datapath/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/version.h>
  40 #include <linux/ethtool.h>
  41 #include <linux/wait.h>
  42 #include <asm/div64.h>
  43 #include <linux/highmem.h>
  44 #include <linux/netfilter_bridge.h>
  45 #include <linux/netfilter_ipv4.h>
  46 #include <linux/inetdevice.h>
  47 #include <linux/list.h>
  48 #include <linux/openvswitch.h>
  49 #include <linux/rculist.h>
  50 #include <linux/dmi.h>
  51 #include <linux/genetlink.h>
  52 #include <net/genetlink.h>
  53 #include <net/genetlink.h>
  54 #include <net/net_namespace.h>
  55 #include <net/netns/generic.h>
  56
  57 #include "datapath.h"
  58 #include "flow.h"
  59 #include "flow_table.h"
  60 #include "flow_netlink.h"
  61 #include "vlan.h"
  62 #include "vport-internal_dev.h"
  63 #include "vport-netdev.h"
  64
  65 int ovs_net_id __read_mostly;
  66
  67 static struct genl_family dp_packet_genl_family;
  68 static struct genl_family dp_flow_genl_family;
  69 static struct genl_family dp_datapath_genl_family;
  70
  71 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
  72         .name = OVS_FLOW_MCGROUP
  73 };
  74
  75 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
  76         .name = OVS_DATAPATH_MCGROUP
  77 };
  78
  79 struct genl_multicast_group ovs_dp_vport_multicast_group = {
  80         .name = OVS_VPORT_MCGROUP
  81 };
  82
  83 /* Check if need to build a reply message.
  84  * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
  85 static bool ovs_must_notify(struct genl_info *info,
  86                             const struct genl_multicast_group *grp)
  87 {
  88         return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
  89                 netlink_has_listeners(genl_info_net(info)->genl_sock, GROUP_ID(grp));
  90 }
  91
  92 static void ovs_notify(struct genl_family *family, struct genl_multicast_group *grp,
  93                        struct sk_buff *skb, struct genl_info *info)
  94 {
  95         genl_notify(family, skb, genl_info_net(info),
  96                     info->snd_portid, GROUP_ID(grp), info->nlhdr, GFP_KERNEL);
  97 }
  98
  99 /**
 100  * DOC: Locking:
 101  *
 102  * All writes e.g. Writes to device state (add/remove datapath, port, set
 103  * operations on vports, etc.), Writes to other state (flow table
 104  * modifications, set miscellaneous datapath parameters, etc.) are protected
 105  * by ovs_lock.
 106  *
 107  * Reads are protected by RCU.
 108  *
 109  * There are a few special cases (mostly stats) that have their own
 110  * synchronization but they nest under all of above and don't interact with
 111  * each other.
 112  *
 113  * The RTNL lock nests inside ovs_mutex.
 114  */
 115
 116 static DEFINE_MUTEX(ovs_mutex);
 117
 118 void ovs_lock(void)
 119 {
 120         mutex_lock(&ovs_mutex);
 121 }
 122
 123 void ovs_unlock(void)
 124 {
 125         mutex_unlock(&ovs_mutex);
 126 }
 127
 128 #ifdef CONFIG_LOCKDEP
 129 int lockdep_ovsl_is_held(void)
 130 {
 131         if (debug_locks)
 132                 return lockdep_is_held(&ovs_mutex);
 133         else
 134                 return 1;
 135 }
 136 #endif
 137
 138 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
 139                              const struct dp_upcall_info *);
 140 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
 141                                   const struct dp_upcall_info *);
 142
 143 /* Must be called with rcu_read_lock or ovs_mutex. */
 144 static struct datapath *get_dp(struct net *net, int dp_ifindex)
 145 {
 146         struct datapath *dp = NULL;
 147         struct net_device *dev;
 148
 149         rcu_read_lock();
 150         dev = dev_get_by_index_rcu(net, dp_ifindex);
 151         if (dev) {
 152                 struct vport *vport = ovs_internal_dev_get_vport(dev);
 153                 if (vport)
 154                         dp = vport->dp;
 155         }
 156         rcu_read_unlock();
 157
 158         return dp;
 159 }
 160
 161 /* Must be called with rcu_read_lock or ovs_mutex. */
 162 const char *ovs_dp_name(const struct datapath *dp)
 163 {
 164         struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
 165         return vport->ops->get_name(vport);
 166 }
 167
 168 static int get_dpifindex(struct datapath *dp)
 169 {
 170         struct vport *local;
 171         int ifindex;
 172
 173         rcu_read_lock();
 174
 175         local = ovs_vport_rcu(dp, OVSP_LOCAL);
 176         if (local)
 177                 ifindex = netdev_vport_priv(local)->dev->ifindex;
 178         else
 179                 ifindex = 0;
 180
 181         rcu_read_unlock();
 182
 183         return ifindex;
 184 }
 185
 186 static void destroy_dp_rcu(struct rcu_head *rcu)
 187 {
 188         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 189
 190         ovs_flow_tbl_destroy(&dp->table);
 191         free_percpu(dp->stats_percpu);
 192         release_net(ovs_dp_get_net(dp));
 193         kfree(dp->ports);
 194         kfree(dp);
 195 }
 196
 197 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
 198                                             u16 port_no)
 199 {
 200         return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
 201 }
 202
 203 /* Called with ovs_mutex or RCU read lock. */
 204 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
 205 {
 206         struct vport *vport;
 207         struct hlist_head *head;
 208
 209         head = vport_hash_bucket(dp, port_no);
 210         hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
 211                 if (vport->port_no == port_no)
 212                         return vport;
 213         }
 214         return NULL;
 215 }
 216
 217 /* Called with ovs_mutex. */
 218 static struct vport *new_vport(const struct vport_parms *parms)
 219 {
 220         struct vport *vport;
 221
 222         vport = ovs_vport_add(parms);
 223         if (!IS_ERR(vport)) {
 224                 struct datapath *dp = parms->dp;
 225                 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
 226
 227                 hlist_add_head_rcu(&vport->dp_hash_node, head);
 228         }
 229         return vport;
 230 }
 231
 232 void ovs_dp_detach_port(struct vport *p)
 233 {
 234         ASSERT_OVSL();
 235
 236         /* First drop references to device. */
 237         hlist_del_rcu(&p->dp_hash_node);
 238
 239         /* Then destroy it. */
 240         ovs_vport_del(p);
 241 }
 242
 243 void ovs_dp_process_packet_with_key(struct sk_buff *skb,
 244                                     struct sw_flow_key *pkt_key,
 245                                     bool recirc)
 246 {
 247         const struct vport *p = OVS_CB(skb)->input_vport;
 248         struct datapath *dp = p->dp;
 249         struct sw_flow *flow;
 250         struct dp_stats_percpu *stats;
 251         u64 *stats_counter;
 252         u32 n_mask_hit;
 253
 254         stats = this_cpu_ptr(dp->stats_percpu);
 255
 256         /* Look up flow. */
 257         flow = ovs_flow_tbl_lookup_stats(&dp->table, pkt_key, skb_get_hash(skb),
 258                                          &n_mask_hit);
 259         if (unlikely(!flow)) {
 260                 struct dp_upcall_info upcall;
 261
 262                 upcall.cmd = OVS_PACKET_CMD_MISS;
 263                 upcall.key = pkt_key;
 264                 upcall.userdata = NULL;
 265                 upcall.portid = ovs_vport_find_upcall_portid(p, skb);
 266                 ovs_dp_upcall(dp, skb, &upcall);
 267                 consume_skb(skb);
 268                 stats_counter = &stats->n_missed;
 269                 goto out;
 270         }
 271
 272         OVS_CB(skb)->pkt_key = pkt_key;
 273         OVS_CB(skb)->flow = flow;
 274
 275         ovs_flow_stats_update(OVS_CB(skb)->flow, pkt_key->tp.flags, skb);
 276         ovs_execute_actions(dp, skb, recirc);
 277         stats_counter = &stats->n_hit;
 278
 279 out:
 280         /* Update datapath statistics. */
 281         u64_stats_update_begin(&stats->sync);
 282         (*stats_counter)++;
 283         stats->n_mask_hit += n_mask_hit;
 284         u64_stats_update_end(&stats->sync);
 285 }
 286
 287 /* Must be called with rcu_read_lock. */
 288 void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
 289 {
 290         int error;
 291         struct sw_flow_key key;
 292
 293         OVS_CB(skb)->input_vport = p;
 294
 295         /* Extract flow from 'skb' into 'key'. */
 296         error = ovs_flow_key_extract(skb, &key);
 297         if (unlikely(error)) {
 298                 kfree_skb(skb);
 299                 return;
 300         }
 301
 302         ovs_dp_process_packet_with_key(skb, &key, false);
 303 }
 304
 305 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
 306                   const struct dp_upcall_info *upcall_info)
 307 {
 308         struct dp_stats_percpu *stats;
 309         int err;
 310
 311         if (upcall_info->portid == 0) {
 312                 err = -ENOTCONN;
 313                 goto err;
 314         }
 315
 316         if (!skb_is_gso(skb))
 317                 err = queue_userspace_packet(dp, skb, upcall_info);
 318         else
 319                 err = queue_gso_packets(dp, skb, upcall_info);
 320         if (err)
 321                 goto err;
 322
 323         return 0;
 324
 325 err:
 326         stats = this_cpu_ptr(dp->stats_percpu);
 327
 328         u64_stats_update_begin(&stats->sync);
 329         stats->n_lost++;
 330         u64_stats_update_end(&stats->sync);
 331
 332         return err;
 333 }
 334
 335 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
 336                              const struct dp_upcall_info *upcall_info)
 337 {
 338         unsigned short gso_type = skb_shinfo(skb)->gso_type;
 339         struct dp_upcall_info later_info;
 340         struct sw_flow_key later_key;
 341         struct sk_buff *segs, *nskb;
 342         struct ovs_skb_cb ovs_cb;
 343         int err;
 344
 345         ovs_cb = *OVS_CB(skb);
 346         segs = __skb_gso_segment(skb, NETIF_F_SG, false);
 347         *OVS_CB(skb) = ovs_cb;
 348         if (IS_ERR(segs))
 349                 return PTR_ERR(segs);
 350         if (segs == NULL)
 351                 return -EINVAL;
 352
 353         /* Queue all of the segments. */
 354         skb = segs;
 355         do {
 356                 *OVS_CB(skb) = ovs_cb;
 357                 err = queue_userspace_packet(dp, skb, upcall_info);
 358                 if (err)
 359                         break;
 360
 361                 if (skb == segs && gso_type & SKB_GSO_UDP) {
 362                         /* The initial flow key extracted by ovs_flow_extract()
 363                          * in this case is for a first fragment, so we need to
 364                          * properly mark later fragments.
 365                          */
 366                         later_key = *upcall_info->key;
 367                         later_key.ip.frag = OVS_FRAG_TYPE_LATER;
 368
 369                         later_info = *upcall_info;
 370                         later_info.key = &later_key;
 371                         upcall_info = &later_info;
 372                 }
 373         } while ((skb = skb->next));
 374
 375         /* Free all of the segments. */
 376         skb = segs;
 377         do {
 378                 nskb = skb->next;
 379                 if (err)
 380                         kfree_skb(skb);
 381                 else
 382                         consume_skb(skb);
 383         } while ((skb = nskb));
 384         return err;
 385 }
 386
 387 static size_t key_attr_size(void)
 388 {
 389         /* Whenever adding new OVS_KEY_ FIELDS, we should consider
 390          * updating this function.  */
 391         BUILD_BUG_ON(OVS_KEY_ATTR_IPV4_TUNNEL != 21);
 392
 393         return    nla_total_size(4)   /* OVS_KEY_ATTR_PRIORITY */
 394                 + nla_total_size(0)   /* OVS_KEY_ATTR_TUNNEL */
 395                   + nla_total_size(8)   /* OVS_TUNNEL_KEY_ATTR_ID */
 396                   + nla_total_size(4)   /* OVS_TUNNEL_KEY_ATTR_IPV4_SRC */
 397                   + nla_total_size(4)   /* OVS_TUNNEL_KEY_ATTR_IPV4_DST */
 398                   + nla_total_size(1)   /* OVS_TUNNEL_KEY_ATTR_TOS */
 399                   + nla_total_size(1)   /* OVS_TUNNEL_KEY_ATTR_TTL */
 400                   + nla_total_size(0)   /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
 401                   + nla_total_size(0)   /* OVS_TUNNEL_KEY_ATTR_CSUM */
 402                 + nla_total_size(4)   /* OVS_KEY_ATTR_IN_PORT */
 403                 + nla_total_size(4)   /* OVS_KEY_ATTR_SKB_MARK */
 404                 + nla_total_size(4)   /* OVS_KEY_ATTR_DP_HASH */
 405                 + nla_total_size(4)   /* OVS_KEY_ATTR_RECIRC_ID */
 406                 + nla_total_size(12)  /* OVS_KEY_ATTR_ETHERNET */
 407                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 408                 + nla_total_size(4)   /* OVS_KEY_ATTR_8021Q */
 409                 + nla_total_size(0)   /* OVS_KEY_ATTR_ENCAP */
 410                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
 411                 + nla_total_size(40)  /* OVS_KEY_ATTR_IPV6 */
 412                 + nla_total_size(2)   /* OVS_KEY_ATTR_ICMPV6 */
 413                 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
 414 }
 415
 416 static size_t upcall_msg_size(const struct nlattr *userdata,
 417                               unsigned int hdrlen)
 418 {
 419         size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
 420                 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
 421                 + nla_total_size(key_attr_size()); /* OVS_PACKET_ATTR_KEY */
 422
 423         /* OVS_PACKET_ATTR_USERDATA */
 424         if (userdata)
 425                 size += NLA_ALIGN(userdata->nla_len);
 426
 427         return size;
 428 }
 429
 430 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
 431                                   const struct dp_upcall_info *upcall_info)
 432 {
 433         struct ovs_header *upcall;
 434         struct sk_buff *nskb = NULL;
 435         struct sk_buff *user_skb; /* to be queued to userspace */
 436         struct nlattr *nla;
 437         struct genl_info info = {
 438 #ifdef HAVE_GENLMSG_NEW_UNICAST
 439                 .dst_sk = ovs_dp_get_net(dp)->genl_sock,
 440 #endif
 441                 .snd_portid = upcall_info->portid,
 442         };
 443         size_t len;
 444         unsigned int hlen;
 445         int err, dp_ifindex;
 446
 447         dp_ifindex = get_dpifindex(dp);
 448         if (!dp_ifindex)
 449                 return -ENODEV;
 450
 451         if (vlan_tx_tag_present(skb)) {
 452                 nskb = skb_clone(skb, GFP_ATOMIC);
 453                 if (!nskb)
 454                         return -ENOMEM;
 455
 456                 nskb = __vlan_put_tag(nskb, nskb->vlan_proto, vlan_tx_tag_get(nskb));
 457                 if (!nskb)
 458                         return -ENOMEM;
 459
 460                 vlan_set_tci(nskb, 0);
 461
 462                 skb = nskb;
 463         }
 464
 465         if (nla_attr_size(skb->len) > USHRT_MAX) {
 466                 err = -EFBIG;
 467                 goto out;
 468         }
 469
 470         /* Complete checksum if needed */
 471         if (skb->ip_summed == CHECKSUM_PARTIAL &&
 472             (err = skb_checksum_help(skb)))
 473                 goto out;
 474
 475         /* Older versions of OVS user space enforce alignment of the last
 476          * Netlink attribute to NLA_ALIGNTO which would require extensive
 477          * padding logic. Only perform zerocopy if padding is not required.
 478          */
 479         if (dp->user_features & OVS_DP_F_UNALIGNED)
 480                 hlen = skb_zerocopy_headlen(skb);
 481         else
 482                 hlen = skb->len;
 483
 484         len = upcall_msg_size(upcall_info->userdata, hlen);
 485         user_skb = genlmsg_new_unicast(len, &info, GFP_ATOMIC);
 486         if (!user_skb) {
 487                 err = -ENOMEM;
 488                 goto out;
 489         }
 490
 491         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 492                              0, upcall_info->cmd);
 493         upcall->dp_ifindex = dp_ifindex;
 494
 495         nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
 496         err = ovs_nla_put_flow(upcall_info->key, upcall_info->key, user_skb);
 497         BUG_ON(err);
 498         nla_nest_end(user_skb, nla);
 499
 500         if (upcall_info->userdata)
 501                 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
 502                           nla_len(upcall_info->userdata),
 503                           nla_data(upcall_info->userdata));
 504
 505         /* Only reserve room for attribute header, packet data is added
 506          * in skb_zerocopy() */
 507         if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
 508                 err = -ENOBUFS;
 509                 goto out;
 510         }
 511         nla->nla_len = nla_attr_size(skb->len);
 512
 513         err = skb_zerocopy(user_skb, skb, skb->len, hlen);
 514         if (err)
 515                 goto out;
 516
 517         /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
 518         if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
 519                 size_t plen = NLA_ALIGN(user_skb->len) - user_skb->len;
 520
 521                 if (plen > 0)
 522                         memset(skb_put(user_skb, plen), 0, plen);
 523         }
 524
 525         ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
 526
 527         err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
 528 out:
 529         if (err)
 530                 skb_tx_error(skb);
 531         kfree_skb(nskb);
 532         return err;
 533 }
 534
 535 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 536 {
 537         struct ovs_header *ovs_header = info->userhdr;
 538         struct nlattr **a = info->attrs;
 539         struct sw_flow_actions *acts;
 540         struct sk_buff *packet;
 541         struct sw_flow *flow;
 542         struct datapath *dp;
 543         struct ethhdr *eth;
 544         struct vport *input_vport;
 545         int len;
 546         int err;
 547
 548         err = -EINVAL;
 549         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 550             !a[OVS_PACKET_ATTR_ACTIONS])
 551                 goto err;
 552
 553         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 554         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 555         err = -ENOMEM;
 556         if (!packet)
 557                 goto err;
 558         skb_reserve(packet, NET_IP_ALIGN);
 559
 560         nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
 561
 562         skb_reset_mac_header(packet);
 563         eth = eth_hdr(packet);
 564
 565         /* Normally, setting the skb 'protocol' field would be handled by a
 566          * call to eth_type_trans(), but it assumes there's a sending
 567          * device, which we may not have. */
 568         if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
 569                 packet->protocol = eth->h_proto;
 570         else
 571                 packet->protocol = htons(ETH_P_802_2);
 572
 573         /* Build an sw_flow for sending this packet. */
 574         flow = ovs_flow_alloc();
 575         err = PTR_ERR(flow);
 576         if (IS_ERR(flow))
 577                 goto err_kfree_skb;
 578
 579         err = ovs_flow_key_extract_userspace(a[OVS_PACKET_ATTR_KEY], packet,
 580                                              &flow->key);
 581         if (err)
 582                 goto err_flow_free;
 583
 584         acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_PACKET_ATTR_ACTIONS]));
 585         err = PTR_ERR(acts);
 586         if (IS_ERR(acts))
 587                 goto err_flow_free;
 588
 589         err = ovs_nla_copy_actions(a[OVS_PACKET_ATTR_ACTIONS],
 590                                    &flow->key, 0, &acts);
 591         rcu_assign_pointer(flow->sf_acts, acts);
 592         if (err)
 593                 goto err_flow_free;
 594
 595         OVS_CB(packet)->flow = flow;
 596         OVS_CB(packet)->pkt_key = &flow->key;
 597         packet->priority = flow->key.phy.priority;
 598         packet->mark = flow->key.phy.skb_mark;
 599
 600         rcu_read_lock();
 601         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 602         err = -ENODEV;
 603         if (!dp)
 604                 goto err_unlock;
 605
 606         input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
 607         if (!input_vport)
 608                 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
 609
 610         if (!input_vport)
 611                 goto err_unlock;
 612
 613         OVS_CB(packet)->input_vport = input_vport;
 614
 615         local_bh_disable();
 616         err = ovs_execute_actions(dp, packet, false);
 617         local_bh_enable();
 618         rcu_read_unlock();
 619
 620         ovs_flow_free(flow, false);
 621         return err;
 622
 623 err_unlock:
 624         rcu_read_unlock();
 625 err_flow_free:
 626         ovs_flow_free(flow, false);
 627 err_kfree_skb:
 628         kfree_skb(packet);
 629 err:
 630         return err;
 631 }
 632
 633 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 634         [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
 635         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 636         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 637 };
 638
 639 static struct genl_ops dp_packet_genl_ops[] = {
 640         { .cmd = OVS_PACKET_CMD_EXECUTE,
 641           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 642           .policy = packet_policy,
 643           .doit = ovs_packet_cmd_execute
 644         }
 645 };
 646
 647 static struct genl_family dp_packet_genl_family = {
 648         .id = GENL_ID_GENERATE,
 649         .hdrsize = sizeof(struct ovs_header),
 650         .name = OVS_PACKET_FAMILY,
 651         .version = OVS_PACKET_VERSION,
 652         .maxattr = OVS_PACKET_ATTR_MAX,
 653         .netnsok = true,
 654         .parallel_ops = true,
 655         .ops = dp_packet_genl_ops,
 656         .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
 657 };
 658
 659 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats,
 660                          struct ovs_dp_megaflow_stats *mega_stats)
 661 {
 662         int i;
 663
 664         memset(mega_stats, 0, sizeof(*mega_stats));
 665
 666         stats->n_flows = ovs_flow_tbl_count(&dp->table);
 667         mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
 668
 669         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 670
 671         for_each_possible_cpu(i) {
 672                 const struct dp_stats_percpu *percpu_stats;
 673                 struct dp_stats_percpu local_stats;
 674                 unsigned int start;
 675
 676                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 677
 678                 do {
 679                         start = u64_stats_fetch_begin_irq(&percpu_stats->sync);
 680                         local_stats = *percpu_stats;
 681                 } while (u64_stats_fetch_retry_irq(&percpu_stats->sync, start));
 682
 683                 stats->n_hit += local_stats.n_hit;
 684                 stats->n_missed += local_stats.n_missed;
 685                 stats->n_lost += local_stats.n_lost;
 686                 mega_stats->n_mask_hit += local_stats.n_mask_hit;
 687         }
 688 }
 689
 690 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
 691 {
 692         return NLMSG_ALIGN(sizeof(struct ovs_header))
 693                 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
 694                 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_MASK */
 695                 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
 696                 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
 697                 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
 698                 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
 699 }
 700
 701 /* Called with ovs_mutex or RCU read lock. */
 702 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
 703                                   struct sk_buff *skb, u32 portid,
 704                                   u32 seq, u32 flags, u8 cmd)
 705 {
 706         const int skb_orig_len = skb->len;
 707         struct nlattr *start;
 708         struct ovs_flow_stats stats;
 709         __be16 tcp_flags;
 710         unsigned long used;
 711         struct ovs_header *ovs_header;
 712         struct nlattr *nla;
 713         int err;
 714
 715         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
 716         if (!ovs_header)
 717                 return -EMSGSIZE;
 718
 719         ovs_header->dp_ifindex = dp_ifindex;
 720
 721         /* Fill flow key. */
 722         nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
 723         if (!nla)
 724                 goto nla_put_failure;
 725
 726         err = ovs_nla_put_flow(&flow->unmasked_key, &flow->unmasked_key, skb);
 727         if (err)
 728                 goto error;
 729         nla_nest_end(skb, nla);
 730
 731         nla = nla_nest_start(skb, OVS_FLOW_ATTR_MASK);
 732         if (!nla)
 733                 goto nla_put_failure;
 734
 735         err = ovs_nla_put_flow(&flow->key, &flow->mask->key, skb);
 736         if (err)
 737                 goto error;
 738
 739         nla_nest_end(skb, nla);
 740
 741         ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
 742
 743         if (used &&
 744             nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
 745                 goto nla_put_failure;
 746
 747         if (stats.n_packets &&
 748             nla_put(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats))
 749                 goto nla_put_failure;
 750
 751         if ((u8)ntohs(tcp_flags) &&
 752              nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
 753                 goto nla_put_failure;
 754
 755         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 756          * this is the first flow to be dumped into 'skb'.  This is unusual for
 757          * Netlink but individual action lists can be longer than
 758          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 759          * The userspace caller can always fetch the actions separately if it
 760          * really wants them.  (Most userspace callers in fact don't care.)
 761          *
 762          * This can only fail for dump operations because the skb is always
 763          * properly sized for single flows.
 764          */
 765         start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
 766         if (start) {
 767                 const struct sw_flow_actions *sf_acts;
 768
 769                 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
 770                 err = ovs_nla_put_actions(sf_acts->actions,
 771                                           sf_acts->actions_len, skb);
 772
 773                 if (!err)
 774                         nla_nest_end(skb, start);
 775                 else {
 776                         if (skb_orig_len)
 777                                 goto error;
 778
 779                         nla_nest_cancel(skb, start);
 780                 }
 781         } else if (skb_orig_len)
 782                 goto nla_put_failure;
 783
 784         return genlmsg_end(skb, ovs_header);
 785
 786 nla_put_failure:
 787         err = -EMSGSIZE;
 788 error:
 789         genlmsg_cancel(skb, ovs_header);
 790         return err;
 791 }
 792
 793 /* May not be called with RCU read lock. */
 794 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
 795                                                struct genl_info *info,
 796                                                bool always)
 797 {
 798         struct sk_buff *skb;
 799
 800         if (!always && !ovs_must_notify(info, &ovs_dp_flow_multicast_group))
 801                 return NULL;
 802
 803         skb = genlmsg_new_unicast(ovs_flow_cmd_msg_size(acts), info, GFP_KERNEL);
 804
 805         if (!skb)
 806                 return ERR_PTR(-ENOMEM);
 807
 808         return skb;
 809 }
 810
 811 /* Called with ovs_mutex. */
 812 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
 813                                                int dp_ifindex,
 814                                                struct genl_info *info, u8 cmd,
 815                                                bool always)
 816 {
 817         struct sk_buff *skb;
 818         int retval;
 819
 820         skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts), info,
 821                                       always);
 822         if (!skb || IS_ERR(skb))
 823                 return skb;
 824
 825         retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
 826                                         info->snd_portid, info->snd_seq, 0,
 827                                         cmd);
 828         BUG_ON(retval < 0);
 829         return skb;
 830 }
 831
 832 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
 833 {
 834         struct nlattr **a = info->attrs;
 835         struct ovs_header *ovs_header = info->userhdr;
 836         struct sw_flow *flow, *new_flow;
 837         struct sw_flow_mask mask;
 838         struct sk_buff *reply;
 839         struct datapath *dp;
 840         struct sw_flow_actions *acts;
 841         struct sw_flow_match match;
 842         int error;
 843
 844         /* Must have key and actions. */
 845         error = -EINVAL;
 846         if (!a[OVS_FLOW_ATTR_KEY])
 847                 goto error;
 848         if (!a[OVS_FLOW_ATTR_ACTIONS])
 849                 goto error;
 850
 851         /* Most of the time we need to allocate a new flow, do it before
 852          * locking. */
 853         new_flow = ovs_flow_alloc();
 854         if (IS_ERR(new_flow)) {
 855                 error = PTR_ERR(new_flow);
 856                 goto error;
 857         }
 858
 859         /* Extract key. */
 860         ovs_match_init(&match, &new_flow->unmasked_key, &mask);
 861         error = ovs_nla_get_match(&match,
 862                                   a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
 863         if (error)
 864                 goto err_kfree_flow;
 865
 866         ovs_flow_mask_key(&new_flow->key, &new_flow->unmasked_key, &mask);
 867
 868         /* Validate actions. */
 869         acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
 870         error = PTR_ERR(acts);
 871         if (IS_ERR(acts))
 872                 goto err_kfree_flow;
 873
 874         error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS], &new_flow->key,
 875                                      0, &acts);
 876         if (error) {
 877                 OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
 878                 goto err_kfree_acts;
 879         }
 880
 881         reply = ovs_flow_cmd_alloc_info(acts, info, false);
 882         if (IS_ERR(reply)) {
 883                 error = PTR_ERR(reply);
 884                 goto err_kfree_acts;
 885         }
 886
 887         ovs_lock();
 888         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
 889         if (unlikely(!dp)) {
 890                 error = -ENODEV;
 891                 goto err_unlock_ovs;
 892         }
 893         /* Check if this is a duplicate flow */
 894         flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->unmasked_key);
 895         if (likely(!flow)) {
 896                 rcu_assign_pointer(new_flow->sf_acts, acts);
 897
 898                 /* Put flow in bucket. */
 899                 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
 900                 if (unlikely(error)) {
 901                         acts = NULL;
 902                         goto err_unlock_ovs;
 903                 }
 904
 905                 if (unlikely(reply)) {
 906                         error = ovs_flow_cmd_fill_info(new_flow,
 907                                                        ovs_header->dp_ifindex,
 908                                                        reply, info->snd_portid,
 909                                                        info->snd_seq, 0,
 910                                                        OVS_FLOW_CMD_NEW);
 911                         BUG_ON(error < 0);
 912                 }
 913                 ovs_unlock();
 914         } else {
 915                 struct sw_flow_actions *old_acts;
 916
 917                 /* Bail out if we're not allowed to modify an existing flow.
 918                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
 919                  * because Generic Netlink treats the latter as a dump
 920                  * request.  We also accept NLM_F_EXCL in case that bug ever
 921                  * gets fixed.
 922                  */
 923                 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
 924                                                          | NLM_F_EXCL))) {
 925                         error = -EEXIST;
 926                         goto err_unlock_ovs;
 927                 }
 928                 /* The unmasked key has to be the same for flow updates. */
 929                 if (unlikely(!ovs_flow_cmp_unmasked_key(flow, &match))) {
 930                         /* Look for any overlapping flow. */
 931                         flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
 932                         if (!flow) {
 933                                 error = -ENOENT;
 934                                 goto err_unlock_ovs;
 935                         }
 936                 }
 937                 /* Update actions. */
 938                 old_acts = ovsl_dereference(flow->sf_acts);
 939                 rcu_assign_pointer(flow->sf_acts, acts);
 940
 941                 if (unlikely(reply)) {
 942                         error = ovs_flow_cmd_fill_info(flow,
 943                                                        ovs_header->dp_ifindex,
 944                                                        reply, info->snd_portid,
 945                                                        info->snd_seq, 0,
 946                                                        OVS_FLOW_CMD_NEW);
 947                         BUG_ON(error < 0);
 948                 }
 949                 ovs_unlock();
 950
 951                 ovs_nla_free_flow_actions(old_acts);
 952                 ovs_flow_free(new_flow, false);
 953         }
 954
 955         if (reply)
 956                 ovs_notify(&dp_flow_genl_family, &ovs_dp_flow_multicast_group, reply, info);
 957         return 0;
 958
 959 err_unlock_ovs:
 960         ovs_unlock();
 961         kfree_skb(reply);
 962 err_kfree_acts:
 963         kfree(acts);
 964 err_kfree_flow:
 965         ovs_flow_free(new_flow, false);
 966 error:
 967         return error;
 968 }
 969
 970 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
 971 {
 972         struct nlattr **a = info->attrs;
 973         struct ovs_header *ovs_header = info->userhdr;
 974         struct sw_flow_key key, masked_key;
 975         struct sw_flow *flow;
 976         struct sw_flow_mask mask;
 977         struct sk_buff *reply = NULL;
 978         struct datapath *dp;
 979         struct sw_flow_actions *old_acts = NULL, *acts = NULL;
 980         struct sw_flow_match match;
 981         int error;
 982
 983         /* Extract key. */
 984         error = -EINVAL;
 985         if (!a[OVS_FLOW_ATTR_KEY])
 986                 goto error;
 987
 988         ovs_match_init(&match, &key, &mask);
 989         error = ovs_nla_get_match(&match,
 990                                   a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
 991         if (error)
 992                 goto error;
 993
 994         /* Validate actions. */
 995         if (a[OVS_FLOW_ATTR_ACTIONS]) {
 996                 acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
 997                 error = PTR_ERR(acts);
 998                 if (IS_ERR(acts))
 999                         goto error;
1000
1001                 ovs_flow_mask_key(&masked_key, &key, &mask);
1002                 error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS],
1003                                              &masked_key, 0, &acts);
1004                 if (error) {
1005                         OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
1006                         goto err_kfree_acts;
1007                 }
1008         }
1009
1010         /* Can allocate before locking if have acts. */
1011         if (acts) {
1012                 reply = ovs_flow_cmd_alloc_info(acts, info, false);
1013                 if (IS_ERR(reply)) {
1014                         error = PTR_ERR(reply);
1015                         goto err_kfree_acts;
1016                 }
1017         }
1018
1019         ovs_lock();
1020         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1021         if (unlikely(!dp)) {
1022                 error = -ENODEV;
1023                 goto err_unlock_ovs;
1024         }
1025         /* Check that the flow exists. */
1026         flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1027         if (unlikely(!flow)) {
1028                 error = -ENOENT;
1029                 goto err_unlock_ovs;
1030         }
1031
1032         /* Update actions, if present. */
1033         if (likely(acts)) {
1034                 old_acts = ovsl_dereference(flow->sf_acts);
1035                 rcu_assign_pointer(flow->sf_acts, acts);
1036
1037                 if (unlikely(reply)) {
1038                         error = ovs_flow_cmd_fill_info(flow,
1039                                                        ovs_header->dp_ifindex,
1040                                                        reply, info->snd_portid,
1041                                                        info->snd_seq, 0,
1042                                                        OVS_FLOW_CMD_NEW);
1043                         BUG_ON(error < 0);
1044                 }
1045         } else {
1046                 /* Could not alloc without acts before locking. */
1047                 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1048                                                 info, OVS_FLOW_CMD_NEW, false);
1049                 if (unlikely(IS_ERR(reply))) {
1050                         error = PTR_ERR(reply);
1051                         goto err_unlock_ovs;
1052                 }
1053         }
1054
1055         /* Clear stats. */
1056         if (a[OVS_FLOW_ATTR_CLEAR])
1057                 ovs_flow_stats_clear(flow);
1058         ovs_unlock();
1059
1060         if (reply)
1061                 ovs_notify(&dp_flow_genl_family, &ovs_dp_flow_multicast_group, reply, info);
1062         if (old_acts)
1063                 ovs_nla_free_flow_actions(old_acts);
1064         return 0;
1065
1066 err_unlock_ovs:
1067         ovs_unlock();
1068         kfree_skb(reply);
1069 err_kfree_acts:
1070         kfree(acts);
1071 error:
1072         return error;
1073 }
1074
1075 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1076 {
1077         struct nlattr **a = info->attrs;
1078         struct ovs_header *ovs_header = info->userhdr;
1079         struct sw_flow_key key;
1080         struct sk_buff *reply;
1081         struct sw_flow *flow;
1082         struct datapath *dp;
1083         struct sw_flow_match match;
1084         int err;
1085
1086         if (!a[OVS_FLOW_ATTR_KEY]) {
1087                 OVS_NLERR("Flow get message rejected, Key attribute missing.\n");
1088                 return -EINVAL;
1089         }
1090
1091         ovs_match_init(&match, &key, NULL);
1092         err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
1093         if (err)
1094                 return err;
1095
1096         ovs_lock();
1097         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1098         if (!dp) {
1099                 err = -ENODEV;
1100                 goto unlock;
1101         }
1102
1103         flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1104         if (!flow) {
1105                 err = -ENOENT;
1106                 goto unlock;
1107         }
1108
1109         reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1110                                         OVS_FLOW_CMD_NEW, true);
1111         if (IS_ERR(reply)) {
1112                 err = PTR_ERR(reply);
1113                 goto unlock;
1114         }
1115
1116         ovs_unlock();
1117         return genlmsg_reply(reply, info);
1118 unlock:
1119         ovs_unlock();
1120         return err;
1121 }
1122
1123 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1124 {
1125         struct nlattr **a = info->attrs;
1126         struct ovs_header *ovs_header = info->userhdr;
1127         struct sw_flow_key key;
1128         struct sk_buff *reply;
1129         struct sw_flow *flow;
1130         struct datapath *dp;
1131         struct sw_flow_match match;
1132         int err;
1133
1134         if (likely(a[OVS_FLOW_ATTR_KEY])) {
1135                 ovs_match_init(&match, &key, NULL);
1136                 err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
1137                 if (unlikely(err))
1138                         return err;
1139         }
1140
1141         ovs_lock();
1142         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1143         if (unlikely(!dp)) {
1144                 err = -ENODEV;
1145                 goto unlock;
1146         }
1147         if (unlikely(!a[OVS_FLOW_ATTR_KEY])) {
1148                 err = ovs_flow_tbl_flush(&dp->table);
1149                 goto unlock;
1150         }
1151         flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1152         if (unlikely(!flow)) {
1153                 err = -ENOENT;
1154                 goto unlock;
1155         }
1156
1157         ovs_flow_tbl_remove(&dp->table, flow);
1158         ovs_unlock();
1159
1160         reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *)flow->sf_acts,
1161                                         info, false);
1162
1163         if (likely(reply)) {
1164                 if (likely(!IS_ERR(reply))) {
1165                         rcu_read_lock(); /* Keep RCU checker happy. */
1166                         err = ovs_flow_cmd_fill_info(flow,
1167                                                      ovs_header->dp_ifindex,
1168                                                      reply, info->snd_portid,
1169                                                      info->snd_seq, 0,
1170                                                      OVS_FLOW_CMD_DEL);
1171                         rcu_read_unlock();
1172                         BUG_ON(err < 0);
1173                         ovs_notify(&dp_flow_genl_family, &ovs_dp_flow_multicast_group, reply, info);
1174                 } else {
1175                         genl_set_err(&dp_flow_genl_family, sock_net(skb->sk), 0,
1176                                      GROUP_ID(&ovs_dp_flow_multicast_group), PTR_ERR(reply));
1177
1178                 }
1179         }
1180
1181         ovs_flow_free(flow, true);
1182         return 0;
1183 unlock:
1184         ovs_unlock();
1185         return err;
1186 }
1187
1188 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1189 {
1190         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1191         struct table_instance *ti;
1192         struct datapath *dp;
1193
1194         rcu_read_lock();
1195         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1196         if (!dp) {
1197                 rcu_read_unlock();
1198                 return -ENODEV;
1199         }
1200
1201         ti = rcu_dereference(dp->table.ti);
1202         for (;;) {
1203                 struct sw_flow *flow;
1204                 u32 bucket, obj;
1205
1206                 bucket = cb->args[0];
1207                 obj = cb->args[1];
1208                 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1209                 if (!flow)
1210                         break;
1211
1212                 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1213                                            NETLINK_CB(cb->skb).portid,
1214                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1215                                            OVS_FLOW_CMD_NEW) < 0)
1216                         break;
1217
1218                 cb->args[0] = bucket;
1219                 cb->args[1] = obj;
1220         }
1221         rcu_read_unlock();
1222         return skb->len;
1223 }
1224
1225 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1226         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1227         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1228         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1229 };
1230
1231 static struct genl_ops dp_flow_genl_ops[] = {
1232         { .cmd = OVS_FLOW_CMD_NEW,
1233           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1234           .policy = flow_policy,
1235           .doit = ovs_flow_cmd_new
1236         },
1237         { .cmd = OVS_FLOW_CMD_DEL,
1238           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1239           .policy = flow_policy,
1240           .doit = ovs_flow_cmd_del
1241         },
1242         { .cmd = OVS_FLOW_CMD_GET,
1243           .flags = 0,               /* OK for unprivileged users. */
1244           .policy = flow_policy,
1245           .doit = ovs_flow_cmd_get,
1246           .dumpit = ovs_flow_cmd_dump
1247         },
1248         { .cmd = OVS_FLOW_CMD_SET,
1249           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1250           .policy = flow_policy,
1251           .doit = ovs_flow_cmd_set,
1252         },
1253 };
1254
1255 static struct genl_family dp_flow_genl_family = {
1256         .id = GENL_ID_GENERATE,
1257         .hdrsize = sizeof(struct ovs_header),
1258         .name = OVS_FLOW_FAMILY,
1259         .version = OVS_FLOW_VERSION,
1260         .maxattr = OVS_FLOW_ATTR_MAX,
1261         .netnsok = true,
1262         .parallel_ops = true,
1263         .ops = dp_flow_genl_ops,
1264         .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1265         .mcgrps = &ovs_dp_flow_multicast_group,
1266         .n_mcgrps = 1,
1267 };
1268
1269 static size_t ovs_dp_cmd_msg_size(void)
1270 {
1271         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1272
1273         msgsize += nla_total_size(IFNAMSIZ);
1274         msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1275         msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1276         msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1277
1278         return msgsize;
1279 }
1280
1281 /* Called with ovs_mutex. */
1282 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1283                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1284 {
1285         struct ovs_header *ovs_header;
1286         struct ovs_dp_stats dp_stats;
1287         struct ovs_dp_megaflow_stats dp_megaflow_stats;
1288         int err;
1289
1290         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1291                                    flags, cmd);
1292         if (!ovs_header)
1293                 goto error;
1294
1295         ovs_header->dp_ifindex = get_dpifindex(dp);
1296
1297         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1298         if (err)
1299                 goto nla_put_failure;
1300
1301         get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1302         if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1303                         &dp_stats))
1304                 goto nla_put_failure;
1305
1306         if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1307                         sizeof(struct ovs_dp_megaflow_stats),
1308                         &dp_megaflow_stats))
1309                 goto nla_put_failure;
1310
1311         if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1312                 goto nla_put_failure;
1313
1314         return genlmsg_end(skb, ovs_header);
1315
1316 nla_put_failure:
1317         genlmsg_cancel(skb, ovs_header);
1318 error:
1319         return -EMSGSIZE;
1320 }
1321
1322 static struct sk_buff *ovs_dp_cmd_alloc_info(struct genl_info *info)
1323 {
1324         return genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info, GFP_KERNEL);
1325 }
1326
1327 /* Called with rcu_read_lock or ovs_mutex. */
1328 static struct datapath *lookup_datapath(struct net *net,
1329                                         struct ovs_header *ovs_header,
1330                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1331 {
1332         struct datapath *dp;
1333
1334         if (!a[OVS_DP_ATTR_NAME])
1335                 dp = get_dp(net, ovs_header->dp_ifindex);
1336         else {
1337                 struct vport *vport;
1338
1339                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1340                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1341         }
1342         return dp ? dp : ERR_PTR(-ENODEV);
1343 }
1344
1345 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1346 {
1347         struct datapath *dp;
1348
1349         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1350         if (IS_ERR(dp))
1351                 return;
1352
1353         WARN(dp->user_features, "Dropping previously announced user features\n");
1354         dp->user_features = 0;
1355 }
1356
1357 static void ovs_dp_change(struct datapath *dp, struct nlattr **a)
1358 {
1359         if (a[OVS_DP_ATTR_USER_FEATURES])
1360                 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1361 }
1362
1363 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1364 {
1365         struct nlattr **a = info->attrs;
1366         struct vport_parms parms;
1367         struct sk_buff *reply;
1368         struct datapath *dp;
1369         struct vport *vport;
1370         struct ovs_net *ovs_net;
1371         int err, i;
1372
1373         err = -EINVAL;
1374         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1375                 goto err;
1376
1377         reply = ovs_dp_cmd_alloc_info(info);
1378         if (!reply)
1379                 return -ENOMEM;
1380
1381         err = -ENOMEM;
1382         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1383         if (dp == NULL)
1384                 goto err_free_reply;
1385
1386         ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1387
1388         /* Allocate table. */
1389         err = ovs_flow_tbl_init(&dp->table);
1390         if (err)
1391                 goto err_free_dp;
1392
1393         dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1394         if (!dp->stats_percpu) {
1395                 err = -ENOMEM;
1396                 goto err_destroy_table;
1397         }
1398
1399         for_each_possible_cpu(i) {
1400                 struct dp_stats_percpu *dpath_stats;
1401                 dpath_stats = per_cpu_ptr(dp->stats_percpu, i);
1402                 u64_stats_init(&dpath_stats->sync);
1403         }
1404
1405         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1406                             GFP_KERNEL);
1407         if (!dp->ports) {
1408                 err = -ENOMEM;
1409                 goto err_destroy_percpu;
1410         }
1411
1412         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1413                 INIT_HLIST_HEAD(&dp->ports[i]);
1414
1415         /* Set up our datapath device. */
1416         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1417         parms.type = OVS_VPORT_TYPE_INTERNAL;
1418         parms.options = NULL;
1419         parms.dp = dp;
1420         parms.port_no = OVSP_LOCAL;
1421         parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1422
1423         ovs_dp_change(dp, a);
1424
1425         /* So far only local changes have been made, now need the lock. */
1426         ovs_lock();
1427
1428         vport = new_vport(&parms);
1429         if (IS_ERR(vport)) {
1430                 err = PTR_ERR(vport);
1431                 if (err == -EBUSY)
1432                         err = -EEXIST;
1433
1434                 if (err == -EEXIST) {
1435                         /* An outdated user space instance that does not understand
1436                          * the concept of user_features has attempted to create a new
1437                          * datapath and is likely to reuse it. Drop all user features.
1438                          */
1439                         if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1440                                 ovs_dp_reset_user_features(skb, info);
1441                 }
1442
1443                 goto err_destroy_ports_array;
1444         }
1445
1446         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1447                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1448         BUG_ON(err < 0);
1449
1450         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1451         list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1452
1453         ovs_unlock();
1454
1455         ovs_notify(&dp_datapath_genl_family, &ovs_dp_datapath_multicast_group, reply, info);
1456         return 0;
1457
1458 err_destroy_ports_array:
1459         ovs_unlock();
1460         kfree(dp->ports);
1461 err_destroy_percpu:
1462         free_percpu(dp->stats_percpu);
1463 err_destroy_table:
1464         ovs_flow_tbl_destroy(&dp->table);
1465 err_free_dp:
1466         release_net(ovs_dp_get_net(dp));
1467         kfree(dp);
1468 err_free_reply:
1469         kfree_skb(reply);
1470 err:
1471         return err;
1472 }
1473
1474 /* Called with ovs_mutex. */
1475 static void __dp_destroy(struct datapath *dp)
1476 {
1477         int i;
1478
1479         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1480                 struct vport *vport;
1481                 struct hlist_node *n;
1482
1483                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1484                         if (vport->port_no != OVSP_LOCAL)
1485                                 ovs_dp_detach_port(vport);
1486         }
1487
1488         list_del_rcu(&dp->list_node);
1489
1490         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1491          * all ports in datapath are destroyed first before freeing datapath.
1492          */
1493         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1494
1495         /* RCU destroy the flow table */
1496         call_rcu(&dp->rcu, destroy_dp_rcu);
1497 }
1498
1499 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1500 {
1501         struct sk_buff *reply;
1502         struct datapath *dp;
1503         int err;
1504
1505         reply = ovs_dp_cmd_alloc_info(info);
1506         if (!reply)
1507                 return -ENOMEM;
1508
1509         ovs_lock();
1510         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1511         err = PTR_ERR(dp);
1512         if (IS_ERR(dp))
1513                 goto err_unlock_free;
1514
1515         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1516                                    info->snd_seq, 0, OVS_DP_CMD_DEL);
1517         BUG_ON(err < 0);
1518
1519         __dp_destroy(dp);
1520
1521         ovs_unlock();
1522         ovs_notify(&dp_datapath_genl_family, &ovs_dp_datapath_multicast_group, reply, info);
1523         return 0;
1524
1525 err_unlock_free:
1526         ovs_unlock();
1527         kfree_skb(reply);
1528         return err;
1529 }
1530
1531 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1532 {
1533         struct sk_buff *reply;
1534         struct datapath *dp;
1535         int err;
1536
1537         reply = ovs_dp_cmd_alloc_info(info);
1538         if (!reply)
1539                 return -ENOMEM;
1540
1541         ovs_lock();
1542         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1543         err = PTR_ERR(dp);
1544         if (IS_ERR(dp))
1545                 goto err_unlock_free;
1546
1547         ovs_dp_change(dp, info->attrs);
1548
1549         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1550                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1551         BUG_ON(err < 0);
1552
1553         ovs_unlock();
1554         ovs_notify(&dp_datapath_genl_family, &ovs_dp_datapath_multicast_group, reply, info);
1555         return 0;
1556
1557 err_unlock_free:
1558         ovs_unlock();
1559         kfree_skb(reply);
1560         return err;
1561 }
1562
1563 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1564 {
1565         struct sk_buff *reply;
1566         struct datapath *dp;
1567         int err;
1568
1569         reply = ovs_dp_cmd_alloc_info(info);
1570         if (!reply)
1571                 return -ENOMEM;
1572
1573         ovs_lock();
1574         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1575         if (IS_ERR(dp)) {
1576                 err = PTR_ERR(dp);
1577                 goto err_unlock_free;
1578         }
1579         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1580                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1581         BUG_ON(err < 0);
1582         ovs_unlock();
1583
1584         return genlmsg_reply(reply, info);
1585
1586 err_unlock_free:
1587         ovs_unlock();
1588         kfree_skb(reply);
1589         return err;
1590 }
1591
1592 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1593 {
1594         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1595         struct datapath *dp;
1596         int skip = cb->args[0];
1597         int i = 0;
1598
1599         ovs_lock();
1600         list_for_each_entry(dp, &ovs_net->dps, list_node) {
1601                 if (i >= skip &&
1602                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1603                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1604                                          OVS_DP_CMD_NEW) < 0)
1605                         break;
1606                 i++;
1607         }
1608         ovs_unlock();
1609
1610         cb->args[0] = i;
1611
1612         return skb->len;
1613 }
1614
1615 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1616         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1617         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1618         [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1619 };
1620
1621 static struct genl_ops dp_datapath_genl_ops[] = {
1622         { .cmd = OVS_DP_CMD_NEW,
1623           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1624           .policy = datapath_policy,
1625           .doit = ovs_dp_cmd_new
1626         },
1627         { .cmd = OVS_DP_CMD_DEL,
1628           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1629           .policy = datapath_policy,
1630           .doit = ovs_dp_cmd_del
1631         },
1632         { .cmd = OVS_DP_CMD_GET,
1633           .flags = 0,               /* OK for unprivileged users. */
1634           .policy = datapath_policy,
1635           .doit = ovs_dp_cmd_get,
1636           .dumpit = ovs_dp_cmd_dump
1637         },
1638         { .cmd = OVS_DP_CMD_SET,
1639           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1640           .policy = datapath_policy,
1641           .doit = ovs_dp_cmd_set,
1642         },
1643 };
1644
1645 static struct genl_family dp_datapath_genl_family = {
1646         .id = GENL_ID_GENERATE,
1647         .hdrsize = sizeof(struct ovs_header),
1648         .name = OVS_DATAPATH_FAMILY,
1649         .version = OVS_DATAPATH_VERSION,
1650         .maxattr = OVS_DP_ATTR_MAX,
1651         .netnsok = true,
1652         .parallel_ops = true,
1653         .ops = dp_datapath_genl_ops,
1654         .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1655         .mcgrps = &ovs_dp_datapath_multicast_group,
1656         .n_mcgrps = 1,
1657 };
1658
1659 /* Called with ovs_mutex or RCU read lock. */
1660 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1661                                    u32 portid, u32 seq, u32 flags, u8 cmd)
1662 {
1663         struct ovs_header *ovs_header;
1664         struct ovs_vport_stats vport_stats;
1665         int err;
1666
1667         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1668                                  flags, cmd);
1669         if (!ovs_header)
1670                 return -EMSGSIZE;
1671
1672         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1673
1674         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1675             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1676             nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)))
1677                 goto nla_put_failure;
1678
1679         ovs_vport_get_stats(vport, &vport_stats);
1680         if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1681                     &vport_stats))
1682                 goto nla_put_failure;
1683
1684         if (ovs_vport_get_upcall_portids(vport, skb))
1685                 goto nla_put_failure;
1686
1687         err = ovs_vport_get_options(vport, skb);
1688         if (err == -EMSGSIZE)
1689                 goto error;
1690
1691         return genlmsg_end(skb, ovs_header);
1692
1693 nla_put_failure:
1694         err = -EMSGSIZE;
1695 error:
1696         genlmsg_cancel(skb, ovs_header);
1697         return err;
1698 }
1699
1700 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1701 {
1702         return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1703 }
1704
1705 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1706 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1707                                          u32 seq, u8 cmd)
1708 {
1709         struct sk_buff *skb;
1710         int retval;
1711
1712         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1713         if (!skb)
1714                 return ERR_PTR(-ENOMEM);
1715
1716         retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1717         BUG_ON(retval < 0);
1718
1719         return skb;
1720 }
1721
1722 /* Called with ovs_mutex or RCU read lock. */
1723 static struct vport *lookup_vport(struct net *net,
1724                                   struct ovs_header *ovs_header,
1725                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1726 {
1727         struct datapath *dp;
1728         struct vport *vport;
1729
1730         if (a[OVS_VPORT_ATTR_NAME]) {
1731                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1732                 if (!vport)
1733                         return ERR_PTR(-ENODEV);
1734                 if (ovs_header->dp_ifindex &&
1735                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1736                         return ERR_PTR(-ENODEV);
1737                 return vport;
1738         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1739                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1740
1741                 if (port_no >= DP_MAX_PORTS)
1742                         return ERR_PTR(-EFBIG);
1743
1744                 dp = get_dp(net, ovs_header->dp_ifindex);
1745                 if (!dp)
1746                         return ERR_PTR(-ENODEV);
1747
1748                 vport = ovs_vport_ovsl_rcu(dp, port_no);
1749                 if (!vport)
1750                         return ERR_PTR(-ENODEV);
1751                 return vport;
1752         } else
1753                 return ERR_PTR(-EINVAL);
1754 }
1755
1756 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1757 {
1758         struct nlattr **a = info->attrs;
1759         struct ovs_header *ovs_header = info->userhdr;
1760         struct vport_parms parms;
1761         struct sk_buff *reply;
1762         struct vport *vport;
1763         struct datapath *dp;
1764         u32 port_no;
1765         int err;
1766
1767         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1768             !a[OVS_VPORT_ATTR_UPCALL_PID])
1769                 return -EINVAL;
1770
1771         port_no = a[OVS_VPORT_ATTR_PORT_NO]
1772                 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
1773         if (port_no >= DP_MAX_PORTS)
1774                 return -EFBIG;
1775
1776         reply = ovs_vport_cmd_alloc_info();
1777         if (!reply)
1778                 return -ENOMEM;
1779
1780         ovs_lock();
1781         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1782         err = -ENODEV;
1783         if (!dp)
1784                 goto exit_unlock_free;
1785
1786         if (port_no) {
1787                 vport = ovs_vport_ovsl(dp, port_no);
1788                 err = -EBUSY;
1789                 if (vport)
1790                         goto exit_unlock_free;
1791         } else {
1792                 for (port_no = 1; ; port_no++) {
1793                         if (port_no >= DP_MAX_PORTS) {
1794                                 err = -EFBIG;
1795                                 goto exit_unlock_free;
1796                         }
1797                         vport = ovs_vport_ovsl(dp, port_no);
1798                         if (!vport)
1799                                 break;
1800                 }
1801         }
1802
1803         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1804         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1805         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1806         parms.dp = dp;
1807         parms.port_no = port_no;
1808         parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
1809
1810         vport = new_vport(&parms);
1811         err = PTR_ERR(vport);
1812         if (IS_ERR(vport))
1813                 goto exit_unlock_free;
1814
1815         err = 0;
1816         if (a[OVS_VPORT_ATTR_STATS])
1817                 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1818
1819         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1820                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1821         BUG_ON(err < 0);
1822         ovs_unlock();
1823
1824         ovs_notify(&dp_vport_genl_family, &ovs_dp_vport_multicast_group, reply, info);
1825         return 0;
1826
1827 exit_unlock_free:
1828         ovs_unlock();
1829         kfree_skb(reply);
1830         return err;
1831 }
1832
1833 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1834 {
1835         struct nlattr **a = info->attrs;
1836         struct sk_buff *reply;
1837         struct vport *vport;
1838         int err;
1839
1840         reply = ovs_vport_cmd_alloc_info();
1841         if (!reply)
1842                 return -ENOMEM;
1843
1844         ovs_lock();
1845         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1846         err = PTR_ERR(vport);
1847         if (IS_ERR(vport))
1848                 goto exit_unlock_free;
1849
1850         if (a[OVS_VPORT_ATTR_TYPE] &&
1851             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
1852                 err = -EINVAL;
1853                 goto exit_unlock_free;
1854         }
1855
1856         if (a[OVS_VPORT_ATTR_OPTIONS]) {
1857                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1858                 if (err)
1859                         goto exit_unlock_free;
1860         }
1861
1862         if (a[OVS_VPORT_ATTR_STATS])
1863                 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1864
1865
1866         if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
1867                 err = ovs_vport_set_upcall_portids(vport,
1868                                                    a[OVS_VPORT_ATTR_UPCALL_PID]);
1869                 if (err)
1870                         goto exit_unlock_free;
1871         }
1872
1873         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1874                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1875         BUG_ON(err < 0);
1876         ovs_unlock();
1877
1878         ovs_notify(&dp_vport_genl_family, &ovs_dp_vport_multicast_group, reply, info);
1879         return 0;
1880
1881 exit_unlock_free:
1882         ovs_unlock();
1883         kfree_skb(reply);
1884         return err;
1885 }
1886
1887 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1888 {
1889         struct nlattr **a = info->attrs;
1890         struct sk_buff *reply;
1891         struct vport *vport;
1892         int err;
1893
1894         reply = ovs_vport_cmd_alloc_info();
1895         if (!reply)
1896                 return -ENOMEM;
1897
1898         ovs_lock();
1899         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1900         err = PTR_ERR(vport);
1901         if (IS_ERR(vport))
1902                 goto exit_unlock_free;
1903
1904         if (vport->port_no == OVSP_LOCAL) {
1905                 err = -EINVAL;
1906                 goto exit_unlock_free;
1907         }
1908
1909         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1910                                       info->snd_seq, 0, OVS_VPORT_CMD_DEL);
1911         BUG_ON(err < 0);
1912         ovs_dp_detach_port(vport);
1913         ovs_unlock();
1914
1915         ovs_notify(&dp_vport_genl_family, &ovs_dp_vport_multicast_group, reply, info);
1916         return 0;
1917
1918 exit_unlock_free:
1919         ovs_unlock();
1920         kfree_skb(reply);
1921         return err;
1922 }
1923
1924 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1925 {
1926         struct nlattr **a = info->attrs;
1927         struct ovs_header *ovs_header = info->userhdr;
1928         struct sk_buff *reply;
1929         struct vport *vport;
1930         int err;
1931
1932         reply = ovs_vport_cmd_alloc_info();
1933         if (!reply)
1934                 return -ENOMEM;
1935
1936         rcu_read_lock();
1937         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1938         err = PTR_ERR(vport);
1939         if (IS_ERR(vport))
1940                 goto exit_unlock_free;
1941         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1942                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1943         BUG_ON(err < 0);
1944         rcu_read_unlock();
1945
1946         return genlmsg_reply(reply, info);
1947
1948 exit_unlock_free:
1949         rcu_read_unlock();
1950         kfree_skb(reply);
1951         return err;
1952 }
1953
1954 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1955 {
1956         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1957         struct datapath *dp;
1958         int bucket = cb->args[0], skip = cb->args[1];
1959         int i, j = 0;
1960
1961         rcu_read_lock();
1962         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1963         if (!dp) {
1964                 rcu_read_unlock();
1965                 return -ENODEV;
1966         }
1967         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
1968                 struct vport *vport;
1969
1970                 j = 0;
1971                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1972                         if (j >= skip &&
1973                             ovs_vport_cmd_fill_info(vport, skb,
1974                                                     NETLINK_CB(cb->skb).portid,
1975                                                     cb->nlh->nlmsg_seq,
1976                                                     NLM_F_MULTI,
1977                                                     OVS_VPORT_CMD_NEW) < 0)
1978                                 goto out;
1979
1980                         j++;
1981                 }
1982                 skip = 0;
1983         }
1984 out:
1985         rcu_read_unlock();
1986
1987         cb->args[0] = i;
1988         cb->args[1] = j;
1989
1990         return skb->len;
1991 }
1992
1993 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1994         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1995         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1996         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1997         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1998         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1999         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2000 };
2001
2002 static struct genl_ops dp_vport_genl_ops[] = {
2003         { .cmd = OVS_VPORT_CMD_NEW,
2004           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2005           .policy = vport_policy,
2006           .doit = ovs_vport_cmd_new
2007         },
2008         { .cmd = OVS_VPORT_CMD_DEL,
2009           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2010           .policy = vport_policy,
2011           .doit = ovs_vport_cmd_del
2012         },
2013         { .cmd = OVS_VPORT_CMD_GET,
2014           .flags = 0,               /* OK for unprivileged users. */
2015           .policy = vport_policy,
2016           .doit = ovs_vport_cmd_get,
2017           .dumpit = ovs_vport_cmd_dump
2018         },
2019         { .cmd = OVS_VPORT_CMD_SET,
2020           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2021           .policy = vport_policy,
2022           .doit = ovs_vport_cmd_set,
2023         },
2024 };
2025
2026 struct genl_family dp_vport_genl_family = {
2027         .id = GENL_ID_GENERATE,
2028         .hdrsize = sizeof(struct ovs_header),
2029         .name = OVS_VPORT_FAMILY,
2030         .version = OVS_VPORT_VERSION,
2031         .maxattr = OVS_VPORT_ATTR_MAX,
2032         .netnsok = true,
2033         .parallel_ops = true,
2034         .ops = dp_vport_genl_ops,
2035         .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2036         .mcgrps = &ovs_dp_vport_multicast_group,
2037         .n_mcgrps = 1,
2038 };
2039
2040 static struct genl_family *dp_genl_families[] = {
2041         &dp_datapath_genl_family,
2042         &dp_vport_genl_family,
2043         &dp_flow_genl_family,
2044         &dp_packet_genl_family,
2045 };
2046
2047 static void dp_unregister_genl(int n_families)
2048 {
2049         int i;
2050
2051         for (i = 0; i < n_families; i++)
2052                 genl_unregister_family(dp_genl_families[i]);
2053 }
2054
2055 static int dp_register_genl(void)
2056 {
2057         int err;
2058         int i;
2059
2060         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2061
2062                 err = genl_register_family(dp_genl_families[i]);
2063                 if (err)
2064                         goto error;
2065         }
2066
2067         return 0;
2068
2069 error:
2070         dp_unregister_genl(i);
2071         return err;
2072 }
2073
2074 static int __net_init ovs_init_net(struct net *net)
2075 {
2076         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2077
2078         INIT_LIST_HEAD(&ovs_net->dps);
2079         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2080         return 0;
2081 }
2082
2083 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2084                                             struct list_head *head)
2085 {
2086         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2087         struct datapath *dp;
2088
2089         list_for_each_entry(dp, &ovs_net->dps, list_node) {
2090                 int i;
2091
2092                 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2093                         struct vport *vport;
2094
2095                         hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2096                                 struct netdev_vport *netdev_vport;
2097
2098                                 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2099                                         continue;
2100
2101                                 netdev_vport = netdev_vport_priv(vport);
2102                                 if (dev_net(netdev_vport->dev) == dnet)
2103                                         list_add(&vport->detach_list, head);
2104                         }
2105                 }
2106         }
2107 }
2108
2109 static void __net_exit ovs_exit_net(struct net *dnet)
2110 {
2111         struct datapath *dp, *dp_next;
2112         struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2113         struct vport *vport, *vport_next;
2114         struct net *net;
2115         LIST_HEAD(head);
2116
2117         ovs_lock();
2118         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2119                 __dp_destroy(dp);
2120
2121         rtnl_lock();
2122         for_each_net(net)
2123                 list_vports_from_net(net, dnet, &head);
2124         rtnl_unlock();
2125
2126         /* Detach all vports from given namespace. */
2127         list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2128                 list_del(&vport->detach_list);
2129                 ovs_dp_detach_port(vport);
2130         }
2131
2132         ovs_unlock();
2133
2134         cancel_work_sync(&ovs_net->dp_notify_work);
2135 }
2136
2137 static struct pernet_operations ovs_net_ops = {
2138         .init = ovs_init_net,
2139         .exit = ovs_exit_net,
2140         .id   = &ovs_net_id,
2141         .size = sizeof(struct ovs_net),
2142 };
2143
2144 DEFINE_COMPAT_PNET_REG_FUNC(device);
2145
2146 static int __init dp_init(void)
2147 {
2148         int err;
2149
2150         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2151
2152         pr_info("Open vSwitch switching datapath %s\n", VERSION);
2153
2154         err = ovs_flow_init();
2155         if (err)
2156                 goto error;
2157
2158         err = ovs_vport_init();
2159         if (err)
2160                 goto error_flow_exit;
2161
2162         err = register_pernet_device(&ovs_net_ops);
2163         if (err)
2164                 goto error_vport_exit;
2165
2166         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2167         if (err)
2168                 goto error_netns_exit;
2169
2170         err = dp_register_genl();
2171         if (err < 0)
2172                 goto error_unreg_notifier;
2173
2174         return 0;
2175
2176 error_unreg_notifier:
2177         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2178 error_netns_exit:
2179         unregister_pernet_device(&ovs_net_ops);
2180 error_vport_exit:
2181         ovs_vport_exit();
2182 error_flow_exit:
2183         ovs_flow_exit();
2184 error:
2185         return err;
2186 }
2187
2188 static void dp_cleanup(void)
2189 {
2190         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2191         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2192         unregister_pernet_device(&ovs_net_ops);
2193         rcu_barrier();
2194         ovs_vport_exit();
2195         ovs_flow_exit();
2196 }
2197
2198 module_init(dp_init);
2199 module_exit(dp_cleanup);
2200
2201 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2202 MODULE_LICENSE("GPL");
2203 MODULE_VERSION(VERSION);