name = OVS_PACKET_FAMILY,
 663         .version = OVS_PACKET_VERSION,
 664         .maxattr = OVS_PACKET_ATTR_MAX,
 665         .netnsok = true,
 666         .parallel_ops = true,
 667         .ops = dp_packet_genl_ops,
 668         .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
 669 };
 670
 671 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
 672                          struct ovs_dp_megaflow_stats *mega_stats)
 673 {
 674         int i;
 675
 676         memset(mega_stats, 0, sizeof(*mega_stats));
 677
 678         stats->n_flows = ovs_flow_tbl_count(&dp->table);
 679         mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
 680
 681         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 682
 683         for_each_possible_cpu(i) {
 684                 const struct dp_stats_percpu *percpu_stats;
 685                 struct dp_stats_percpu local_stats;
 686                 unsigned int start;
 687
 688                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 689
 690                 do {
 691                         start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
 692                         local_stats = *percpu_stats;
 693                 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
 694
 695                 stats->n_hit += local_stats.n_hit;
 696                 stats->n_missed += local_stats.n_missed;
 697                 stats->n_lost += local_stats.n_lost;
 698                 mega_stats->n_mask_hit += local_stats.n_mask_hit;
 699         }
 700 }
 701
 702 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
 703 {
 704         return ovs_identifier_is_ufid(sfid) &&
 705                !(ufid_flags & OVS_UFID_F_OMIT_KEY);
 706 }
 707
 708 static bool should_fill_mask(uint32_t ufid_flags)
 709 {
 710         return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
 711 }
 712
 713 static bool should_fill_actions(uint32_t ufid_flags)
 714 {
 715         return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
 716 }
 717
 718 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
 719                                     const struct sw_flow_id *sfid,
 720                                     uint32_t ufid_flags)
 721 {
 722         size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
 723
 724         /* OVS_FLOW_ATTR_UFID */
 725         if (sfid && ovs_identifier_is_ufid(sfid))
 726                 len += nla_total_size(sfid->ufid_len);
 727
 728         /* OVS_FLOW_ATTR_KEY */
 729         if (!sfid || should_fill_key(sfid, ufid_flags))
 730                 len += nla_total_size(ovs_key_attr_size());
 731
 732         /* OVS_FLOW_ATTR_MASK */
 733         if (should_fill_mask(ufid_flags))
 734                 len += nla_total_size(ovs_key_attr_size());
 735
 736         /* OVS_FLOW_ATTR_ACTIONS */
 737         if (should_fill_actions(ufid_flags))
 738                 len += nla_total_size(acts->orig_len);
 739
 740         return len
 741                 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
 742                 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
 743                 + nla_total_size(8); /* OVS_FLOW_ATTR_USED */
 744 }
 745
 746 /* Called with ovs_mutex or RCU read lock. */
 747 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
 748                                    struct sk_buff *skb)
 749 {
 750         struct ovs_flow_stats stats;
 751         __be16 tcp_flags;
 752         unsigned long used;
 753
 754         ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
 755
 756         if (used &&
 757             nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
 758                 return -EMSGSIZE;
 759
 760         if (stats.n_packets &&
 761             nla_put(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats))
 762                 return -EMSGSIZE;
 763
 764         if ((u8)ntohs(tcp_flags) &&
 765              nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
 766                 return -EMSGSIZE;
 767
 768         return 0;
 769 }
 770
 771 /* Called with ovs_mutex or RCU read lock. */
 772 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
 773                                      struct sk_buff *skb, int skb_orig_len)
 774 {
 775         struct nlattr *start;
 776         int err;
 777
 778         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 779          * this is the first flow to be dumped into 'skb'.  This is unusual for
 780          * Netlink but individual action lists can be longer than
 781          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 782          * The userspace caller can always fetch the actions separately if it
 783          * really wants them.  (Most userspace callers in fact don't care.)
 784          *
 785          * This can only fail for dump operations because the skb is always
 786          * properly sized for single flows.
 787          */
 788         start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
 789         if (start) {
 790                 const struct sw_flow_actions *sf_acts;
 791
 792                 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
 793                 err = ovs_nla_put_actions(sf_acts->actions,
 794                                           sf_acts->actions_len, skb);
 795
 796                 if (!err)
 797                         nla_nest_end(skb, start);
 798                 else {
 799                         if (skb_orig_len)
 800                                 return err;
 801
 802                         nla_nest_cancel(skb, start);
 803                 }
 804         } else if (skb_orig_len) {
 805                 return -EMSGSIZE;
 806         }
 807
 808         return 0;
 809 }
 810
 811 /* Called with ovs_mutex or RCU read lock. */
 812 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
 813                                   struct sk_buff *skb, u32 portid,
 814                                   u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
 815 {
 816         const int skb_orig_len = skb->len;
 817         struct ovs_header *ovs_header;
 818         int err;
 819
 820         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
 821                                  flags, cmd);
 822         if (!ovs_header)
 823                 return -EMSGSIZE;
 824
 825         ovs_header->dp_ifindex = dp_ifindex;
 826
 827         err = ovs_nla_put_identifier(flow, skb);
 828         if (err)
 829                 goto error;
 830
 831         if (should_fill_key(&flow->id, ufid_flags)) {
 832                 err = ovs_nla_put_masked_key(flow, skb);
 833                 if (err)
 834                         goto error;
 835         }
 836
 837         if (should_fill_mask(ufid_flags)) {
 838                 err = ovs_nla_put_mask(flow, skb);
 839                 if (err)
 840                         goto error;
 841         }
 842
 843         err = ovs_flow_cmd_fill_stats(flow, skb);
 844         if (err)
 845                 goto error;
 846
 847         if (should_fill_actions(ufid_flags)) {
 848                 err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
 849                 if (err)
 850                         goto error;
 851         }
 852
 853         genlmsg_end(skb, ovs_header);
 854         return 0;
 855
 856 error:
 857         genlmsg_cancel(skb, ovs_header);
 858         return err;
 859 }
 860
 861 /* May not be called with RCU read lock. */
 862 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
 863                                                const struct sw_flow_id *sfid,
 864                                                struct genl_info *info,
 865                                                bool always,
 866                                                uint32_t ufid_flags)
 867 {
 868         struct sk_buff *skb;
 869         size_t len;
 870
 871         if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
 872                 return NULL;
 873
 874         len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
 875         skb = genlmsg_new(len, GFP_KERNEL);
 876         if (!skb)
 877                 return ERR_PTR(-ENOMEM);
 878
 879         return skb;
 880 }
 881
 882 /* Called with ovs_mutex. */
 883 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
 884                                                int dp_ifindex,
 885                                                struct genl_info *info, u8 cmd,
 886                                                bool always, u32 ufid_flags)
 887 {
 888         struct sk_buff *skb;
 889         int retval;
 890
 891         skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
 892                                       &flow->id, info, always, ufid_flags);
 893         if (IS_ERR_OR_NULL(skb))
 894                 return skb;
 895
 896         retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
 897                                         info->snd_portid, info->snd_seq, 0,
 898                                         cmd, ufid_flags);
 899         BUG_ON(retval < 0);
 900         return skb;
 901 }
 902
 903 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
 904 {
 905         struct net *net = sock_net(skb->sk);
 906         struct nlattr **a = info->attrs;
 907         struct ovs_header *ovs_header = info->userhdr;
 908         struct sw_flow *flow = NULL, *new_flow;
 909         struct sw_flow_mask mask;
 910         struct sk_buff *reply;
 911         struct datapath *dp;
 912         struct sw_flow_key key;
 913         struct sw_flow_actions *acts;
 914         struct sw_flow_match match;
 915         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
 916         int error;
 917         bool log = !a[OVS_FLOW_ATTR_PROBE];
 918
 919         /* Must have key and actions. */
 920         error = -EINVAL;
 921         if (!a[OVS_FLOW_ATTR_KEY]) {
 922                 OVS_NLERR(log, "Flow key attr not present in new flow.");
 923                 goto error;
 924         }
 925         if (!a[OVS_FLOW_ATTR_ACTIONS]) {
 926                 OVS_NLERR(log, "Flow actions attr not present in new flow.");
 927                 goto error;
 928         }
 929
 930         /* Most of the time we need to allocate a new flow, do it before
 931          * locking.
 932          */
 933         new_flow = ovs_flow_alloc();
 934         if (IS_ERR(new_flow)) {
 935                 error = PTR_ERR(new_flow);
 936                 goto error;
 937         }
 938
 939         /* Extract key. */
 940         ovs_match_init(&match, &key, &mask);
 941         error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
 942                                   a[OVS_FLOW_ATTR_MASK], log);
 943         if (error)
 944                 goto err_kfree_flow;
 945
 946         ovs_flow_mask_key(&new_flow->key, &key, true, &mask);
 947
 948         /* Extract flow identifier. */
 949         error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
 950                                        &key, log);
 951         if (error)
 952                 goto err_kfree_flow;
 953
 954         /* Validate actions. */
 955         error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
 956                                      &new_flow->key, &acts, log);
 957         if (error) {
 958                 OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
 959                 goto err_kfree_flow;
 960         }
 961
 962         reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
 963                                         ufid_flags);
 964         if (IS_ERR(reply)) {
 965                 error = PTR_ERR(reply);
 966                 goto err_kfree_acts;
 967         }
 968
 969         ovs_lock();
 970         dp = get_dp(net, ovs_header->dp_ifindex);
 971         if (unlikely(!dp)) {
 972                 error = -ENODEV;
 973                 goto err_unlock_ovs;
 974         }
 975
 976         /* Check if this is a duplicate flow */
 977         if (ovs_identifier_is_ufid(&new_flow->id))
 978                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
 979         if (!flow)
 980                 flow = ovs_flow_tbl_lookup(&dp->table, &key);
 981         if (likely(!flow)) {
 982                 rcu_assign_pointer(new_flow->sf_acts, acts);
 983
 984                 /* Put flow in bucket. */
 985                 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
 986                 if (unlikely(error)) {
 987                         acts = NULL;
 988                         goto err_unlock_ovs;
 989                 }
 990
 991                 if (unlikely(reply)) {
 992                         error = ovs_flow_cmd_fill_info(new_flow,
 993                                                        ovs_header->dp_ifindex,
 994                                                        reply, info->snd_portid,
 995                                                        info->snd_seq, 0,
 996                                                        OVS_FLOW_CMD_NEW,
 997                                                        ufid_flags);
 998                         BUG_ON(error < 0);
 999                 }
1000                 ovs_unlock();
1001         } else {
1002                 struct sw_flow_actions *old_acts;
1003
1004                 /* Bail out if we're not allowed to modify an existing flow.
1005                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1006                  * because Generic Netlink treats the latter as a dump
1007                  * request.  We also accept NLM_F_EXCL in case that bug ever
1008                  * gets fixed.
1009                  */
1010                 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1011                                                          | NLM_F_EXCL))) {
1012                         error = -EEXIST;
1013                         goto err_unlock_ovs;
1014                 }
1015                 /* The flow identifier has to be the same for flow updates.
1016                  * Look for any overlapping flow.
1017                  */
1018                 if (unlikely(!ovs_flow_cmp(flow, &match))) {
1019                         if (ovs_identifier_is_key(&flow->id))
1020                                 flow = ovs_flow_tbl_lookup_exact(&dp->table,
1021                                                                  &match);
1022                         else /* UFID matches but key is different */
1023                                 flow = NULL;
1024                         if (!flow) {
1025                                 error = -ENOENT;
1026                                 goto err_unlock_ovs;
1027                         }
1028                 }
1029                 /* Update actions. */
1030                 old_acts = ovsl_dereference(flow->sf_acts);
1031                 rcu_assign_pointer(flow->sf_acts, acts);
1032
1033                 if (unlikely(reply)) {
1034                         error = ovs_flow_cmd_fill_info(flow,
1035                                                        ovs_header->dp_ifindex,
1036                                                        reply, info->snd_portid,
1037                                                        info->snd_seq, 0,
1038                                                        OVS_FLOW_CMD_NEW,
1039                                                        ufid_flags);
1040                         BUG_ON(error < 0);
1041                 }
1042                 ovs_unlock();
1043
1044                 ovs_nla_free_flow_actions_rcu(old_acts);
1045                 ovs_flow_free(new_flow, false);
1046         }
1047
1048         if (reply)
1049                 ovs_notify(&dp_flow_genl_family, reply, info);
1050         return 0;
1051
1052 err_unlock_ovs:
1053         ovs_unlock();
1054         kfree_skb(reply);
1055 err_kfree_acts:
1056         ovs_nla_free_flow_actions(acts);
1057 err_kfree_flow:
1058         ovs_flow_free(new_flow, false);
1059 error:
1060         return error;
1061 }
1062
1063 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1064 static struct sw_flow_actions *get_flow_actions(struct net *net,
1065                                                 const struct nlattr *a,
1066                                                 const struct sw_flow_key *key,
1067                                                 const struct sw_flow_mask *mask,
1068                                                 bool log)
1069 {
1070         struct sw_flow_actions *acts;
1071         struct sw_flow_key masked_key;
1072         int error;
1073
1074         ovs_flow_mask_key(&masked_key, key, true, mask);
1075         error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1076         if (error) {
1077                 OVS_NLERR(log,
1078                           "Actions may not be safe on all matching packets");
1079                 return ERR_PTR(error);
1080         }
1081
1082         return acts;
1083 }
1084
1085 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1086 {
1087         struct net *net = sock_net(skb->sk);
1088         struct nlattr **a = info->attrs;
1089         struct ovs_header *ovs_header = info->userhdr;
1090         struct sw_flow_key key;
1091         struct sw_flow *flow;
1092         struct sw_flow_mask mask;
1093         struct sk_buff *reply = NULL;
1094         struct datapath *dp;
1095         struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1096         struct sw_flow_match match;
1097         struct sw_flow_id sfid;
1098         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1099         int error;
1100         bool log = !a[OVS_FLOW_ATTR_PROBE];
1101         bool ufid_present;
1102
1103         /* Extract key. */
1104         error = -EINVAL;
1105         if (!a[OVS_FLOW_ATTR_KEY]) {
1106                 OVS_NLERR(log, "Flow key attribute not present in set flow.");
1107                 goto error;
1108         }
1109
1110         ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1111         ovs_match_init(&match, &key, &mask);
1112         error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1113                                   a[OVS_FLOW_ATTR_MASK], log);
1114         if (error)
1115                 goto error;
1116
1117         /* Validate actions. */
1118         if (a[OVS_FLOW_ATTR_ACTIONS]) {
1119                 acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], &key,
1120                                         &mask, log);
1121                 if (IS_ERR(acts)) {
1122                         error = PTR_ERR(acts);
1123                         goto error;
1124                 }
1125
1126                 /* Can allocate before locking if have acts. */
1127                 reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1128                                                 ufid_flags);
1129                 if (IS_ERR(reply)) {
1130                         error = PTR_ERR(reply);
1131                         goto err_kfree_acts;
1132                 }
1133         }
1134
1135         ovs_lock();
1136         dp = get_dp(net, ovs_header->dp_ifindex);
1137         if (unlikely(!dp)) {
1138                 error = -ENODEV;
1139                 goto err_unlock_ovs;
1140         }
1141         /* Check that the flow exists. */
1142         if (ufid_present)
1143                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1144         else
1145                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1146         if (unlikely(!flow)) {
1147                 error = -ENOENT;
1148                 goto err_unlock_ovs;
1149         }
1150
1151         /* Update actions, if present. */
1152         if (likely(acts)) {
1153                 old_acts = ovsl_dereference(flow->sf_acts);
1154                 rcu_assign_pointer(flow->sf_acts, acts);
1155
1156                 if (unlikely(reply)) {
1157                         error = ovs_flow_cmd_fill_info(flow,
1158                                                        ovs_header->dp_ifindex,
1159                                                        reply, info->snd_portid,
1160                                                        info->snd_seq, 0,
1161                                                        OVS_FLOW_CMD_NEW,
1162                                                        ufid_flags);
1163                         BUG_ON(error < 0);
1164                 }
1165         } else {
1166                 /* Could not alloc without acts before locking. */
1167                 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1168                                                 info, OVS_FLOW_CMD_NEW, false,
1169                                                 ufid_flags);
1170
1171                 if (IS_ERR(reply)) {
1172                         error = PTR_ERR(reply);
1173                         goto err_unlock_ovs;
1174                 }
1175         }
1176
1177         /* Clear stats. */
1178         if (a[OVS_FLOW_ATTR_CLEAR])
1179                 ovs_flow_stats_clear(flow);
1180         ovs_unlock();
1181
1182         if (reply)
1183                 ovs_notify(&dp_flow_genl_family, reply, info);
1184         if (old_acts)
1185                 ovs_nla_free_flow_actions_rcu(old_acts);
1186
1187         return 0;
1188
1189 err_unlock_ovs:
1190         ovs_unlock();
1191         kfree_skb(reply);
1192 err_kfree_acts:
1193         ovs_nla_free_flow_actions(acts);
1194 error:
1195         return error;
1196 }
1197
1198 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1199 {
1200         struct nlattr **a = info->attrs;
1201         struct ovs_header *ovs_header = info->userhdr;
1202         struct net *net = sock_net(skb->sk);
1203         struct sw_flow_key key;
1204         struct sk_buff *reply;
1205         struct sw_flow *flow;
1206         struct datapath *dp;
1207         struct sw_flow_match match;
1208         struct sw_flow_id ufid;
1209         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1210         int err = 0;
1211         bool log = !a[OVS_FLOW_ATTR_PROBE];
1212         bool ufid_present;
1213
1214         ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1215         if (a[OVS_FLOW_ATTR_KEY]) {
1216                 ovs_match_init(&match, &key, NULL);
1217                 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1218                                         log);
1219         } else if (!ufid_present) {
1220                 OVS_NLERR(log,
1221                           "Flow get message rejected, Key attribute missing.");
1222                 err = -EINVAL;
1223         }
1224         if (err)
1225                 return err;
1226
1227         ovs_lock();
1228         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1229         if (!dp) {
1230                 err = -ENODEV;
1231                 goto unlock;
1232         }
1233
1234         if (ufid_present)
1235                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1236         else
1237                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1238         if (!flow) {
1239                 err = -ENOENT;
1240                 goto unlock;
1241         }
1242
1243         reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1244                                         OVS_FLOW_CMD_NEW, true, ufid_flags);
1245         if (IS_ERR(reply)) {
1246                 err = PTR_ERR(reply);
1247                 goto unlock;
1248         }
1249
1250         ovs_unlock();
1251         return genlmsg_reply(reply, info);
1252 unlock:
1253         ovs_unlock();
1254         return err;
1255 }
1256
1257 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1258 {
1259         struct nlattr **a = info->attrs;
1260         struct ovs_header *ovs_header = info->userhdr;
1261         struct net *net = sock_net(skb->sk);
1262         struct sw_flow_key key;
1263         struct sk_buff *reply;
1264         struct sw_flow *flow = NULL;
1265         struct datapath *dp;
1266         struct sw_flow_match match;
1267         struct sw_flow_id ufid;
1268         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1269         int err;
1270         bool log = !a[OVS_FLOW_ATTR_PROBE];
1271         bool ufid_present;
1272
1273         ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1274         if (a[OVS_FLOW_ATTR_KEY]) {
1275                 ovs_match_init(&match, &key, NULL);
1276                 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1277                                         NULL, log);
1278                 if (unlikely(err))
1279                         return err;
1280         }
1281
1282         ovs_lock();
1283         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1284         if (unlikely(!dp)) {
1285                 err = -ENODEV;
1286                 goto unlock;
1287         }
1288
1289         if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1290                 err = ovs_flow_tbl_flush(&dp->table);
1291                 goto unlock;
1292         }
1293
1294         if (ufid_present)
1295                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1296         else
1297                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1298         if (unlikely(!flow)) {
1299                 err = -ENOENT;
1300                 goto unlock;
1301         }
1302
1303         ovs_flow_tbl_remove(&dp->table, flow);
1304         ovs_unlock();
1305
1306         reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1307                                         &flow->id, info, false, ufid_flags);
1308         if (likely(reply)) {
1309                 if (likely(!IS_ERR(reply))) {
1310                         rcu_read_lock();        /*To keep RCU checker happy. */
1311                         err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1312                                                      reply, info->snd_portid,
1313                                                      info->snd_seq, 0,
1314                                                      OVS_FLOW_CMD_DEL,
1315                                                      ufid_flags);
1316                         rcu_read_unlock();
1317                         BUG_ON(err < 0);
1318
1319                         ovs_notify(&dp_flow_genl_family, reply, info);
1320                 } else {
1321                         netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
1322                 }
1323         }
1324
1325         ovs_flow_free(flow, true);
1326         return 0;
1327 unlock:
1328         ovs_unlock();
1329         return err;
1330 }
1331
1332 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1333 {
1334         struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1335         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1336         struct table_instance *ti;
1337         struct datapath *dp;
1338         u32 ufid_flags;
1339         int err;
1340
1341         err = genlmsg_parse(cb->nlh, &dp_flow_genl_family, a,
1342                             OVS_FLOW_ATTR_MAX, flow_policy);
1343         if (err)
1344                 return err;
1345         ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1346
1347         rcu_read_lock();
1348         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1349         if (!dp) {
1350                 rcu_read_unlock();
1351                 return -ENODEV;
1352         }
1353
1354         ti = rcu_dereference(dp->table.ti);
1355         for (;;) {
1356                 struct sw_flow *flow;
1357                 u32 bucket, obj;
1358
1359                 bucket = cb->args[0];
1360                 obj = cb->args[1];
1361                 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1362                 if (!flow)
1363                         break;
1364
1365                 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1366                                            NETLINK_CB(cb->skb).portid,
1367                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1368                                            OVS_FLOW_CMD_NEW, ufid_flags) < 0)
1369                         break;
1370
1371                 cb->args[0] = bucket;
1372                 cb->args[1] = obj;
1373         }
1374         rcu_read_unlock();
1375         return skb->len;
1376 }
1377
1378 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1379         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1380         [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1381         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1382         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1383         [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1384         [OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1385         [OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1386 };
1387
1388 static const struct genl_ops dp_flow_genl_ops[] = {
1389         { .cmd = OVS_FLOW_CMD_NEW,
1390           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1391           .policy = flow_policy,
1392           .doit = ovs_flow_cmd_new
1393         },
1394         { .cmd = OVS_FLOW_CMD_DEL,
1395           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1396           .policy = flow_policy,
1397           .doit = ovs_flow_cmd_del
1398         },
1399         { .cmd = OVS_FLOW_CMD_GET,
1400           .flags = 0,               /* OK for unprivileged users. */
1401           .policy = flow_policy,
1402           .doit = ovs_flow_cmd_get,
1403           .dumpit = ovs_flow_cmd_dump
1404         },
1405         { .cmd = OVS_FLOW_CMD_SET,
1406           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1407           .policy = flow_policy,
1408           .doit = ovs_flow_cmd_set,
1409         },
1410 };
1411
1412 static struct genl_family dp_flow_genl_family = {
1413         .id = GENL_ID_GENERATE,
1414         .hdrsize = sizeof(struct ovs_header),
1415         .name = OVS_FLOW_FAMILY,
1416         .version = OVS_FLOW_VERSION,
1417         .maxattr = OVS_FLOW_ATTR_MAX,
1418         .netnsok = true,
1419         .parallel_ops = true,
1420         .ops = dp_flow_genl_ops,
1421         .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1422         .mcgrps = &ovs_dp_flow_multicast_group,
1423         .n_mcgrps = 1,
1424 };
1425
1426 static size_t ovs_dp_cmd_msg_size(void)
1427 {
1428         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1429
1430         msgsize += nla_total_size(IFNAMSIZ);
1431         msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1432         msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1433         msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1434
1435         return msgsize;
1436 }
1437
1438 /* Called with ovs_mutex. */
1439 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1440                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1441 {
1442         struct ovs_header *ovs_header;
1443         struct ovs_dp_stats dp_stats;
1444         struct ovs_dp_megaflow_stats dp_megaflow_stats;
1445         int err;
1446
1447         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1448                                    flags, cmd);
1449         if (!ovs_header)
1450                 goto error;
1451
1452         ovs_header->dp_ifindex = get_dpifindex(dp);
1453
1454         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1455         if (err)
1456                 goto nla_put_failure;
1457
1458         get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1459         if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1460                         &dp_stats))
1461                 goto nla_put_failure;
1462
1463         if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1464                         sizeof(struct ovs_dp_megaflow_stats),
1465                         &dp_megaflow_stats))
1466                 goto nla_put_failure;
1467
1468         if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1469                 goto nla_put_failure;
1470
1471         genlmsg_end(skb, ovs_header);
1472         return 0;
1473
1474 nla_put_failure:
1475         genlmsg_cancel(skb, ovs_header);
1476 error:
1477         return -EMSGSIZE;
1478 }
1479
1480 static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1481 {
1482         return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1483 }
1484
1485 /* Called with rcu_read_lock or ovs_mutex. */
1486 static struct datapath *lookup_datapath(struct net *net,
1487                                         const struct ovs_header *ovs_header,
1488                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1489 {
1490         struct datapath *dp;
1491
1492         if (!a[OVS_DP_ATTR_NAME])
1493                 dp = get_dp(net, ovs_header->dp_ifindex);
1494         else {
1495                 struct vport *vport;
1496
1497                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1498                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1499         }
1500         return dp ? dp : ERR_PTR(-ENODEV);
1501 }
1502
1503 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1504 {
1505         struct datapath *dp;
1506
1507         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1508         if (IS_ERR(dp))
1509                 return;
1510
1511         WARN(dp->user_features, "Dropping previously announced user features\n");
1512         dp->user_features = 0;
1513 }
1514
1515 static void ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1516 {
1517         if (a[OVS_DP_ATTR_USER_FEATURES])
1518                 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1519 }
1520
1521 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1522 {
1523         struct nlattr **a = info->attrs;
1524         struct vport_parms parms;
1525         struct sk_buff *reply;
1526         struct datapath *dp;
1527         struct vport *vport;
1528         struct ovs_net *ovs_net;
1529         int err, i;
1530
1531         err = -EINVAL;
1532         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1533                 goto err;
1534
1535         reply = ovs_dp_cmd_alloc_info();
1536         if (!reply)
1537                 return -ENOMEM;
1538
1539         err = -ENOMEM;
1540         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1541         if (dp == NULL)
1542                 goto err_free_reply;
1543
1544         ovs_dp_set_net(dp, sock_net(skb->sk));
1545
1546         /* Allocate table. */
1547         err = ovs_flow_tbl_init(&dp->table);
1548         if (err)
1549                 goto err_free_dp;
1550
1551         dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1552         if (!dp->stats_percpu) {
1553                 err = -ENOMEM;
1554                 goto err_destroy_table;
1555         }
1556
1557         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1558                             GFP_KERNEL);
1559         if (!dp->ports) {
1560                 err = -ENOMEM;
1561                 goto err_destroy_percpu;
1562         }
1563
1564         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1565                 INIT_HLIST_HEAD(&dp->ports[i]);
1566
1567         /* Set up our datapath device. */
1568         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1569         parms.type = OVS_VPORT_TYPE_INTERNAL;
1570         parms.options = NULL;
1571         parms.dp = dp;
1572         parms.port_no = OVSP_LOCAL;
1573         parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1574
1575         ovs_dp_change(dp, a);
1576
1577         /* So far only local changes have been made, now need the lock. */
1578         ovs_lock();
1579
1580         vport = new_vport(&parms);
1581         if (IS_ERR(vport)) {
1582                 err = PTR_ERR(vport);
1583                 if (err == -EBUSY)
1584                         err = -EEXIST;
1585
1586                 if (err == -EEXIST) {
1587                         /* An outdated user space instance that does not understand
1588                          * the concept of user_features has attempted to create a new
1589                          * datapath and is likely to reuse it. Drop all user features.
1590                          */
1591                         if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1592                                 ovs_dp_reset_user_features(skb, info);
1593                 }
1594
1595                 goto err_destroy_ports_array;
1596         }
1597
1598         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1599                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1600         BUG_ON(err < 0);
1601
1602         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1603         list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1604
1605         ovs_unlock();
1606
1607         ovs_notify(&dp_datapath_genl_family, reply, info);
1608         return 0;
1609
1610 err_destroy_ports_array:
1611         ovs_unlock();
1612         kfree(dp->ports);
1613 err_destroy_percpu:
1614         free_percpu(dp->stats_percpu);
1615 err_destroy_table:
1616         ovs_flow_tbl_destroy(&dp->table);
1617 err_free_dp:
1618         kfree(dp);
1619 err_free_reply:
1620         kfree_skb(reply);
1621 err:
1622         return err;
1623 }
1624
1625 /* Called with ovs_mutex. */
1626 static void __dp_destroy(struct datapath *dp)
1627 {
1628         int i;
1629
1630         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1631                 struct vport *vport;
1632                 struct hlist_node *n;
1633
1634                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1635                         if (vport->port_no != OVSP_LOCAL)
1636                                 ovs_dp_detach_port(vport);
1637         }
1638
1639         list_del_rcu(&dp->list_node);
1640
1641         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1642          * all ports in datapath are destroyed first before freeing datapath.
1643          */
1644         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1645
1646         /* RCU destroy the flow table */
1647         call_rcu(&dp->rcu, destroy_dp_rcu);
1648 }
1649
1650 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1651 {
1652         struct sk_buff *reply;
1653         struct datapath *dp;
1654         int err;
1655
1656         reply = ovs_dp_cmd_alloc_info();
1657         if (!reply)
1658                 return -ENOMEM;
1659
1660         ovs_lock();
1661         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1662         err = PTR_ERR(dp);
1663         if (IS_ERR(dp))
1664                 goto err_unlock_free;
1665
1666         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1667                                    info->snd_seq, 0, OVS_DP_CMD_DEL);
1668         BUG_ON(err < 0);
1669
1670         __dp_destroy(dp);
1671         ovs_unlock();
1672
1673         ovs_notify(&dp_datapath_genl_family, reply, info);
1674
1675         return 0;
1676
1677 err_unlock_free:
1678         ovs_unlock();
1679         kfree_skb(reply);
1680         return err;
1681 }
1682
1683 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1684 {
1685         struct sk_buff *reply;
1686         struct datapath *dp;
1687         int err;
1688
1689         reply = ovs_dp_cmd_alloc_info();
1690         if (!reply)
1691                 return -ENOMEM;
1692
1693         ovs_lock();
1694         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1695         err = PTR_ERR(dp);
1696         if (IS_ERR(dp))
1697                 goto err_unlock_free;
1698
1699         ovs_dp_change(dp, info->attrs);
1700
1701         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1702                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1703         BUG_ON(err < 0);
1704
1705         ovs_unlock();
1706         ovs_notify(&dp_datapath_genl_family, reply, info);
1707
1708         return 0;
1709
1710 err_unlock_free:
1711         ovs_unlock();
1712         kfree_skb(reply);
1713         return err;
1714 }
1715
1716 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1717 {
1718         struct sk_buff *reply;
1719         struct datapath *dp;
1720         int err;
1721
1722         reply = ovs_dp_cmd_alloc_info();
1723         if (!reply)
1724                 return -ENOMEM;
1725
1726         ovs_lock();
1727         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1728         if (IS_ERR(dp)) {
1729                 err = PTR_ERR(dp);
1730                 goto err_unlock_free;
1731         }
1732         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1733                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1734         BUG_ON(err < 0);
1735         ovs_unlock();
1736
1737         return genlmsg_reply(reply, info);
1738
1739 err_unlock_free:
1740         ovs_unlock();
1741         kfree_skb(reply);
1742         return err;
1743 }
1744
1745 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1746 {
1747         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1748         struct datapath *dp;
1749         int skip = cb->args[0];
1750         int i = 0;
1751
1752         ovs_lock();
1753         list_for_each_entry(dp, &ovs_net->dps, list_node) {
1754                 if (i >= skip &&
1755                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1756                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1757                                          OVS_DP_CMD_NEW) < 0)
1758                         break;
1759                 i++;
1760         }
1761         ovs_unlock();
1762
1763         cb->args[0] = i;
1764
1765         return skb->len;
1766 }
1767
1768 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1769         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1770         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1771         [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1772 };
1773
1774 static const struct genl_ops dp_datapath_genl_ops[] = {
1775         { .cmd = OVS_DP_CMD_NEW,
1776           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1777           .policy = datapath_policy,
1778           .doit = ovs_dp_cmd_new
1779         },
1780         { .cmd = OVS_DP_CMD_DEL,
1781           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1782           .policy = datapath_policy,
1783           .doit = ovs_dp_cmd_del
1784         },
1785         { .cmd = OVS_DP_CMD_GET,
1786           .flags = 0,               /* OK for unprivileged users. */
1787           .policy = datapath_policy,
1788           .doit = ovs_dp_cmd_get,
1789           .dumpit = ovs_dp_cmd_dump
1790         },
1791         { .cmd = OVS_DP_CMD_SET,
1792           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1793           .policy = datapath_policy,
1794           .doit = ovs_dp_cmd_set,
1795         },
1796 };
1797
1798 static struct genl_family dp_datapath_genl_family = {
1799         .id = GENL_ID_GENERATE,
1800         .hdrsize = sizeof(struct ovs_header),
1801         .name = OVS_DATAPATH_FAMILY,
1802         .version = OVS_DATAPATH_VERSION,
1803         .maxattr = OVS_DP_ATTR_MAX,
1804         .netnsok = true,
1805         .parallel_ops = true,
1806         .ops = dp_datapath_genl_ops,
1807         .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1808         .mcgrps = &ovs_dp_datapath_multicast_group,
1809         .n_mcgrps = 1,
1810 };
1811
1812 /* Called with ovs_mutex or RCU read lock. */
1813 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1814                                    u32 portid, u32 seq, u32 flags, u8 cmd)
1815 {
1816         struct ovs_header *ovs_header;
1817         struct ovs_vport_stats vport_stats;
1818         int err;
1819
1820         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1821                                  flags, cmd);
1822         if (!ovs_header)
1823                 return -EMSGSIZE;
1824
1825         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1826
1827         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1828             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1829             nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1830                            ovs_vport_name(vport)))
1831                 goto nla_put_failure;
1832
1833         ovs_vport_get_stats(vport, &vport_stats);
1834         if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1835                     &vport_stats))
1836                 goto nla_put_failure;
1837
1838         if (ovs_vport_get_upcall_portids(vport, skb))
1839                 goto nla_put_failure;
1840
1841         err = ovs_vport_get_options(vport, skb);
1842         if (err == -EMSGSIZE)
1843                 goto error;
1844
1845         genlmsg_end(skb, ovs_header);
1846         return 0;
1847
1848 nla_put_failure:
1849         err = -EMSGSIZE;
1850 error:
1851         genlmsg_cancel(skb, ovs_header);
1852         return err;
1853 }
1854
1855 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1856 {
1857         return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1858 }
1859
1860 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1861 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1862                                          u32 seq, u8 cmd)
1863 {
1864         struct sk_buff *skb;
1865         int retval;
1866
1867         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1868         if (!skb)
1869                 return ERR_PTR(-ENOMEM);
1870
1871         retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1872         BUG_ON(retval < 0);
1873
1874         return skb;
1875 }
1876
1877 /* Called with ovs_mutex or RCU read lock. */
1878 static struct vport *lookup_vport(struct net *net,
1879                                   const struct ovs_header *ovs_header,
1880                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1881 {
1882         struct datapath *dp;
1883         struct vport *vport;
1884
1885         if (a[OVS_VPORT_ATTR_NAME]) {
1886                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1887                 if (!vport)
1888                         return ERR_PTR(-ENODEV);
1889                 if (ovs_header->dp_ifindex &&
1890                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1891                         return ERR_PTR(-ENODEV);
1892                 return vport;
1893         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1894                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1895
1896                 if (port_no >= DP_MAX_PORTS)
1897                         return ERR_PTR(-EFBIG);
1898
1899                 dp = get_dp(net, ovs_header->dp_ifindex);
1900                 if (!dp)
1901                         return ERR_PTR(-ENODEV);
1902
1903                 vport = ovs_vport_ovsl_rcu(dp, port_no);
1904                 if (!vport)
1905                         return ERR_PTR(-ENODEV);
1906                 return vport;
1907         } else
1908                 return ERR_PTR(-EINVAL);
1909 }
1910
1911 /* Called with ovs_mutex */
1912 static void update_headroom(struct datapath *dp)
1913 {
1914         unsigned dev_headroom, max_headroom = 0;
1915         struct net_device *dev;
1916         struct vport *vport;
1917         int i;
1918
1919         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1920                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1921                         dev = vport->dev;
1922                         dev_headroom = netdev_get_fwd_headroom(dev);
1923                         if (dev_headroom > max_headroom)
1924                                 max_headroom = dev_headroom;
1925                 }
1926         }
1927
1928         dp->max_headroom = max_headroom;
1929         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1930                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node)
1931                         netdev_set_rx_headroom(vport->dev, max_headroom);
1932 }
1933
1934 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1935 {
1936         struct nlattr **a = info->attrs;
1937         struct ovs_header *ovs_header = info->userhdr;
1938         struct vport_parms parms;
1939         struct sk_buff *reply;
1940         struct vport *vport;
1941         struct datapath *dp;
1942         u32 port_no;
1943         int err;
1944
1945         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1946             !a[OVS_VPORT_ATTR_UPCALL_PID])
1947                 return -EINVAL;
1948
1949         port_no = a[OVS_VPORT_ATTR_PORT_NO]
1950                 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
1951         if (port_no >= DP_MAX_PORTS)
1952                 return -EFBIG;
1953
1954         reply = ovs_vport_cmd_alloc_info();
1955         if (!reply)
1956                 return -ENOMEM;
1957
1958         ovs_lock();
1959 restart:
1960         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1961         err = -ENODEV;
1962         if (!dp)
1963                 goto exit_unlock_free;
1964
1965         if (port_no) {
1966                 vport = ovs_vport_ovsl(dp, port_no);
1967                 err = -EBUSY;
1968                 if (vport)
1969                         goto exit_unlock_free;
1970         } else {
1971                 for (port_no = 1; ; port_no++) {
1972                         if (port_no >= DP_MAX_PORTS) {
1973                                 err = -EFBIG;
1974                                 goto exit_unlock_free;
1975                         }
1976                         vport = ovs_vport_ovsl(dp, port_no);
1977                         if (!vport)
1978                                 break;
1979                 }
1980         }
1981
1982         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1983         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1984         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1985         parms.dp = dp;
1986         parms.port_no = port_no;
1987         parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
1988
1989         vport = new_vport(&parms);
1990         err = PTR_ERR(vport);
1991         if (IS_ERR(vport)) {
1992                 if (err == -EAGAIN)
1993                         goto restart;
1994                 goto exit_unlock_free;
1995         }
1996
1997         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1998                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1999
2000         if (netdev_get_fwd_headroom(vport->dev) > dp->max_headroom)
2001                 update_headroom(dp);
2002         else
2003                 netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2004
2005         BUG_ON(err < 0);
2006         ovs_unlock();
2007
2008         ovs_notify(&dp_vport_genl_family, reply, info);
2009         return 0;
2010
2011 exit_unlock_free:
2012         ovs_unlock();
2013         kfree_skb(reply);
2014         return err;
2015 }
2016
2017 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2018 {
2019         struct nlattr **a = info->attrs;
2020         struct sk_buff *reply;
2021         struct vport *vport;
2022         int err;
2023
2024         reply = ovs_vport_cmd_alloc_info();
2025         if (!reply)
2026                 return -ENOMEM;
2027
2028         ovs_lock();
2029         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2030         err = PTR_ERR(vport);
2031         if (IS_ERR(vport))
2032                 goto exit_unlock_free;
2033
2034         if (a[OVS_VPORT_ATTR_TYPE] &&
2035             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2036                 err = -EINVAL;
2037                 goto exit_unlock_free;
2038         }
2039
2040         if (a[OVS_VPORT_ATTR_OPTIONS]) {
2041                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2042                 if (err)
2043                         goto exit_unlock_free;
2044         }
2045
2046
2047         if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2048                 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2049
2050                 err = ovs_vport_set_upcall_portids(vport, ids);
2051                 if (err)
2052                         goto exit_unlock_free;
2053         }
2054
2055         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2056                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2057         BUG_ON(err < 0);
2058
2059         ovs_unlock();
2060         ovs_notify(&dp_vport_genl_family, reply, info);
2061         return 0;
2062
2063 exit_unlock_free:
2064         ovs_unlock();
2065         kfree_skb(reply);
2066         return err;
2067 }
2068
2069 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2070 {
2071         bool must_update_headroom = false;
2072         struct nlattr **a = info->attrs;
2073         struct sk_buff *reply;
2074         struct datapath *dp;
2075         struct vport *vport;
2076         int err;
2077
2078         reply = ovs_vport_cmd_alloc_info();
2079         if (!reply)
2080                 return -ENOMEM;
2081
2082         ovs_lock();
2083         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2084         err = PTR_ERR(vport);
2085         if (IS_ERR(vport))
2086                 goto exit_unlock_free;
2087
2088         if (vport->port_no == OVSP_LOCAL) {
2089                 err = -EINVAL;
2090                 goto exit_unlock_free;
2091         }
2092
2093         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2094                                       info->snd_seq, 0, OVS_VPORT_CMD_DEL);
2095         BUG_ON(err < 0);
2096
2097         /* the vport deletion may trigger dp headroom update */
2098         dp = vport->dp;
2099         if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2100                 must_update_headroom = true;
2101         netdev_reset_rx_headroom(vport->dev);
2102         ovs_dp_detach_port(vport);
2103
2104         if (must_update_headroom)
2105                 update_headroom(dp);
2106         ovs_unlock();
2107
2108         ovs_notify(&dp_vport_genl_family, reply, info);
2109         return 0;
2110
2111 exit_unlock_free:
2112         ovs_unlock();
2113         kfree_skb(reply);
2114         return err;
2115 }
2116
2117 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2118 {
2119         struct nlattr **a = info->attrs;
2120         struct ovs_header *ovs_header = info->userhdr;
2121         struct sk_buff *reply;
2122         struct vport *vport;
2123         int err;
2124
2125         reply = ovs_vport_cmd_alloc_info();
2126         if (!reply)
2127                 return -ENOMEM;
2128
2129         rcu_read_lock();
2130         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2131         err = PTR_ERR(vport);
2132         if (IS_ERR(vport))
2133                 goto exit_unlock_free;
2134         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2135                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2136         BUG_ON(err < 0);
2137         rcu_read_unlock();
2138
2139         return genlmsg_reply(reply, info);
2140
2141 exit_unlock_free:
2142         rcu_read_unlock();
2143         kfree_skb(reply);
2144         return err;
2145 }
2146
2147 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2148 {
2149         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2150         struct datapath *dp;
2151         int bucket = cb->args[0], skip = cb->args[1];
2152         int i, j = 0;
2153
2154         rcu_read_lock();
2155         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2156         if (!dp) {
2157                 rcu_read_unlock();
2158                 return -ENODEV;
2159         }
2160         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2161                 struct vport *vport;
2162
2163                 j = 0;
2164                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2165                         if (j >= skip &&
2166                             ovs_vport_cmd_fill_info(vport, skb,
2167                                                     NETLINK_CB(cb->skb).portid,
2168                                                     cb->nlh->nlmsg_seq,
2169                                                     NLM_F_MULTI,
2170                                                     OVS_VPORT_CMD_NEW) < 0)
2171                                 goto out;
2172
2173                         j++;
2174                 }
2175                 skip = 0;
2176         }
2177 out:
2178         rcu_read_unlock();
2179
2180         cb->args[0] = i;
2181         cb->args[1] = j;
2182
2183         return skb->len;
2184 }
2185
2186 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2187         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2188         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2189         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2190         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2191         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
2192         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2193 };
2194
2195 static const struct genl_ops dp_vport_genl_ops[] = {
2196         { .cmd = OVS_VPORT_CMD_NEW,
2197           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2198           .policy = vport_policy,
2199           .doit = ovs_vport_cmd_new
2200         },
2201         { .cmd = OVS_VPORT_CMD_DEL,
2202           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2203           .policy = vport_policy,
2204           .doit = ovs_vport_cmd_del
2205         },
2206         { .cmd = OVS_VPORT_CMD_GET,
2207           .flags = 0,               /* OK for unprivileged users. */
2208           .policy = vport_policy,
2209           .doit = ovs_vport_cmd_get,
2210           .dumpit = ovs_vport_cmd_dump
2211         },
2212         { .cmd = OVS_VPORT_CMD_SET,
2213           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2214           .policy = vport_policy,
2215           .doit = ovs_vport_cmd_set,
2216         },
2217 };
2218
2219 struct genl_family dp_vport_genl_family = {
2220         .id = GENL_ID_GENERATE,
2221         .hdrsize = sizeof(struct ovs_header),
2222         .name = OVS_VPORT_FAMILY,
2223         .version = OVS_VPORT_VERSION,
2224         .maxattr = OVS_VPORT_ATTR_MAX,
2225         .netnsok = true,
2226         .parallel_ops = true,
2227         .ops = dp_vport_genl_ops,
2228         .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2229         .mcgrps = &ovs_dp_vport_multicast_group,
2230         .n_mcgrps = 1,
2231 };
2232
2233 static struct genl_family * const dp_genl_families[] = {
2234         &dp_datapath_genl_family,
2235         &dp_vport_genl_family,
2236         &dp_flow_genl_family,
2237         &dp_packet_genl_family,
2238 };
2239
2240 static void dp_unregister_genl(int n_families)
2241 {
2242         int i;
2243
2244         for (i = 0; i < n_families; i++)
2245                 genl_unregister_family(dp_genl_families[i]);
2246 }
2247
2248 static int dp_register_genl(void)
2249 {
2250         int err;
2251         int i;
2252
2253         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2254
2255                 err = genl_register_family(dp_genl_families[i]);
2256                 if (err)
2257                         goto error;
2258         }
2259
2260         return 0;
2261
2262 error:
2263         dp_unregister_genl(i);
2264         return err;
2265 }
2266
2267 static int __net_init ovs_init_net(struct net *net)
2268 {
2269         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2270
2271         INIT_LIST_HEAD(&ovs_net->dps);
2272         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2273         ovs_ct_init(net);
2274         return 0;
2275 }
2276
2277 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2278                                             struct list_head *head)
2279 {
2280         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2281         struct datapath *dp;
2282
2283         list_for_each_entry(dp, &ovs_net->dps, list_node) {
2284                 int i;
2285
2286                 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2287                         struct vport *vport;
2288
2289                         hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2290                                 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2291                                         continue;
2292
2293                                 if (dev_net(vport->dev) == dnet)
2294                                         list_add(&vport->detach_list, head);
2295                         }
2296                 }
2297         }
2298 }
2299
2300 static void __net_exit ovs_exit_net(struct net *dnet)
2301 {
2302         struct datapath *dp, *dp_next;
2303         struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2304         struct vport *vport, *vport_next;
2305         struct net *net;
2306         LIST_HEAD(head);
2307
2308         ovs_ct_exit(dnet);
2309         ovs_lock();
2310         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2311                 __dp_destroy(dp);
2312
2313         rtnl_lock();
2314         for_each_net(net)
2315                 list_vports_from_net(net, dnet, &head);
2316         rtnl_unlock();
2317
2318         /* Detach all vports from given namespace. */
2319         list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2320                 list_del(&vport->detach_list);
2321                 ovs_dp_detach_port(vport);
2322         }
2323
2324         ovs_unlock();
2325
2326         cancel_work_sync(&ovs_net->dp_notify_work);
2327 }
2328
2329 static struct pernet_operations ovs_net_ops = {
2330         .init = ovs_init_net,
2331         .exit = ovs_exit_net,
2332         .id   = &ovs_net_id,
2333         .size = sizeof(struct ovs_net),
2334 };
2335
2336 static int __init dp_init(void)
2337 {
2338         int err;
2339
2340         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2341
2342         pr_info("Open vSwitch switching datapath\n");
2343
2344         err = action_fifos_init();
2345         if (err)
2346                 goto error;
2347
2348         err = ovs_internal_dev_rtnl_link_register();
2349         if (err)
2350                 goto error_action_fifos_exit;
2351
2352         err = ovs_flow_init();
2353         if (err)
2354                 goto error_unreg_rtnl_link;
2355
2356         err = ovs_vport_init();
2357         if (err)
2358                 goto error_flow_exit;
2359
2360         err = register_pernet_device(&ovs_net_ops);
2361         if (err)
2362                 goto error_vport_exit;
2363
2364         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2365         if (err)
2366                 goto error_netns_exit;
2367
2368         err = ovs_netdev_init();
2369         if (err)
2370                 goto error_unreg_notifier;
2371
2372         err = dp_register_genl();
2373         if (err < 0)
2374                 goto error_unreg_netdev;
2375
2376         return 0;
2377
2378 error_unreg_netdev:
2379         ovs_netdev_exit();
2380 error_unreg_notifier:
2381         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2382 error_netns_exit:
2383         unregister_pernet_device(&ovs_net_ops);
2384 error_vport_exit:
2385         ovs_vport_exit();
2386 error_flow_exit:
2387         ovs_flow_exit();
2388 error_unreg_rtnl_link:
2389         ovs_internal_dev_rtnl_link_unregister();
2390 error_action_fifos_exit:
2391         action_fifos_exit();
2392 error:
2393         return err;
2394 }
2395
2396 static void dp_cleanup(void)
2397 {
2398         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2399         ovs_netdev_exit();
2400         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2401         unregister_pernet_device(&ovs_net_ops);
2402         rcu_barrier();
2403         ovs_vport_exit();
2404         ovs_flow_exit();
2405         ovs_internal_dev_rtnl_link_unregister();
2406         action_fifos_exit();
2407 }
2408
2409 module_init(dp_init);
2410 module_exit(dp_cleanup);
2411
2412 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2413 MODULE_LICENSE("GPL");