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 = 0;
1100         bool log = !a[OVS_FLOW_ATTR_PROBE];
1101         bool ufid_present;
1102
1103         ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1104         if (a[OVS_FLOW_ATTR_KEY]) {
1105                 ovs_match_init(&match, &key, &mask);
1106                 error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1107                                           a[OVS_FLOW_ATTR_MASK], log);
1108         } else if (!ufid_present) {
1109                 OVS_NLERR(log,
1110                           "Flow set message rejected, Key attribute missing.");
1111                 error = -EINVAL;
1112         }
1113         if (error)
1114                 goto error;
1115
1116         /* Validate actions. */
1117         if (a[OVS_FLOW_ATTR_ACTIONS]) {
1118                 if (!a[OVS_FLOW_ATTR_KEY]) {
1119                         OVS_NLERR(log,
1120                                   "Flow key attribute not present in set flow.");
1121                         error = -EINVAL;
1122                         goto error;
1123                 }
1124
1125                 acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], &key,
1126                                         &mask, log);
1127                 if (IS_ERR(acts)) {
1128                         error = PTR_ERR(acts);
1129                         goto error;
1130                 }
1131
1132                 /* Can allocate before locking if have acts. */
1133                 reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1134                                                 ufid_flags);
1135                 if (IS_ERR(reply)) {
1136                         error = PTR_ERR(reply);
1137                         goto err_kfree_acts;
1138                 }
1139         }
1140
1141         ovs_lock();
1142         dp = get_dp(net, ovs_header->dp_ifindex);
1143         if (unlikely(!dp)) {
1144                 error = -ENODEV;
1145                 goto err_unlock_ovs;
1146         }
1147         /* Check that the flow exists. */
1148         if (ufid_present)
1149                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1150         else
1151                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1152         if (unlikely(!flow)) {
1153                 error = -ENOENT;
1154                 goto err_unlock_ovs;
1155         }
1156
1157         /* Update actions, if present. */
1158         if (likely(acts)) {
1159                 old_acts = ovsl_dereference(flow->sf_acts);
1160                 rcu_assign_pointer(flow->sf_acts, acts);
1161
1162                 if (unlikely(reply)) {
1163                         error = ovs_flow_cmd_fill_info(flow,
1164                                                        ovs_header->dp_ifindex,
1165                                                        reply, info->snd_portid,
1166                                                        info->snd_seq, 0,
1167                                                        OVS_FLOW_CMD_NEW,
1168                                                        ufid_flags);
1169                         BUG_ON(error < 0);
1170                 }
1171         } else {
1172                 /* Could not alloc without acts before locking. */
1173                 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1174                                                 info, OVS_FLOW_CMD_NEW, false,
1175                                                 ufid_flags);
1176
1177                 if (IS_ERR(reply)) {
1178                         error = PTR_ERR(reply);
1179                         goto err_unlock_ovs;
1180                 }
1181         }
1182
1183         /* Clear stats. */
1184         if (a[OVS_FLOW_ATTR_CLEAR])
1185                 ovs_flow_stats_clear(flow);
1186         ovs_unlock();
1187
1188         if (reply)
1189                 ovs_notify(&dp_flow_genl_family, reply, info);
1190         if (old_acts)
1191                 ovs_nla_free_flow_actions_rcu(old_acts);
1192
1193         return 0;
1194
1195 err_unlock_ovs:
1196         ovs_unlock();
1197         kfree_skb(reply);
1198 err_kfree_acts:
1199         ovs_nla_free_flow_actions(acts);
1200 error:
1201         return error;
1202 }
1203
1204 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1205 {
1206         struct nlattr **a = info->attrs;
1207         struct ovs_header *ovs_header = info->userhdr;
1208         struct net *net = sock_net(skb->sk);
1209         struct sw_flow_key key;
1210         struct sk_buff *reply;
1211         struct sw_flow *flow;
1212         struct datapath *dp;
1213         struct sw_flow_match match;
1214         struct sw_flow_id ufid;
1215         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1216         int err = 0;
1217         bool log = !a[OVS_FLOW_ATTR_PROBE];
1218         bool ufid_present;
1219
1220         ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1221         if (a[OVS_FLOW_ATTR_KEY]) {
1222                 ovs_match_init(&match, &key, NULL);
1223                 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1224                                         log);
1225         } else if (!ufid_present) {
1226                 OVS_NLERR(log,
1227                           "Flow get message rejected, Key attribute missing.");
1228                 err = -EINVAL;
1229         }
1230         if (err)
1231                 return err;
1232
1233         ovs_lock();
1234         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1235         if (!dp) {
1236                 err = -ENODEV;
1237                 goto unlock;
1238         }
1239
1240         if (ufid_present)
1241                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1242         else
1243                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1244         if (!flow) {
1245                 err = -ENOENT;
1246                 goto unlock;
1247         }
1248
1249         reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1250                                         OVS_FLOW_CMD_NEW, true, ufid_flags);
1251         if (IS_ERR(reply)) {
1252                 err = PTR_ERR(reply);
1253                 goto unlock;
1254         }
1255
1256         ovs_unlock();
1257         return genlmsg_reply(reply, info);
1258 unlock:
1259         ovs_unlock();
1260         return err;
1261 }
1262
1263 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1264 {
1265         struct nlattr **a = info->attrs;
1266         struct ovs_header *ovs_header = info->userhdr;
1267         struct net *net = sock_net(skb->sk);
1268         struct sw_flow_key key;
1269         struct sk_buff *reply;
1270         struct sw_flow *flow = NULL;
1271         struct datapath *dp;
1272         struct sw_flow_match match;
1273         struct sw_flow_id ufid;
1274         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1275         int err;
1276         bool log = !a[OVS_FLOW_ATTR_PROBE];
1277         bool ufid_present;
1278
1279         ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1280         if (a[OVS_FLOW_ATTR_KEY]) {
1281                 ovs_match_init(&match, &key, NULL);
1282                 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1283                                         NULL, log);
1284                 if (unlikely(err))
1285                         return err;
1286         }
1287
1288         ovs_lock();
1289         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1290         if (unlikely(!dp)) {
1291                 err = -ENODEV;
1292                 goto unlock;
1293         }
1294
1295         if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1296                 err = ovs_flow_tbl_flush(&dp->table);
1297                 goto unlock;
1298         }
1299
1300         if (ufid_present)
1301                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1302         else
1303                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1304         if (unlikely(!flow)) {
1305                 err = -ENOENT;
1306                 goto unlock;
1307         }
1308
1309         ovs_flow_tbl_remove(&dp->table, flow);
1310         ovs_unlock();
1311
1312         reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1313                                         &flow->id, info, false, ufid_flags);
1314         if (likely(reply)) {
1315                 if (likely(!IS_ERR(reply))) {
1316                         rcu_read_lock();        /*To keep RCU checker happy. */
1317                         err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1318                                                      reply, info->snd_portid,
1319                                                      info->snd_seq, 0,
1320                                                      OVS_FLOW_CMD_DEL,
1321                                                      ufid_flags);
1322                         rcu_read_unlock();
1323                         BUG_ON(err < 0);
1324
1325                         ovs_notify(&dp_flow_genl_family, reply, info);
1326                 } else {
1327                         netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
1328                 }
1329         }
1330
1331         ovs_flow_free(flow, true);
1332         return 0;
1333 unlock:
1334         ovs_unlock();
1335         return err;
1336 }
1337
1338 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1339 {
1340         struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1341         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1342         struct table_instance *ti;
1343         struct datapath *dp;
1344         u32 ufid_flags;
1345         int err;
1346
1347         err = genlmsg_parse(cb->nlh, &dp_flow_genl_family, a,
1348                             OVS_FLOW_ATTR_MAX, flow_policy);
1349         if (err)
1350                 return err;
1351         ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1352
1353         rcu_read_lock();
1354         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1355         if (!dp) {
1356                 rcu_read_unlock();
1357                 return -ENODEV;
1358         }
1359
1360         ti = rcu_dereference(dp->table.ti);
1361         for (;;) {
1362                 struct sw_flow *flow;
1363                 u32 bucket, obj;
1364
1365                 bucket = cb->args[0];
1366                 obj = cb->args[1];
1367                 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1368                 if (!flow)
1369                         break;
1370
1371                 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1372                                            NETLINK_CB(cb->skb).portid,
1373                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1374                                            OVS_FLOW_CMD_NEW, ufid_flags) < 0)
1375                         break;
1376
1377                 cb->args[0] = bucket;
1378                 cb->args[1] = obj;
1379         }
1380         rcu_read_unlock();
1381         return skb->len;
1382 }
1383
1384 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1385         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1386         [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1387         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1388         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1389         [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1390         [OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1391         [OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1392 };
1393
1394 static const struct genl_ops dp_flow_genl_ops[] = {
1395         { .cmd = OVS_FLOW_CMD_NEW,
1396           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1397           .policy = flow_policy,
1398           .doit = ovs_flow_cmd_new
1399         },
1400         { .cmd = OVS_FLOW_CMD_DEL,
1401           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1402           .policy = flow_policy,
1403           .doit = ovs_flow_cmd_del
1404         },
1405         { .cmd = OVS_FLOW_CMD_GET,
1406           .flags = 0,               /* OK for unprivileged users. */
1407           .policy = flow_policy,
1408           .doit = ovs_flow_cmd_get,
1409           .dumpit = ovs_flow_cmd_dump
1410         },
1411         { .cmd = OVS_FLOW_CMD_SET,
1412           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1413           .policy = flow_policy,
1414           .doit = ovs_flow_cmd_set,
1415         },
1416 };
1417
1418 static struct genl_family dp_flow_genl_family = {
1419         .id = GENL_ID_GENERATE,
1420         .hdrsize = sizeof(struct ovs_header),
1421         .name = OVS_FLOW_FAMILY,
1422         .version = OVS_FLOW_VERSION,
1423         .maxattr = OVS_FLOW_ATTR_MAX,
1424         .netnsok = true,
1425         .parallel_ops = true,
1426         .ops = dp_flow_genl_ops,
1427         .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1428         .mcgrps = &ovs_dp_flow_multicast_group,
1429         .n_mcgrps = 1,
1430 };
1431
1432 static size_t ovs_dp_cmd_msg_size(void)
1433 {
1434         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1435
1436         msgsize += nla_total_size(IFNAMSIZ);
1437         msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1438         msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1439         msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1440
1441         return msgsize;
1442 }
1443
1444 /* Called with ovs_mutex. */
1445 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1446                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1447 {
1448         struct ovs_header *ovs_header;
1449         struct ovs_dp_stats dp_stats;
1450         struct ovs_dp_megaflow_stats dp_megaflow_stats;
1451         int err;
1452
1453         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1454                                    flags, cmd);
1455         if (!ovs_header)
1456                 goto error;
1457
1458         ovs_header->dp_ifindex = get_dpifindex(dp);
1459
1460         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1461         if (err)
1462                 goto nla_put_failure;
1463
1464         get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1465         if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1466                         &dp_stats))
1467                 goto nla_put_failure;
1468
1469         if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1470                         sizeof(struct ovs_dp_megaflow_stats),
1471                         &dp_megaflow_stats))
1472                 goto nla_put_failure;
1473
1474         if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1475                 goto nla_put_failure;
1476
1477         genlmsg_end(skb, ovs_header);
1478         return 0;
1479
1480 nla_put_failure:
1481         genlmsg_cancel(skb, ovs_header);
1482 error:
1483         return -EMSGSIZE;
1484 }
1485
1486 static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1487 {
1488         return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1489 }
1490
1491 /* Called with rcu_read_lock or ovs_mutex. */
1492 static struct datapath *lookup_datapath(struct net *net,
1493                                         const struct ovs_header *ovs_header,
1494                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1495 {
1496         struct datapath *dp;
1497
1498         if (!a[OVS_DP_ATTR_NAME])
1499                 dp = get_dp(net, ovs_header->dp_ifindex);
1500         else {
1501                 struct vport *vport;
1502
1503                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1504                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1505         }
1506         return dp ? dp : ERR_PTR(-ENODEV);
1507 }
1508
1509 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1510 {
1511         struct datapath *dp;
1512
1513         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1514         if (IS_ERR(dp))
1515                 return;
1516
1517         WARN(dp->user_features, "Dropping previously announced user features\n");
1518         dp->user_features = 0;
1519 }
1520
1521 static void ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1522 {
1523         if (a[OVS_DP_ATTR_USER_FEATURES])
1524                 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1525 }
1526
1527 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1528 {
1529         struct nlattr **a = info->attrs;
1530         struct vport_parms parms;
1531         struct sk_buff *reply;
1532         struct datapath *dp;
1533         struct vport *vport;
1534         struct ovs_net *ovs_net;
1535         int err, i;
1536
1537         err = -EINVAL;
1538         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1539                 goto err;
1540
1541         reply = ovs_dp_cmd_alloc_info();
1542         if (!reply)
1543                 return -ENOMEM;
1544
1545         err = -ENOMEM;
1546         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1547         if (dp == NULL)
1548                 goto err_free_reply;
1549
1550         ovs_dp_set_net(dp, sock_net(skb->sk));
1551
1552         /* Allocate table. */
1553         err = ovs_flow_tbl_init(&dp->table);
1554         if (err)
1555                 goto err_free_dp;
1556
1557         dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1558         if (!dp->stats_percpu) {
1559                 err = -ENOMEM;
1560                 goto err_destroy_table;
1561         }
1562
1563         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1564                             GFP_KERNEL);
1565         if (!dp->ports) {
1566                 err = -ENOMEM;
1567                 goto err_destroy_percpu;
1568         }
1569
1570         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1571                 INIT_HLIST_HEAD(&dp->ports[i]);
1572
1573         /* Set up our datapath device. */
1574         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1575         parms.type = OVS_VPORT_TYPE_INTERNAL;
1576         parms.options = NULL;
1577         parms.dp = dp;
1578         parms.port_no = OVSP_LOCAL;
1579         parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1580
1581         ovs_dp_change(dp, a);
1582
1583         /* So far only local changes have been made, now need the lock. */
1584         ovs_lock();
1585
1586         vport = new_vport(&parms);
1587         if (IS_ERR(vport)) {
1588                 err = PTR_ERR(vport);
1589                 if (err == -EBUSY)
1590                         err = -EEXIST;
1591
1592                 if (err == -EEXIST) {
1593                         /* An outdated user space instance that does not understand
1594                          * the concept of user_features has attempted to create a new
1595                          * datapath and is likely to reuse it. Drop all user features.
1596                          */
1597                         if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1598                                 ovs_dp_reset_user_features(skb, info);
1599                 }
1600
1601                 goto err_destroy_ports_array;
1602         }
1603
1604         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1605                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1606         BUG_ON(err < 0);
1607
1608         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1609         list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1610
1611         ovs_unlock();
1612
1613         ovs_notify(&dp_datapath_genl_family, reply, info);
1614         return 0;
1615
1616 err_destroy_ports_array:
1617         ovs_unlock();
1618         kfree(dp->ports);
1619 err_destroy_percpu:
1620         free_percpu(dp->stats_percpu);
1621 err_destroy_table:
1622         ovs_flow_tbl_destroy(&dp->table);
1623 err_free_dp:
1624         kfree(dp);
1625 err_free_reply:
1626         kfree_skb(reply);
1627 err:
1628         return err;
1629 }
1630
1631 /* Called with ovs_mutex. */
1632 static void __dp_destroy(struct datapath *dp)
1633 {
1634         int i;
1635
1636         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1637                 struct vport *vport;
1638                 struct hlist_node *n;
1639
1640                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1641                         if (vport->port_no != OVSP_LOCAL)
1642                                 ovs_dp_detach_port(vport);
1643         }
1644
1645         list_del_rcu(&dp->list_node);
1646
1647         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1648          * all ports in datapath are destroyed first before freeing datapath.
1649          */
1650         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1651
1652         /* RCU destroy the flow table */
1653         call_rcu(&dp->rcu, destroy_dp_rcu);
1654 }
1655
1656 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1657 {
1658         struct sk_buff *reply;
1659         struct datapath *dp;
1660         int err;
1661
1662         reply = ovs_dp_cmd_alloc_info();
1663         if (!reply)
1664                 return -ENOMEM;
1665
1666         ovs_lock();
1667         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1668         err = PTR_ERR(dp);
1669         if (IS_ERR(dp))
1670                 goto err_unlock_free;
1671
1672         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1673                                    info->snd_seq, 0, OVS_DP_CMD_DEL);
1674         BUG_ON(err < 0);
1675
1676         __dp_destroy(dp);
1677         ovs_unlock();
1678
1679         ovs_notify(&dp_datapath_genl_family, reply, info);
1680
1681         return 0;
1682
1683 err_unlock_free:
1684         ovs_unlock();
1685         kfree_skb(reply);
1686         return err;
1687 }
1688
1689 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1690 {
1691         struct sk_buff *reply;
1692         struct datapath *dp;
1693         int err;
1694
1695         reply = ovs_dp_cmd_alloc_info();
1696         if (!reply)
1697                 return -ENOMEM;
1698
1699         ovs_lock();
1700         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1701         err = PTR_ERR(dp);
1702         if (IS_ERR(dp))
1703                 goto err_unlock_free;
1704
1705         ovs_dp_change(dp, info->attrs);
1706
1707         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1708                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1709         BUG_ON(err < 0);
1710
1711         ovs_unlock();
1712         ovs_notify(&dp_datapath_genl_family, reply, info);
1713
1714         return 0;
1715
1716 err_unlock_free:
1717         ovs_unlock();
1718         kfree_skb(reply);
1719         return err;
1720 }
1721
1722 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1723 {
1724         struct sk_buff *reply;
1725         struct datapath *dp;
1726         int err;
1727
1728         reply = ovs_dp_cmd_alloc_info();
1729         if (!reply)
1730                 return -ENOMEM;
1731
1732         ovs_lock();
1733         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1734         if (IS_ERR(dp)) {
1735                 err = PTR_ERR(dp);
1736                 goto err_unlock_free;
1737         }
1738         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1739                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1740         BUG_ON(err < 0);
1741         ovs_unlock();
1742
1743         return genlmsg_reply(reply, info);
1744
1745 err_unlock_free:
1746         ovs_unlock();
1747         kfree_skb(reply);
1748         return err;
1749 }
1750
1751 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1752 {
1753         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1754         struct datapath *dp;
1755         int skip = cb->args[0];
1756         int i = 0;
1757
1758         ovs_lock();
1759         list_for_each_entry(dp, &ovs_net->dps, list_node) {
1760                 if (i >= skip &&
1761                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1762                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1763                                          OVS_DP_CMD_NEW) < 0)
1764                         break;
1765                 i++;
1766         }
1767         ovs_unlock();
1768
1769         cb->args[0] = i;
1770
1771         return skb->len;
1772 }
1773
1774 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1775         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1776         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1777         [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1778 };
1779
1780 static const struct genl_ops dp_datapath_genl_ops[] = {
1781         { .cmd = OVS_DP_CMD_NEW,
1782           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1783           .policy = datapath_policy,
1784           .doit = ovs_dp_cmd_new
1785         },
1786         { .cmd = OVS_DP_CMD_DEL,
1787           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1788           .policy = datapath_policy,
1789           .doit = ovs_dp_cmd_del
1790         },
1791         { .cmd = OVS_DP_CMD_GET,
1792           .flags = 0,               /* OK for unprivileged users. */
1793           .policy = datapath_policy,
1794           .doit = ovs_dp_cmd_get,
1795           .dumpit = ovs_dp_cmd_dump
1796         },
1797         { .cmd = OVS_DP_CMD_SET,
1798           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1799           .policy = datapath_policy,
1800           .doit = ovs_dp_cmd_set,
1801         },
1802 };
1803
1804 static struct genl_family dp_datapath_genl_family = {
1805         .id = GENL_ID_GENERATE,
1806         .hdrsize = sizeof(struct ovs_header),
1807         .name = OVS_DATAPATH_FAMILY,
1808         .version = OVS_DATAPATH_VERSION,
1809         .maxattr = OVS_DP_ATTR_MAX,
1810         .netnsok = true,
1811         .parallel_ops = true,
1812         .ops = dp_datapath_genl_ops,
1813         .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1814         .mcgrps = &ovs_dp_datapath_multicast_group,
1815         .n_mcgrps = 1,
1816 };
1817
1818 /* Called with ovs_mutex or RCU read lock. */
1819 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1820                                    u32 portid, u32 seq, u32 flags, u8 cmd)
1821 {
1822         struct ovs_header *ovs_header;
1823         struct ovs_vport_stats vport_stats;
1824         int err;
1825
1826         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1827                                  flags, cmd);
1828         if (!ovs_header)
1829                 return -EMSGSIZE;
1830
1831         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1832
1833         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1834             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1835             nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1836                            ovs_vport_name(vport)))
1837                 goto nla_put_failure;
1838
1839         ovs_vport_get_stats(vport, &vport_stats);
1840         if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1841                     &vport_stats))
1842                 goto nla_put_failure;
1843
1844         if (ovs_vport_get_upcall_portids(vport, skb))
1845                 goto nla_put_failure;
1846
1847         err = ovs_vport_get_options(vport, skb);
1848         if (err == -EMSGSIZE)
1849                 goto error;
1850
1851         genlmsg_end(skb, ovs_header);
1852         return 0;
1853
1854 nla_put_failure:
1855         err = -EMSGSIZE;
1856 error:
1857         genlmsg_cancel(skb, ovs_header);
1858         return err;
1859 }
1860
1861 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1862 {
1863         return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1864 }
1865
1866 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1867 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1868                                          u32 seq, u8 cmd)
1869 {
1870         struct sk_buff *skb;
1871         int retval;
1872
1873         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1874         if (!skb)
1875                 return ERR_PTR(-ENOMEM);
1876
1877         retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1878         BUG_ON(retval < 0);
1879
1880         return skb;
1881 }
1882
1883 /* Called with ovs_mutex or RCU read lock. */
1884 static struct vport *lookup_vport(struct net *net,
1885                                   const struct ovs_header *ovs_header,
1886                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1887 {
1888         struct datapath *dp;
1889         struct vport *vport;
1890
1891         if (a[OVS_VPORT_ATTR_NAME]) {
1892                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1893                 if (!vport)
1894                         return ERR_PTR(-ENODEV);
1895                 if (ovs_header->dp_ifindex &&
1896                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1897                         return ERR_PTR(-ENODEV);
1898                 return vport;
1899         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1900                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1901
1902                 if (port_no >= DP_MAX_PORTS)
1903                         return ERR_PTR(-EFBIG);
1904
1905                 dp = get_dp(net, ovs_header->dp_ifindex);
1906                 if (!dp)
1907                         return ERR_PTR(-ENODEV);
1908
1909                 vport = ovs_vport_ovsl_rcu(dp, port_no);
1910                 if (!vport)
1911                         return ERR_PTR(-ENODEV);
1912                 return vport;
1913         } else
1914                 return ERR_PTR(-EINVAL);
1915 }
1916
1917 /* Called with ovs_mutex */
1918 static void update_headroom(struct datapath *dp)
1919 {
1920         unsigned dev_headroom, max_headroom = 0;
1921         struct net_device *dev;
1922         struct vport *vport;
1923         int i;
1924
1925         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1926                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1927                         dev = vport->dev;
1928                         dev_headroom = netdev_get_fwd_headroom(dev);
1929                         if (dev_headroom > max_headroom)
1930                                 max_headroom = dev_headroom;
1931                 }
1932         }
1933
1934         dp->max_headroom = max_headroom;
1935         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1936                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node)
1937                         netdev_set_rx_headroom(vport->dev, max_headroom);
1938 }
1939
1940 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1941 {
1942         struct nlattr **a = info->attrs;
1943         struct ovs_header *ovs_header = info->userhdr;
1944         struct vport_parms parms;
1945         struct sk_buff *reply;
1946         struct vport *vport;
1947         struct datapath *dp;
1948         u32 port_no;
1949         int err;
1950
1951         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1952             !a[OVS_VPORT_ATTR_UPCALL_PID])
1953                 return -EINVAL;
1954
1955         port_no = a[OVS_VPORT_ATTR_PORT_NO]
1956                 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
1957         if (port_no >= DP_MAX_PORTS)
1958                 return -EFBIG;
1959
1960         reply = ovs_vport_cmd_alloc_info();
1961         if (!reply)
1962                 return -ENOMEM;
1963
1964         ovs_lock();
1965 restart:
1966         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1967         err = -ENODEV;
1968         if (!dp)
1969                 goto exit_unlock_free;
1970
1971         if (port_no) {
1972                 vport = ovs_vport_ovsl(dp, port_no);
1973                 err = -EBUSY;
1974                 if (vport)
1975                         goto exit_unlock_free;
1976         } else {
1977                 for (port_no = 1; ; port_no++) {
1978                         if (port_no >= DP_MAX_PORTS) {
1979                                 err = -EFBIG;
1980                                 goto exit_unlock_free;
1981                         }
1982                         vport = ovs_vport_ovsl(dp, port_no);
1983                         if (!vport)
1984                                 break;
1985                 }
1986         }
1987
1988         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1989         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1990         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1991         parms.dp = dp;
1992         parms.port_no = port_no;
1993         parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
1994
1995         vport = new_vport(&parms);
1996         err = PTR_ERR(vport);
1997         if (IS_ERR(vport)) {
1998                 if (err == -EAGAIN)
1999                         goto restart;
2000                 goto exit_unlock_free;
2001         }
2002
2003         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2004                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2005
2006         if (netdev_get_fwd_headroom(vport->dev) > dp->max_headroom)
2007                 update_headroom(dp);
2008         else
2009                 netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2010
2011         BUG_ON(err < 0);
2012         ovs_unlock();
2013
2014         ovs_notify(&dp_vport_genl_family, reply, info);
2015         return 0;
2016
2017 exit_unlock_free:
2018         ovs_unlock();
2019         kfree_skb(reply);
2020         return err;
2021 }
2022
2023 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2024 {
2025         struct nlattr **a = info->attrs;
2026         struct sk_buff *reply;
2027         struct vport *vport;
2028         int err;
2029
2030         reply = ovs_vport_cmd_alloc_info();
2031         if (!reply)
2032                 return -ENOMEM;
2033
2034         ovs_lock();
2035         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2036         err = PTR_ERR(vport);
2037         if (IS_ERR(vport))
2038                 goto exit_unlock_free;
2039
2040         if (a[OVS_VPORT_ATTR_TYPE] &&
2041             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2042                 err = -EINVAL;
2043                 goto exit_unlock_free;
2044         }
2045
2046         if (a[OVS_VPORT_ATTR_OPTIONS]) {
2047                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2048                 if (err)
2049                         goto exit_unlock_free;
2050         }
2051
2052
2053         if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2054                 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2055
2056                 err = ovs_vport_set_upcall_portids(vport, ids);
2057                 if (err)
2058                         goto exit_unlock_free;
2059         }
2060
2061         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2062                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2063         BUG_ON(err < 0);
2064
2065         ovs_unlock();
2066         ovs_notify(&dp_vport_genl_family, reply, info);
2067         return 0;
2068
2069 exit_unlock_free:
2070         ovs_unlock();
2071         kfree_skb(reply);
2072         return err;
2073 }
2074
2075 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2076 {
2077         bool must_update_headroom = false;
2078         struct nlattr **a = info->attrs;
2079         struct sk_buff *reply;
2080         struct datapath *dp;
2081         struct vport *vport;
2082         int err;
2083
2084         reply = ovs_vport_cmd_alloc_info();
2085         if (!reply)
2086                 return -ENOMEM;
2087
2088         ovs_lock();
2089         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2090         err = PTR_ERR(vport);
2091         if (IS_ERR(vport))
2092                 goto exit_unlock_free;
2093
2094         if (vport->port_no == OVSP_LOCAL) {
2095                 err = -EINVAL;
2096                 goto exit_unlock_free;
2097         }
2098
2099         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2100                                       info->snd_seq, 0, OVS_VPORT_CMD_DEL);
2101         BUG_ON(err < 0);
2102
2103         /* the vport deletion may trigger dp headroom update */
2104         dp = vport->dp;
2105         if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2106                 must_update_headroom = true;
2107         netdev_reset_rx_headroom(vport->dev);
2108         ovs_dp_detach_port(vport);
2109
2110         if (must_update_headroom)
2111                 update_headroom(dp);
2112         ovs_unlock();
2113
2114         ovs_notify(&dp_vport_genl_family, reply, info);
2115         return 0;
2116
2117 exit_unlock_free:
2118         ovs_unlock();
2119         kfree_skb(reply);
2120         return err;
2121 }
2122
2123 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2124 {
2125         struct nlattr **a = info->attrs;
2126         struct ovs_header *ovs_header = info->userhdr;
2127         struct sk_buff *reply;
2128         struct vport *vport;
2129         int err;
2130
2131         reply = ovs_vport_cmd_alloc_info();
2132         if (!reply)
2133                 return -ENOMEM;
2134
2135         rcu_read_lock();
2136         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2137         err = PTR_ERR(vport);
2138         if (IS_ERR(vport))
2139                 goto exit_unlock_free;
2140         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2141                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2142         BUG_ON(err < 0);
2143         rcu_read_unlock();
2144
2145         return genlmsg_reply(reply, info);
2146
2147 exit_unlock_free:
2148         rcu_read_unlock();
2149         kfree_skb(reply);
2150         return err;
2151 }
2152
2153 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2154 {
2155         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2156         struct datapath *dp;
2157         int bucket = cb->args[0], skip = cb->args[1];
2158         int i, j = 0;
2159
2160         rcu_read_lock();
2161         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2162         if (!dp) {
2163                 rcu_read_unlock();
2164                 return -ENODEV;
2165         }
2166         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2167                 struct vport *vport;
2168
2169                 j = 0;
2170                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2171                         if (j >= skip &&
2172                             ovs_vport_cmd_fill_info(vport, skb,
2173                                                     NETLINK_CB(cb->skb).portid,
2174                                                     cb->nlh->nlmsg_seq,
2175                                                     NLM_F_MULTI,
2176                                                     OVS_VPORT_CMD_NEW) < 0)
2177                                 goto out;
2178
2179                         j++;
2180                 }
2181                 skip = 0;
2182         }
2183 out:
2184         rcu_read_unlock();
2185
2186         cb->args[0] = i;
2187         cb->args[1] = j;
2188
2189         return skb->len;
2190 }
2191
2192 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2193         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2194         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2195         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2196         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2197         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
2198         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2199 };
2200
2201 static const struct genl_ops dp_vport_genl_ops[] = {
2202         { .cmd = OVS_VPORT_CMD_NEW,
2203           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2204           .policy = vport_policy,
2205           .doit = ovs_vport_cmd_new
2206         },
2207         { .cmd = OVS_VPORT_CMD_DEL,
2208           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2209           .policy = vport_policy,
2210           .doit = ovs_vport_cmd_del
2211         },
2212         { .cmd = OVS_VPORT_CMD_GET,
2213           .flags = 0,               /* OK for unprivileged users. */
2214           .policy = vport_policy,
2215           .doit = ovs_vport_cmd_get,
2216           .dumpit = ovs_vport_cmd_dump
2217         },
2218         { .cmd = OVS_VPORT_CMD_SET,
2219           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2220           .policy = vport_policy,
2221           .doit = ovs_vport_cmd_set,
2222         },
2223 };
2224
2225 struct genl_family dp_vport_genl_family = {
2226         .id = GENL_ID_GENERATE,
2227         .hdrsize = sizeof(struct ovs_header),
2228         .name = OVS_VPORT_FAMILY,
2229         .version = OVS_VPORT_VERSION,
2230         .maxattr = OVS_VPORT_ATTR_MAX,
2231         .netnsok = true,
2232         .parallel_ops = true,
2233         .ops = dp_vport_genl_ops,
2234         .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2235         .mcgrps = &ovs_dp_vport_multicast_group,
2236         .n_mcgrps = 1,
2237 };
2238
2239 static struct genl_family * const dp_genl_families[] = {
2240         &dp_datapath_genl_family,
2241         &dp_vport_genl_family,
2242         &dp_flow_genl_family,
2243         &dp_packet_genl_family,
2244 };
2245
2246 static void dp_unregister_genl(int n_families)
2247 {
2248         int i;
2249
2250         for (i = 0; i < n_families; i++)
2251                 genl_unregister_family(dp_genl_families[i]);
2252 }
2253
2254 static int dp_register_genl(void)
2255 {
2256         int err;
2257         int i;
2258
2259         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2260
2261                 err = genl_register_family(dp_genl_families[i]);
2262                 if (err)
2263                         goto error;
2264         }
2265
2266         return 0;
2267
2268 error:
2269         dp_unregister_genl(i);
2270         return err;
2271 }
2272
2273 static int __net_init ovs_init_net(struct net *net)
2274 {
2275         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2276
2277         INIT_LIST_HEAD(&ovs_net->dps);
2278         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2279         ovs_ct_init(net);
2280         return 0;
2281 }
2282
2283 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2284                                             struct list_head *head)
2285 {
2286         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2287         struct datapath *dp;
2288
2289         list_for_each_entry(dp, &ovs_net->dps, list_node) {
2290                 int i;
2291
2292                 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2293                         struct vport *vport;
2294
2295                         hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2296                                 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2297                                         continue;
2298
2299                                 if (dev_net(vport->dev) == dnet)
2300                                         list_add(&vport->detach_list, head);
2301                         }
2302                 }
2303         }
2304 }
2305
2306 static void __net_exit ovs_exit_net(struct net *dnet)
2307 {
2308         struct datapath *dp, *dp_next;
2309         struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2310         struct vport *vport, *vport_next;
2311         struct net *net;
2312         LIST_HEAD(head);
2313
2314         ovs_ct_exit(dnet);
2315         ovs_lock();
2316         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2317                 __dp_destroy(dp);
2318
2319         rtnl_lock();
2320         for_each_net(net)
2321                 list_vports_from_net(net, dnet, &head);
2322         rtnl_unlock();
2323
2324         /* Detach all vports from given namespace. */
2325         list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2326                 list_del(&vport->detach_list);
2327                 ovs_dp_detach_port(vport);
2328         }
2329
2330         ovs_unlock();
2331
2332         cancel_work_sync(&ovs_net->dp_notify_work);
2333 }
2334
2335 static struct pernet_operations ovs_net_ops = {
2336         .init = ovs_init_net,
2337         .exit = ovs_exit_net,
2338         .id   = &ovs_net_id,
2339         .size = sizeof(struct ovs_net),
2340 };
2341
2342 static int __init dp_init(void)
2343 {
2344         int err;
2345
2346         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2347
2348         pr_info("Open vSwitch switching datapath\n");
2349
2350         err = action_fifos_init();
2351         if (err)
2352                 goto error;
2353
2354         err = ovs_internal_dev_rtnl_link_register();
2355         if (err)
2356                 goto error_action_fifos_exit;
2357
2358         err = ovs_flow_init();
2359         if (err)
2360                 goto error_unreg_rtnl_link;
2361
2362         err = ovs_vport_init();
2363         if (err)
2364                 goto error_flow_exit;
2365
2366         err = register_pernet_device(&ovs_net_ops);
2367         if (err)
2368                 goto error_vport_exit;
2369
2370         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2371         if (err)
2372                 goto error_netns_exit;
2373
2374         err = ovs_netdev_init();
2375         if (err)
2376                 goto error_unreg_notifier;
2377
2378         err = dp_register_genl();
2379         if (err < 0)
2380                 goto error_unreg_netdev;
2381
2382         return 0;
2383
2384 error_unreg_netdev:
2385         ovs_netdev_exit();
2386 error_unreg_notifier:
2387         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2388 error_netns_exit:
2389         unregister_pernet_device(&ovs_net_ops);
2390 error_vport_exit:
2391         ovs_vport_exit();
2392 error_flow_exit:
2393         ovs_flow_exit();
2394 error_unreg_rtnl_link:
2395         ovs_internal_dev_rtnl_link_unregister();
2396 error_action_fifos_exit:
2397         action_fifos_exit();
2398 error:
2399         return err;
2400 }
2401
2402 static void dp_cleanup(void)
2403 {
2404         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2405         ovs_netdev_exit();
2406         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2407         unregister_pernet_device(&ovs_net_ops);
2408         rcu_barrier();
2409         ovs_vport_exit();
2410         ovs_flow_exit();
2411         ovs_internal_dev_rtnl_link_unregister();
2412         action_fifos_exit();
2413 }
2414
2415 module_init(dp_init);
2416 module_exit(dp_cleanup);
2417
2418 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2419 MODULE_LICENSE("GPL");