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                              uint32_t cutlen);
 142 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
 143                                   const struct sw_flow_key *,
 144                                   const struct dp_upcall_info *,
 145                                   uint32_t cutlen);
 146
 147 /* Must be called with rcu_read_lock. */
 148 static struct datapath *get_dp_rcu(struct net *net, int dp_ifindex)
 149 {
 150         struct net_device *dev = dev_get_by_index_rcu(net, dp_ifindex);
 151
 152         if (dev) {
 153                 struct vport *vport = ovs_internal_dev_get_vport(dev);
 154                 if (vport)
 155                         return vport->dp;
 156         }
 157
 158         return NULL;
 159 }
 160
 161 /* The caller must hold either ovs_mutex or rcu_read_lock to keep the
 162  * returned dp pointer valid.
 163  */
 164 static inline struct datapath *get_dp(struct net *net, int dp_ifindex)
 165 {
 166         struct datapath *dp;
 167
 168         WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_ovsl_is_held());
 169         rcu_read_lock();
 170         dp = get_dp_rcu(net, dp_ifindex);
 171         rcu_read_unlock();
 172
 173         return dp;
 174 }
 175
 176 /* Must be called with rcu_read_lock or ovs_mutex. */
 177 const char *ovs_dp_name(const struct datapath *dp)
 178 {
 179         struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
 180         return ovs_vport_name(vport);
 181 }
 182
 183 static int get_dpifindex(const struct datapath *dp)
 184 {
 185         struct vport *local;
 186         int ifindex;
 187
 188         rcu_read_lock();
 189
 190         local = ovs_vport_rcu(dp, OVSP_LOCAL);
 191         if (local)
 192                 ifindex = local->dev->ifindex;
 193         else
 194                 ifindex = 0;
 195
 196         rcu_read_unlock();
 197
 198         return ifindex;
 199 }
 200
 201 static void destroy_dp_rcu(struct rcu_head *rcu)
 202 {
 203         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 204
 205         ovs_flow_tbl_destroy(&dp->table);
 206         free_percpu(dp->stats_percpu);
 207         kfree(dp->ports);
 208         kfree(dp);
 209 }
 210
 211 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
 212                                             u16 port_no)
 213 {
 214         return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
 215 }
 216
 217 /* Called with ovs_mutex or RCU read lock. */
 218 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
 219 {
 220         struct vport *vport;
 221         struct hlist_head *head;
 222
 223         head = vport_hash_bucket(dp, port_no);
 224         hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
 225                 if (vport->port_no == port_no)
 226                         return vport;
 227         }
 228         return NULL;
 229 }
 230
 231 /* Called with ovs_mutex. */
 232 static struct vport *new_vport(const struct vport_parms *parms)
 233 {
 234         struct vport *vport;
 235
 236         vport = ovs_vport_add(parms);
 237         if (!IS_ERR(vport)) {
 238                 struct datapath *dp = parms->dp;
 239                 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
 240
 241                 hlist_add_head_rcu(&vport->dp_hash_node, head);
 242         }
 243         return vport;
 244 }
 245
 246 void ovs_dp_detach_port(struct vport *p)
 247 {
 248         ASSERT_OVSL();
 249
 250         /* First drop references to device. */
 251         hlist_del_rcu(&p->dp_hash_node);
 252
 253         /* Then destroy it. */
 254         ovs_vport_del(p);
 255 }
 256
 257 /* Must be called with rcu_read_lock. */
 258 void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
 259 {
 260         const struct vport *p = OVS_CB(skb)->input_vport;
 261         struct datapath *dp = p->dp;
 262         struct sw_flow *flow;
 263         struct sw_flow_actions *sf_acts;
 264         struct dp_stats_percpu *stats;
 265         u64 *stats_counter;
 266         u32 n_mask_hit;
 267
 268         stats = this_cpu_ptr(dp->stats_percpu);
 269
 270         /* Look up flow. */
 271         flow = ovs_flow_tbl_lookup_stats(&dp->table, key, &n_mask_hit);
 272         if (unlikely(!flow)) {
 273                 struct dp_upcall_info upcall;
 274                 int error;
 275
 276                 memset(&upcall, 0, sizeof(upcall));
 277                 upcall.cmd = OVS_PACKET_CMD_MISS;
 278                 upcall.portid = ovs_vport_find_upcall_portid(p, skb);
 279                 upcall.mru = OVS_CB(skb)->mru;
 280                 error = ovs_dp_upcall(dp, skb, key, &upcall, 0);
 281                 if (unlikely(error))
 282                         kfree_skb(skb);
 283                 else
 284                         consume_skb(skb);
 285                 stats_counter = &stats->n_missed;
 286                 goto out;
 287         }
 288
 289         ovs_flow_stats_update(flow, key->tp.flags, skb);
 290         sf_acts = rcu_dereference(flow->sf_acts);
 291         ovs_execute_actions(dp, skb, sf_acts, key);
 292
 293         stats_counter = &stats->n_hit;
 294
 295 out:
 296         /* Update datapath statistics. */
 297         u64_stats_update_begin(&stats->syncp);
 298         (*stats_counter)++;
 299         stats->n_mask_hit += n_mask_hit;
 300         u64_stats_update_end(&stats->syncp);
 301 }
 302
 303 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
 304                   const struct sw_flow_key *key,
 305                   const struct dp_upcall_info *upcall_info,
 306                   uint32_t cutlen)
 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, key, upcall_info, cutlen);
 318         else
 319                 err = queue_gso_packets(dp, skb, key, upcall_info, cutlen);
 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->syncp);
 329         stats->n_lost++;
 330         u64_stats_update_end(&stats->syncp);
 331
 332         return err;
 333 }
 334
 335 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
 336                              const struct sw_flow_key *key,
 337                              const struct dp_upcall_info *upcall_info,
 338                                  uint32_t cutlen)
 339 {
 340         unsigned short gso_type = skb_shinfo(skb)->gso_type;
 341         struct sw_flow_key later_key;
 342         struct sk_buff *segs, *nskb;
 343         int err;
 344
 345         BUILD_BUG_ON(sizeof(*OVS_CB(skb)) > SKB_SGO_CB_OFFSET);
 346         segs = __skb_gso_segment(skb, NETIF_F_SG, false);
 347         if (IS_ERR(segs))
 348                 return PTR_ERR(segs);
 349         if (segs == NULL)
 350                 return -EINVAL;
 351
 352         if (gso_type & SKB_GSO_UDP) {
 353                 /* The initial flow key extracted by ovs_flow_key_extract()
 354                  * in this case is for a first fragment, so we need to
 355                  * properly mark later fragments.
 356                  */
 357                 later_key = *key;
 358                 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
 359         }
 360
 361         /* Queue all of the segments. */
 362         skb = segs;
 363         do {
 364                 if (gso_type & SKB_GSO_UDP && skb != segs)
 365                         key = &later_key;
 366
 367                 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
 368                 if (err)
 369                         break;
 370
 371         } while ((skb = skb->next));
 372
 373         /* Free all of the segments. */
 374         skb = segs;
 375         do {
 376                 nskb = skb->next;
 377                 if (err)
 378                         kfree_skb(skb);
 379                 else
 380                         consume_skb(skb);
 381         } while ((skb = nskb));
 382         return err;
 383 }
 384
 385 static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
 386                               unsigned int hdrlen)
 387 {
 388         size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
 389                 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
 390                 + nla_total_size(ovs_key_attr_size()) /* OVS_PACKET_ATTR_KEY */
 391                 + nla_total_size(sizeof(unsigned int)); /* OVS_PACKET_ATTR_LEN */
 392
 393         /* OVS_PACKET_ATTR_USERDATA */
 394         if (upcall_info->userdata)
 395                 size += NLA_ALIGN(upcall_info->userdata->nla_len);
 396
 397         /* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
 398         if (upcall_info->egress_tun_info)
 399                 size += nla_total_size(ovs_tun_key_attr_size());
 400
 401         /* OVS_PACKET_ATTR_ACTIONS */
 402         if (upcall_info->actions_len)
 403                 size += nla_total_size(upcall_info->actions_len);
 404
 405         /* OVS_PACKET_ATTR_MRU */
 406         if (upcall_info->mru)
 407                 size += nla_total_size(sizeof(upcall_info->mru));
 408
 409         return size;
 410 }
 411
 412 static void pad_packet(struct datapath *dp, struct sk_buff *skb)
 413 {
 414         if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
 415                 size_t plen = NLA_ALIGN(skb->len) - skb->len;
 416
 417                 if (plen > 0)
 418                         memset(skb_put(skb, plen), 0, plen);
 419         }
 420 }
 421
 422 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
 423                                   const struct sw_flow_key *key,
 424                                   const struct dp_upcall_info *upcall_info,
 425                                   uint32_t cutlen)
 426 {
 427         struct ovs_header *upcall;
 428         struct sk_buff *nskb = NULL;
 429         struct sk_buff *user_skb = NULL; /* to be queued to userspace */
 430         struct nlattr *nla;
 431         size_t len;
 432         unsigned int hlen;
 433         int err, dp_ifindex;
 434
 435         dp_ifindex = get_dpifindex(dp);
 436         if (!dp_ifindex)
 437                 return -ENODEV;
 438
 439         if (skb_vlan_tag_present(skb)) {
 440                 nskb = skb_clone(skb, GFP_ATOMIC);
 441                 if (!nskb)
 442                         return -ENOMEM;
 443
 444                 nskb = __vlan_hwaccel_push_inside(nskb);
 445                 if (!nskb)
 446                         return -ENOMEM;
 447
 448                 skb = nskb;
 449         }
 450
 451         if (nla_attr_size(skb->len) > USHRT_MAX) {
 452                 err = -EFBIG;
 453                 goto out;
 454         }
 455
 456         /* Complete checksum if needed */
 457         if (skb->ip_summed == CHECKSUM_PARTIAL &&
 458             (err = skb_checksum_help(skb)))
 459                 goto out;
 460
 461         /* Older versions of OVS user space enforce alignment of the last
 462          * Netlink attribute to NLA_ALIGNTO which would require extensive
 463          * padding logic. Only perform zerocopy if padding is not required.
 464          */
 465         if (dp->user_features & OVS_DP_F_UNALIGNED)
 466                 hlen = skb_zerocopy_headlen(skb);
 467         else
 468                 hlen = skb->len;
 469
 470         len = upcall_msg_size(upcall_info, hlen - cutlen);
 471         user_skb = genlmsg_new(len, GFP_ATOMIC);
 472         if (!user_skb) {
 473                 err = -ENOMEM;
 474                 goto out;
 475         }
 476
 477         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 478                              0, upcall_info->cmd);
 479         upcall->dp_ifindex = dp_ifindex;
 480
 481         err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
 482         BUG_ON(err);
 483
 484         if (upcall_info->userdata)
 485                 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
 486                           nla_len(upcall_info->userdata),
 487                           nla_data(upcall_info->userdata));
 488
 489         if (upcall_info->egress_tun_info) {
 490                 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_EGRESS_TUN_KEY);
 491                 err = ovs_nla_put_tunnel_info(user_skb,
 492                                               upcall_info->egress_tun_info);
 493                 BUG_ON(err);
 494                 nla_nest_end(user_skb, nla);
 495         }
 496
 497         if (upcall_info->actions_len) {
 498                 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_ACTIONS);
 499                 err = ovs_nla_put_actions(upcall_info->actions,
 500                                           upcall_info->actions_len,
 501                                           user_skb);
 502                 if (!err)
 503                         nla_nest_end(user_skb, nla);
 504                 else
 505                         nla_nest_cancel(user_skb, nla);
 506         }
 507
 508         /* Add OVS_PACKET_ATTR_MRU */
 509         if (upcall_info->mru) {
 510                 if (nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU,
 511                                 upcall_info->mru)) {
 512                         err = -ENOBUFS;
 513                         goto out;
 514                 }
 515                 pad_packet(dp, user_skb);
 516         }
 517
 518         /* Add OVS_PACKET_ATTR_LEN when packet is truncated */
 519         if (cutlen > 0) {
 520                 if (nla_put_u32(user_skb, OVS_PACKET_ATTR_LEN,
 521                                 skb->len)) {
 522                         err = -ENOBUFS;
 523                         goto out;
 524                 }
 525                 pad_packet(dp, user_skb);
 526         }
 527
 528         /* Only reserve room for attribute header, packet data is added
 529          * in skb_zerocopy() */
 530         if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
 531                 err = -ENOBUFS;
 532                 goto out;
 533         }
 534         nla->nla_len = nla_attr_size(skb->len - cutlen);
 535
 536         err = skb_zerocopy(user_skb, skb, skb->len - cutlen, hlen);
 537         if (err)
 538                 goto out;
 539
 540         /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
 541         pad_packet(dp, user_skb);
 542
 543         ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
 544
 545         err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
 546         user_skb = NULL;
 547 out:
 548         if (err)
 549                 skb_tx_error(skb);
 550         kfree_skb(user_skb);
 551         kfree_skb(nskb);
 552         return err;
 553 }
 554
 555 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 556 {
 557         struct ovs_header *ovs_header = info->userhdr;
 558         struct net *net = sock_net(skb->sk);
 559         struct nlattr **a = info->attrs;
 560         struct sw_flow_actions *acts;
 561         struct sk_buff *packet;
 562         struct sw_flow *flow;
 563         struct sw_flow_actions *sf_acts;
 564         struct datapath *dp;
 565         struct ethhdr *eth;
 566         struct vport *input_vport;
 567         u16 mru = 0;
 568         int len;
 569         int err;
 570         bool log = !a[OVS_PACKET_ATTR_PROBE];
 571
 572         err = -EINVAL;
 573         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 574             !a[OVS_PACKET_ATTR_ACTIONS])
 575                 goto err;
 576
 577         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 578         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 579         err = -ENOMEM;
 580         if (!packet)
 581                 goto err;
 582         skb_reserve(packet, NET_IP_ALIGN);
 583
 584         nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
 585
 586         skb_reset_mac_header(packet);
 587         eth = eth_hdr(packet);
 588
 589         /* Normally, setting the skb 'protocol' field would be handled by a
 590          * call to eth_type_trans(), but it assumes there's a sending
 591          * device, which we may not have. */
 592         if (eth_proto_is_802_3(eth->h_proto))
 593                 packet->protocol = eth->h_proto;
 594         else
 595                 packet->protocol = htons(ETH_P_802_2);
 596
 597         /* Set packet's mru */
 598         if (a[OVS_PACKET_ATTR_MRU]) {
 599                 mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
 600                 packet->ignore_df = 1;
 601         }
 602         OVS_CB(packet)->mru = mru;
 603
 604         /* Build an sw_flow for sending this packet. */
 605         flow = ovs_flow_alloc();
 606         err = PTR_ERR(flow);
 607         if (IS_ERR(flow))
 608                 goto err_kfree_skb;
 609
 610         err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
 611                                              packet, &flow->key, log);
 612         if (err)
 613                 goto err_flow_free;
 614
 615         err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
 616                                    &flow->key, &acts, log);
 617         if (err)
 618                 goto err_flow_free;
 619
 620         rcu_assign_pointer(flow->sf_acts, acts);
 621         packet->priority = flow->key.phy.priority;
 622         packet->mark = flow->key.phy.skb_mark;
 623
 624         rcu_read_lock();
 625         dp = get_dp_rcu(net, ovs_header->dp_ifindex);
 626         err = -ENODEV;
 627         if (!dp)
 628                 goto err_unlock;
 629
 630         input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
 631         if (!input_vport)
 632                 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
 633
 634         if (!input_vport)
 635                 goto err_unlock;
 636
 637         packet->dev = input_vport->dev;
 638         OVS_CB(packet)->input_vport = input_vport;
 639         sf_acts = rcu_dereference(flow->sf_acts);
 640
 641         local_bh_disable();
 642         err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
 643         local_bh_enable();
 644         rcu_read_unlock();
 645
 646         ovs_flow_free(flow, false);
 647         return err;
 648
 649 err_unlock:
 650         rcu_read_unlock();
 651 err_flow_free:
 652         ovs_flow_free(flow, false);
 653 err_kfree_skb:
 654         kfree_skb(packet);
 655 err:
 656         return err;
 657 }
 658
 659 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 660         [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
 661         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 662         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 663         [OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
 664         [OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
 665 };
 666
 667 static const struct genl_ops dp_packet_genl_ops[] = {
 668         { .cmd = OVS_PACKET_CMD_EXECUTE,
 669           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 670           .policy = packet_policy,
 671           .doit = ovs_packet_cmd_execute
 672         }
 673 };
 674
 675 static struct genl_family dp_packet_genl_family = {
 676         .id = GENL_ID_GENERATE,
 677         .hdrsize = sizeof(struct ovs_header),
 678         .name = OVS_PACKET_FAMILY,
 679         .version = OVS_PACKET_VERSION,
 680         .maxattr = OVS_PACKET_ATTR_MAX,
 681         .netnsok = true,
 682         .parallel_ops = true,
 683         .ops = dp_packet_genl_ops,
 684         .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
 685 };
 686
 687 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
 688                          struct ovs_dp_megaflow_stats *mega_stats)
 689 {
 690         int i;
 691
 692         memset(mega_stats, 0, sizeof(*mega_stats));
 693
 694         stats->n_flows = ovs_flow_tbl_count(&dp->table);
 695         mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
 696
 697         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 698
 699         for_each_possible_cpu(i) {
 700                 const struct dp_stats_percpu *percpu_stats;
 701                 struct dp_stats_percpu local_stats;
 702                 unsigned int start;
 703
 704                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 705
 706                 do {
 707                         start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
 708                         local_stats = *percpu_stats;
 709                 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
 710
 711                 stats->n_hit += local_stats.n_hit;
 712                 stats->n_missed += local_stats.n_missed;
 713                 stats->n_lost += local_stats.n_lost;
 714                 mega_stats->n_mask_hit += local_stats.n_mask_hit;
 715         }
 716 }
 717
 718 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
 719 {
 720         return ovs_identifier_is_ufid(sfid) &&
 721                !(ufid_flags & OVS_UFID_F_OMIT_KEY);
 722 }
 723
 724 static bool should_fill_mask(uint32_t ufid_flags)
 725 {
 726         return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
 727 }
 728
 729 static bool should_fill_actions(uint32_t ufid_flags)
 730 {
 731         return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
 732 }
 733
 734 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
 735                                     const struct sw_flow_id *sfid,
 736                                     uint32_t ufid_flags)
 737 {
 738         size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
 739
 740         /* OVS_FLOW_ATTR_UFID */
 741         if (sfid && ovs_identifier_is_ufid(sfid))
 742                 len += nla_total_size(sfid->ufid_len);
 743
 744         /* OVS_FLOW_ATTR_KEY */
 745         if (!sfid || should_fill_key(sfid, ufid_flags))
 746                 len += nla_total_size(ovs_key_attr_size());
 747
 748         /* OVS_FLOW_ATTR_MASK */
 749         if (should_fill_mask(ufid_flags))
 750                 len += nla_total_size(ovs_key_attr_size());
 751
 752         /* OVS_FLOW_ATTR_ACTIONS */
 753         if (should_fill_actions(ufid_flags))
 754                 len += nla_total_size(acts->orig_len);
 755
 756         return len
 757                 + nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
 758                 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
 759                 + nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */
 760 }
 761
 762 /* Called with ovs_mutex or RCU read lock. */
 763 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
 764                                    struct sk_buff *skb)
 765 {
 766         struct ovs_flow_stats stats;
 767         __be16 tcp_flags;
 768         unsigned long used;
 769
 770         ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
 771
 772         if (used &&
 773             nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
 774                               OVS_FLOW_ATTR_PAD))
 775                 return -EMSGSIZE;
 776
 777         if (stats.n_packets &&
 778             nla_put_64bit(skb, OVS_FLOW_ATTR_STATS,
 779                           sizeof(struct ovs_flow_stats), &stats,
 780                           OVS_FLOW_ATTR_PAD))
 781                 return -EMSGSIZE;
 782
 783         if ((u8)ntohs(tcp_flags) &&
 784              nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
 785                 return -EMSGSIZE;
 786
 787         return 0;
 788 }
 789
 790 /* Called with ovs_mutex or RCU read lock. */
 791 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
 792                                      struct sk_buff *skb, int skb_orig_len)
 793 {
 794         struct nlattr *start;
 795         int err;
 796
 797         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 798          * this is the first flow to be dumped into 'skb'.  This is unusual for
 799          * Netlink but individual action lists can be longer than
 800          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 801          * The userspace caller can always fetch the actions separately if it
 802          * really wants them.  (Most userspace callers in fact don't care.)
 803          *
 804          * This can only fail for dump operations because the skb is always
 805          * properly sized for single flows.
 806          */
 807         start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
 808         if (start) {
 809                 const struct sw_flow_actions *sf_acts;
 810
 811                 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
 812                 err = ovs_nla_put_actions(sf_acts->actions,
 813                                           sf_acts->actions_len, skb);
 814
 815                 if (!err)
 816                         nla_nest_end(skb, start);
 817                 else {
 818                         if (skb_orig_len)
 819                                 return err;
 820
 821                         nla_nest_cancel(skb, start);
 822                 }
 823         } else if (skb_orig_len) {
 824                 return -EMSGSIZE;
 825         }
 826
 827         return 0;
 828 }
 829
 830 /* Called with ovs_mutex or RCU read lock. */
 831 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
 832                                   struct sk_buff *skb, u32 portid,
 833                                   u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
 834 {
 835         const int skb_orig_len = skb->len;
 836         struct ovs_header *ovs_header;
 837         int err;
 838
 839         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
 840                                  flags, cmd);
 841         if (!ovs_header)
 842                 return -EMSGSIZE;
 843
 844         ovs_header->dp_ifindex = dp_ifindex;
 845
 846         err = ovs_nla_put_identifier(flow, skb);
 847         if (err)
 848                 goto error;
 849
 850         if (should_fill_key(&flow->id, ufid_flags)) {
 851                 err = ovs_nla_put_masked_key(flow, skb);
 852                 if (err)
 853                         goto error;
 854         }
 855
 856         if (should_fill_mask(ufid_flags)) {
 857                 err = ovs_nla_put_mask(flow, skb);
 858                 if (err)
 859                         goto error;
 860         }
 861
 862         err = ovs_flow_cmd_fill_stats(flow, skb);
 863         if (err)
 864                 goto error;
 865
 866         if (should_fill_actions(ufid_flags)) {
 867                 err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
 868                 if (err)
 869                         goto error;
 870         }
 871
 872         genlmsg_end(skb, ovs_header);
 873         return 0;
 874
 875 error:
 876         genlmsg_cancel(skb, ovs_header);
 877         return err;
 878 }
 879
 880 /* May not be called with RCU read lock. */
 881 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
 882                                                const struct sw_flow_id *sfid,
 883                                                struct genl_info *info,
 884                                                bool always,
 885                                                uint32_t ufid_flags)
 886 {
 887         struct sk_buff *skb;
 888         size_t len;
 889
 890         if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
 891                 return NULL;
 892
 893         len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
 894         skb = genlmsg_new(len, GFP_KERNEL);
 895         if (!skb)
 896                 return ERR_PTR(-ENOMEM);
 897
 898         return skb;
 899 }
 900
 901 /* Called with ovs_mutex. */
 902 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
 903                                                int dp_ifindex,
 904                                                struct genl_info *info, u8 cmd,
 905                                                bool always, u32 ufid_flags)
 906 {
 907         struct sk_buff *skb;
 908         int retval;
 909
 910         skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
 911                                       &flow->id, info, always, ufid_flags);
 912         if (IS_ERR_OR_NULL(skb))
 913                 return skb;
 914
 915         retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
 916                                         info->snd_portid, info->snd_seq, 0,
 917                                         cmd, ufid_flags);
 918         BUG_ON(retval < 0);
 919         return skb;
 920 }
 921
 922 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
 923 {
 924         struct net *net = sock_net(skb->sk);
 925         struct nlattr **a = info->attrs;
 926         struct ovs_header *ovs_header = info->userhdr;
 927         struct sw_flow *flow = NULL, *new_flow;
 928         struct sw_flow_mask mask;
 929         struct sk_buff *reply;
 930         struct datapath *dp;
 931         struct sw_flow_actions *acts;
 932         struct sw_flow_match match;
 933         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
 934         int error;
 935         bool log = !a[OVS_FLOW_ATTR_PROBE];
 936
 937         /* Must have key and actions. */
 938         error = -EINVAL;
 939         if (!a[OVS_FLOW_ATTR_KEY]) {
 940                 OVS_NLERR(log, "Flow key attr not present in new flow.");
 941                 goto error;
 942         }
 943         if (!a[OVS_FLOW_ATTR_ACTIONS]) {
 944                 OVS_NLERR(log, "Flow actions attr not present in new flow.");
 945                 goto error;
 946         }
 947
 948         /* Most of the time we need to allocate a new flow, do it before
 949          * locking.
 950          */
 951         new_flow = ovs_flow_alloc();
 952         if (IS_ERR(new_flow)) {
 953                 error = PTR_ERR(new_flow);
 954                 goto error;
 955         }
 956
 957         /* Extract key. */
 958         ovs_match_init(&match, &new_flow->key, false, &mask);
 959         error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
 960                                   a[OVS_FLOW_ATTR_MASK], log);
 961         if (error)
 962                 goto err_kfree_flow;
 963
 964         /* Extract flow identifier. */
 965         error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
 966                                        &new_flow->key, log);
 967         if (error)
 968                 goto err_kfree_flow;
 969
 970         /* unmasked key is needed to match when ufid is not used. */
 971         if (ovs_identifier_is_key(&new_flow->id))
 972                 match.key = new_flow->id.unmasked_key;
 973
 974         ovs_flow_mask_key(&new_flow->key, &new_flow->key, true, &mask);
 975
 976         /* Validate actions. */
 977         error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
 978                                      &new_flow->key, &acts, log);
 979         if (error) {
 980                 OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
 981                 goto err_kfree_flow;
 982         }
 983
 984         reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
 985                                         ufid_flags);
 986         if (IS_ERR(reply)) {
 987                 error = PTR_ERR(reply);
 988                 goto err_kfree_acts;
 989         }
 990
 991         ovs_lock();
 992         dp = get_dp(net, ovs_header->dp_ifindex);
 993         if (unlikely(!dp)) {
 994                 error = -ENODEV;
 995                 goto err_unlock_ovs;
 996         }
 997
 998         /* Check if this is a duplicate flow */
 999         if (ovs_identifier_is_ufid(&new_flow->id))
1000                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
1001         if (!flow)
1002                 flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->key);
1003         if (likely(!flow)) {
1004                 rcu_assign_pointer(new_flow->sf_acts, acts);
1005
1006                 /* Put flow in bucket. */
1007                 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
1008                 if (unlikely(error)) {
1009                         acts = NULL;
1010                         goto err_unlock_ovs;
1011                 }
1012
1013                 if (unlikely(reply)) {
1014                         error = ovs_flow_cmd_fill_info(new_flow,
1015                                                        ovs_header->dp_ifindex,
1016                                                        reply, info->snd_portid,
1017                                                        info->snd_seq, 0,
1018                                                        OVS_FLOW_CMD_NEW,
1019                                                        ufid_flags);
1020                         BUG_ON(error < 0);
1021                 }
1022                 ovs_unlock();
1023         } else {
1024                 struct sw_flow_actions *old_acts;
1025
1026                 /* Bail out if we're not allowed to modify an existing flow.
1027                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1028                  * because Generic Netlink treats the latter as a dump
1029                  * request.  We also accept NLM_F_EXCL in case that bug ever
1030                  * gets fixed.
1031                  */
1032                 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1033                                                          | NLM_F_EXCL))) {
1034                         error = -EEXIST;
1035                         goto err_unlock_ovs;
1036                 }
1037                 /* The flow identifier has to be the same for flow updates.
1038                  * Look for any overlapping flow.
1039                  */
1040                 if (unlikely(!ovs_flow_cmp(flow, &match))) {
1041                         if (ovs_identifier_is_key(&flow->id))
1042                                 flow = ovs_flow_tbl_lookup_exact(&dp->table,
1043                                                                  &match);
1044                         else /* UFID matches but key is different */
1045                                 flow = NULL;
1046                         if (!flow) {
1047                                 error = -ENOENT;
1048                                 goto err_unlock_ovs;
1049                         }
1050                 }
1051                 /* Update actions. */
1052                 old_acts = ovsl_dereference(flow->sf_acts);
1053                 rcu_assign_pointer(flow->sf_acts, acts);
1054
1055                 if (unlikely(reply)) {
1056                         error = ovs_flow_cmd_fill_info(flow,
1057                                                        ovs_header->dp_ifindex,
1058                                                        reply, info->snd_portid,
1059                                                        info->snd_seq, 0,
1060                                                        OVS_FLOW_CMD_NEW,
1061                                                        ufid_flags);
1062                         BUG_ON(error < 0);
1063                 }
1064                 ovs_unlock();
1065
1066                 ovs_nla_free_flow_actions_rcu(old_acts);
1067                 ovs_flow_free(new_flow, false);
1068         }
1069
1070         if (reply)
1071                 ovs_notify(&dp_flow_genl_family, reply, info);
1072         return 0;
1073
1074 err_unlock_ovs:
1075         ovs_unlock();
1076         kfree_skb(reply);
1077 err_kfree_acts:
1078         ovs_nla_free_flow_actions(acts);
1079 err_kfree_flow:
1080         ovs_flow_free(new_flow, false);
1081 error:
1082         return error;
1083 }
1084
1085 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1086 static struct sw_flow_actions *get_flow_actions(struct net *net,
1087                                                 const struct nlattr *a,
1088                                                 const struct sw_flow_key *key,
1089                                                 const struct sw_flow_mask *mask,
1090                                                 bool log)
1091 {
1092         struct sw_flow_actions *acts;
1093         struct sw_flow_key masked_key;
1094         int error;
1095
1096         ovs_flow_mask_key(&masked_key, key, true, mask);
1097         error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1098         if (error) {
1099                 OVS_NLERR(log,
1100                           "Actions may not be safe on all matching packets");
1101                 return ERR_PTR(error);
1102         }
1103
1104         return acts;
1105 }
1106
1107 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1108 {
1109         struct net *net = sock_net(skb->sk);
1110         struct nlattr **a = info->attrs;
1111         struct ovs_header *ovs_header = info->userhdr;
1112         struct sw_flow_key key;
1113         struct sw_flow *flow;
1114         struct sw_flow_mask mask;
1115         struct sk_buff *reply = NULL;
1116         struct datapath *dp;
1117         struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1118         struct sw_flow_match match;
1119         struct sw_flow_id sfid;
1120         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1121         int error = 0;
1122         bool log = !a[OVS_FLOW_ATTR_PROBE];
1123         bool ufid_present;
1124
1125         ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1126         if (a[OVS_FLOW_ATTR_KEY]) {
1127                 ovs_match_init(&match, &key, true, &mask);
1128                 error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1129                                           a[OVS_FLOW_ATTR_MASK], log);
1130         } else if (!ufid_present) {
1131                 OVS_NLERR(log,
1132                           "Flow set message rejected, Key attribute missing.");
1133                 error = -EINVAL;
1134         }
1135         if (error)
1136                 goto error;
1137
1138         /* Validate actions. */
1139         if (a[OVS_FLOW_ATTR_ACTIONS]) {
1140                 if (!a[OVS_FLOW_ATTR_KEY]) {
1141                         OVS_NLERR(log,
1142                                   "Flow key attribute not present in set flow.");
1143                         error = -EINVAL;
1144                         goto error;
1145                 }
1146
1147                 acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], &key,
1148                                         &mask, log);
1149                 if (IS_ERR(acts)) {
1150                         error = PTR_ERR(acts);
1151                         goto error;
1152                 }
1153
1154                 /* Can allocate before locking if have acts. */
1155                 reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1156                                                 ufid_flags);
1157                 if (IS_ERR(reply)) {
1158                         error = PTR_ERR(reply);
1159                         goto err_kfree_acts;
1160                 }
1161         }
1162
1163         ovs_lock();
1164         dp = get_dp(net, ovs_header->dp_ifindex);
1165         if (unlikely(!dp)) {
1166                 error = -ENODEV;
1167                 goto err_unlock_ovs;
1168         }
1169         /* Check that the flow exists. */
1170         if (ufid_present)
1171                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1172         else
1173                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1174         if (unlikely(!flow)) {
1175                 error = -ENOENT;
1176                 goto err_unlock_ovs;
1177         }
1178
1179         /* Update actions, if present. */
1180         if (likely(acts)) {
1181                 old_acts = ovsl_dereference(flow->sf_acts);
1182                 rcu_assign_pointer(flow->sf_acts, acts);
1183
1184                 if (unlikely(reply)) {
1185                         error = ovs_flow_cmd_fill_info(flow,
1186                                                        ovs_header->dp_ifindex,
1187                                                        reply, info->snd_portid,
1188                                                        info->snd_seq, 0,
1189                                                        OVS_FLOW_CMD_NEW,
1190                                                        ufid_flags);
1191                         BUG_ON(error < 0);
1192                 }
1193         } else {
1194                 /* Could not alloc without acts before locking. */
1195                 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1196                                                 info, OVS_FLOW_CMD_NEW, false,
1197                                                 ufid_flags);
1198
1199                 if (IS_ERR(reply)) {
1200                         error = PTR_ERR(reply);
1201                         goto err_unlock_ovs;
1202                 }
1203         }
1204
1205         /* Clear stats. */
1206         if (a[OVS_FLOW_ATTR_CLEAR])
1207                 ovs_flow_stats_clear(flow);
1208         ovs_unlock();
1209
1210         if (reply)
1211                 ovs_notify(&dp_flow_genl_family, reply, info);
1212         if (old_acts)
1213                 ovs_nla_free_flow_actions_rcu(old_acts);
1214
1215         return 0;
1216
1217 err_unlock_ovs:
1218         ovs_unlock();
1219         kfree_skb(reply);
1220 err_kfree_acts:
1221         ovs_nla_free_flow_actions(acts);
1222 error:
1223         return error;
1224 }
1225
1226 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1227 {
1228         struct nlattr **a = info->attrs;
1229         struct ovs_header *ovs_header = info->userhdr;
1230         struct net *net = sock_net(skb->sk);
1231         struct sw_flow_key key;
1232         struct sk_buff *reply;
1233         struct sw_flow *flow;
1234         struct datapath *dp;
1235         struct sw_flow_match match;
1236         struct sw_flow_id ufid;
1237         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1238         int err = 0;
1239         bool log = !a[OVS_FLOW_ATTR_PROBE];
1240         bool ufid_present;
1241
1242         ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1243         if (a[OVS_FLOW_ATTR_KEY]) {
1244                 ovs_match_init(&match, &key, true, NULL);
1245                 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1246                                         log);
1247         } else if (!ufid_present) {
1248                 OVS_NLERR(log,
1249                           "Flow get message rejected, Key attribute missing.");
1250                 err = -EINVAL;
1251         }
1252         if (err)
1253                 return err;
1254
1255         ovs_lock();
1256         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1257         if (!dp) {
1258                 err = -ENODEV;
1259                 goto unlock;
1260         }
1261
1262         if (ufid_present)
1263                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1264         else
1265                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1266         if (!flow) {
1267                 err = -ENOENT;
1268                 goto unlock;
1269         }
1270
1271         reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1272                                         OVS_FLOW_CMD_NEW, true, ufid_flags);
1273         if (IS_ERR(reply)) {
1274                 err = PTR_ERR(reply);
1275                 goto unlock;
1276         }
1277
1278         ovs_unlock();
1279         return genlmsg_reply(reply, info);
1280 unlock:
1281         ovs_unlock();
1282         return err;
1283 }
1284
1285 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1286 {
1287         struct nlattr **a = info->attrs;
1288         struct ovs_header *ovs_header = info->userhdr;
1289         struct net *net = sock_net(skb->sk);
1290         struct sw_flow_key key;
1291         struct sk_buff *reply;
1292         struct sw_flow *flow = NULL;
1293         struct datapath *dp;
1294         struct sw_flow_match match;
1295         struct sw_flow_id ufid;
1296         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1297         int err;
1298         bool log = !a[OVS_FLOW_ATTR_PROBE];
1299         bool ufid_present;
1300
1301         ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1302         if (a[OVS_FLOW_ATTR_KEY]) {
1303                 ovs_match_init(&match, &key, true, NULL);
1304                 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1305                                         NULL, log);
1306                 if (unlikely(err))
1307                         return err;
1308         }
1309
1310         ovs_lock();
1311         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1312         if (unlikely(!dp)) {
1313                 err = -ENODEV;
1314                 goto unlock;
1315         }
1316
1317         if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1318                 err = ovs_flow_tbl_flush(&dp->table);
1319                 goto unlock;
1320         }
1321
1322         if (ufid_present)
1323                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1324         else
1325                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1326         if (unlikely(!flow)) {
1327                 err = -ENOENT;
1328                 goto unlock;
1329         }
1330
1331         ovs_flow_tbl_remove(&dp->table, flow);
1332         ovs_unlock();
1333
1334         reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1335                                         &flow->id, info, false, ufid_flags);
1336         if (likely(reply)) {
1337                 if (likely(!IS_ERR(reply))) {
1338                         rcu_read_lock();        /*To keep RCU checker happy. */
1339                         err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1340                                                      reply, info->snd_portid,
1341                                                      info->snd_seq, 0,
1342                                                      OVS_FLOW_CMD_DEL,
1343                                                      ufid_flags);
1344                         rcu_read_unlock();
1345                         BUG_ON(err < 0);
1346
1347                         ovs_notify(&dp_flow_genl_family, reply, info);
1348                 } else {
1349                         netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
1350                 }
1351         }
1352
1353         ovs_flow_free(flow, true);
1354         return 0;
1355 unlock:
1356         ovs_unlock();
1357         return err;
1358 }
1359
1360 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1361 {
1362         struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1363         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1364         struct table_instance *ti;
1365         struct datapath *dp;
1366         u32 ufid_flags;
1367         int err;
1368
1369         err = genlmsg_parse(cb->nlh, &dp_flow_genl_family, a,
1370                             OVS_FLOW_ATTR_MAX, flow_policy);
1371         if (err)
1372                 return err;
1373         ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1374
1375         rcu_read_lock();
1376         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1377         if (!dp) {
1378                 rcu_read_unlock();
1379                 return -ENODEV;
1380         }
1381
1382         ti = rcu_dereference(dp->table.ti);
1383         for (;;) {
1384                 struct sw_flow *flow;
1385                 u32 bucket, obj;
1386
1387                 bucket = cb->args[0];
1388                 obj = cb->args[1];
1389                 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1390                 if (!flow)
1391                         break;
1392
1393                 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1394                                            NETLINK_CB(cb->skb).portid,
1395                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1396                                            OVS_FLOW_CMD_NEW, ufid_flags) < 0)
1397                         break;
1398
1399                 cb->args[0] = bucket;
1400                 cb->args[1] = obj;
1401         }
1402         rcu_read_unlock();
1403         return skb->len;
1404 }
1405
1406 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1407         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1408         [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1409         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1410         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1411         [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1412         [OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1413         [OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1414 };
1415
1416 static const struct genl_ops dp_flow_genl_ops[] = {
1417         { .cmd = OVS_FLOW_CMD_NEW,
1418           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1419           .policy = flow_policy,
1420           .doit = ovs_flow_cmd_new
1421         },
1422         { .cmd = OVS_FLOW_CMD_DEL,
1423           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1424           .policy = flow_policy,
1425           .doit = ovs_flow_cmd_del
1426         },
1427         { .cmd = OVS_FLOW_CMD_GET,
1428           .flags = 0,               /* OK for unprivileged users. */
1429           .policy = flow_policy,
1430           .doit = ovs_flow_cmd_get,
1431           .dumpit = ovs_flow_cmd_dump
1432         },
1433         { .cmd = OVS_FLOW_CMD_SET,
1434           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1435           .policy = flow_policy,
1436           .doit = ovs_flow_cmd_set,
1437         },
1438 };
1439
1440 static struct genl_family dp_flow_genl_family = {
1441         .id = GENL_ID_GENERATE,
1442         .hdrsize = sizeof(struct ovs_header),
1443         .name = OVS_FLOW_FAMILY,
1444         .version = OVS_FLOW_VERSION,
1445         .maxattr = OVS_FLOW_ATTR_MAX,
1446         .netnsok = true,
1447         .parallel_ops = true,
1448         .ops = dp_flow_genl_ops,
1449         .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1450         .mcgrps = &ovs_dp_flow_multicast_group,
1451         .n_mcgrps = 1,
1452 };
1453
1454 static size_t ovs_dp_cmd_msg_size(void)
1455 {
1456         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1457
1458         msgsize += nla_total_size(IFNAMSIZ);
1459         msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1460         msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
1461         msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1462
1463         return msgsize;
1464 }
1465
1466 /* Called with ovs_mutex. */
1467 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1468                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1469 {
1470         struct ovs_header *ovs_header;
1471         struct ovs_dp_stats dp_stats;
1472         struct ovs_dp_megaflow_stats dp_megaflow_stats;
1473         int err;
1474
1475         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1476                                    flags, cmd);
1477         if (!ovs_header)
1478                 goto error;
1479
1480         ovs_header->dp_ifindex = get_dpifindex(dp);
1481
1482         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1483         if (err)
1484                 goto nla_put_failure;
1485
1486         get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1487         if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1488                           &dp_stats, OVS_DP_ATTR_PAD))
1489                 goto nla_put_failure;
1490
1491         if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1492                           sizeof(struct ovs_dp_megaflow_stats),
1493                           &dp_megaflow_stats, OVS_DP_ATTR_PAD))
1494                 goto nla_put_failure;
1495
1496         if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1497                 goto nla_put_failure;
1498
1499         genlmsg_end(skb, ovs_header);
1500         return 0;
1501
1502 nla_put_failure:
1503         genlmsg_cancel(skb, ovs_header);
1504 error:
1505         return -EMSGSIZE;
1506 }
1507
1508 static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1509 {
1510         return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1511 }
1512
1513 /* Called with rcu_read_lock or ovs_mutex. */
1514 static struct datapath *lookup_datapath(struct net *net,
1515                                         const struct ovs_header *ovs_header,
1516                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1517 {
1518         struct datapath *dp;
1519
1520         if (!a[OVS_DP_ATTR_NAME])
1521                 dp = get_dp(net, ovs_header->dp_ifindex);
1522         else {
1523                 struct vport *vport;
1524
1525                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1526                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1527         }
1528         return dp ? dp : ERR_PTR(-ENODEV);
1529 }
1530
1531 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1532 {
1533         struct datapath *dp;
1534
1535         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1536         if (IS_ERR(dp))
1537                 return;
1538
1539         WARN(dp->user_features, "Dropping previously announced user features\n");
1540         dp->user_features = 0;
1541 }
1542
1543 static void ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1544 {
1545         if (a[OVS_DP_ATTR_USER_FEATURES])
1546                 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1547 }
1548
1549 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1550 {
1551         struct nlattr **a = info->attrs;
1552         struct vport_parms parms;
1553         struct sk_buff *reply;
1554         struct datapath *dp;
1555         struct vport *vport;
1556         struct ovs_net *ovs_net;
1557         int err, i;
1558
1559         err = -EINVAL;
1560         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1561                 goto err;
1562
1563         reply = ovs_dp_cmd_alloc_info();
1564         if (!reply)
1565                 return -ENOMEM;
1566
1567         err = -ENOMEM;
1568         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1569         if (dp == NULL)
1570                 goto err_free_reply;
1571
1572         ovs_dp_set_net(dp, sock_net(skb->sk));
1573
1574         /* Allocate table. */
1575         err = ovs_flow_tbl_init(&dp->table);
1576         if (err)
1577                 goto err_free_dp;
1578
1579         dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1580         if (!dp->stats_percpu) {
1581                 err = -ENOMEM;
1582                 goto err_destroy_table;
1583         }
1584
1585         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1586                             GFP_KERNEL);
1587         if (!dp->ports) {
1588                 err = -ENOMEM;
1589                 goto err_destroy_percpu;
1590         }
1591
1592         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1593                 INIT_HLIST_HEAD(&dp->ports[i]);
1594
1595         /* Set up our datapath device. */
1596         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1597         parms.type = OVS_VPORT_TYPE_INTERNAL;
1598         parms.options = NULL;
1599         parms.dp = dp;
1600         parms.port_no = OVSP_LOCAL;
1601         parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1602
1603         ovs_dp_change(dp, a);
1604
1605         /* So far only local changes have been made, now need the lock. */
1606         ovs_lock();
1607
1608         vport = new_vport(&parms);
1609         if (IS_ERR(vport)) {
1610                 err = PTR_ERR(vport);
1611                 if (err == -EBUSY)
1612                         err = -EEXIST;
1613
1614                 if (err == -EEXIST) {
1615                         /* An outdated user space instance that does not understand
1616                          * the concept of user_features has attempted to create a new
1617                          * datapath and is likely to reuse it. Drop all user features.
1618                          */
1619                         if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1620                                 ovs_dp_reset_user_features(skb, info);
1621                 }
1622
1623                 goto err_destroy_ports_array;
1624         }
1625
1626         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1627                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1628         BUG_ON(err < 0);
1629
1630         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1631         list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1632
1633         ovs_unlock();
1634
1635         ovs_notify(&dp_datapath_genl_family, reply, info);
1636         return 0;
1637
1638 err_destroy_ports_array:
1639         ovs_unlock();
1640         kfree(dp->ports);
1641 err_destroy_percpu:
1642         free_percpu(dp->stats_percpu);
1643 err_destroy_table:
1644         ovs_flow_tbl_destroy(&dp->table);
1645 err_free_dp:
1646         kfree(dp);
1647 err_free_reply:
1648         kfree_skb(reply);
1649 err:
1650         return err;
1651 }
1652
1653 /* Called with ovs_mutex. */
1654 static void __dp_destroy(struct datapath *dp)
1655 {
1656         int i;
1657
1658         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1659                 struct vport *vport;
1660                 struct hlist_node *n;
1661
1662                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1663                         if (vport->port_no != OVSP_LOCAL)
1664                                 ovs_dp_detach_port(vport);
1665         }
1666
1667         list_del_rcu(&dp->list_node);
1668
1669         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1670          * all ports in datapath are destroyed first before freeing datapath.
1671          */
1672         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1673
1674         /* RCU destroy the flow table */
1675         call_rcu(&dp->rcu, destroy_dp_rcu);
1676 }
1677
1678 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1679 {
1680         struct sk_buff *reply;
1681         struct datapath *dp;
1682         int err;
1683
1684         reply = ovs_dp_cmd_alloc_info();
1685         if (!reply)
1686                 return -ENOMEM;
1687
1688         ovs_lock();
1689         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1690         err = PTR_ERR(dp);
1691         if (IS_ERR(dp))
1692                 goto err_unlock_free;
1693
1694         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1695                                    info->snd_seq, 0, OVS_DP_CMD_DEL);
1696         BUG_ON(err < 0);
1697
1698         __dp_destroy(dp);
1699         ovs_unlock();
1700
1701         ovs_notify(&dp_datapath_genl_family, reply, info);
1702
1703         return 0;
1704
1705 err_unlock_free:
1706         ovs_unlock();
1707         kfree_skb(reply);
1708         return err;
1709 }
1710
1711 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1712 {
1713         struct sk_buff *reply;
1714         struct datapath *dp;
1715         int err;
1716
1717         reply = ovs_dp_cmd_alloc_info();
1718         if (!reply)
1719                 return -ENOMEM;
1720
1721         ovs_lock();
1722         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1723         err = PTR_ERR(dp);
1724         if (IS_ERR(dp))
1725                 goto err_unlock_free;
1726
1727         ovs_dp_change(dp, info->attrs);
1728
1729         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1730                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1731         BUG_ON(err < 0);
1732
1733         ovs_unlock();
1734         ovs_notify(&dp_datapath_genl_family, reply, info);
1735
1736         return 0;
1737
1738 err_unlock_free:
1739         ovs_unlock();
1740         kfree_skb(reply);
1741         return err;
1742 }
1743
1744 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1745 {
1746         struct sk_buff *reply;
1747         struct datapath *dp;
1748         int err;
1749
1750         reply = ovs_dp_cmd_alloc_info();
1751         if (!reply)
1752                 return -ENOMEM;
1753
1754         ovs_lock();
1755         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1756         if (IS_ERR(dp)) {
1757                 err = PTR_ERR(dp);
1758                 goto err_unlock_free;
1759         }
1760         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1761                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1762         BUG_ON(err < 0);
1763         ovs_unlock();
1764
1765         return genlmsg_reply(reply, info);
1766
1767 err_unlock_free:
1768         ovs_unlock();
1769         kfree_skb(reply);
1770         return err;
1771 }
1772
1773 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1774 {
1775         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1776         struct datapath *dp;
1777         int skip = cb->args[0];
1778         int i = 0;
1779
1780         ovs_lock();
1781         list_for_each_entry(dp, &ovs_net->dps, list_node) {
1782                 if (i >= skip &&
1783                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1784                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1785                                          OVS_DP_CMD_NEW) < 0)
1786                         break;
1787                 i++;
1788         }
1789         ovs_unlock();
1790
1791         cb->args[0] = i;
1792
1793         return skb->len;
1794 }
1795
1796 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1797         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1798         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1799         [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1800 };
1801
1802 static const struct genl_ops dp_datapath_genl_ops[] = {
1803         { .cmd = OVS_DP_CMD_NEW,
1804           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1805           .policy = datapath_policy,
1806           .doit = ovs_dp_cmd_new
1807         },
1808         { .cmd = OVS_DP_CMD_DEL,
1809           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1810           .policy = datapath_policy,
1811           .doit = ovs_dp_cmd_del
1812         },
1813         { .cmd = OVS_DP_CMD_GET,
1814           .flags = 0,               /* OK for unprivileged users. */
1815           .policy = datapath_policy,
1816           .doit = ovs_dp_cmd_get,
1817           .dumpit = ovs_dp_cmd_dump
1818         },
1819         { .cmd = OVS_DP_CMD_SET,
1820           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1821           .policy = datapath_policy,
1822           .doit = ovs_dp_cmd_set,
1823         },
1824 };
1825
1826 static struct genl_family dp_datapath_genl_family = {
1827         .id = GENL_ID_GENERATE,
1828         .hdrsize = sizeof(struct ovs_header),
1829         .name = OVS_DATAPATH_FAMILY,
1830         .version = OVS_DATAPATH_VERSION,
1831         .maxattr = OVS_DP_ATTR_MAX,
1832         .netnsok = true,
1833         .parallel_ops = true,
1834         .ops = dp_datapath_genl_ops,
1835         .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1836         .mcgrps = &ovs_dp_datapath_multicast_group,
1837         .n_mcgrps = 1,
1838 };
1839
1840 /* Called with ovs_mutex or RCU read lock. */
1841 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1842                                    u32 portid, u32 seq, u32 flags, u8 cmd)
1843 {
1844         struct ovs_header *ovs_header;
1845         struct ovs_vport_stats vport_stats;
1846         int err;
1847
1848         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1849                                  flags, cmd);
1850         if (!ovs_header)
1851                 return -EMSGSIZE;
1852
1853         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1854
1855         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1856             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1857             nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1858                            ovs_vport_name(vport)))
1859                 goto nla_put_failure;
1860
1861         ovs_vport_get_stats(vport, &vport_stats);
1862         if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
1863                           sizeof(struct ovs_vport_stats), &vport_stats,
1864                           OVS_VPORT_ATTR_PAD))
1865                 goto nla_put_failure;
1866
1867         if (ovs_vport_get_upcall_portids(vport, skb))
1868                 goto nla_put_failure;
1869
1870         err = ovs_vport_get_options(vport, skb);
1871         if (err == -EMSGSIZE)
1872                 goto error;
1873
1874         genlmsg_end(skb, ovs_header);
1875         return 0;
1876
1877 nla_put_failure:
1878         err = -EMSGSIZE;
1879 error:
1880         genlmsg_cancel(skb, ovs_header);
1881         return err;
1882 }
1883
1884 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1885 {
1886         return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1887 }
1888
1889 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1890 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1891                                          u32 seq, u8 cmd)
1892 {
1893         struct sk_buff *skb;
1894         int retval;
1895
1896         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1897         if (!skb)
1898                 return ERR_PTR(-ENOMEM);
1899
1900         retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1901         BUG_ON(retval < 0);
1902
1903         return skb;
1904 }
1905
1906 /* Called with ovs_mutex or RCU read lock. */
1907 static struct vport *lookup_vport(struct net *net,
1908                                   const struct ovs_header *ovs_header,
1909                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1910 {
1911         struct datapath *dp;
1912         struct vport *vport;
1913
1914         if (a[OVS_VPORT_ATTR_NAME]) {
1915                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1916                 if (!vport)
1917                         return ERR_PTR(-ENODEV);
1918                 if (ovs_header->dp_ifindex &&
1919                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1920                         return ERR_PTR(-ENODEV);
1921                 return vport;
1922         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1923                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1924
1925                 if (port_no >= DP_MAX_PORTS)
1926                         return ERR_PTR(-EFBIG);
1927
1928                 dp = get_dp(net, ovs_header->dp_ifindex);
1929                 if (!dp)
1930                         return ERR_PTR(-ENODEV);
1931
1932                 vport = ovs_vport_ovsl_rcu(dp, port_no);
1933                 if (!vport)
1934                         return ERR_PTR(-ENODEV);
1935                 return vport;
1936         } else
1937                 return ERR_PTR(-EINVAL);
1938 }
1939
1940 /* Called with ovs_mutex */
1941 static void update_headroom(struct datapath *dp)
1942 {
1943         unsigned dev_headroom, max_headroom = 0;
1944         struct net_device *dev;
1945         struct vport *vport;
1946         int i;
1947
1948         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1949                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1950                         dev = vport->dev;
1951                         dev_headroom = netdev_get_fwd_headroom(dev);
1952                         if (dev_headroom > max_headroom)
1953                                 max_headroom = dev_headroom;
1954                 }
1955         }
1956
1957         dp->max_headroom = max_headroom;
1958         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1959                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node)
1960                         netdev_set_rx_headroom(vport->dev, max_headroom);
1961 }
1962
1963 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1964 {
1965         struct nlattr **a = info->attrs;
1966         struct ovs_header *ovs_header = info->userhdr;
1967         struct vport_parms parms;
1968         struct sk_buff *reply;
1969         struct vport *vport;
1970         struct datapath *dp;
1971         u32 port_no;
1972         int err;
1973
1974         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1975             !a[OVS_VPORT_ATTR_UPCALL_PID])
1976                 return -EINVAL;
1977
1978         port_no = a[OVS_VPORT_ATTR_PORT_NO]
1979                 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
1980         if (port_no >= DP_MAX_PORTS)
1981                 return -EFBIG;
1982
1983         reply = ovs_vport_cmd_alloc_info();
1984         if (!reply)
1985                 return -ENOMEM;
1986
1987         ovs_lock();
1988 restart:
1989         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1990         err = -ENODEV;
1991         if (!dp)
1992                 goto exit_unlock_free;
1993
1994         if (port_no) {
1995                 vport = ovs_vport_ovsl(dp, port_no);
1996                 err = -EBUSY;
1997                 if (vport)
1998                         goto exit_unlock_free;
1999         } else {
2000                 for (port_no = 1; ; port_no++) {
2001                         if (port_no >= DP_MAX_PORTS) {
2002                                 err = -EFBIG;
2003                                 goto exit_unlock_free;
2004                         }
2005                         vport = ovs_vport_ovsl(dp, port_no);
2006                         if (!vport)
2007                                 break;
2008                 }
2009         }
2010
2011         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2012         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2013         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2014         parms.dp = dp;
2015         parms.port_no = port_no;
2016         parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
2017
2018         vport = new_vport(&parms);
2019         err = PTR_ERR(vport);
2020         if (IS_ERR(vport)) {
2021                 if (err == -EAGAIN)
2022                         goto restart;
2023                 goto exit_unlock_free;
2024         }
2025
2026         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2027                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2028
2029         if (netdev_get_fwd_headroom(vport->dev) > dp->max_headroom)
2030                 update_headroom(dp);
2031         else
2032                 netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2033
2034         BUG_ON(err < 0);
2035         ovs_unlock();
2036
2037         ovs_notify(&dp_vport_genl_family, reply, info);
2038         return 0;
2039
2040 exit_unlock_free:
2041         ovs_unlock();
2042         kfree_skb(reply);
2043         return err;
2044 }
2045
2046 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2047 {
2048         struct nlattr **a = info->attrs;
2049         struct sk_buff *reply;
2050         struct vport *vport;
2051         int err;
2052
2053         reply = ovs_vport_cmd_alloc_info();
2054         if (!reply)
2055                 return -ENOMEM;
2056
2057         ovs_lock();
2058         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2059         err = PTR_ERR(vport);
2060         if (IS_ERR(vport))
2061                 goto exit_unlock_free;
2062
2063         if (a[OVS_VPORT_ATTR_TYPE] &&
2064             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2065                 err = -EINVAL;
2066                 goto exit_unlock_free;
2067         }
2068
2069         if (a[OVS_VPORT_ATTR_OPTIONS]) {
2070                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2071                 if (err)
2072                         goto exit_unlock_free;
2073         }
2074
2075
2076         if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2077                 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2078
2079                 err = ovs_vport_set_upcall_portids(vport, ids);
2080                 if (err)
2081                         goto exit_unlock_free;
2082         }
2083
2084         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2085                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2086         BUG_ON(err < 0);
2087
2088         ovs_unlock();
2089         ovs_notify(&dp_vport_genl_family, reply, info);
2090         return 0;
2091
2092 exit_unlock_free:
2093         ovs_unlock();
2094         kfree_skb(reply);
2095         return err;
2096 }
2097
2098 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2099 {
2100         bool must_update_headroom = false;
2101         struct nlattr **a = info->attrs;
2102         struct sk_buff *reply;
2103         struct datapath *dp;
2104         struct vport *vport;
2105         int err;
2106
2107         reply = ovs_vport_cmd_alloc_info();
2108         if (!reply)
2109                 return -ENOMEM;
2110
2111         ovs_lock();
2112         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2113         err = PTR_ERR(vport);
2114         if (IS_ERR(vport))
2115                 goto exit_unlock_free;
2116
2117         if (vport->port_no == OVSP_LOCAL) {
2118                 err = -EINVAL;
2119                 goto exit_unlock_free;
2120         }
2121
2122         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2123                                       info->snd_seq, 0, OVS_VPORT_CMD_DEL);
2124         BUG_ON(err < 0);
2125
2126         /* the vport deletion may trigger dp headroom update */
2127         dp = vport->dp;
2128         if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2129                 must_update_headroom = true;
2130         netdev_reset_rx_headroom(vport->dev);
2131         ovs_dp_detach_port(vport);
2132
2133         if (must_update_headroom)
2134                 update_headroom(dp);
2135         ovs_unlock();
2136
2137         ovs_notify(&dp_vport_genl_family, reply, info);
2138         return 0;
2139
2140 exit_unlock_free:
2141         ovs_unlock();
2142         kfree_skb(reply);
2143         return err;
2144 }
2145
2146 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2147 {
2148         struct nlattr **a = info->attrs;
2149         struct ovs_header *ovs_header = info->userhdr;
2150         struct sk_buff *reply;
2151         struct vport *vport;
2152         int err;
2153
2154         reply = ovs_vport_cmd_alloc_info();
2155         if (!reply)
2156                 return -ENOMEM;
2157
2158         rcu_read_lock();
2159         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2160         err = PTR_ERR(vport);
2161         if (IS_ERR(vport))
2162                 goto exit_unlock_free;
2163         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2164                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2165         BUG_ON(err < 0);
2166         rcu_read_unlock();
2167
2168         return genlmsg_reply(reply, info);
2169
2170 exit_unlock_free:
2171         rcu_read_unlock();
2172         kfree_skb(reply);
2173         return err;
2174 }
2175
2176 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2177 {
2178         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2179         struct datapath *dp;
2180         int bucket = cb->args[0], skip = cb->args[1];
2181         int i, j = 0;
2182
2183         rcu_read_lock();
2184         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2185         if (!dp) {
2186                 rcu_read_unlock();
2187                 return -ENODEV;
2188         }
2189         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2190                 struct vport *vport;
2191
2192                 j = 0;
2193                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2194                         if (j >= skip &&
2195                             ovs_vport_cmd_fill_info(vport, skb,
2196                                                     NETLINK_CB(cb->skb).portid,
2197                                                     cb->nlh->nlmsg_seq,
2198                                                     NLM_F_MULTI,
2199                                                     OVS_VPORT_CMD_NEW) < 0)
2200                                 goto out;
2201
2202                         j++;
2203                 }
2204                 skip = 0;
2205         }
2206 out:
2207         rcu_read_unlock();
2208
2209         cb->args[0] = i;
2210         cb->args[1] = j;
2211
2212         return skb->len;
2213 }
2214
2215 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2216         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2217         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2218         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2219         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2220         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
2221         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2222 };
2223
2224 static const struct genl_ops dp_vport_genl_ops[] = {
2225         { .cmd = OVS_VPORT_CMD_NEW,
2226           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2227           .policy = vport_policy,
2228           .doit = ovs_vport_cmd_new
2229         },
2230         { .cmd = OVS_VPORT_CMD_DEL,
2231           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2232           .policy = vport_policy,
2233           .doit = ovs_vport_cmd_del
2234         },
2235         { .cmd = OVS_VPORT_CMD_GET,
2236           .flags = 0,               /* OK for unprivileged users. */
2237           .policy = vport_policy,
2238           .doit = ovs_vport_cmd_get,
2239           .dumpit = ovs_vport_cmd_dump
2240         },
2241         { .cmd = OVS_VPORT_CMD_SET,
2242           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2243           .policy = vport_policy,
2244           .doit = ovs_vport_cmd_set,
2245         },
2246 };
2247
2248 struct genl_family dp_vport_genl_family = {
2249         .id = GENL_ID_GENERATE,
2250         .hdrsize = sizeof(struct ovs_header),
2251         .name = OVS_VPORT_FAMILY,
2252         .version = OVS_VPORT_VERSION,
2253         .maxattr = OVS_VPORT_ATTR_MAX,
2254         .netnsok = true,
2255         .parallel_ops = true,
2256         .ops = dp_vport_genl_ops,
2257         .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2258         .mcgrps = &ovs_dp_vport_multicast_group,
2259         .n_mcgrps = 1,
2260 };
2261
2262 static struct genl_family * const dp_genl_families[] = {
2263         &dp_datapath_genl_family,
2264         &dp_vport_genl_family,
2265         &dp_flow_genl_family,
2266         &dp_packet_genl_family,
2267 };
2268
2269 static void dp_unregister_genl(int n_families)
2270 {
2271         int i;
2272
2273         for (i = 0; i < n_families; i++)
2274                 genl_unregister_family(dp_genl_families[i]);
2275 }
2276
2277 static int dp_register_genl(void)
2278 {
2279         int err;
2280         int i;
2281
2282         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2283
2284                 err = genl_register_family(dp_genl_families[i]);
2285                 if (err)
2286                         goto error;
2287         }
2288
2289         return 0;
2290
2291 error:
2292         dp_unregister_genl(i);
2293         return err;
2294 }
2295
2296 static int __net_init ovs_init_net(struct net *net)
2297 {
2298         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2299
2300         INIT_LIST_HEAD(&ovs_net->dps);
2301         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2302         ovs_ct_init(net);
2303         return 0;
2304 }
2305
2306 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2307                                             struct list_head *head)
2308 {
2309         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2310         struct datapath *dp;
2311
2312         list_for_each_entry(dp, &ovs_net->dps, list_node) {
2313                 int i;
2314
2315                 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2316                         struct vport *vport;
2317
2318                         hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2319                                 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2320                                         continue;
2321
2322                                 if (dev_net(vport->dev) == dnet)
2323                                         list_add(&vport->detach_list, head);
2324                         }
2325                 }
2326         }
2327 }
2328
2329 static void __net_exit ovs_exit_net(struct net *dnet)
2330 {
2331         struct datapath *dp, *dp_next;
2332         struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2333         struct vport *vport, *vport_next;
2334         struct net *net;
2335         LIST_HEAD(head);
2336
2337         ovs_ct_exit(dnet);
2338         ovs_lock();
2339         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2340                 __dp_destroy(dp);
2341
2342         rtnl_lock();
2343         for_each_net(net)
2344                 list_vports_from_net(net, dnet, &head);
2345         rtnl_unlock();
2346
2347         /* Detach all vports from given namespace. */
2348         list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2349                 list_del(&vport->detach_list);
2350                 ovs_dp_detach_port(vport);
2351         }
2352
2353         ovs_unlock();
2354
2355         cancel_work_sync(&ovs_net->dp_notify_work);
2356 }
2357
2358 static struct pernet_operations ovs_net_ops = {
2359         .init = ovs_init_net,
2360         .exit = ovs_exit_net,
2361         .id   = &ovs_net_id,
2362         .size = sizeof(struct ovs_net),
2363 };
2364
2365 static int __init dp_init(void)
2366 {
2367         int err;
2368
2369         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2370
2371         pr_info("Open vSwitch switching datapath\n");
2372
2373         err = action_fifos_init();
2374         if (err)
2375                 goto error;
2376
2377         err = ovs_internal_dev_rtnl_link_register();
2378         if (err)
2379                 goto error_action_fifos_exit;
2380
2381         err = ovs_flow_init();
2382         if (err)
2383                 goto error_unreg_rtnl_link;
2384
2385         err = ovs_vport_init();
2386         if (err)
2387                 goto error_flow_exit;
2388
2389         err = register_pernet_device(&ovs_net_ops);
2390         if (err)
2391                 goto error_vport_exit;
2392
2393         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2394         if (err)
2395                 goto error_netns_exit;
2396
2397         err = ovs_netdev_init();
2398         if (err)
2399                 goto error_unreg_notifier;
2400
2401         err = dp_register_genl();
2402         if (err < 0)
2403                 goto error_unreg_netdev;
2404
2405         return 0;
2406
2407 error_unreg_netdev:
2408         ovs_netdev_exit();
2409 error_unreg_notifier:
2410         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2411 error_netns_exit:
2412         unregister_pernet_device(&ovs_net_ops);
2413 error_vport_exit:
2414         ovs_vport_exit();
2415 error_flow_exit:
2416         ovs_flow_exit();
2417 error_unreg_rtnl_link:
2418         ovs_internal_dev_rtnl_link_unregister();
2419 error_action_fifos_exit:
2420         action_fifos_exit();
2421 error:
2422         return err;
2423 }
2424
2425 static void dp_cleanup(void)
2426 {
2427         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2428         ovs_netdev_exit();
2429         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2430         unregister_pernet_device(&ovs_net_ops);
2431         rcu_barrier();
2432         ovs_vport_exit();
2433         ovs_flow_exit();
2434         ovs_internal_dev_rtnl_link_unregister();
2435         action_fifos_exit();
2436 }
2437
2438 module_init(dp_init);
2439 module_exit(dp_cleanup);
2440
2441 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2442 MODULE_LICENSE("GPL");
2443 MODULE_ALIAS_GENL_FAMILY(OVS_DATAPATH_FAMILY);
2444 MODULE_ALIAS_GENL_FAMILY(OVS_VPORT_FAMILY);
2445 MODULE_ALIAS_GENL_FAMILY(OVS_FLOW_FAMILY);
2446 MODULE_ALIAS_GENL_FAMILY(OVS_PACKET_FAMILY);