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_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
 742                 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
 743                 + nla_total_size_64bit(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_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
 758                               OVS_FLOW_ATTR_PAD))
 759                 return -EMSGSIZE;
 760
 761         if (stats.n_packets &&
 762             nla_put_64bit(skb, OVS_FLOW_ATTR_STATS,
 763                           sizeof(struct ovs_flow_stats), &stats,
 764                           OVS_FLOW_ATTR_PAD))
 765                 return -EMSGSIZE;
 766
 767         if ((u8)ntohs(tcp_flags) &&
 768              nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
 769                 return -EMSGSIZE;
 770
 771         return 0;
 772 }
 773
 774 /* Called with ovs_mutex or RCU read lock. */
 775 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
 776                                      struct sk_buff *skb, int skb_orig_len)
 777 {
 778         struct nlattr *start;
 779         int err;
 780
 781         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 782          * this is the first flow to be dumped into 'skb'.  This is unusual for
 783          * Netlink but individual action lists can be longer than
 784          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 785          * The userspace caller can always fetch the actions separately if it
 786          * really wants them.  (Most userspace callers in fact don't care.)
 787          *
 788          * This can only fail for dump operations because the skb is always
 789          * properly sized for single flows.
 790          */
 791         start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
 792         if (start) {
 793                 const struct sw_flow_actions *sf_acts;
 794
 795                 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
 796                 err = ovs_nla_put_actions(sf_acts->actions,
 797                                           sf_acts->actions_len, skb);
 798
 799                 if (!err)
 800                         nla_nest_end(skb, start);
 801                 else {
 802                         if (skb_orig_len)
 803                                 return err;
 804
 805                         nla_nest_cancel(skb, start);
 806                 }
 807         } else if (skb_orig_len) {
 808                 return -EMSGSIZE;
 809         }
 810
 811         return 0;
 812 }
 813
 814 /* Called with ovs_mutex or RCU read lock. */
 815 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
 816                                   struct sk_buff *skb, u32 portid,
 817                                   u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
 818 {
 819         const int skb_orig_len = skb->len;
 820         struct ovs_header *ovs_header;
 821         int err;
 822
 823         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
 824                                  flags, cmd);
 825         if (!ovs_header)
 826                 return -EMSGSIZE;
 827
 828         ovs_header->dp_ifindex = dp_ifindex;
 829
 830         err = ovs_nla_put_identifier(flow, skb);
 831         if (err)
 832                 goto error;
 833
 834         if (should_fill_key(&flow->id, ufid_flags)) {
 835                 err = ovs_nla_put_masked_key(flow, skb);
 836                 if (err)
 837                         goto error;
 838         }
 839
 840         if (should_fill_mask(ufid_flags)) {
 841                 err = ovs_nla_put_mask(flow, skb);
 842                 if (err)
 843                         goto error;
 844         }
 845
 846         err = ovs_flow_cmd_fill_stats(flow, skb);
 847         if (err)
 848                 goto error;
 849
 850         if (should_fill_actions(ufid_flags)) {
 851                 err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
 852                 if (err)
 853                         goto error;
 854         }
 855
 856         genlmsg_end(skb, ovs_header);
 857         return 0;
 858
 859 error:
 860         genlmsg_cancel(skb, ovs_header);
 861         return err;
 862 }
 863
 864 /* May not be called with RCU read lock. */
 865 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
 866                                                const struct sw_flow_id *sfid,
 867                                                struct genl_info *info,
 868                                                bool always,
 869                                                uint32_t ufid_flags)
 870 {
 871         struct sk_buff *skb;
 872         size_t len;
 873
 874         if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
 875                 return NULL;
 876
 877         len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
 878         skb = genlmsg_new(len, GFP_KERNEL);
 879         if (!skb)
 880                 return ERR_PTR(-ENOMEM);
 881
 882         return skb;
 883 }
 884
 885 /* Called with ovs_mutex. */
 886 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
 887                                                int dp_ifindex,
 888                                                struct genl_info *info, u8 cmd,
 889                                                bool always, u32 ufid_flags)
 890 {
 891         struct sk_buff *skb;
 892         int retval;
 893
 894         skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
 895                                       &flow->id, info, always, ufid_flags);
 896         if (IS_ERR_OR_NULL(skb))
 897                 return skb;
 898
 899         retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
 900                                         info->snd_portid, info->snd_seq, 0,
 901                                         cmd, ufid_flags);
 902         BUG_ON(retval < 0);
 903         return skb;
 904 }
 905
 906 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
 907 {
 908         struct net *net = sock_net(skb->sk);
 909         struct nlattr **a = info->attrs;
 910         struct ovs_header *ovs_header = info->userhdr;
 911         struct sw_flow *flow = NULL, *new_flow;
 912         struct sw_flow_mask mask;
 913         struct sk_buff *reply;
 914         struct datapath *dp;
 915         struct sw_flow_key key;
 916         struct sw_flow_actions *acts;
 917         struct sw_flow_match match;
 918         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
 919         int error;
 920         bool log = !a[OVS_FLOW_ATTR_PROBE];
 921
 922         /* Must have key and actions. */
 923         error = -EINVAL;
 924         if (!a[OVS_FLOW_ATTR_KEY]) {
 925                 OVS_NLERR(log, "Flow key attr not present in new flow.");
 926                 goto error;
 927         }
 928         if (!a[OVS_FLOW_ATTR_ACTIONS]) {
 929                 OVS_NLERR(log, "Flow actions attr not present in new flow.");
 930                 goto error;
 931         }
 932
 933         /* Most of the time we need to allocate a new flow, do it before
 934          * locking.
 935          */
 936         new_flow = ovs_flow_alloc();
 937         if (IS_ERR(new_flow)) {
 938                 error = PTR_ERR(new_flow);
 939                 goto error;
 940         }
 941
 942         /* Extract key. */
 943         ovs_match_init(&match, &key, &mask);
 944         error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
 945                                   a[OVS_FLOW_ATTR_MASK], log);
 946         if (error)
 947                 goto err_kfree_flow;
 948
 949         ovs_flow_mask_key(&new_flow->key, &key, true, &mask);
 950
 951         /* Extract flow identifier. */
 952         error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
 953                                        &key, log);
 954         if (error)
 955                 goto err_kfree_flow;
 956
 957         /* Validate actions. */
 958         error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
 959                                      &new_flow->key, &acts, log);
 960         if (error) {
 961                 OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
 962                 goto err_kfree_flow;
 963         }
 964
 965         reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
 966                                         ufid_flags);
 967         if (IS_ERR(reply)) {
 968                 error = PTR_ERR(reply);
 969                 goto err_kfree_acts;
 970         }
 971
 972         ovs_lock();
 973         dp = get_dp(net, ovs_header->dp_ifindex);
 974         if (unlikely(!dp)) {
 975                 error = -ENODEV;
 976                 goto err_unlock_ovs;
 977         }
 978
 979         /* Check if this is a duplicate flow */
 980         if (ovs_identifier_is_ufid(&new_flow->id))
 981                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
 982         if (!flow)
 983                 flow = ovs_flow_tbl_lookup(&dp->table, &key);
 984         if (likely(!flow)) {
 985                 rcu_assign_pointer(new_flow->sf_acts, acts);
 986
 987                 /* Put flow in bucket. */
 988                 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
 989                 if (unlikely(error)) {
 990                         acts = NULL;
 991                         goto err_unlock_ovs;
 992                 }
 993
 994                 if (unlikely(reply)) {
 995                         error = ovs_flow_cmd_fill_info(new_flow,
 996                                                        ovs_header->dp_ifindex,
 997                                                        reply, info->snd_portid,
 998                                                        info->snd_seq, 0,
 999                                                        OVS_FLOW_CMD_NEW,
1000                                                        ufid_flags);
1001                         BUG_ON(error < 0);
1002                 }
1003                 ovs_unlock();
1004         } else {
1005                 struct sw_flow_actions *old_acts;
1006
1007                 /* Bail out if we're not allowed to modify an existing flow.
1008                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1009                  * because Generic Netlink treats the latter as a dump
1010                  * request.  We also accept NLM_F_EXCL in case that bug ever
1011                  * gets fixed.
1012                  */
1013                 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1014                                                          | NLM_F_EXCL))) {
1015                         error = -EEXIST;
1016                         goto err_unlock_ovs;
1017                 }
1018                 /* The flow identifier has to be the same for flow updates.
1019                  * Look for any overlapping flow.
1020                  */
1021                 if (unlikely(!ovs_flow_cmp(flow, &match))) {
1022                         if (ovs_identifier_is_key(&flow->id))
1023                                 flow = ovs_flow_tbl_lookup_exact(&dp->table,
1024                                                                  &match);
1025                         else /* UFID matches but key is different */
1026                                 flow = NULL;
1027                         if (!flow) {
1028                                 error = -ENOENT;
1029                                 goto err_unlock_ovs;
1030                         }
1031                 }
1032                 /* Update actions. */
1033                 old_acts = ovsl_dereference(flow->sf_acts);
1034                 rcu_assign_pointer(flow->sf_acts, acts);
1035
1036                 if (unlikely(reply)) {
1037                         error = ovs_flow_cmd_fill_info(flow,
1038                                                        ovs_header->dp_ifindex,
1039                                                        reply, info->snd_portid,
1040                                                        info->snd_seq, 0,
1041                                                        OVS_FLOW_CMD_NEW,
1042                                                        ufid_flags);
1043                         BUG_ON(error < 0);
1044                 }
1045                 ovs_unlock();
1046
1047                 ovs_nla_free_flow_actions_rcu(old_acts);
1048                 ovs_flow_free(new_flow, false);
1049         }
1050
1051         if (reply)
1052                 ovs_notify(&dp_flow_genl_family, reply, info);
1053         return 0;
1054
1055 err_unlock_ovs:
1056         ovs_unlock();
1057         kfree_skb(reply);
1058 err_kfree_acts:
1059         ovs_nla_free_flow_actions(acts);
1060 err_kfree_flow:
1061         ovs_flow_free(new_flow, false);
1062 error:
1063         return error;
1064 }
1065
1066 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1067 static struct sw_flow_actions *get_flow_actions(struct net *net,
1068                                                 const struct nlattr *a,
1069                                                 const struct sw_flow_key *key,
1070                                                 const struct sw_flow_mask *mask,
1071                                                 bool log)
1072 {
1073         struct sw_flow_actions *acts;
1074         struct sw_flow_key masked_key;
1075         int error;
1076
1077         ovs_flow_mask_key(&masked_key, key, true, mask);
1078         error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1079         if (error) {
1080                 OVS_NLERR(log,
1081                           "Actions may not be safe on all matching packets");
1082                 return ERR_PTR(error);
1083         }
1084
1085         return acts;
1086 }
1087
1088 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1089 {
1090         struct net *net = sock_net(skb->sk);
1091         struct nlattr **a = info->attrs;
1092         struct ovs_header *ovs_header = info->userhdr;
1093         struct sw_flow_key key;
1094         struct sw_flow *flow;
1095         struct sw_flow_mask mask;
1096         struct sk_buff *reply = NULL;
1097         struct datapath *dp;
1098         struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1099         struct sw_flow_match match;
1100         struct sw_flow_id sfid;
1101         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1102         int error = 0;
1103         bool log = !a[OVS_FLOW_ATTR_PROBE];
1104         bool ufid_present;
1105
1106         ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1107         if (a[OVS_FLOW_ATTR_KEY]) {
1108                 ovs_match_init(&match, &key, &mask);
1109                 error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1110                                           a[OVS_FLOW_ATTR_MASK], log);
1111         } else if (!ufid_present) {
1112                 OVS_NLERR(log,
1113                           "Flow set message rejected, Key attribute missing.");
1114                 error = -EINVAL;
1115         }
1116         if (error)
1117                 goto error;
1118
1119         /* Validate actions. */
1120         if (a[OVS_FLOW_ATTR_ACTIONS]) {
1121                 if (!a[OVS_FLOW_ATTR_KEY]) {
1122                         OVS_NLERR(log,
1123                                   "Flow key attribute not present in set flow.");
1124                         error = -EINVAL;
1125                         goto error;
1126                 }
1127
1128                 acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], &key,
1129                                         &mask, log);
1130                 if (IS_ERR(acts)) {
1131                         error = PTR_ERR(acts);
1132                         goto error;
1133                 }
1134
1135                 /* Can allocate before locking if have acts. */
1136                 reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1137                                                 ufid_flags);
1138                 if (IS_ERR(reply)) {
1139                         error = PTR_ERR(reply);
1140                         goto err_kfree_acts;
1141                 }
1142         }
1143
1144         ovs_lock();
1145         dp = get_dp(net, ovs_header->dp_ifindex);
1146         if (unlikely(!dp)) {
1147                 error = -ENODEV;
1148                 goto err_unlock_ovs;
1149         }
1150         /* Check that the flow exists. */
1151         if (ufid_present)
1152                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1153         else
1154                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1155         if (unlikely(!flow)) {
1156                 error = -ENOENT;
1157                 goto err_unlock_ovs;
1158         }
1159
1160         /* Update actions, if present. */
1161         if (likely(acts)) {
1162                 old_acts = ovsl_dereference(flow->sf_acts);
1163                 rcu_assign_pointer(flow->sf_acts, acts);
1164
1165                 if (unlikely(reply)) {
1166                         error = ovs_flow_cmd_fill_info(flow,
1167                                                        ovs_header->dp_ifindex,
1168                                                        reply, info->snd_portid,
1169                                                        info->snd_seq, 0,
1170                                                        OVS_FLOW_CMD_NEW,
1171                                                        ufid_flags);
1172                         BUG_ON(error < 0);
1173                 }
1174         } else {
1175                 /* Could not alloc without acts before locking. */
1176                 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1177                                                 info, OVS_FLOW_CMD_NEW, false,
1178                                                 ufid_flags);
1179
1180                 if (IS_ERR(reply)) {
1181                         error = PTR_ERR(reply);
1182                         goto err_unlock_ovs;
1183                 }
1184         }
1185
1186         /* Clear stats. */
1187         if (a[OVS_FLOW_ATTR_CLEAR])
1188                 ovs_flow_stats_clear(flow);
1189         ovs_unlock();
1190
1191         if (reply)
1192                 ovs_notify(&dp_flow_genl_family, reply, info);
1193         if (old_acts)
1194                 ovs_nla_free_flow_actions_rcu(old_acts);
1195
1196         return 0;
1197
1198 err_unlock_ovs:
1199         ovs_unlock();
1200         kfree_skb(reply);
1201 err_kfree_acts:
1202         ovs_nla_free_flow_actions(acts);
1203 error:
1204         return error;
1205 }
1206
1207 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1208 {
1209         struct nlattr **a = info->attrs;
1210         struct ovs_header *ovs_header = info->userhdr;
1211         struct net *net = sock_net(skb->sk);
1212         struct sw_flow_key key;
1213         struct sk_buff *reply;
1214         struct sw_flow *flow;
1215         struct datapath *dp;
1216         struct sw_flow_match match;
1217         struct sw_flow_id ufid;
1218         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1219         int err = 0;
1220         bool log = !a[OVS_FLOW_ATTR_PROBE];
1221         bool ufid_present;
1222
1223         ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1224         if (a[OVS_FLOW_ATTR_KEY]) {
1225                 ovs_match_init(&match, &key, NULL);
1226                 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1227                                         log);
1228         } else if (!ufid_present) {
1229                 OVS_NLERR(log,
1230                           "Flow get message rejected, Key attribute missing.");
1231                 err = -EINVAL;
1232         }
1233         if (err)
1234                 return err;
1235
1236         ovs_lock();
1237         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1238         if (!dp) {
1239                 err = -ENODEV;
1240                 goto unlock;
1241         }
1242
1243         if (ufid_present)
1244                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1245         else
1246                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1247         if (!flow) {
1248                 err = -ENOENT;
1249                 goto unlock;
1250         }
1251
1252         reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1253                                         OVS_FLOW_CMD_NEW, true, ufid_flags);
1254         if (IS_ERR(reply)) {
1255                 err = PTR_ERR(reply);
1256                 goto unlock;
1257         }
1258
1259         ovs_unlock();
1260         return genlmsg_reply(reply, info);
1261 unlock:
1262         ovs_unlock();
1263         return err;
1264 }
1265
1266 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1267 {
1268         struct nlattr **a = info->attrs;
1269         struct ovs_header *ovs_header = info->userhdr;
1270         struct net *net = sock_net(skb->sk);
1271         struct sw_flow_key key;
1272         struct sk_buff *reply;
1273         struct sw_flow *flow = NULL;
1274         struct datapath *dp;
1275         struct sw_flow_match match;
1276         struct sw_flow_id ufid;
1277         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1278         int err;
1279         bool log = !a[OVS_FLOW_ATTR_PROBE];
1280         bool ufid_present;
1281
1282         ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1283         if (a[OVS_FLOW_ATTR_KEY]) {
1284                 ovs_match_init(&match, &key, NULL);
1285                 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1286                                         NULL, log);
1287                 if (unlikely(err))
1288                         return err;
1289         }
1290
1291         ovs_lock();
1292         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1293         if (unlikely(!dp)) {
1294                 err = -ENODEV;
1295                 goto unlock;
1296         }
1297
1298         if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1299                 err = ovs_flow_tbl_flush(&dp->table);
1300                 goto unlock;
1301         }
1302
1303         if (ufid_present)
1304                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1305         else
1306                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1307         if (unlikely(!flow)) {
1308                 err = -ENOENT;
1309                 goto unlock;
1310         }
1311
1312         ovs_flow_tbl_remove(&dp->table, flow);
1313         ovs_unlock();
1314
1315         reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1316                                         &flow->id, info, false, ufid_flags);
1317         if (likely(reply)) {
1318                 if (likely(!IS_ERR(reply))) {
1319                         rcu_read_lock();        /*To keep RCU checker happy. */
1320                         err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1321                                                      reply, info->snd_portid,
1322                                                      info->snd_seq, 0,
1323                                                      OVS_FLOW_CMD_DEL,
1324                                                      ufid_flags);
1325                         rcu_read_unlock();
1326                         BUG_ON(err < 0);
1327
1328                         ovs_notify(&dp_flow_genl_family, reply, info);
1329                 } else {
1330                         netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
1331                 }
1332         }
1333
1334         ovs_flow_free(flow, true);
1335         return 0;
1336 unlock:
1337         ovs_unlock();
1338         return err;
1339 }
1340
1341 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1342 {
1343         struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1344         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1345         struct table_instance *ti;
1346         struct datapath *dp;
1347         u32 ufid_flags;
1348         int err;
1349
1350         err = genlmsg_parse(cb->nlh, &dp_flow_genl_family, a,
1351                             OVS_FLOW_ATTR_MAX, flow_policy);
1352         if (err)
1353                 return err;
1354         ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1355
1356         rcu_read_lock();
1357         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1358         if (!dp) {
1359                 rcu_read_unlock();
1360                 return -ENODEV;
1361         }
1362
1363         ti = rcu_dereference(dp->table.ti);
1364         for (;;) {
1365                 struct sw_flow *flow;
1366                 u32 bucket, obj;
1367
1368                 bucket = cb->args[0];
1369                 obj = cb->args[1];
1370                 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1371                 if (!flow)
1372                         break;
1373
1374                 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1375                                            NETLINK_CB(cb->skb).portid,
1376                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1377                                            OVS_FLOW_CMD_NEW, ufid_flags) < 0)
1378                         break;
1379
1380                 cb->args[0] = bucket;
1381                 cb->args[1] = obj;
1382         }
1383         rcu_read_unlock();
1384         return skb->len;
1385 }
1386
1387 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1388         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1389         [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1390         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1391         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1392         [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1393         [OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1394         [OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1395 };
1396
1397 static const struct genl_ops dp_flow_genl_ops[] = {
1398         { .cmd = OVS_FLOW_CMD_NEW,
1399           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1400           .policy = flow_policy,
1401           .doit = ovs_flow_cmd_new
1402         },
1403         { .cmd = OVS_FLOW_CMD_DEL,
1404           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1405           .policy = flow_policy,
1406           .doit = ovs_flow_cmd_del
1407         },
1408         { .cmd = OVS_FLOW_CMD_GET,
1409           .flags = 0,               /* OK for unprivileged users. */
1410           .policy = flow_policy,
1411           .doit = ovs_flow_cmd_get,
1412           .dumpit = ovs_flow_cmd_dump
1413         },
1414         { .cmd = OVS_FLOW_CMD_SET,
1415           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1416           .policy = flow_policy,
1417           .doit = ovs_flow_cmd_set,
1418         },
1419 };
1420
1421 static struct genl_family dp_flow_genl_family = {
1422         .id = GENL_ID_GENERATE,
1423         .hdrsize = sizeof(struct ovs_header),
1424         .name = OVS_FLOW_FAMILY,
1425         .version = OVS_FLOW_VERSION,
1426         .maxattr = OVS_FLOW_ATTR_MAX,
1427         .netnsok = true,
1428         .parallel_ops = true,
1429         .ops = dp_flow_genl_ops,
1430         .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1431         .mcgrps = &ovs_dp_flow_multicast_group,
1432         .n_mcgrps = 1,
1433 };
1434
1435 static size_t ovs_dp_cmd_msg_size(void)
1436 {
1437         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1438
1439         msgsize += nla_total_size(IFNAMSIZ);
1440         msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1441         msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
1442         msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1443
1444         return msgsize;
1445 }
1446
1447 /* Called with ovs_mutex. */
1448 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1449                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1450 {
1451         struct ovs_header *ovs_header;
1452         struct ovs_dp_stats dp_stats;
1453         struct ovs_dp_megaflow_stats dp_megaflow_stats;
1454         int err;
1455
1456         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1457                                    flags, cmd);
1458         if (!ovs_header)
1459                 goto error;
1460
1461         ovs_header->dp_ifindex = get_dpifindex(dp);
1462
1463         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1464         if (err)
1465                 goto nla_put_failure;
1466
1467         get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1468         if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1469                           &dp_stats, OVS_DP_ATTR_PAD))
1470                 goto nla_put_failure;
1471
1472         if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1473                           sizeof(struct ovs_dp_megaflow_stats),
1474                           &dp_megaflow_stats, OVS_DP_ATTR_PAD))
1475                 goto nla_put_failure;
1476
1477         if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1478                 goto nla_put_failure;
1479
1480         genlmsg_end(skb, ovs_header);
1481         return 0;
1482
1483 nla_put_failure:
1484         genlmsg_cancel(skb, ovs_header);
1485 error:
1486         return -EMSGSIZE;
1487 }
1488
1489 static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1490 {
1491         return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1492 }
1493
1494 /* Called with rcu_read_lock or ovs_mutex. */
1495 static struct datapath *lookup_datapath(struct net *net,
1496                                         const struct ovs_header *ovs_header,
1497                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1498 {
1499         struct datapath *dp;
1500
1501         if (!a[OVS_DP_ATTR_NAME])
1502                 dp = get_dp(net, ovs_header->dp_ifindex);
1503         else {
1504                 struct vport *vport;
1505
1506                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1507                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1508         }
1509         return dp ? dp : ERR_PTR(-ENODEV);
1510 }
1511
1512 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1513 {
1514         struct datapath *dp;
1515
1516         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1517         if (IS_ERR(dp))
1518                 return;
1519
1520         WARN(dp->user_features, "Dropping previously announced user features\n");
1521         dp->user_features = 0;
1522 }
1523
1524 static void ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1525 {
1526         if (a[OVS_DP_ATTR_USER_FEATURES])
1527                 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1528 }
1529
1530 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1531 {
1532         struct nlattr **a = info->attrs;
1533         struct vport_parms parms;
1534         struct sk_buff *reply;
1535         struct datapath *dp;
1536         struct vport *vport;
1537         struct ovs_net *ovs_net;
1538         int err, i;
1539
1540         err = -EINVAL;
1541         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1542                 goto err;
1543
1544         reply = ovs_dp_cmd_alloc_info();
1545         if (!reply)
1546                 return -ENOMEM;
1547
1548         err = -ENOMEM;
1549         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1550         if (dp == NULL)
1551                 goto err_free_reply;
1552
1553         ovs_dp_set_net(dp, sock_net(skb->sk));
1554
1555         /* Allocate table. */
1556         err = ovs_flow_tbl_init(&dp->table);
1557         if (err)
1558                 goto err_free_dp;
1559
1560         dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1561         if (!dp->stats_percpu) {
1562                 err = -ENOMEM;
1563                 goto err_destroy_table;
1564         }
1565
1566         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1567                             GFP_KERNEL);
1568         if (!dp->ports) {
1569                 err = -ENOMEM;
1570                 goto err_destroy_percpu;
1571         }
1572
1573         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1574                 INIT_HLIST_HEAD(&dp->ports[i]);
1575
1576         /* Set up our datapath device. */
1577         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1578         parms.type = OVS_VPORT_TYPE_INTERNAL;
1579         parms.options = NULL;
1580         parms.dp = dp;
1581         parms.port_no = OVSP_LOCAL;
1582         parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1583
1584         ovs_dp_change(dp, a);
1585
1586         /* So far only local changes have been made, now need the lock. */
1587         ovs_lock();
1588
1589         vport = new_vport(&parms);
1590         if (IS_ERR(vport)) {
1591                 err = PTR_ERR(vport);
1592                 if (err == -EBUSY)
1593                         err = -EEXIST;
1594
1595                 if (err == -EEXIST) {
1596                         /* An outdated user space instance that does not understand
1597                          * the concept of user_features has attempted to create a new
1598                          * datapath and is likely to reuse it. Drop all user features.
1599                          */
1600                         if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1601                                 ovs_dp_reset_user_features(skb, info);
1602                 }
1603
1604                 goto err_destroy_ports_array;
1605         }
1606
1607         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1608                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1609         BUG_ON(err < 0);
1610
1611         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1612         list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1613
1614         ovs_unlock();
1615
1616         ovs_notify(&dp_datapath_genl_family, reply, info);
1617         return 0;
1618
1619 err_destroy_ports_array:
1620         ovs_unlock();
1621         kfree(dp->ports);
1622 err_destroy_percpu:
1623         free_percpu(dp->stats_percpu);
1624 err_destroy_table:
1625         ovs_flow_tbl_destroy(&dp->table);
1626 err_free_dp:
1627         kfree(dp);
1628 err_free_reply:
1629         kfree_skb(reply);
1630 err:
1631         return err;
1632 }
1633
1634 /* Called with ovs_mutex. */
1635 static void __dp_destroy(struct datapath *dp)
1636 {
1637         int i;
1638
1639         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1640                 struct vport *vport;
1641                 struct hlist_node *n;
1642
1643                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1644                         if (vport->port_no != OVSP_LOCAL)
1645                                 ovs_dp_detach_port(vport);
1646         }
1647
1648         list_del_rcu(&dp->list_node);
1649
1650         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1651          * all ports in datapath are destroyed first before freeing datapath.
1652          */
1653         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1654
1655         /* RCU destroy the flow table */
1656         call_rcu(&dp->rcu, destroy_dp_rcu);
1657 }
1658
1659 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1660 {
1661         struct sk_buff *reply;
1662         struct datapath *dp;
1663         int err;
1664
1665         reply = ovs_dp_cmd_alloc_info();
1666         if (!reply)
1667                 return -ENOMEM;
1668
1669         ovs_lock();
1670         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1671         err = PTR_ERR(dp);
1672         if (IS_ERR(dp))
1673                 goto err_unlock_free;
1674
1675         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1676                                    info->snd_seq, 0, OVS_DP_CMD_DEL);
1677         BUG_ON(err < 0);
1678
1679         __dp_destroy(dp);
1680         ovs_unlock();
1681
1682         ovs_notify(&dp_datapath_genl_family, reply, info);
1683
1684         return 0;
1685
1686 err_unlock_free:
1687         ovs_unlock();
1688         kfree_skb(reply);
1689         return err;
1690 }
1691
1692 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1693 {
1694         struct sk_buff *reply;
1695         struct datapath *dp;
1696         int err;
1697
1698         reply = ovs_dp_cmd_alloc_info();
1699         if (!reply)
1700                 return -ENOMEM;
1701
1702         ovs_lock();
1703         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1704         err = PTR_ERR(dp);
1705         if (IS_ERR(dp))
1706                 goto err_unlock_free;
1707
1708         ovs_dp_change(dp, info->attrs);
1709
1710         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1711                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1712         BUG_ON(err < 0);
1713
1714         ovs_unlock();
1715         ovs_notify(&dp_datapath_genl_family, reply, info);
1716
1717         return 0;
1718
1719 err_unlock_free:
1720         ovs_unlock();
1721         kfree_skb(reply);
1722         return err;
1723 }
1724
1725 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1726 {
1727         struct sk_buff *reply;
1728         struct datapath *dp;
1729         int err;
1730
1731         reply = ovs_dp_cmd_alloc_info();
1732         if (!reply)
1733                 return -ENOMEM;
1734
1735         ovs_lock();
1736         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1737         if (IS_ERR(dp)) {
1738                 err = PTR_ERR(dp);
1739                 goto err_unlock_free;
1740         }
1741         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1742                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1743         BUG_ON(err < 0);
1744         ovs_unlock();
1745
1746         return genlmsg_reply(reply, info);
1747
1748 err_unlock_free:
1749         ovs_unlock();
1750         kfree_skb(reply);
1751         return err;
1752 }
1753
1754 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1755 {
1756         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1757         struct datapath *dp;
1758         int skip = cb->args[0];
1759         int i = 0;
1760
1761         ovs_lock();
1762         list_for_each_entry(dp, &ovs_net->dps, list_node) {
1763                 if (i >= skip &&
1764                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1765                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1766                                          OVS_DP_CMD_NEW) < 0)
1767                         break;
1768                 i++;
1769         }
1770         ovs_unlock();
1771
1772         cb->args[0] = i;
1773
1774         return skb->len;
1775 }
1776
1777 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1778         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1779         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1780         [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1781 };
1782
1783 static const struct genl_ops dp_datapath_genl_ops[] = {
1784         { .cmd = OVS_DP_CMD_NEW,
1785           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1786           .policy = datapath_policy,
1787           .doit = ovs_dp_cmd_new
1788         },
1789         { .cmd = OVS_DP_CMD_DEL,
1790           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1791           .policy = datapath_policy,
1792           .doit = ovs_dp_cmd_del
1793         },
1794         { .cmd = OVS_DP_CMD_GET,
1795           .flags = 0,               /* OK for unprivileged users. */
1796           .policy = datapath_policy,
1797           .doit = ovs_dp_cmd_get,
1798           .dumpit = ovs_dp_cmd_dump
1799         },
1800         { .cmd = OVS_DP_CMD_SET,
1801           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1802           .policy = datapath_policy,
1803           .doit = ovs_dp_cmd_set,
1804         },
1805 };
1806
1807 static struct genl_family dp_datapath_genl_family = {
1808         .id = GENL_ID_GENERATE,
1809         .hdrsize = sizeof(struct ovs_header),
1810         .name = OVS_DATAPATH_FAMILY,
1811         .version = OVS_DATAPATH_VERSION,
1812         .maxattr = OVS_DP_ATTR_MAX,
1813         .netnsok = true,
1814         .parallel_ops = true,
1815         .ops = dp_datapath_genl_ops,
1816         .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1817         .mcgrps = &ovs_dp_datapath_multicast_group,
1818         .n_mcgrps = 1,
1819 };
1820
1821 /* Called with ovs_mutex or RCU read lock. */
1822 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1823                                    u32 portid, u32 seq, u32 flags, u8 cmd)
1824 {
1825         struct ovs_header *ovs_header;
1826         struct ovs_vport_stats vport_stats;
1827         int err;
1828
1829         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1830                                  flags, cmd);
1831         if (!ovs_header)
1832                 return -EMSGSIZE;
1833
1834         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1835
1836         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1837             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1838             nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1839                            ovs_vport_name(vport)))
1840                 goto nla_put_failure;
1841
1842         ovs_vport_get_stats(vport, &vport_stats);
1843         if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
1844                           sizeof(struct ovs_vport_stats), &vport_stats,
1845                           OVS_VPORT_ATTR_PAD))
1846                 goto nla_put_failure;
1847
1848         if (ovs_vport_get_upcall_portids(vport, skb))
1849                 goto nla_put_failure;
1850
1851         err = ovs_vport_get_options(vport, skb);
1852         if (err == -EMSGSIZE)
1853                 goto error;
1854
1855         genlmsg_end(skb, ovs_header);
1856         return 0;
1857
1858 nla_put_failure:
1859         err = -EMSGSIZE;
1860 error:
1861         genlmsg_cancel(skb, ovs_header);
1862         return err;
1863 }
1864
1865 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1866 {
1867         return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1868 }
1869
1870 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1871 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1872                                          u32 seq, u8 cmd)
1873 {
1874         struct sk_buff *skb;
1875         int retval;
1876
1877         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1878         if (!skb)
1879                 return ERR_PTR(-ENOMEM);
1880
1881         retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1882         BUG_ON(retval < 0);
1883
1884         return skb;
1885 }
1886
1887 /* Called with ovs_mutex or RCU read lock. */
1888 static struct vport *lookup_vport(struct net *net,
1889                                   const struct ovs_header *ovs_header,
1890                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1891 {
1892         struct datapath *dp;
1893         struct vport *vport;
1894
1895         if (a[OVS_VPORT_ATTR_NAME]) {
1896                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1897                 if (!vport)
1898                         return ERR_PTR(-ENODEV);
1899                 if (ovs_header->dp_ifindex &&
1900                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1901                         return ERR_PTR(-ENODEV);
1902                 return vport;
1903         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1904                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1905
1906                 if (port_no >= DP_MAX_PORTS)
1907                         return ERR_PTR(-EFBIG);
1908
1909                 dp = get_dp(net, ovs_header->dp_ifindex);
1910                 if (!dp)
1911                         return ERR_PTR(-ENODEV);
1912
1913                 vport = ovs_vport_ovsl_rcu(dp, port_no);
1914                 if (!vport)
1915                         return ERR_PTR(-ENODEV);
1916                 return vport;
1917         } else
1918                 return ERR_PTR(-EINVAL);
1919 }
1920
1921 /* Called with ovs_mutex */
1922 static void update_headroom(struct datapath *dp)
1923 {
1924         unsigned dev_headroom, max_headroom = 0;
1925         struct net_device *dev;
1926         struct vport *vport;
1927         int i;
1928
1929         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1930                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1931                         dev = vport->dev;
1932                         dev_headroom = netdev_get_fwd_headroom(dev);
1933                         if (dev_headroom > max_headroom)
1934                                 max_headroom = dev_headroom;
1935                 }
1936         }
1937
1938         dp->max_headroom = max_headroom;
1939         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1940                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node)
1941                         netdev_set_rx_headroom(vport->dev, max_headroom);
1942 }
1943
1944 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1945 {
1946         struct nlattr **a = info->attrs;
1947         struct ovs_header *ovs_header = info->userhdr;
1948         struct vport_parms parms;
1949         struct sk_buff *reply;
1950         struct vport *vport;
1951         struct datapath *dp;
1952         u32 port_no;
1953         int err;
1954
1955         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1956             !a[OVS_VPORT_ATTR_UPCALL_PID])
1957                 return -EINVAL;
1958
1959         port_no = a[OVS_VPORT_ATTR_PORT_NO]
1960                 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
1961         if (port_no >= DP_MAX_PORTS)
1962                 return -EFBIG;
1963
1964         reply = ovs_vport_cmd_alloc_info();
1965         if (!reply)
1966                 return -ENOMEM;
1967
1968         ovs_lock();
1969 restart:
1970         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1971         err = -ENODEV;
1972         if (!dp)
1973                 goto exit_unlock_free;
1974
1975         if (port_no) {
1976                 vport = ovs_vport_ovsl(dp, port_no);
1977                 err = -EBUSY;
1978                 if (vport)
1979                         goto exit_unlock_free;
1980         } else {
1981                 for (port_no = 1; ; port_no++) {
1982                         if (port_no >= DP_MAX_PORTS) {
1983                                 err = -EFBIG;
1984                                 goto exit_unlock_free;
1985                         }
1986                         vport = ovs_vport_ovsl(dp, port_no);
1987                         if (!vport)
1988                                 break;
1989                 }
1990         }
1991
1992         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1993         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1994         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1995         parms.dp = dp;
1996         parms.port_no = port_no;
1997         parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
1998
1999         vport = new_vport(&parms);
2000         err = PTR_ERR(vport);
2001         if (IS_ERR(vport)) {
2002                 if (err == -EAGAIN)
2003                         goto restart;
2004                 goto exit_unlock_free;
2005         }
2006
2007         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2008                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2009
2010         if (netdev_get_fwd_headroom(vport->dev) > dp->max_headroom)
2011                 update_headroom(dp);
2012         else
2013                 netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2014
2015         BUG_ON(err < 0);
2016         ovs_unlock();
2017
2018         ovs_notify(&dp_vport_genl_family, reply, info);
2019         return 0;
2020
2021 exit_unlock_free:
2022         ovs_unlock();
2023         kfree_skb(reply);
2024         return err;
2025 }
2026
2027 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2028 {
2029         struct nlattr **a = info->attrs;
2030         struct sk_buff *reply;
2031         struct vport *vport;
2032         int err;
2033
2034         reply = ovs_vport_cmd_alloc_info();
2035         if (!reply)
2036                 return -ENOMEM;
2037
2038         ovs_lock();
2039         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2040         err = PTR_ERR(vport);
2041         if (IS_ERR(vport))
2042                 goto exit_unlock_free;
2043
2044         if (a[OVS_VPORT_ATTR_TYPE] &&
2045             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2046                 err = -EINVAL;
2047                 goto exit_unlock_free;
2048         }
2049
2050         if (a[OVS_VPORT_ATTR_OPTIONS]) {
2051                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2052                 if (err)
2053                         goto exit_unlock_free;
2054         }
2055
2056
2057         if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2058                 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2059
2060                 err = ovs_vport_set_upcall_portids(vport, ids);
2061                 if (err)
2062                         goto exit_unlock_free;
2063         }
2064
2065         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2066                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2067         BUG_ON(err < 0);
2068
2069         ovs_unlock();
2070         ovs_notify(&dp_vport_genl_family, reply, info);
2071         return 0;
2072
2073 exit_unlock_free:
2074         ovs_unlock();
2075         kfree_skb(reply);
2076         return err;
2077 }
2078
2079 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2080 {
2081         bool must_update_headroom = false;
2082         struct nlattr **a = info->attrs;
2083         struct sk_buff *reply;
2084         struct datapath *dp;
2085         struct vport *vport;
2086         int err;
2087
2088         reply = ovs_vport_cmd_alloc_info();
2089         if (!reply)
2090                 return -ENOMEM;
2091
2092         ovs_lock();
2093         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2094         err = PTR_ERR(vport);
2095         if (IS_ERR(vport))
2096                 goto exit_unlock_free;
2097
2098         if (vport->port_no == OVSP_LOCAL) {
2099                 err = -EINVAL;
2100                 goto exit_unlock_free;
2101         }
2102
2103         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2104                                       info->snd_seq, 0, OVS_VPORT_CMD_DEL);
2105         BUG_ON(err < 0);
2106
2107         /* the vport deletion may trigger dp headroom update */
2108         dp = vport->dp;
2109         if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2110                 must_update_headroom = true;
2111         netdev_reset_rx_headroom(vport->dev);
2112         ovs_dp_detach_port(vport);
2113
2114         if (must_update_headroom)
2115                 update_headroom(dp);
2116         ovs_unlock();
2117
2118         ovs_notify(&dp_vport_genl_family, reply, info);
2119         return 0;
2120
2121 exit_unlock_free:
2122         ovs_unlock();
2123         kfree_skb(reply);
2124         return err;
2125 }
2126
2127 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2128 {
2129         struct nlattr **a = info->attrs;
2130         struct ovs_header *ovs_header = info->userhdr;
2131         struct sk_buff *reply;
2132         struct vport *vport;
2133         int err;
2134
2135         reply = ovs_vport_cmd_alloc_info();
2136         if (!reply)
2137                 return -ENOMEM;
2138
2139         rcu_read_lock();
2140         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2141         err = PTR_ERR(vport);
2142         if (IS_ERR(vport))
2143                 goto exit_unlock_free;
2144         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2145                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2146         BUG_ON(err < 0);
2147         rcu_read_unlock();
2148
2149         return genlmsg_reply(reply, info);
2150
2151 exit_unlock_free:
2152         rcu_read_unlock();
2153         kfree_skb(reply);
2154         return err;
2155 }
2156
2157 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2158 {
2159         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2160         struct datapath *dp;
2161         int bucket = cb->args[0], skip = cb->args[1];
2162         int i, j = 0;
2163
2164         rcu_read_lock();
2165         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2166         if (!dp) {
2167                 rcu_read_unlock();
2168                 return -ENODEV;
2169         }
2170         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2171                 struct vport *vport;
2172
2173                 j = 0;
2174                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2175                         if (j >= skip &&
2176                             ovs_vport_cmd_fill_info(vport, skb,
2177                                                     NETLINK_CB(cb->skb).portid,
2178                                                     cb->nlh->nlmsg_seq,
2179                                                     NLM_F_MULTI,
2180                                                     OVS_VPORT_CMD_NEW) < 0)
2181                                 goto out;
2182
2183                         j++;
2184                 }
2185                 skip = 0;
2186         }
2187 out:
2188         rcu_read_unlock();
2189
2190         cb->args[0] = i;
2191         cb->args[1] = j;
2192
2193         return skb->len;
2194 }
2195
2196 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2197         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2198         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2199         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2200         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2201         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
2202         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2203 };
2204
2205 static const struct genl_ops dp_vport_genl_ops[] = {
2206         { .cmd = OVS_VPORT_CMD_NEW,
2207           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2208           .policy = vport_policy,
2209           .doit = ovs_vport_cmd_new
2210         },
2211         { .cmd = OVS_VPORT_CMD_DEL,
2212           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2213           .policy = vport_policy,
2214           .doit = ovs_vport_cmd_del
2215         },
2216         { .cmd = OVS_VPORT_CMD_GET,
2217           .flags = 0,               /* OK for unprivileged users. */
2218           .policy = vport_policy,
2219           .doit = ovs_vport_cmd_get,
2220           .dumpit = ovs_vport_cmd_dump
2221         },
2222         { .cmd = OVS_VPORT_CMD_SET,
2223           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2224           .policy = vport_policy,
2225           .doit = ovs_vport_cmd_set,
2226         },
2227 };
2228
2229 struct genl_family dp_vport_genl_family = {
2230         .id = GENL_ID_GENERATE,
2231         .hdrsize = sizeof(struct ovs_header),
2232         .name = OVS_VPORT_FAMILY,
2233         .version = OVS_VPORT_VERSION,
2234         .maxattr = OVS_VPORT_ATTR_MAX,
2235         .netnsok = true,
2236         .parallel_ops = true,
2237         .ops = dp_vport_genl_ops,
2238         .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2239         .mcgrps = &ovs_dp_vport_multicast_group,
2240         .n_mcgrps = 1,
2241 };
2242
2243 static struct genl_family * const dp_genl_families[] = {
2244         &dp_datapath_genl_family,
2245         &dp_vport_genl_family,
2246         &dp_flow_genl_family,
2247         &dp_packet_genl_family,
2248 };
2249
2250 static void dp_unregister_genl(int n_families)
2251 {
2252         int i;
2253
2254         for (i = 0; i < n_families; i++)
2255                 genl_unregister_family(dp_genl_families[i]);
2256 }
2257
2258 static int dp_register_genl(void)
2259 {
2260         int err;
2261         int i;
2262
2263         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2264
2265                 err = genl_register_family(dp_genl_families[i]);
2266                 if (err)
2267                         goto error;
2268         }
2269
2270         return 0;
2271
2272 error:
2273         dp_unregister_genl(i);
2274         return err;
2275 }
2276
2277 static int __net_init ovs_init_net(struct net *net)
2278 {
2279         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2280
2281         INIT_LIST_HEAD(&ovs_net->dps);
2282         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2283         ovs_ct_init(net);
2284         return 0;
2285 }
2286
2287 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2288                                             struct list_head *head)
2289 {
2290         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2291         struct datapath *dp;
2292
2293         list_for_each_entry(dp, &ovs_net->dps, list_node) {
2294                 int i;
2295
2296                 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2297                         struct vport *vport;
2298
2299                         hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2300                                 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2301                                         continue;
2302
2303                                 if (dev_net(vport->dev) == dnet)
2304                                         list_add(&vport->detach_list, head);
2305                         }
2306                 }
2307         }
2308 }
2309
2310 static void __net_exit ovs_exit_net(struct net *dnet)
2311 {
2312         struct datapath *dp, *dp_next;
2313         struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2314         struct vport *vport, *vport_next;
2315         struct net *net;
2316         LIST_HEAD(head);
2317
2318         ovs_ct_exit(dnet);
2319         ovs_lock();
2320         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2321                 __dp_destroy(dp);
2322
2323         rtnl_lock();
2324         for_each_net(net)
2325                 list_vports_from_net(net, dnet, &head);
2326         rtnl_unlock();
2327
2328         /* Detach all vports from given namespace. */
2329         list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2330                 list_del(&vport->detach_list);
2331                 ovs_dp_detach_port(vport);
2332         }
2333
2334         ovs_unlock();
2335
2336         cancel_work_sync(&ovs_net->dp_notify_work);
2337 }
2338
2339 static struct pernet_operations ovs_net_ops = {
2340         .init = ovs_init_net,
2341         .exit = ovs_exit_net,
2342         .id   = &ovs_net_id,
2343         .size = sizeof(struct ovs_net),
2344 };
2345
2346 static int __init dp_init(void)
2347 {
2348         int err;
2349
2350         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2351
2352         pr_info("Open vSwitch switching datapath\n");
2353
2354         err = action_fifos_init();
2355         if (err)
2356                 goto error;
2357
2358         err = ovs_internal_dev_rtnl_link_register();
2359         if (err)
2360                 goto error_action_fifos_exit;
2361
2362         err = ovs_flow_init();
2363         if (err)
2364                 goto error_unreg_rtnl_link;
2365
2366         err = ovs_vport_init();
2367         if (err)
2368                 goto error_flow_exit;
2369
2370         err = register_pernet_device(&ovs_net_ops);
2371         if (err)
2372                 goto error_vport_exit;
2373
2374         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2375         if (err)
2376                 goto error_netns_exit;
2377
2378         err = ovs_netdev_init();
2379         if (err)
2380                 goto error_unreg_notifier;
2381
2382         err = dp_register_genl();
2383         if (err < 0)
2384                 goto error_unreg_netdev;
2385
2386         return 0;
2387
2388 error_unreg_netdev:
2389         ovs_netdev_exit();
2390 error_unreg_notifier:
2391         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2392 error_netns_exit:
2393         unregister_pernet_device(&ovs_net_ops);
2394 error_vport_exit:
2395         ovs_vport_exit();
2396 error_flow_exit:
2397         ovs_flow_exit();
2398 error_unreg_rtnl_link:
2399         ovs_internal_dev_rtnl_link_unregister();
2400 error_action_fifos_exit:
2401         action_fifos_exit();
2402 error:
2403         return err;
2404 }
2405
2406 static void dp_cleanup(void)
2407 {
2408         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2409         ovs_netdev_exit();
2410         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2411         unregister_pernet_device(&ovs_net_ops);
2412         rcu_barrier();
2413         ovs_vport_exit();
2414         ovs_flow_exit();
2415         ovs_internal_dev_rtnl_link_unregister();
2416         action_fifos_exit();
2417 }
2418
2419 module_init(dp_init);
2420 module_exit(dp_cleanup);
2421
2422 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2423 MODULE_LICENSE("GPL");