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         struct ovs_skb_cb ovs_cb;
 340         int err;
 341
 342         ovs_cb = *OVS_CB(skb);
 343         segs = __skb_gso_segment(skb, NETIF_F_SG, false);
 344         *OVS_CB(skb) = ovs_cb;
 345         if (IS_ERR(segs))
 346                 return PTR_ERR(segs);
 347         if (segs == NULL)
 348                 return -EINVAL;
 349
 350         if (gso_type & SKB_GSO_UDP) {
 351                 /* The initial flow key extracted by ovs_flow_key_extract()
 352                  * in this case is for a first fragment, so we need to
 353                  * properly mark later fragments.
 354                  */
 355                 later_key = *key;
 356                 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
 357         }
 358
 359         /* Queue all of the segments. */
 360         skb = segs;
 361         do {
 362                 *OVS_CB(skb) = ovs_cb;
 363                 if (gso_type & SKB_GSO_UDP && skb != segs)
 364                         key = &later_key;
 365
 366                 err = queue_userspace_packet(dp, skb, key, upcall_info);
 367                 if (err)
 368                         break;
 369
 370         } while ((skb = skb->next));
 371
 372         /* Free all of the segments. */
 373         skb = segs;
 374         do {
 375                 nskb = skb->next;
 376                 if (err)
 377                         kfree_skb(skb);
 378                 else
 379                         consume_skb(skb);
 380         } while ((skb = nskb));
 381         return err;
 382 }
 383
 384 static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
 385                               unsigned int hdrlen)
 386 {
 387         size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
 388                 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
 389                 + nla_total_size(ovs_key_attr_size()); /* OVS_PACKET_ATTR_KEY */
 390
 391         /* OVS_PACKET_ATTR_USERDATA */
 392         if (upcall_info->userdata)
 393                 size += NLA_ALIGN(upcall_info->userdata->nla_len);
 394
 395         /* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
 396         if (upcall_info->egress_tun_info)
 397                 size += nla_total_size(ovs_tun_key_attr_size());
 398
 399         /* OVS_PACKET_ATTR_ACTIONS */
 400         if (upcall_info->actions_len)
 401                 size += nla_total_size(upcall_info->actions_len);
 402
 403         /* OVS_PACKET_ATTR_MRU */
 404         if (upcall_info->mru)
 405                 size += nla_total_size(sizeof(upcall_info->mru));
 406
 407         return size;
 408 }
 409
 410 static void pad_packet(struct datapath *dp, struct sk_buff *skb)
 411 {
 412         if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
 413                 size_t plen = NLA_ALIGN(skb->len) - skb->len;
 414
 415                 if (plen > 0)
 416                         memset(skb_put(skb, plen), 0, plen);
 417         }
 418 }
 419
 420 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
 421                                   const struct sw_flow_key *key,
 422                                   const struct dp_upcall_info *upcall_info)
 423 {
 424         struct ovs_header *upcall;
 425         struct sk_buff *nskb = NULL;
 426         struct sk_buff *user_skb = NULL; /* to be queued to userspace */
 427         struct nlattr *nla;
 428         struct genl_info info = {
 429                 .dst_sk = ovs_dp_get_net(dp)->genl_sock,
 430                 .snd_portid = upcall_info->portid,
 431         };
 432         size_t len;
 433         unsigned int hlen;
 434         int err, dp_ifindex;
 435
 436         dp_ifindex = get_dpifindex(dp);
 437         if (!dp_ifindex)
 438                 return -ENODEV;
 439
 440         if (skb_vlan_tag_present(skb)) {
 441                 nskb = skb_clone(skb, GFP_ATOMIC);
 442                 if (!nskb)
 443                         return -ENOMEM;
 444
 445                 nskb = __vlan_hwaccel_push_inside(nskb);
 446                 if (!nskb)
 447                         return -ENOMEM;
 448
 449                 skb = nskb;
 450         }
 451
 452         if (nla_attr_size(skb->len) > USHRT_MAX) {
 453                 err = -EFBIG;
 454                 goto out;
 455         }
 456
 457         /* Complete checksum if needed */
 458         if (skb->ip_summed == CHECKSUM_PARTIAL &&
 459             (err = skb_checksum_help(skb)))
 460                 goto out;
 461
 462         /* Older versions of OVS user space enforce alignment of the last
 463          * Netlink attribute to NLA_ALIGNTO which would require extensive
 464          * padding logic. Only perform zerocopy if padding is not required.
 465          */
 466         if (dp->user_features & OVS_DP_F_UNALIGNED)
 467                 hlen = skb_zerocopy_headlen(skb);
 468         else
 469                 hlen = skb->len;
 470
 471         len = upcall_msg_size(upcall_info, hlen);
 472         user_skb = genlmsg_new_unicast(len, &info, GFP_ATOMIC);
 473         if (!user_skb) {
 474                 err = -ENOMEM;
 475                 goto out;
 476         }
 477
 478         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 479                              0, upcall_info->cmd);
 480         upcall->dp_ifindex = dp_ifindex;
 481
 482         err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
 483         BUG_ON(err);
 484
 485         if (upcall_info->userdata)
 486                 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
 487                           nla_len(upcall_info->userdata),
 488                           nla_data(upcall_info->userdata));
 489
 490         if (upcall_info->egress_tun_info) {
 491                 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_EGRESS_TUN_KEY);
 492                 err = ovs_nla_put_tunnel_info(user_skb,
 493                                               upcall_info->egress_tun_info);
 494                 BUG_ON(err);
 495                 nla_nest_end(user_skb, nla);
 496         }
 497
 498         if (upcall_info->actions_len) {
 499                 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_ACTIONS);
 500                 err = ovs_nla_put_actions(upcall_info->actions,
 501                                           upcall_info->actions_len,
 502                                           user_skb);
 503                 if (!err)
 504                         nla_nest_end(user_skb, nla);
 505                 else
 506                         nla_nest_cancel(user_skb, nla);
 507         }
 508
 509         /* Add OVS_PACKET_ATTR_MRU */
 510         if (upcall_info->mru) {
 511                 if (nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU,
 512                                 upcall_info->mru)) {
 513                         err = -ENOBUFS;
 514                         goto out;
 515                 }
 516                 pad_packet(dp, user_skb);
 517         }
 518
 519         /* Only reserve room for attribute header, packet data is added
 520          * in skb_zerocopy() */
 521         if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
 522                 err = -ENOBUFS;
 523                 goto out;
 524         }
 525         nla->nla_len = nla_attr_size(skb->len);
 526
 527         err = skb_zerocopy(user_skb, skb, skb->len, hlen);
 528         if (err)
 529                 goto out;
 530
 531         /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
 532         pad_packet(dp, user_skb);
 533
 534         ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
 535
 536         err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
 537         user_skb = NULL;
 538 out:
 539         if (err)
 540                 skb_tx_error(skb);
 541         kfree_skb(user_skb);
 542         kfree_skb(nskb);
 543         return err;
 544 }
 545
 546 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 547 {
 548         struct ovs_header *ovs_header = info->userhdr;
 549         struct net *net = sock_net(skb->sk);
 550         struct nlattr **a = info->attrs;
 551         struct sw_flow_actions *acts;
 552         struct sk_buff *packet;
 553         struct sw_flow *flow;
 554         struct sw_flow_actions *sf_acts;
 555         struct datapath *dp;
 556         struct ethhdr *eth;
 557         struct vport *input_vport;
 558         u16 mru = 0;
 559         int len;
 560         int err;
 561         bool log = !a[OVS_PACKET_ATTR_PROBE];
 562
 563         err = -EINVAL;
 564         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 565             !a[OVS_PACKET_ATTR_ACTIONS])
 566                 goto err;
 567
 568         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 569         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 570         err = -ENOMEM;
 571         if (!packet)
 572                 goto err;
 573         skb_reserve(packet, NET_IP_ALIGN);
 574
 575         nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
 576
 577         skb_reset_mac_header(packet);
 578         eth = eth_hdr(packet);
 579
 580         /* Normally, setting the skb 'protocol' field would be handled by a
 581          * call to eth_type_trans(), but it assumes there's a sending
 582          * device, which we may not have. */
 583         if (eth_proto_is_802_3(eth->h_proto))
 584                 packet->protocol = eth->h_proto;
 585         else
 586                 packet->protocol = htons(ETH_P_802_2);
 587
 588         /* Set packet's mru */
 589         if (a[OVS_PACKET_ATTR_MRU]) {
 590                 mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
 591                 packet->ignore_df = 1;
 592         }
 593         OVS_CB(packet)->mru = mru;
 594
 595         /* Build an sw_flow for sending this packet. */
 596         flow = ovs_flow_alloc();
 597         err = PTR_ERR(flow);
 598         if (IS_ERR(flow))
 599                 goto err_kfree_skb;
 600
 601         err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
 602                                              packet, &flow->key, log);
 603         if (err)
 604                 goto err_flow_free;
 605
 606         err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
 607                                    &flow->key, &acts, log);
 608         if (err)
 609                 goto err_flow_free;
 610
 611         rcu_assign_pointer(flow->sf_acts, acts);
 612         packet->priority = flow->key.phy.priority;
 613         packet->mark = flow->key.phy.skb_mark;
 614
 615         rcu_read_lock();
 616         dp = get_dp_rcu(net, ovs_header->dp_ifindex);
 617         err = -ENODEV;
 618         if (!dp)
 619                 goto err_unlock;
 620
 621         input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
 622         if (!input_vport)
 623                 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
 624
 625         if (!input_vport)
 626                 goto err_unlock;
 627
 628         packet->dev = input_vport->dev;
 629         OVS_CB(packet)->input_vport = input_vport;
 630         sf_acts = rcu_dereference(flow->sf_acts);
 631
 632         local_bh_disable();
 633         err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
 634         local_bh_enable();
 635         rcu_read_unlock();
 636
 637         ovs_flow_free(flow, false);
 638         return err;
 639
 640 err_unlock:
 641         rcu_read_unlock();
 642 err_flow_free:
 643         ovs_flow_free(flow, false);
 644 err_kfree_skb:
 645         kfree_skb(packet);
 646 err:
 647         return err;
 648 }
 649
 650 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 651         [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
 652         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 653         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 654         [OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
 655         [OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
 656 };
 657
 658 static const struct genl_ops dp_packet_genl_ops[] = {
 659         { .cmd = OVS_PACKET_CMD_EXECUTE,
 660           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 661           .policy = packet_policy,
 662           .doit = ovs_packet_cmd_execute
 663         }
 664 };
 665
 666 static struct genl_family dp_packet_genl_family = {
 667         .id = GENL_ID_GENERATE,
 668         .hdrsize = sizeof(struct ovs_header),
 669         .name = OVS_PACKET_FAMILY,
 670         .version = OVS_PACKET_VERSION,
 671         .maxattr = OVS_PACKET_ATTR_MAX,
 672         .netnsok = true,
 673         .parallel_ops = true,
 674         .ops = dp_packet_genl_ops,
 675         .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
 676 };
 677
 678 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
 679                          struct ovs_dp_megaflow_stats *mega_stats)
 680 {
 681         int i;
 682
 683         memset(mega_stats, 0, sizeof(*mega_stats));
 684
 685         stats->n_flows = ovs_flow_tbl_count(&dp->table);
 686         mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
 687
 688         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 689
 690         for_each_possible_cpu(i) {
 691                 const struct dp_stats_percpu *percpu_stats;
 692                 struct dp_stats_percpu local_stats;
 693                 unsigned int start;
 694
 695                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 696
 697                 do {
 698                         start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
 699                         local_stats = *percpu_stats;
 700                 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
 701
 702                 stats->n_hit += local_stats.n_hit;
 703                 stats->n_missed += local_stats.n_missed;
 704                 stats->n_lost += local_stats.n_lost;
 705                 mega_stats->n_mask_hit += local_stats.n_mask_hit;
 706         }
 707 }
 708
 709 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
 710 {
 711         return ovs_identifier_is_ufid(sfid) &&
 712                !(ufid_flags & OVS_UFID_F_OMIT_KEY);
 713 }
 714
 715 static bool should_fill_mask(uint32_t ufid_flags)
 716 {
 717         return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
 718 }
 719
 720 static bool should_fill_actions(uint32_t ufid_flags)
 721 {
 722         return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
 723 }
 724
 725 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
 726                                     const struct sw_flow_id *sfid,
 727                                     uint32_t ufid_flags)
 728 {
 729         size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
 730
 731         /* OVS_FLOW_ATTR_UFID */
 732         if (sfid && ovs_identifier_is_ufid(sfid))
 733                 len += nla_total_size(sfid->ufid_len);
 734
 735         /* OVS_FLOW_ATTR_KEY */
 736         if (!sfid || should_fill_key(sfid, ufid_flags))
 737                 len += nla_total_size(ovs_key_attr_size());
 738
 739         /* OVS_FLOW_ATTR_MASK */
 740         if (should_fill_mask(ufid_flags))
 741                 len += nla_total_size(ovs_key_attr_size());
 742
 743         /* OVS_FLOW_ATTR_ACTIONS */
 744         if (should_fill_actions(ufid_flags))
 745                 len += nla_total_size(acts->orig_len);
 746
 747         return len
 748                 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
 749                 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
 750                 + nla_total_size(8); /* OVS_FLOW_ATTR_USED */
 751 }
 752
 753 /* Called with ovs_mutex or RCU read lock. */
 754 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
 755                                    struct sk_buff *skb)
 756 {
 757         struct ovs_flow_stats stats;
 758         __be16 tcp_flags;
 759         unsigned long used;
 760
 761         ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
 762
 763         if (used &&
 764             nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
 765                 return -EMSGSIZE;
 766
 767         if (stats.n_packets &&
 768             nla_put(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats))
 769                 return -EMSGSIZE;
 770
 771         if ((u8)ntohs(tcp_flags) &&
 772              nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
 773                 return -EMSGSIZE;
 774
 775         return 0;
 776 }
 777
 778 /* Called with ovs_mutex or RCU read lock. */
 779 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
 780                                      struct sk_buff *skb, int skb_orig_len)
 781 {
 782         struct nlattr *start;
 783         int err;
 784
 785         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 786          * this is the first flow to be dumped into 'skb'.  This is unusual for
 787          * Netlink but individual action lists can be longer than
 788          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 789          * The userspace caller can always fetch the actions separately if it
 790          * really wants them.  (Most userspace callers in fact don't care.)
 791          *
 792          * This can only fail for dump operations because the skb is always
 793          * properly sized for single flows.
 794          */
 795         start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
 796         if (start) {
 797                 const struct sw_flow_actions *sf_acts;
 798
 799                 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
 800                 err = ovs_nla_put_actions(sf_acts->actions,
 801                                           sf_acts->actions_len, skb);
 802
 803                 if (!err)
 804                         nla_nest_end(skb, start);
 805                 else {
 806                         if (skb_orig_len)
 807                                 return err;
 808
 809                         nla_nest_cancel(skb, start);
 810                 }
 811         } else if (skb_orig_len) {
 812                 return -EMSGSIZE;
 813         }
 814
 815         return 0;
 816 }
 817
 818 /* Called with ovs_mutex or RCU read lock. */
 819 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
 820                                   struct sk_buff *skb, u32 portid,
 821                                   u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
 822 {
 823         const int skb_orig_len = skb->len;
 824         struct ovs_header *ovs_header;
 825         int err;
 826
 827         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
 828                                  flags, cmd);
 829         if (!ovs_header)
 830                 return -EMSGSIZE;
 831
 832         ovs_header->dp_ifindex = dp_ifindex;
 833
 834         err = ovs_nla_put_identifier(flow, skb);
 835         if (err)
 836                 goto error;
 837
 838         if (should_fill_key(&flow->id, ufid_flags)) {
 839                 err = ovs_nla_put_masked_key(flow, skb);
 840                 if (err)
 841                         goto error;
 842         }
 843
 844         if (should_fill_mask(ufid_flags)) {
 845                 err = ovs_nla_put_mask(flow, skb);
 846                 if (err)
 847                         goto error;
 848         }
 849
 850         err = ovs_flow_cmd_fill_stats(flow, skb);
 851         if (err)
 852                 goto error;
 853
 854         if (should_fill_actions(ufid_flags)) {
 855                 err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
 856                 if (err)
 857                         goto error;
 858         }
 859
 860         genlmsg_end(skb, ovs_header);
 861         return 0;
 862
 863 error:
 864         genlmsg_cancel(skb, ovs_header);
 865         return err;
 866 }
 867
 868 /* May not be called with RCU read lock. */
 869 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
 870                                                const struct sw_flow_id *sfid,
 871                                                struct genl_info *info,
 872                                                bool always,
 873                                                uint32_t ufid_flags)
 874 {
 875         struct sk_buff *skb;
 876         size_t len;
 877
 878         if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
 879                 return NULL;
 880
 881         len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
 882         skb = genlmsg_new_unicast(len, info, GFP_KERNEL);
 883         if (!skb)
 884                 return ERR_PTR(-ENOMEM);
 885
 886         return skb;
 887 }
 888
 889 /* Called with ovs_mutex. */
 890 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
 891                                                int dp_ifindex,
 892                                                struct genl_info *info, u8 cmd,
 893                                                bool always, u32 ufid_flags)
 894 {
 895         struct sk_buff *skb;
 896         int retval;
 897
 898         skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
 899                                       &flow->id, info, always, ufid_flags);
 900         if (IS_ERR_OR_NULL(skb))
 901                 return skb;
 902
 903         retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
 904                                         info->snd_portid, info->snd_seq, 0,
 905                                         cmd, ufid_flags);
 906         BUG_ON(retval < 0);
 907         return skb;
 908 }
 909
 910 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
 911 {
 912         struct net *net = sock_net(skb->sk);
 913         struct nlattr **a = info->attrs;
 914         struct ovs_header *ovs_header = info->userhdr;
 915         struct sw_flow *flow = NULL, *new_flow;
 916         struct sw_flow_mask mask;
 917         struct sk_buff *reply;
 918         struct datapath *dp;
 919         struct sw_flow_key key;
 920         struct sw_flow_actions *acts;
 921         struct sw_flow_match match;
 922         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
 923         int error;
 924         bool log = !a[OVS_FLOW_ATTR_PROBE];
 925
 926         /* Must have key and actions. */
 927         error = -EINVAL;
 928         if (!a[OVS_FLOW_ATTR_KEY]) {
 929                 OVS_NLERR(log, "Flow key attr not present in new flow.");
 930                 goto error;
 931         }
 932         if (!a[OVS_FLOW_ATTR_ACTIONS]) {
 933                 OVS_NLERR(log, "Flow actions attr not present in new flow.");
 934                 goto error;
 935         }
 936
 937         /* Most of the time we need to allocate a new flow, do it before
 938          * locking.
 939          */
 940         new_flow = ovs_flow_alloc();
 941         if (IS_ERR(new_flow)) {
 942                 error = PTR_ERR(new_flow);
 943                 goto error;
 944         }
 945
 946         /* Extract key. */
 947         ovs_match_init(&match, &key, &mask);
 948         error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
 949                                   a[OVS_FLOW_ATTR_MASK], log);
 950         if (error)
 951                 goto err_kfree_flow;
 952
 953         ovs_flow_mask_key(&new_flow->key, &key, true, &mask);
 954
 955         /* Extract flow identifier. */
 956         error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
 957                                        &key, log);
 958         if (error)
 959                 goto err_kfree_flow;
 960
 961         /* Validate actions. */
 962         error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
 963                                      &new_flow->key, &acts, log);
 964         if (error) {
 965                 OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
 966                 goto err_kfree_flow;
 967         }
 968
 969         reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
 970                                         ufid_flags);
 971         if (IS_ERR(reply)) {
 972                 error = PTR_ERR(reply);
 973                 goto err_kfree_acts;
 974         }
 975
 976         ovs_lock();
 977         dp = get_dp(net, ovs_header->dp_ifindex);
 978         if (unlikely(!dp)) {
 979                 error = -ENODEV;
 980                 goto err_unlock_ovs;
 981         }
 982
 983         /* Check if this is a duplicate flow */
 984         if (ovs_identifier_is_ufid(&new_flow->id))
 985                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
 986         if (!flow)
 987                 flow = ovs_flow_tbl_lookup(&dp->table, &key);
 988         if (likely(!flow)) {
 989                 rcu_assign_pointer(new_flow->sf_acts, acts);
 990
 991                 /* Put flow in bucket. */
 992                 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
 993                 if (unlikely(error)) {
 994                         acts = NULL;
 995                         goto err_unlock_ovs;
 996                 }
 997
 998                 if (unlikely(reply)) {
 999                         error = ovs_flow_cmd_fill_info(new_flow,
1000                                                        ovs_header->dp_ifindex,
1001                                                        reply, info->snd_portid,
1002                                                        info->snd_seq, 0,
1003                                                        OVS_FLOW_CMD_NEW,
1004                                                        ufid_flags);
1005                         BUG_ON(error < 0);
1006                 }
1007                 ovs_unlock();
1008         } else {
1009                 struct sw_flow_actions *old_acts;
1010
1011                 /* Bail out if we're not allowed to modify an existing flow.
1012                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1013                  * because Generic Netlink treats the latter as a dump
1014                  * request.  We also accept NLM_F_EXCL in case that bug ever
1015                  * gets fixed.
1016                  */
1017                 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1018                                                          | NLM_F_EXCL))) {
1019                         error = -EEXIST;
1020                         goto err_unlock_ovs;
1021                 }
1022                 /* The flow identifier has to be the same for flow updates.
1023                  * Look for any overlapping flow.
1024                  */
1025                 if (unlikely(!ovs_flow_cmp(flow, &match))) {
1026                         if (ovs_identifier_is_key(&flow->id))
1027                                 flow = ovs_flow_tbl_lookup_exact(&dp->table,
1028                                                                  &match);
1029                         else /* UFID matches but key is different */
1030                                 flow = NULL;
1031                         if (!flow) {
1032                                 error = -ENOENT;
1033                                 goto err_unlock_ovs;
1034                         }
1035                 }
1036                 /* Update actions. */
1037                 old_acts = ovsl_dereference(flow->sf_acts);
1038                 rcu_assign_pointer(flow->sf_acts, acts);
1039
1040                 if (unlikely(reply)) {
1041                         error = ovs_flow_cmd_fill_info(flow,
1042                                                        ovs_header->dp_ifindex,
1043                                                        reply, info->snd_portid,
1044                                                        info->snd_seq, 0,
1045                                                        OVS_FLOW_CMD_NEW,
1046                                                        ufid_flags);
1047                         BUG_ON(error < 0);
1048                 }
1049                 ovs_unlock();
1050
1051                 ovs_nla_free_flow_actions_rcu(old_acts);
1052                 ovs_flow_free(new_flow, false);
1053         }
1054
1055         if (reply)
1056                 ovs_notify(&dp_flow_genl_family, reply, info);
1057         return 0;
1058
1059 err_unlock_ovs:
1060         ovs_unlock();
1061         kfree_skb(reply);
1062 err_kfree_acts:
1063         ovs_nla_free_flow_actions(acts);
1064 err_kfree_flow:
1065         ovs_flow_free(new_flow, false);
1066 error:
1067         return error;
1068 }
1069
1070 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1071 static struct sw_flow_actions *get_flow_actions(struct net *net,
1072                                                 const struct nlattr *a,
1073                                                 const struct sw_flow_key *key,
1074                                                 const struct sw_flow_mask *mask,
1075                                                 bool log)
1076 {
1077         struct sw_flow_actions *acts;
1078         struct sw_flow_key masked_key;
1079         int error;
1080
1081         ovs_flow_mask_key(&masked_key, key, true, mask);
1082         error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1083         if (error) {
1084                 OVS_NLERR(log,
1085                           "Actions may not be safe on all matching packets");
1086                 return ERR_PTR(error);
1087         }
1088
1089         return acts;
1090 }
1091
1092 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1093 {
1094         struct net *net = sock_net(skb->sk);
1095         struct nlattr **a = info->attrs;
1096         struct ovs_header *ovs_header = info->userhdr;
1097         struct sw_flow_key key;
1098         struct sw_flow *flow;
1099         struct sw_flow_mask mask;
1100         struct sk_buff *reply = NULL;
1101         struct datapath *dp;
1102         struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1103         struct sw_flow_match match;
1104         struct sw_flow_id sfid;
1105         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1106         int error;
1107         bool log = !a[OVS_FLOW_ATTR_PROBE];
1108         bool ufid_present;
1109
1110         /* Extract key. */
1111         error = -EINVAL;
1112         if (!a[OVS_FLOW_ATTR_KEY]) {
1113                 OVS_NLERR(log, "Flow key attribute not present in set flow.");
1114                 goto error;
1115         }
1116
1117         ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1118         ovs_match_init(&match, &key, &mask);
1119         error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1120                                   a[OVS_FLOW_ATTR_MASK], log);
1121         if (error)
1122                 goto error;
1123
1124         /* Validate actions. */
1125         if (a[OVS_FLOW_ATTR_ACTIONS]) {
1126                 acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], &key,
1127                                         &mask, log);
1128                 if (IS_ERR(acts)) {
1129                         error = PTR_ERR(acts);
1130                         goto error;
1131                 }
1132
1133                 /* Can allocate before locking if have acts. */
1134                 reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1135                                                 ufid_flags);
1136                 if (IS_ERR(reply)) {
1137                         error = PTR_ERR(reply);
1138                         goto err_kfree_acts;
1139                 }
1140         }
1141
1142         ovs_lock();
1143         dp = get_dp(net, ovs_header->dp_ifindex);
1144         if (unlikely(!dp)) {
1145                 error = -ENODEV;
1146                 goto err_unlock_ovs;
1147         }
1148         /* Check that the flow exists. */
1149         if (ufid_present)
1150                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1151         else
1152                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1153         if (unlikely(!flow)) {
1154                 error = -ENOENT;
1155                 goto err_unlock_ovs;
1156         }
1157
1158         /* Update actions, if present. */
1159         if (likely(acts)) {
1160                 old_acts = ovsl_dereference(flow->sf_acts);
1161                 rcu_assign_pointer(flow->sf_acts, acts);
1162
1163                 if (unlikely(reply)) {
1164                         error = ovs_flow_cmd_fill_info(flow,
1165                                                        ovs_header->dp_ifindex,
1166                                                        reply, info->snd_portid,
1167                                                        info->snd_seq, 0,
1168                                                        OVS_FLOW_CMD_NEW,
1169                                                        ufid_flags);
1170                         BUG_ON(error < 0);
1171                 }
1172         } else {
1173                 /* Could not alloc without acts before locking. */
1174                 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1175                                                 info, OVS_FLOW_CMD_NEW, false,
1176                                                 ufid_flags);
1177
1178                 if (IS_ERR(reply)) {
1179                         error = PTR_ERR(reply);
1180                         goto err_unlock_ovs;
1181                 }
1182         }
1183
1184         /* Clear stats. */
1185         if (a[OVS_FLOW_ATTR_CLEAR])
1186                 ovs_flow_stats_clear(flow);
1187         ovs_unlock();
1188
1189         if (reply)
1190                 ovs_notify(&dp_flow_genl_family, reply, info);
1191         if (old_acts)
1192                 ovs_nla_free_flow_actions_rcu(old_acts);
1193
1194         return 0;
1195
1196 err_unlock_ovs:
1197         ovs_unlock();
1198         kfree_skb(reply);
1199 err_kfree_acts:
1200         ovs_nla_free_flow_actions(acts);
1201 error:
1202         return error;
1203 }
1204
1205 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1206 {
1207         struct nlattr **a = info->attrs;
1208         struct ovs_header *ovs_header = info->userhdr;
1209         struct net *net = sock_net(skb->sk);
1210         struct sw_flow_key key;
1211         struct sk_buff *reply;
1212         struct sw_flow *flow;
1213         struct datapath *dp;
1214         struct sw_flow_match match;
1215         struct sw_flow_id ufid;
1216         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1217         int err = 0;
1218         bool log = !a[OVS_FLOW_ATTR_PROBE];
1219         bool ufid_present;
1220
1221         ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1222         if (a[OVS_FLOW_ATTR_KEY]) {
1223                 ovs_match_init(&match, &key, NULL);
1224                 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1225                                         log);
1226         } else if (!ufid_present) {
1227                 OVS_NLERR(log,
1228                           "Flow get message rejected, Key attribute missing.");
1229                 err = -EINVAL;
1230         }
1231         if (err)
1232                 return err;
1233
1234         ovs_lock();
1235         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1236         if (!dp) {
1237                 err = -ENODEV;
1238                 goto unlock;
1239         }
1240
1241         if (ufid_present)
1242                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1243         else
1244                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1245         if (!flow) {
1246                 err = -ENOENT;
1247                 goto unlock;
1248         }
1249
1250         reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1251                                         OVS_FLOW_CMD_NEW, true, ufid_flags);
1252         if (IS_ERR(reply)) {
1253                 err = PTR_ERR(reply);
1254                 goto unlock;
1255         }
1256
1257         ovs_unlock();
1258         return genlmsg_reply(reply, info);
1259 unlock:
1260         ovs_unlock();
1261         return err;
1262 }
1263
1264 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1265 {
1266         struct nlattr **a = info->attrs;
1267         struct ovs_header *ovs_header = info->userhdr;
1268         struct net *net = sock_net(skb->sk);
1269         struct sw_flow_key key;
1270         struct sk_buff *reply;
1271         struct sw_flow *flow = NULL;
1272         struct datapath *dp;
1273         struct sw_flow_match match;
1274         struct sw_flow_id ufid;
1275         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1276         int err;
1277         bool log = !a[OVS_FLOW_ATTR_PROBE];
1278         bool ufid_present;
1279
1280         ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1281         if (a[OVS_FLOW_ATTR_KEY]) {
1282                 ovs_match_init(&match, &key, NULL);
1283                 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1284                                         NULL, log);
1285                 if (unlikely(err))
1286                         return err;
1287         }
1288
1289         ovs_lock();
1290         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1291         if (unlikely(!dp)) {
1292                 err = -ENODEV;
1293                 goto unlock;
1294         }
1295
1296         if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1297                 err = ovs_flow_tbl_flush(&dp->table);
1298                 goto unlock;
1299         }
1300
1301         if (ufid_present)
1302                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1303         else
1304                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1305         if (unlikely(!flow)) {
1306                 err = -ENOENT;
1307                 goto unlock;
1308         }
1309
1310         ovs_flow_tbl_remove(&dp->table, flow);
1311         ovs_unlock();
1312
1313         reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1314                                         &flow->id, info, false, ufid_flags);
1315         if (likely(reply)) {
1316                 if (likely(!IS_ERR(reply))) {
1317                         rcu_read_lock();        /*To keep RCU checker happy. */
1318                         err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1319                                                      reply, info->snd_portid,
1320                                                      info->snd_seq, 0,
1321                                                      OVS_FLOW_CMD_DEL,
1322                                                      ufid_flags);
1323                         rcu_read_unlock();
1324                         BUG_ON(err < 0);
1325
1326                         ovs_notify(&dp_flow_genl_family, reply, info);
1327                 } else {
1328                         netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
1329                 }
1330         }
1331
1332         ovs_flow_free(flow, true);
1333         return 0;
1334 unlock:
1335         ovs_unlock();
1336         return err;
1337 }
1338
1339 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1340 {
1341         struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1342         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1343         struct table_instance *ti;
1344         struct datapath *dp;
1345         u32 ufid_flags;
1346         int err;
1347
1348         err = genlmsg_parse(cb->nlh, &dp_flow_genl_family, a,
1349                             OVS_FLOW_ATTR_MAX, flow_policy);
1350         if (err)
1351                 return err;
1352         ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1353
1354         rcu_read_lock();
1355         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1356         if (!dp) {
1357                 rcu_read_unlock();
1358                 return -ENODEV;
1359         }
1360
1361         ti = rcu_dereference(dp->table.ti);
1362         for (;;) {
1363                 struct sw_flow *flow;
1364                 u32 bucket, obj;
1365
1366                 bucket = cb->args[0];
1367                 obj = cb->args[1];
1368                 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1369                 if (!flow)
1370                         break;
1371
1372                 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1373                                            NETLINK_CB(cb->skb).portid,
1374                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1375                                            OVS_FLOW_CMD_NEW, ufid_flags) < 0)
1376                         break;
1377
1378                 cb->args[0] = bucket;
1379                 cb->args[1] = obj;
1380         }
1381         rcu_read_unlock();
1382         return skb->len;
1383 }
1384
1385 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1386         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1387         [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1388         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1389         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1390         [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1391         [OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1392         [OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1393 };
1394
1395 static const struct genl_ops dp_flow_genl_ops[] = {
1396         { .cmd = OVS_FLOW_CMD_NEW,
1397           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1398           .policy = flow_policy,
1399           .doit = ovs_flow_cmd_new
1400         },
1401         { .cmd = OVS_FLOW_CMD_DEL,
1402           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1403           .policy = flow_policy,
1404           .doit = ovs_flow_cmd_del
1405         },
1406         { .cmd = OVS_FLOW_CMD_GET,
1407           .flags = 0,               /* OK for unprivileged users. */
1408           .policy = flow_policy,
1409           .doit = ovs_flow_cmd_get,
1410           .dumpit = ovs_flow_cmd_dump
1411         },
1412         { .cmd = OVS_FLOW_CMD_SET,
1413           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1414           .policy = flow_policy,
1415           .doit = ovs_flow_cmd_set,
1416         },
1417 };
1418
1419 static struct genl_family dp_flow_genl_family = {
1420         .id = GENL_ID_GENERATE,
1421         .hdrsize = sizeof(struct ovs_header),
1422         .name = OVS_FLOW_FAMILY,
1423         .version = OVS_FLOW_VERSION,
1424         .maxattr = OVS_FLOW_ATTR_MAX,
1425         .netnsok = true,
1426         .parallel_ops = true,
1427         .ops = dp_flow_genl_ops,
1428         .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1429         .mcgrps = &ovs_dp_flow_multicast_group,
1430         .n_mcgrps = 1,
1431 };
1432
1433 static size_t ovs_dp_cmd_msg_size(void)
1434 {
1435         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1436
1437         msgsize += nla_total_size(IFNAMSIZ);
1438         msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1439         msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1440         msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1441
1442         return msgsize;
1443 }
1444
1445 /* Called with ovs_mutex. */
1446 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1447                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1448 {
1449         struct ovs_header *ovs_header;
1450         struct ovs_dp_stats dp_stats;
1451         struct ovs_dp_megaflow_stats dp_megaflow_stats;
1452         int err;
1453
1454         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1455                                    flags, cmd);
1456         if (!ovs_header)
1457                 goto error;
1458
1459         ovs_header->dp_ifindex = get_dpifindex(dp);
1460
1461         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1462         if (err)
1463                 goto nla_put_failure;
1464
1465         get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1466         if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1467                         &dp_stats))
1468                 goto nla_put_failure;
1469
1470         if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1471                         sizeof(struct ovs_dp_megaflow_stats),
1472                         &dp_megaflow_stats))
1473                 goto nla_put_failure;
1474
1475         if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1476                 goto nla_put_failure;
1477
1478         genlmsg_end(skb, ovs_header);
1479         return 0;
1480
1481 nla_put_failure:
1482         genlmsg_cancel(skb, ovs_header);
1483 error:
1484         return -EMSGSIZE;
1485 }
1486
1487 static struct sk_buff *ovs_dp_cmd_alloc_info(struct genl_info *info)
1488 {
1489         return genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info, GFP_KERNEL);
1490 }
1491
1492 /* Called with rcu_read_lock or ovs_mutex. */
1493 static struct datapath *lookup_datapath(struct net *net,
1494                                         const struct ovs_header *ovs_header,
1495                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1496 {
1497         struct datapath *dp;
1498
1499         if (!a[OVS_DP_ATTR_NAME])
1500                 dp = get_dp(net, ovs_header->dp_ifindex);
1501         else {
1502                 struct vport *vport;
1503
1504                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1505                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1506         }
1507         return dp ? dp : ERR_PTR(-ENODEV);
1508 }
1509
1510 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1511 {
1512         struct datapath *dp;
1513
1514         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1515         if (IS_ERR(dp))
1516                 return;
1517
1518         WARN(dp->user_features, "Dropping previously announced user features\n");
1519         dp->user_features = 0;
1520 }
1521
1522 static void ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1523 {
1524         if (a[OVS_DP_ATTR_USER_FEATURES])
1525                 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1526 }
1527
1528 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1529 {
1530         struct nlattr **a = info->attrs;
1531         struct vport_parms parms;
1532         struct sk_buff *reply;
1533         struct datapath *dp;
1534         struct vport *vport;
1535         struct ovs_net *ovs_net;
1536         int err, i;
1537
1538         err = -EINVAL;
1539         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1540                 goto err;
1541
1542         reply = ovs_dp_cmd_alloc_info(info);
1543         if (!reply)
1544                 return -ENOMEM;
1545
1546         err = -ENOMEM;
1547         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1548         if (dp == NULL)
1549                 goto err_free_reply;
1550
1551         ovs_dp_set_net(dp, sock_net(skb->sk));
1552
1553         /* Allocate table. */
1554         err = ovs_flow_tbl_init(&dp->table);
1555         if (err)
1556                 goto err_free_dp;
1557
1558         dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1559         if (!dp->stats_percpu) {
1560                 err = -ENOMEM;
1561                 goto err_destroy_table;
1562         }
1563
1564         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1565                             GFP_KERNEL);
1566         if (!dp->ports) {
1567                 err = -ENOMEM;
1568                 goto err_destroy_percpu;
1569         }
1570
1571         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1572                 INIT_HLIST_HEAD(&dp->ports[i]);
1573
1574         /* Set up our datapath device. */
1575         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1576         parms.type = OVS_VPORT_TYPE_INTERNAL;
1577         parms.options = NULL;
1578         parms.dp = dp;
1579         parms.port_no = OVSP_LOCAL;
1580         parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1581
1582         ovs_dp_change(dp, a);
1583
1584         /* So far only local changes have been made, now need the lock. */
1585         ovs_lock();
1586
1587         vport = new_vport(&parms);
1588         if (IS_ERR(vport)) {
1589                 err = PTR_ERR(vport);
1590                 if (err == -EBUSY)
1591                         err = -EEXIST;
1592
1593                 if (err == -EEXIST) {
1594                         /* An outdated user space instance that does not understand
1595                          * the concept of user_features has attempted to create a new
1596                          * datapath and is likely to reuse it. Drop all user features.
1597                          */
1598                         if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1599                                 ovs_dp_reset_user_features(skb, info);
1600                 }
1601
1602                 goto err_destroy_ports_array;
1603         }
1604
1605         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1606                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1607         BUG_ON(err < 0);
1608
1609         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1610         list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1611
1612         ovs_unlock();
1613
1614         ovs_notify(&dp_datapath_genl_family, reply, info);
1615         return 0;
1616
1617 err_destroy_ports_array:
1618         ovs_unlock();
1619         kfree(dp->ports);
1620 err_destroy_percpu:
1621         free_percpu(dp->stats_percpu);
1622 err_destroy_table:
1623         ovs_flow_tbl_destroy(&dp->table);
1624 err_free_dp:
1625         kfree(dp);
1626 err_free_reply:
1627         kfree_skb(reply);
1628 err:
1629         return err;
1630 }
1631
1632 /* Called with ovs_mutex. */
1633 static void __dp_destroy(struct datapath *dp)
1634 {
1635         int i;
1636
1637         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1638                 struct vport *vport;
1639                 struct hlist_node *n;
1640
1641                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1642                         if (vport->port_no != OVSP_LOCAL)
1643                                 ovs_dp_detach_port(vport);
1644         }
1645
1646         list_del_rcu(&dp->list_node);
1647
1648         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1649          * all ports in datapath are destroyed first before freeing datapath.
1650          */
1651         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1652
1653         /* RCU destroy the flow table */
1654         call_rcu(&dp->rcu, destroy_dp_rcu);
1655 }
1656
1657 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1658 {
1659         struct sk_buff *reply;
1660         struct datapath *dp;
1661         int err;
1662
1663         reply = ovs_dp_cmd_alloc_info(info);
1664         if (!reply)
1665                 return -ENOMEM;
1666
1667         ovs_lock();
1668         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1669         err = PTR_ERR(dp);
1670         if (IS_ERR(dp))
1671                 goto err_unlock_free;
1672
1673         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1674                                    info->snd_seq, 0, OVS_DP_CMD_DEL);
1675         BUG_ON(err < 0);
1676
1677         __dp_destroy(dp);
1678         ovs_unlock();
1679
1680         ovs_notify(&dp_datapath_genl_family, reply, info);
1681
1682         return 0;
1683
1684 err_unlock_free:
1685         ovs_unlock();
1686         kfree_skb(reply);
1687         return err;
1688 }
1689
1690 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1691 {
1692         struct sk_buff *reply;
1693         struct datapath *dp;
1694         int err;
1695
1696         reply = ovs_dp_cmd_alloc_info(info);
1697         if (!reply)
1698                 return -ENOMEM;
1699
1700         ovs_lock();
1701         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1702         err = PTR_ERR(dp);
1703         if (IS_ERR(dp))
1704                 goto err_unlock_free;
1705
1706         ovs_dp_change(dp, info->attrs);
1707
1708         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1709                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1710         BUG_ON(err < 0);
1711
1712         ovs_unlock();
1713         ovs_notify(&dp_datapath_genl_family, reply, info);
1714
1715         return 0;
1716
1717 err_unlock_free:
1718         ovs_unlock();
1719         kfree_skb(reply);
1720         return err;
1721 }
1722
1723 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1724 {
1725         struct sk_buff *reply;
1726         struct datapath *dp;
1727         int err;
1728
1729         reply = ovs_dp_cmd_alloc_info(info);
1730         if (!reply)
1731                 return -ENOMEM;
1732
1733         ovs_lock();
1734         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1735         if (IS_ERR(dp)) {
1736                 err = PTR_ERR(dp);
1737                 goto err_unlock_free;
1738         }
1739         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1740                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1741         BUG_ON(err < 0);
1742         ovs_unlock();
1743
1744         return genlmsg_reply(reply, info);
1745
1746 err_unlock_free:
1747         ovs_unlock();
1748         kfree_skb(reply);
1749         return err;
1750 }
1751
1752 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1753 {
1754         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1755         struct datapath *dp;
1756         int skip = cb->args[0];
1757         int i = 0;
1758
1759         ovs_lock();
1760         list_for_each_entry(dp, &ovs_net->dps, list_node) {
1761                 if (i >= skip &&
1762                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1763                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1764                                          OVS_DP_CMD_NEW) < 0)
1765                         break;
1766                 i++;
1767         }
1768         ovs_unlock();
1769
1770         cb->args[0] = i;
1771
1772         return skb->len;
1773 }
1774
1775 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1776         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1777         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1778         [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1779 };
1780
1781 static const struct genl_ops dp_datapath_genl_ops[] = {
1782         { .cmd = OVS_DP_CMD_NEW,
1783           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1784           .policy = datapath_policy,
1785           .doit = ovs_dp_cmd_new
1786         },
1787         { .cmd = OVS_DP_CMD_DEL,
1788           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1789           .policy = datapath_policy,
1790           .doit = ovs_dp_cmd_del
1791         },
1792         { .cmd = OVS_DP_CMD_GET,
1793           .flags = 0,               /* OK for unprivileged users. */
1794           .policy = datapath_policy,
1795           .doit = ovs_dp_cmd_get,
1796           .dumpit = ovs_dp_cmd_dump
1797         },
1798         { .cmd = OVS_DP_CMD_SET,
1799           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1800           .policy = datapath_policy,
1801           .doit = ovs_dp_cmd_set,
1802         },
1803 };
1804
1805 static struct genl_family dp_datapath_genl_family = {
1806         .id = GENL_ID_GENERATE,
1807         .hdrsize = sizeof(struct ovs_header),
1808         .name = OVS_DATAPATH_FAMILY,
1809         .version = OVS_DATAPATH_VERSION,
1810         .maxattr = OVS_DP_ATTR_MAX,
1811         .netnsok = true,
1812         .parallel_ops = true,
1813         .ops = dp_datapath_genl_ops,
1814         .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1815         .mcgrps = &ovs_dp_datapath_multicast_group,
1816         .n_mcgrps = 1,
1817 };
1818
1819 /* Called with ovs_mutex or RCU read lock. */
1820 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1821                                    u32 portid, u32 seq, u32 flags, u8 cmd)
1822 {
1823         struct ovs_header *ovs_header;
1824         struct ovs_vport_stats vport_stats;
1825         int err;
1826
1827         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1828                                  flags, cmd);
1829         if (!ovs_header)
1830                 return -EMSGSIZE;
1831
1832         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1833
1834         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1835             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1836             nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1837                            ovs_vport_name(vport)))
1838                 goto nla_put_failure;
1839
1840         ovs_vport_get_stats(vport, &vport_stats);
1841         if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1842                     &vport_stats))
1843                 goto nla_put_failure;
1844
1845         if (ovs_vport_get_upcall_portids(vport, skb))
1846                 goto nla_put_failure;
1847
1848         err = ovs_vport_get_options(vport, skb);
1849         if (err == -EMSGSIZE)
1850                 goto error;
1851
1852         genlmsg_end(skb, ovs_header);
1853         return 0;
1854
1855 nla_put_failure:
1856         err = -EMSGSIZE;
1857 error:
1858         genlmsg_cancel(skb, ovs_header);
1859         return err;
1860 }
1861
1862 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1863 {
1864         return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1865 }
1866
1867 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1868 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1869                                          u32 seq, u8 cmd)
1870 {
1871         struct sk_buff *skb;
1872         int retval;
1873
1874         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1875         if (!skb)
1876                 return ERR_PTR(-ENOMEM);
1877
1878         retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1879         BUG_ON(retval < 0);
1880
1881         return skb;
1882 }
1883
1884 /* Called with ovs_mutex or RCU read lock. */
1885 static struct vport *lookup_vport(struct net *net,
1886                                   const struct ovs_header *ovs_header,
1887                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1888 {
1889         struct datapath *dp;
1890         struct vport *vport;
1891
1892         if (a[OVS_VPORT_ATTR_NAME]) {
1893                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1894                 if (!vport)
1895                         return ERR_PTR(-ENODEV);
1896                 if (ovs_header->dp_ifindex &&
1897                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1898                         return ERR_PTR(-ENODEV);
1899                 return vport;
1900         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1901                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1902
1903                 if (port_no >= DP_MAX_PORTS)
1904                         return ERR_PTR(-EFBIG);
1905
1906                 dp = get_dp(net, ovs_header->dp_ifindex);
1907                 if (!dp)
1908                         return ERR_PTR(-ENODEV);
1909
1910                 vport = ovs_vport_ovsl_rcu(dp, port_no);
1911                 if (!vport)
1912                         return ERR_PTR(-ENODEV);
1913                 return vport;
1914         } else
1915                 return ERR_PTR(-EINVAL);
1916 }
1917
1918 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1919 {
1920         struct nlattr **a = info->attrs;
1921         struct ovs_header *ovs_header = info->userhdr;
1922         struct vport_parms parms;
1923         struct sk_buff *reply;
1924         struct vport *vport;
1925         struct datapath *dp;
1926         u32 port_no;
1927         int err;
1928
1929         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1930             !a[OVS_VPORT_ATTR_UPCALL_PID])
1931                 return -EINVAL;
1932
1933         port_no = a[OVS_VPORT_ATTR_PORT_NO]
1934                 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
1935         if (port_no >= DP_MAX_PORTS)
1936                 return -EFBIG;
1937
1938         reply = ovs_vport_cmd_alloc_info();
1939         if (!reply)
1940                 return -ENOMEM;
1941
1942         ovs_lock();
1943 restart:
1944         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1945         err = -ENODEV;
1946         if (!dp)
1947                 goto exit_unlock_free;
1948
1949         if (port_no) {
1950                 vport = ovs_vport_ovsl(dp, port_no);
1951                 err = -EBUSY;
1952                 if (vport)
1953                         goto exit_unlock_free;
1954         } else {
1955                 for (port_no = 1; ; port_no++) {
1956                         if (port_no >= DP_MAX_PORTS) {
1957                                 err = -EFBIG;
1958                                 goto exit_unlock_free;
1959                         }
1960                         vport = ovs_vport_ovsl(dp, port_no);
1961                         if (!vport)
1962                                 break;
1963                 }
1964         }
1965
1966         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1967         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1968         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1969         parms.dp = dp;
1970         parms.port_no = port_no;
1971         parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
1972
1973         vport = new_vport(&parms);
1974         err = PTR_ERR(vport);
1975         if (IS_ERR(vport)) {
1976                 if (err == -EAGAIN)
1977                         goto restart;
1978                 goto exit_unlock_free;
1979         }
1980
1981         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1982                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1983         BUG_ON(err < 0);
1984         ovs_unlock();
1985
1986         ovs_notify(&dp_vport_genl_family, reply, info);
1987         return 0;
1988
1989 exit_unlock_free:
1990         ovs_unlock();
1991         kfree_skb(reply);
1992         return err;
1993 }
1994
1995 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1996 {
1997         struct nlattr **a = info->attrs;
1998         struct sk_buff *reply;
1999         struct vport *vport;
2000         int err;
2001
2002         reply = ovs_vport_cmd_alloc_info();
2003         if (!reply)
2004                 return -ENOMEM;
2005
2006         ovs_lock();
2007         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2008         err = PTR_ERR(vport);
2009         if (IS_ERR(vport))
2010                 goto exit_unlock_free;
2011
2012         if (a[OVS_VPORT_ATTR_TYPE] &&
2013             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2014                 err = -EINVAL;
2015                 goto exit_unlock_free;
2016         }
2017
2018         if (a[OVS_VPORT_ATTR_OPTIONS]) {
2019                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2020                 if (err)
2021                         goto exit_unlock_free;
2022         }
2023
2024
2025         if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2026                 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2027
2028                 err = ovs_vport_set_upcall_portids(vport, ids);
2029                 if (err)
2030                         goto exit_unlock_free;
2031         }
2032
2033         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2034                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2035         BUG_ON(err < 0);
2036
2037         ovs_unlock();
2038         ovs_notify(&dp_vport_genl_family, reply, info);
2039         return 0;
2040
2041 exit_unlock_free:
2042         ovs_unlock();
2043         kfree_skb(reply);
2044         return err;
2045 }
2046
2047 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2048 {
2049         struct nlattr **a = info->attrs;
2050         struct sk_buff *reply;
2051         struct vport *vport;
2052         int err;
2053
2054         reply = ovs_vport_cmd_alloc_info();
2055         if (!reply)
2056                 return -ENOMEM;
2057
2058         ovs_lock();
2059         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2060         err = PTR_ERR(vport);
2061         if (IS_ERR(vport))
2062                 goto exit_unlock_free;
2063
2064         if (vport->port_no == OVSP_LOCAL) {
2065                 err = -EINVAL;
2066                 goto exit_unlock_free;
2067         }
2068
2069         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2070                                       info->snd_seq, 0, OVS_VPORT_CMD_DEL);
2071         BUG_ON(err < 0);
2072         ovs_dp_detach_port(vport);
2073         ovs_unlock();
2074
2075         ovs_notify(&dp_vport_genl_family, reply, info);
2076         return 0;
2077
2078 exit_unlock_free:
2079         ovs_unlock();
2080         kfree_skb(reply);
2081         return err;
2082 }
2083
2084 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2085 {
2086         struct nlattr **a = info->attrs;
2087         struct ovs_header *ovs_header = info->userhdr;
2088         struct sk_buff *reply;
2089         struct vport *vport;
2090         int err;
2091
2092         reply = ovs_vport_cmd_alloc_info();
2093         if (!reply)
2094                 return -ENOMEM;
2095
2096         rcu_read_lock();
2097         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2098         err = PTR_ERR(vport);
2099         if (IS_ERR(vport))
2100                 goto exit_unlock_free;
2101         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2102                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2103         BUG_ON(err < 0);
2104         rcu_read_unlock();
2105
2106         return genlmsg_reply(reply, info);
2107
2108 exit_unlock_free:
2109         rcu_read_unlock();
2110         kfree_skb(reply);
2111         return err;
2112 }
2113
2114 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2115 {
2116         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2117         struct datapath *dp;
2118         int bucket = cb->args[0], skip = cb->args[1];
2119         int i, j = 0;
2120
2121         rcu_read_lock();
2122         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2123         if (!dp) {
2124                 rcu_read_unlock();
2125                 return -ENODEV;
2126         }
2127         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2128                 struct vport *vport;
2129
2130                 j = 0;
2131                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2132                         if (j >= skip &&
2133                             ovs_vport_cmd_fill_info(vport, skb,
2134                                                     NETLINK_CB(cb->skb).portid,
2135                                                     cb->nlh->nlmsg_seq,
2136                                                     NLM_F_MULTI,
2137                                                     OVS_VPORT_CMD_NEW) < 0)
2138                                 goto out;
2139
2140                         j++;
2141                 }
2142                 skip = 0;
2143         }
2144 out:
2145         rcu_read_unlock();
2146
2147         cb->args[0] = i;
2148         cb->args[1] = j;
2149
2150         return skb->len;
2151 }
2152
2153 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2154         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2155         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2156         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2157         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2158         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
2159         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2160 };
2161
2162 static const struct genl_ops dp_vport_genl_ops[] = {
2163         { .cmd = OVS_VPORT_CMD_NEW,
2164           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2165           .policy = vport_policy,
2166           .doit = ovs_vport_cmd_new
2167         },
2168         { .cmd = OVS_VPORT_CMD_DEL,
2169           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2170           .policy = vport_policy,
2171           .doit = ovs_vport_cmd_del
2172         },
2173         { .cmd = OVS_VPORT_CMD_GET,
2174           .flags = 0,               /* OK for unprivileged users. */
2175           .policy = vport_policy,
2176           .doit = ovs_vport_cmd_get,
2177           .dumpit = ovs_vport_cmd_dump
2178         },
2179         { .cmd = OVS_VPORT_CMD_SET,
2180           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2181           .policy = vport_policy,
2182           .doit = ovs_vport_cmd_set,
2183         },
2184 };
2185
2186 struct genl_family dp_vport_genl_family = {
2187         .id = GENL_ID_GENERATE,
2188         .hdrsize = sizeof(struct ovs_header),
2189         .name = OVS_VPORT_FAMILY,
2190         .version = OVS_VPORT_VERSION,
2191         .maxattr = OVS_VPORT_ATTR_MAX,
2192         .netnsok = true,
2193         .parallel_ops = true,
2194         .ops = dp_vport_genl_ops,
2195         .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2196         .mcgrps = &ovs_dp_vport_multicast_group,
2197         .n_mcgrps = 1,
2198 };
2199
2200 static struct genl_family * const dp_genl_families[] = {
2201         &dp_datapath_genl_family,
2202         &dp_vport_genl_family,
2203         &dp_flow_genl_family,
2204         &dp_packet_genl_family,
2205 };
2206
2207 static void dp_unregister_genl(int n_families)
2208 {
2209         int i;
2210
2211         for (i = 0; i < n_families; i++)
2212                 genl_unregister_family(dp_genl_families[i]);
2213 }
2214
2215 static int dp_register_genl(void)
2216 {
2217         int err;
2218         int i;
2219
2220         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2221
2222                 err = genl_register_family(dp_genl_families[i]);
2223                 if (err)
2224                         goto error;
2225         }
2226
2227         return 0;
2228
2229 error:
2230         dp_unregister_genl(i);
2231         return err;
2232 }
2233
2234 static int __net_init ovs_init_net(struct net *net)
2235 {
2236         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2237
2238         INIT_LIST_HEAD(&ovs_net->dps);
2239         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2240         ovs_ct_init(net);
2241         return 0;
2242 }
2243
2244 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2245                                             struct list_head *head)
2246 {
2247         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2248         struct datapath *dp;
2249
2250         list_for_each_entry(dp, &ovs_net->dps, list_node) {
2251                 int i;
2252
2253                 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2254                         struct vport *vport;
2255
2256                         hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2257                                 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2258                                         continue;
2259
2260                                 if (dev_net(vport->dev) == dnet)
2261                                         list_add(&vport->detach_list, head);
2262                         }
2263                 }
2264         }
2265 }
2266
2267 static void __net_exit ovs_exit_net(struct net *dnet)
2268 {
2269         struct datapath *dp, *dp_next;
2270         struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2271         struct vport *vport, *vport_next;
2272         struct net *net;
2273         LIST_HEAD(head);
2274
2275         ovs_ct_exit(dnet);
2276         ovs_lock();
2277         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2278                 __dp_destroy(dp);
2279
2280         rtnl_lock();
2281         for_each_net(net)
2282                 list_vports_from_net(net, dnet, &head);
2283         rtnl_unlock();
2284
2285         /* Detach all vports from given namespace. */
2286         list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2287                 list_del(&vport->detach_list);
2288                 ovs_dp_detach_port(vport);
2289         }
2290
2291         ovs_unlock();
2292
2293         cancel_work_sync(&ovs_net->dp_notify_work);
2294 }
2295
2296 static struct pernet_operations ovs_net_ops = {
2297         .init = ovs_init_net,
2298         .exit = ovs_exit_net,
2299         .id   = &ovs_net_id,
2300         .size = sizeof(struct ovs_net),
2301 };
2302
2303 static int __init dp_init(void)
2304 {
2305         int err;
2306
2307         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2308
2309         pr_info("Open vSwitch switching datapath\n");
2310
2311         err = action_fifos_init();
2312         if (err)
2313                 goto error;
2314
2315         err = ovs_internal_dev_rtnl_link_register();
2316         if (err)
2317                 goto error_action_fifos_exit;
2318
2319         err = ovs_flow_init();
2320         if (err)
2321                 goto error_unreg_rtnl_link;
2322
2323         err = ovs_vport_init();
2324         if (err)
2325                 goto error_flow_exit;
2326
2327         err = register_pernet_device(&ovs_net_ops);
2328         if (err)
2329                 goto error_vport_exit;
2330
2331         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2332         if (err)
2333                 goto error_netns_exit;
2334
2335         err = ovs_netdev_init();
2336         if (err)
2337                 goto error_unreg_notifier;
2338
2339         err = dp_register_genl();
2340         if (err < 0)
2341                 goto error_unreg_netdev;
2342
2343         return 0;
2344
2345 error_unreg_netdev:
2346         ovs_netdev_exit();
2347 error_unreg_notifier:
2348         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2349 error_netns_exit:
2350         unregister_pernet_device(&ovs_net_ops);
2351 error_vport_exit:
2352         ovs_vport_exit();
2353 error_flow_exit:
2354         ovs_flow_exit();
2355 error_unreg_rtnl_link:
2356         ovs_internal_dev_rtnl_link_unregister();
2357 error_action_fifos_exit:
2358         action_fifos_exit();
2359 error:
2360         return err;
2361 }
2362
2363 static void dp_cleanup(void)
2364 {
2365         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2366         ovs_netdev_exit();
2367         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2368         unregister_pernet_device(&ovs_net_ops);
2369         rcu_barrier();
2370         ovs_vport_exit();
2371         ovs_flow_exit();
2372         ovs_internal_dev_rtnl_link_unregister();
2373         action_fifos_exit();
2374 }
2375
2376 module_init(dp_init);
2377 module_exit(dp_cleanup);
2378
2379 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2380 MODULE_LICENSE("GPL");