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 unsigned int ovs_net_id __read_mostly;
  62
  63 static struct genl_family dp_packet_genl_family;
  64 static struct genl_family dp_flow_genl_family;
  65 static struct genl_family dp_datapath_genl_family;
  66
  67 static const struct nla_policy flow_policy[];
  68
  69 static const struct genl_multicast_group ovs_dp_flow_multicast_group = {
  70         .name = OVS_FLOW_MCGROUP,
  71 };
  72
  73 static const struct genl_multicast_group ovs_dp_datapath_multicast_group = {
  74         .name = OVS_DATAPATH_MCGROUP,
  75 };
  76
  77 static const struct genl_multicast_group ovs_dp_vport_multicast_group = {
  78         .name = OVS_VPORT_MCGROUP,
  79 };
  80
  81 /* Check if need to build a reply message.
  82  * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
  83 static bool ovs_must_notify(struct genl_family *family, struct genl_info *info,
  84                             unsigned int group)
  85 {
  86         return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
  87                genl_has_listeners(family, genl_info_net(info), group);
  88 }
  89
  90 static void ovs_notify(struct genl_family *family,
  91                        struct sk_buff *skb, struct genl_info *info)
  92 {
  93         genl_notify(family, skb, info, 0, GFP_KERNEL);
  94 }
  95
  96 /**
  97  * DOC: Locking:
  98  *
  99  * All writes e.g. Writes to device state (add/remove datapath, port, set
 100  * operations on vports, etc.), Writes to other state (flow table
 101  * modifications, set miscellaneous datapath parameters, etc.) are protected
 102  * by ovs_lock.
 103  *
 104  * Reads are protected by RCU.
 105  *
 106  * There are a few special cases (mostly stats) that have their own
 107  * synchronization but they nest under all of above and don't interact with
 108  * each other.
 109  *
 110  * The RTNL lock nests inside ovs_mutex.
 111  */
 112
 113 static DEFINE_MUTEX(ovs_mutex);
 114
 115 void ovs_lock(void)
 116 {
 117         mutex_lock(&ovs_mutex);
 118 }
 119
 120 void ovs_unlock(void)
 121 {
 122         mutex_unlock(&ovs_mutex);
 123 }
 124
 125 #ifdef CONFIG_LOCKDEP
 126 int lockdep_ovsl_is_held(void)
 127 {
 128         if (debug_locks)
 129                 return lockdep_is_held(&ovs_mutex);
 130         else
 131                 return 1;
 132 }
 133 #endif
 134
 135 static struct vport *new_vport(const struct vport_parms *);
 136 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
 137                              const struct sw_flow_key *,
 138                              const struct dp_upcall_info *,
 139                              uint32_t cutlen);
 140 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
 141                                   const struct sw_flow_key *,
 142                                   const struct dp_upcall_info *,
 143                                   uint32_t cutlen);
 144
 145 /* Must be called with rcu_read_lock. */
 146 static struct datapath *get_dp_rcu(struct net *net, int dp_ifindex)
 147 {
 148         struct net_device *dev = dev_get_by_index_rcu(net, dp_ifindex);
 149
 150         if (dev) {
 151                 struct vport *vport = ovs_internal_dev_get_vport(dev);
 152                 if (vport)
 153                         return vport->dp;
 154         }
 155
 156         return NULL;
 157 }
 158
 159 /* The caller must hold either ovs_mutex or rcu_read_lock to keep the
 160  * returned dp pointer valid.
 161  */
 162 static inline struct datapath *get_dp(struct net *net, int dp_ifindex)
 163 {
 164         struct datapath *dp;
 165
 166         WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_ovsl_is_held());
 167         rcu_read_lock();
 168         dp = get_dp_rcu(net, dp_ifindex);
 169         rcu_read_unlock();
 170
 171         return dp;
 172 }
 173
 174 /* Must be called with rcu_read_lock or ovs_mutex. */
 175 const char *ovs_dp_name(const struct datapath *dp)
 176 {
 177         struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
 178         return ovs_vport_name(vport);
 179 }
 180
 181 static int get_dpifindex(const struct datapath *dp)
 182 {
 183         struct vport *local;
 184         int ifindex;
 185
 186         rcu_read_lock();
 187
 188         local = ovs_vport_rcu(dp, OVSP_LOCAL);
 189         if (local)
 190                 ifindex = local->dev->ifindex;
 191         else
 192                 ifindex = 0;
 193
 194         rcu_read_unlock();
 195
 196         return ifindex;
 197 }
 198
 199 static void destroy_dp_rcu(struct rcu_head *rcu)
 200 {
 201         struct datapath *dp = container_of(rcu, struct datapath, rcu);
 202
 203         ovs_flow_tbl_destroy(&dp->table);
 204         free_percpu(dp->stats_percpu);
 205         kfree(dp->ports);
 206         kfree(dp);
 207 }
 208
 209 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
 210                                             u16 port_no)
 211 {
 212         return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
 213 }
 214
 215 /* Called with ovs_mutex or RCU read lock. */
 216 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
 217 {
 218         struct vport *vport;
 219         struct hlist_head *head;
 220
 221         head = vport_hash_bucket(dp, port_no);
 222         hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
 223                 if (vport->port_no == port_no)
 224                         return vport;
 225         }
 226         return NULL;
 227 }
 228
 229 /* Called with ovs_mutex. */
 230 static struct vport *new_vport(const struct vport_parms *parms)
 231 {
 232         struct vport *vport;
 233
 234         vport = ovs_vport_add(parms);
 235         if (!IS_ERR(vport)) {
 236                 struct datapath *dp = parms->dp;
 237                 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
 238
 239                 hlist_add_head_rcu(&vport->dp_hash_node, head);
 240         }
 241         return vport;
 242 }
 243
 244 void ovs_dp_detach_port(struct vport *p)
 245 {
 246         ASSERT_OVSL();
 247
 248         /* First drop references to device. */
 249         hlist_del_rcu(&p->dp_hash_node);
 250
 251         /* Then destroy it. */
 252         ovs_vport_del(p);
 253 }
 254
 255 /* Must be called with rcu_read_lock. */
 256 void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
 257 {
 258         const struct vport *p = OVS_CB(skb)->input_vport;
 259         struct datapath *dp = p->dp;
 260         struct sw_flow *flow;
 261         struct sw_flow_actions *sf_acts;
 262         struct dp_stats_percpu *stats;
 263         u64 *stats_counter;
 264         u32 n_mask_hit;
 265
 266         stats = this_cpu_ptr(dp->stats_percpu);
 267
 268         /* Look up flow. */
 269         flow = ovs_flow_tbl_lookup_stats(&dp->table, key, &n_mask_hit);
 270         if (unlikely(!flow)) {
 271                 struct dp_upcall_info upcall;
 272                 int error;
 273
 274                 memset(&upcall, 0, sizeof(upcall));
 275                 upcall.cmd = OVS_PACKET_CMD_MISS;
 276                 upcall.portid = ovs_vport_find_upcall_portid(p, skb);
 277                 upcall.mru = OVS_CB(skb)->mru;
 278                 error = ovs_dp_upcall(dp, skb, key, &upcall, 0);
 279                 if (unlikely(error))
 280                         kfree_skb(skb);
 281                 else
 282                         consume_skb(skb);
 283                 stats_counter = &stats->n_missed;
 284                 goto out;
 285         }
 286
 287         ovs_flow_stats_update(flow, key->tp.flags, skb);
 288         sf_acts = rcu_dereference(flow->sf_acts);
 289         ovs_execute_actions(dp, skb, sf_acts, key);
 290
 291         stats_counter = &stats->n_hit;
 292
 293 out:
 294         /* Update datapath statistics. */
 295         u64_stats_update_begin(&stats->syncp);
 296         (*stats_counter)++;
 297         stats->n_mask_hit += n_mask_hit;
 298         u64_stats_update_end(&stats->syncp);
 299 }
 300
 301 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
 302                   const struct sw_flow_key *key,
 303                   const struct dp_upcall_info *upcall_info,
 304                   uint32_t cutlen)
 305 {
 306         struct dp_stats_percpu *stats;
 307         int err;
 308
 309         if (upcall_info->portid == 0) {
 310                 err = -ENOTCONN;
 311                 goto err;
 312         }
 313
 314         if (!skb_is_gso(skb))
 315                 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
 316         else
 317                 err = queue_gso_packets(dp, skb, key, upcall_info, cutlen);
 318         if (err)
 319                 goto err;
 320
 321         return 0;
 322
 323 err:
 324         stats = this_cpu_ptr(dp->stats_percpu);
 325
 326         u64_stats_update_begin(&stats->syncp);
 327         stats->n_lost++;
 328         u64_stats_update_end(&stats->syncp);
 329
 330         return err;
 331 }
 332
 333 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
 334                              const struct sw_flow_key *key,
 335                              const struct dp_upcall_info *upcall_info,
 336                                  uint32_t cutlen)
 337 {
 338         unsigned short gso_type = skb_shinfo(skb)->gso_type;
 339         struct sw_flow_key later_key;
 340         struct sk_buff *segs, *nskb;
 341         int err;
 342
 343         BUILD_BUG_ON(sizeof(*OVS_CB(skb)) > SKB_SGO_CB_OFFSET);
 344         segs = __skb_gso_segment(skb, NETIF_F_SG, false);
 345         if (IS_ERR(segs))
 346                 return PTR_ERR(segs);
 347         if (segs == NULL)
 348                 return -EINVAL;
 349
 350         if (gso_type & SKB_GSO_UDP) {
 351                 /* The initial flow key extracted by ovs_flow_key_extract()
 352                  * in this case is for a first fragment, so we need to
 353                  * properly mark later fragments.
 354                  */
 355                 later_key = *key;
 356                 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
 357         }
 358
 359         /* Queue all of the segments. */
 360         skb = segs;
 361         do {
 362                 if (gso_type & SKB_GSO_UDP && skb != segs)
 363                         key = &later_key;
 364
 365                 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
 366                 if (err)
 367                         break;
 368
 369         } while ((skb = skb->next));
 370
 371         /* Free all of the segments. */
 372         skb = segs;
 373         do {
 374                 nskb = skb->next;
 375                 if (err)
 376                         kfree_skb(skb);
 377                 else
 378                         consume_skb(skb);
 379         } while ((skb = nskb));
 380         return err;
 381 }
 382
 383 static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
 384                               unsigned int hdrlen)
 385 {
 386         size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
 387                 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
 388                 + nla_total_size(ovs_key_attr_size()) /* OVS_PACKET_ATTR_KEY */
 389                 + nla_total_size(sizeof(unsigned int)); /* OVS_PACKET_ATTR_LEN */
 390
 391         /* OVS_PACKET_ATTR_USERDATA */
 392         if (upcall_info->userdata)
 393                 size += NLA_ALIGN(upcall_info->userdata->nla_len);
 394
 395         /* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
 396         if (upcall_info->egress_tun_info)
 397                 size += nla_total_size(ovs_tun_key_attr_size());
 398
 399         /* OVS_PACKET_ATTR_ACTIONS */
 400         if (upcall_info->actions_len)
 401                 size += nla_total_size(upcall_info->actions_len);
 402
 403         /* OVS_PACKET_ATTR_MRU */
 404         if (upcall_info->mru)
 405                 size += nla_total_size(sizeof(upcall_info->mru));
 406
 407         return size;
 408 }
 409
 410 static void pad_packet(struct datapath *dp, struct sk_buff *skb)
 411 {
 412         if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
 413                 size_t plen = NLA_ALIGN(skb->len) - skb->len;
 414
 415                 if (plen > 0)
 416                         memset(skb_put(skb, plen), 0, plen);
 417         }
 418 }
 419
 420 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
 421                                   const struct sw_flow_key *key,
 422                                   const struct dp_upcall_info *upcall_info,
 423                                   uint32_t cutlen)
 424 {
 425         struct ovs_header *upcall;
 426         struct sk_buff *nskb = NULL;
 427         struct sk_buff *user_skb = NULL; /* to be queued to userspace */
 428         struct nlattr *nla;
 429         size_t len;
 430         unsigned int hlen;
 431         int err, dp_ifindex;
 432
 433         dp_ifindex = get_dpifindex(dp);
 434         if (!dp_ifindex)
 435                 return -ENODEV;
 436
 437         if (skb_vlan_tag_present(skb)) {
 438                 nskb = skb_clone(skb, GFP_ATOMIC);
 439                 if (!nskb)
 440                         return -ENOMEM;
 441
 442                 nskb = __vlan_hwaccel_push_inside(nskb);
 443                 if (!nskb)
 444                         return -ENOMEM;
 445
 446                 skb = nskb;
 447         }
 448
 449         if (nla_attr_size(skb->len) > USHRT_MAX) {
 450                 err = -EFBIG;
 451                 goto out;
 452         }
 453
 454         /* Complete checksum if needed */
 455         if (skb->ip_summed == CHECKSUM_PARTIAL &&
 456             (err = skb_checksum_help(skb)))
 457                 goto out;
 458
 459         /* Older versions of OVS user space enforce alignment of the last
 460          * Netlink attribute to NLA_ALIGNTO which would require extensive
 461          * padding logic. Only perform zerocopy if padding is not required.
 462          */
 463         if (dp->user_features & OVS_DP_F_UNALIGNED)
 464                 hlen = skb_zerocopy_headlen(skb);
 465         else
 466                 hlen = skb->len;
 467
 468         len = upcall_msg_size(upcall_info, hlen - cutlen);
 469         user_skb = genlmsg_new(len, GFP_ATOMIC);
 470         if (!user_skb) {
 471                 err = -ENOMEM;
 472                 goto out;
 473         }
 474
 475         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 476                              0, upcall_info->cmd);
 477         upcall->dp_ifindex = dp_ifindex;
 478
 479         err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
 480         BUG_ON(err);
 481
 482         if (upcall_info->userdata)
 483                 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
 484                           nla_len(upcall_info->userdata),
 485                           nla_data(upcall_info->userdata));
 486
 487         if (upcall_info->egress_tun_info) {
 488                 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_EGRESS_TUN_KEY);
 489                 err = ovs_nla_put_tunnel_info(user_skb,
 490                                               upcall_info->egress_tun_info);
 491                 BUG_ON(err);
 492                 nla_nest_end(user_skb, nla);
 493         }
 494
 495         if (upcall_info->actions_len) {
 496                 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_ACTIONS);
 497                 err = ovs_nla_put_actions(upcall_info->actions,
 498                                           upcall_info->actions_len,
 499                                           user_skb);
 500                 if (!err)
 501                         nla_nest_end(user_skb, nla);
 502                 else
 503                         nla_nest_cancel(user_skb, nla);
 504         }
 505
 506         /* Add OVS_PACKET_ATTR_MRU */
 507         if (upcall_info->mru) {
 508                 if (nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU,
 509                                 upcall_info->mru)) {
 510                         err = -ENOBUFS;
 511                         goto out;
 512                 }
 513                 pad_packet(dp, user_skb);
 514         }
 515
 516         /* Add OVS_PACKET_ATTR_LEN when packet is truncated */
 517         if (cutlen > 0) {
 518                 if (nla_put_u32(user_skb, OVS_PACKET_ATTR_LEN,
 519                                 skb->len)) {
 520                         err = -ENOBUFS;
 521                         goto out;
 522                 }
 523                 pad_packet(dp, user_skb);
 524         }
 525
 526         /* Only reserve room for attribute header, packet data is added
 527          * in skb_zerocopy() */
 528         if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
 529                 err = -ENOBUFS;
 530                 goto out;
 531         }
 532         nla->nla_len = nla_attr_size(skb->len - cutlen);
 533
 534         err = skb_zerocopy(user_skb, skb, skb->len - cutlen, hlen);
 535         if (err)
 536                 goto out;
 537
 538         /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
 539         pad_packet(dp, user_skb);
 540
 541         ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
 542
 543         err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
 544         user_skb = NULL;
 545 out:
 546         if (err)
 547                 skb_tx_error(skb);
 548         kfree_skb(user_skb);
 549         kfree_skb(nskb);
 550         return err;
 551 }
 552
 553 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 554 {
 555         struct ovs_header *ovs_header = info->userhdr;
 556         struct net *net = sock_net(skb->sk);
 557         struct nlattr **a = info->attrs;
 558         struct sw_flow_actions *acts;
 559         struct sk_buff *packet;
 560         struct sw_flow *flow;
 561         struct sw_flow_actions *sf_acts;
 562         struct datapath *dp;
 563         struct vport *input_vport;
 564         u16 mru = 0;
 565         int len;
 566         int err;
 567         bool log = !a[OVS_PACKET_ATTR_PROBE];
 568
 569         err = -EINVAL;
 570         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 571             !a[OVS_PACKET_ATTR_ACTIONS])
 572                 goto err;
 573
 574         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 575         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 576         err = -ENOMEM;
 577         if (!packet)
 578                 goto err;
 579         skb_reserve(packet, NET_IP_ALIGN);
 580
 581         nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
 582
 583         /* Set packet's mru */
 584         if (a[OVS_PACKET_ATTR_MRU]) {
 585                 mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
 586                 packet->ignore_df = 1;
 587         }
 588         OVS_CB(packet)->mru = mru;
 589
 590         /* Build an sw_flow for sending this packet. */
 591         flow = ovs_flow_alloc();
 592         err = PTR_ERR(flow);
 593         if (IS_ERR(flow))
 594                 goto err_kfree_skb;
 595
 596         err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
 597                                              packet, &flow->key, log);
 598         if (err)
 599                 goto err_flow_free;
 600
 601         err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
 602                                    &flow->key, &acts, log);
 603         if (err)
 604                 goto err_flow_free;
 605
 606         rcu_assign_pointer(flow->sf_acts, acts);
 607         packet->priority = flow->key.phy.priority;
 608         packet->mark = flow->key.phy.skb_mark;
 609
 610         rcu_read_lock();
 611         dp = get_dp_rcu(net, ovs_header->dp_ifindex);
 612         err = -ENODEV;
 613         if (!dp)
 614                 goto err_unlock;
 615
 616         input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
 617         if (!input_vport)
 618                 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
 619
 620         if (!input_vport)
 621                 goto err_unlock;
 622
 623         packet->dev = input_vport->dev;
 624         OVS_CB(packet)->input_vport = input_vport;
 625         sf_acts = rcu_dereference(flow->sf_acts);
 626
 627         local_bh_disable();
 628         err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
 629         local_bh_enable();
 630         rcu_read_unlock();
 631
 632         ovs_flow_free(flow, false);
 633         return err;
 634
 635 err_unlock:
 636         rcu_read_unlock();
 637 err_flow_free:
 638         ovs_flow_free(flow, false);
 639 err_kfree_skb:
 640         kfree_skb(packet);
 641 err:
 642         return err;
 643 }
 644
 645 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 646         [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
 647         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 648         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 649         [OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
 650         [OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
 651 };
 652
 653 static const struct genl_ops dp_packet_genl_ops[] = {
 654         { .cmd = OVS_PACKET_CMD_EXECUTE,
 655           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 656           .policy = packet_policy,
 657           .doit = ovs_packet_cmd_execute
 658         }
 659 };
 660
 661 static struct genl_family dp_packet_genl_family __ro_after_init = {
 662         .hdrsize = sizeof(struct ovs_header),
 663         .name = OVS_PACKET_FAMILY,
 664         .version = OVS_PACKET_VERSION,
 665         .maxattr = OVS_PACKET_ATTR_MAX,
 666         .netnsok = true,
 667         .parallel_ops = true,
 668         .ops = dp_packet_genl_ops,
 669         .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
 670         .module = THIS_MODULE,
 671 };
 672
 673 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
 674                          struct ovs_dp_megaflow_stats *mega_stats)
 675 {
 676         int i;
 677
 678         memset(mega_stats, 0, sizeof(*mega_stats));
 679
 680         stats->n_flows = ovs_flow_tbl_count(&dp->table);
 681         mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
 682
 683         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 684
 685         for_each_possible_cpu(i) {
 686                 const struct dp_stats_percpu *percpu_stats;
 687                 struct dp_stats_percpu local_stats;
 688                 unsigned int start;
 689
 690                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 691
 692                 do {
 693                         start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
 694                         local_stats = *percpu_stats;
 695                 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
 696
 697                 stats->n_hit += local_stats.n_hit;
 698                 stats->n_missed += local_stats.n_missed;
 699                 stats->n_lost += local_stats.n_lost;
 700                 mega_stats->n_mask_hit += local_stats.n_mask_hit;
 701         }
 702 }
 703
 704 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
 705 {
 706         return ovs_identifier_is_ufid(sfid) &&
 707                !(ufid_flags & OVS_UFID_F_OMIT_KEY);
 708 }
 709
 710 static bool should_fill_mask(uint32_t ufid_flags)
 711 {
 712         return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
 713 }
 714
 715 static bool should_fill_actions(uint32_t ufid_flags)
 716 {
 717         return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
 718 }
 719
 720 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
 721                                     const struct sw_flow_id *sfid,
 722                                     uint32_t ufid_flags)
 723 {
 724         size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
 725
 726         /* OVS_FLOW_ATTR_UFID */
 727         if (sfid && ovs_identifier_is_ufid(sfid))
 728                 len += nla_total_size(sfid->ufid_len);
 729
 730         /* OVS_FLOW_ATTR_KEY */
 731         if (!sfid || should_fill_key(sfid, ufid_flags))
 732                 len += nla_total_size(ovs_key_attr_size());
 733
 734         /* OVS_FLOW_ATTR_MASK */
 735         if (should_fill_mask(ufid_flags))
 736                 len += nla_total_size(ovs_key_attr_size());
 737
 738         /* OVS_FLOW_ATTR_ACTIONS */
 739         if (should_fill_actions(ufid_flags))
 740                 len += nla_total_size(acts->orig_len);
 741
 742         return len
 743                 + nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
 744                 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
 745                 + nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */
 746 }
 747
 748 /* Called with ovs_mutex or RCU read lock. */
 749 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
 750                                    struct sk_buff *skb)
 751 {
 752         struct ovs_flow_stats stats;
 753         __be16 tcp_flags;
 754         unsigned long used;
 755
 756         ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
 757
 758         if (used &&
 759             nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
 760                               OVS_FLOW_ATTR_PAD))
 761                 return -EMSGSIZE;
 762
 763         if (stats.n_packets &&
 764             nla_put_64bit(skb, OVS_FLOW_ATTR_STATS,
 765                           sizeof(struct ovs_flow_stats), &stats,
 766                           OVS_FLOW_ATTR_PAD))
 767                 return -EMSGSIZE;
 768
 769         if ((u8)ntohs(tcp_flags) &&
 770              nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
 771                 return -EMSGSIZE;
 772
 773         return 0;
 774 }
 775
 776 /* Called with ovs_mutex or RCU read lock. */
 777 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
 778                                      struct sk_buff *skb, int skb_orig_len)
 779 {
 780         struct nlattr *start;
 781         int err;
 782
 783         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 784          * this is the first flow to be dumped into 'skb'.  This is unusual for
 785          * Netlink but individual action lists can be longer than
 786          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 787          * The userspace caller can always fetch the actions separately if it
 788          * really wants them.  (Most userspace callers in fact don't care.)
 789          *
 790          * This can only fail for dump operations because the skb is always
 791          * properly sized for single flows.
 792          */
 793         start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
 794         if (start) {
 795                 const struct sw_flow_actions *sf_acts;
 796
 797                 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
 798                 err = ovs_nla_put_actions(sf_acts->actions,
 799                                           sf_acts->actions_len, skb);
 800
 801                 if (!err)
 802                         nla_nest_end(skb, start);
 803                 else {
 804                         if (skb_orig_len)
 805                                 return err;
 806
 807                         nla_nest_cancel(skb, start);
 808                 }
 809         } else if (skb_orig_len) {
 810                 return -EMSGSIZE;
 811         }
 812
 813         return 0;
 814 }
 815
 816 /* Called with ovs_mutex or RCU read lock. */
 817 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
 818                                   struct sk_buff *skb, u32 portid,
 819                                   u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
 820 {
 821         const int skb_orig_len = skb->len;
 822         struct ovs_header *ovs_header;
 823         int err;
 824
 825         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
 826                                  flags, cmd);
 827         if (!ovs_header)
 828                 return -EMSGSIZE;
 829
 830         ovs_header->dp_ifindex = dp_ifindex;
 831
 832         err = ovs_nla_put_identifier(flow, skb);
 833         if (err)
 834                 goto error;
 835
 836         if (should_fill_key(&flow->id, ufid_flags)) {
 837                 err = ovs_nla_put_masked_key(flow, skb);
 838                 if (err)
 839                         goto error;
 840         }
 841
 842         if (should_fill_mask(ufid_flags)) {
 843                 err = ovs_nla_put_mask(flow, skb);
 844                 if (err)
 845                         goto error;
 846         }
 847
 848         err = ovs_flow_cmd_fill_stats(flow, skb);
 849         if (err)
 850                 goto error;
 851
 852         if (should_fill_actions(ufid_flags)) {
 853                 err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
 854                 if (err)
 855                         goto error;
 856         }
 857
 858         genlmsg_end(skb, ovs_header);
 859         return 0;
 860
 861 error:
 862         genlmsg_cancel(skb, ovs_header);
 863         return err;
 864 }
 865
 866 /* May not be called with RCU read lock. */
 867 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
 868                                                const struct sw_flow_id *sfid,
 869                                                struct genl_info *info,
 870                                                bool always,
 871                                                uint32_t ufid_flags)
 872 {
 873         struct sk_buff *skb;
 874         size_t len;
 875
 876         if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
 877                 return NULL;
 878
 879         len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
 880         skb = genlmsg_new(len, GFP_KERNEL);
 881         if (!skb)
 882                 return ERR_PTR(-ENOMEM);
 883
 884         return skb;
 885 }
 886
 887 /* Called with ovs_mutex. */
 888 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
 889                                                int dp_ifindex,
 890                                                struct genl_info *info, u8 cmd,
 891                                                bool always, u32 ufid_flags)
 892 {
 893         struct sk_buff *skb;
 894         int retval;
 895
 896         skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
 897                                       &flow->id, info, always, ufid_flags);
 898         if (IS_ERR_OR_NULL(skb))
 899                 return skb;
 900
 901         retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
 902                                         info->snd_portid, info->snd_seq, 0,
 903                                         cmd, ufid_flags);
 904         BUG_ON(retval < 0);
 905         return skb;
 906 }
 907
 908 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
 909 {
 910         struct net *net = sock_net(skb->sk);
 911         struct nlattr **a = info->attrs;
 912         struct ovs_header *ovs_header = info->userhdr;
 913         struct sw_flow *flow = NULL, *new_flow;
 914         struct sw_flow_mask mask;
 915         struct sk_buff *reply;
 916         struct datapath *dp;
 917         struct sw_flow_actions *acts;
 918         struct sw_flow_match match;
 919         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
 920         int error;
 921         bool log = !a[OVS_FLOW_ATTR_PROBE];
 922
 923         /* Must have key and actions. */
 924         error = -EINVAL;
 925         if (!a[OVS_FLOW_ATTR_KEY]) {
 926                 OVS_NLERR(log, "Flow key attr not present in new flow.");
 927                 goto error;
 928         }
 929         if (!a[OVS_FLOW_ATTR_ACTIONS]) {
 930                 OVS_NLERR(log, "Flow actions attr not present in new flow.");
 931                 goto error;
 932         }
 933
 934         /* Most of the time we need to allocate a new flow, do it before
 935          * locking.
 936          */
 937         new_flow = ovs_flow_alloc();
 938         if (IS_ERR(new_flow)) {
 939                 error = PTR_ERR(new_flow);
 940                 goto error;
 941         }
 942
 943         /* Extract key. */
 944         ovs_match_init(&match, &new_flow->key, false, &mask);
 945         error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
 946                                   a[OVS_FLOW_ATTR_MASK], log);
 947         if (error)
 948                 goto err_kfree_flow;
 949
 950         /* Extract flow identifier. */
 951         error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
 952                                        &new_flow->key, log);
 953         if (error)
 954                 goto err_kfree_flow;
 955
 956         /* unmasked key is needed to match when ufid is not used. */
 957         if (ovs_identifier_is_key(&new_flow->id))
 958                 match.key = new_flow->id.unmasked_key;
 959
 960         ovs_flow_mask_key(&new_flow->key, &new_flow->key, true, &mask);
 961
 962         /* Validate actions. */
 963         error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
 964                                      &new_flow->key, &acts, log);
 965         if (error) {
 966                 OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
 967                 goto err_kfree_flow;
 968         }
 969
 970         reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
 971                                         ufid_flags);
 972         if (IS_ERR(reply)) {
 973                 error = PTR_ERR(reply);
 974                 goto err_kfree_acts;
 975         }
 976
 977         ovs_lock();
 978         dp = get_dp(net, ovs_header->dp_ifindex);
 979         if (unlikely(!dp)) {
 980                 error = -ENODEV;
 981                 goto err_unlock_ovs;
 982         }
 983
 984         /* Check if this is a duplicate flow */
 985         if (ovs_identifier_is_ufid(&new_flow->id))
 986                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
 987         if (!flow)
 988                 flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->key);
 989         if (likely(!flow)) {
 990                 rcu_assign_pointer(new_flow->sf_acts, acts);
 991
 992                 /* Put flow in bucket. */
 993                 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
 994                 if (unlikely(error)) {
 995                         acts = NULL;
 996                         goto err_unlock_ovs;
 997                 }
 998
 999                 if (unlikely(reply)) {
1000                         error = ovs_flow_cmd_fill_info(new_flow,
1001                                                        ovs_header->dp_ifindex,
1002                                                        reply, info->snd_portid,
1003                                                        info->snd_seq, 0,
1004                                                        OVS_FLOW_CMD_NEW,
1005                                                        ufid_flags);
1006                         BUG_ON(error < 0);
1007                 }
1008                 ovs_unlock();
1009         } else {
1010                 struct sw_flow_actions *old_acts;
1011
1012                 /* Bail out if we're not allowed to modify an existing flow.
1013                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1014                  * because Generic Netlink treats the latter as a dump
1015                  * request.  We also accept NLM_F_EXCL in case that bug ever
1016                  * gets fixed.
1017                  */
1018                 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1019                                                          | NLM_F_EXCL))) {
1020                         error = -EEXIST;
1021                         goto err_unlock_ovs;
1022                 }
1023                 /* The flow identifier has to be the same for flow updates.
1024                  * Look for any overlapping flow.
1025                  */
1026                 if (unlikely(!ovs_flow_cmp(flow, &match))) {
1027                         if (ovs_identifier_is_key(&flow->id))
1028                                 flow = ovs_flow_tbl_lookup_exact(&dp->table,
1029                                                                  &match);
1030                         else /* UFID matches but key is different */
1031                                 flow = NULL;
1032                         if (!flow) {
1033                                 error = -ENOENT;
1034                                 goto err_unlock_ovs;
1035                         }
1036                 }
1037                 /* Update actions. */
1038                 old_acts = ovsl_dereference(flow->sf_acts);
1039                 rcu_assign_pointer(flow->sf_acts, acts);
1040
1041                 if (unlikely(reply)) {
1042                         error = ovs_flow_cmd_fill_info(flow,
1043                                                        ovs_header->dp_ifindex,
1044                                                        reply, info->snd_portid,
1045                                                        info->snd_seq, 0,
1046                                                        OVS_FLOW_CMD_NEW,
1047                                                        ufid_flags);
1048                         BUG_ON(error < 0);
1049                 }
1050                 ovs_unlock();
1051
1052                 ovs_nla_free_flow_actions_rcu(old_acts);
1053                 ovs_flow_free(new_flow, false);
1054         }
1055
1056         if (reply)
1057                 ovs_notify(&dp_flow_genl_family, reply, info);
1058         return 0;
1059
1060 err_unlock_ovs:
1061         ovs_unlock();
1062         kfree_skb(reply);
1063 err_kfree_acts:
1064         ovs_nla_free_flow_actions(acts);
1065 err_kfree_flow:
1066         ovs_flow_free(new_flow, false);
1067 error:
1068         return error;
1069 }
1070
1071 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1072 static struct sw_flow_actions *get_flow_actions(struct net *net,
1073                                                 const struct nlattr *a,
1074                                                 const struct sw_flow_key *key,
1075                                                 const struct sw_flow_mask *mask,
1076                                                 bool log)
1077 {
1078         struct sw_flow_actions *acts;
1079         struct sw_flow_key masked_key;
1080         int error;
1081
1082         ovs_flow_mask_key(&masked_key, key, true, mask);
1083         error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1084         if (error) {
1085                 OVS_NLERR(log,
1086                           "Actions may not be safe on all matching packets");
1087                 return ERR_PTR(error);
1088         }
1089
1090         return acts;
1091 }
1092
1093 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1094 {
1095         struct net *net = sock_net(skb->sk);
1096         struct nlattr **a = info->attrs;
1097         struct ovs_header *ovs_header = info->userhdr;
1098         struct sw_flow_key key;
1099         struct sw_flow *flow;
1100         struct sw_flow_mask mask;
1101         struct sk_buff *reply = NULL;
1102         struct datapath *dp;
1103         struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1104         struct sw_flow_match match;
1105         struct sw_flow_id sfid;
1106         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1107         int error = 0;
1108         bool log = !a[OVS_FLOW_ATTR_PROBE];
1109         bool ufid_present;
1110
1111         ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1112         if (a[OVS_FLOW_ATTR_KEY]) {
1113                 ovs_match_init(&match, &key, true, &mask);
1114                 error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1115                                           a[OVS_FLOW_ATTR_MASK], log);
1116         } else if (!ufid_present) {
1117                 OVS_NLERR(log,
1118                           "Flow set message rejected, Key attribute missing.");
1119                 error = -EINVAL;
1120         }
1121         if (error)
1122                 goto error;
1123
1124         /* Validate actions. */
1125         if (a[OVS_FLOW_ATTR_ACTIONS]) {
1126                 if (!a[OVS_FLOW_ATTR_KEY]) {
1127                         OVS_NLERR(log,
1128                                   "Flow key attribute not present in set flow.");
1129                         error = -EINVAL;
1130                         goto error;
1131                 }
1132
1133                 acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], &key,
1134                                         &mask, log);
1135                 if (IS_ERR(acts)) {
1136                         error = PTR_ERR(acts);
1137                         goto error;
1138                 }
1139
1140                 /* Can allocate before locking if have acts. */
1141                 reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1142                                                 ufid_flags);
1143                 if (IS_ERR(reply)) {
1144                         error = PTR_ERR(reply);
1145                         goto err_kfree_acts;
1146                 }
1147         }
1148
1149         ovs_lock();
1150         dp = get_dp(net, ovs_header->dp_ifindex);
1151         if (unlikely(!dp)) {
1152                 error = -ENODEV;
1153                 goto err_unlock_ovs;
1154         }
1155         /* Check that the flow exists. */
1156         if (ufid_present)
1157                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1158         else
1159                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1160         if (unlikely(!flow)) {
1161                 error = -ENOENT;
1162                 goto err_unlock_ovs;
1163         }
1164
1165         /* Update actions, if present. */
1166         if (likely(acts)) {
1167                 old_acts = ovsl_dereference(flow->sf_acts);
1168                 rcu_assign_pointer(flow->sf_acts, acts);
1169
1170                 if (unlikely(reply)) {
1171                         error = ovs_flow_cmd_fill_info(flow,
1172                                                        ovs_header->dp_ifindex,
1173                                                        reply, info->snd_portid,
1174                                                        info->snd_seq, 0,
1175                                                        OVS_FLOW_CMD_NEW,
1176                                                        ufid_flags);
1177                         BUG_ON(error < 0);
1178                 }
1179         } else {
1180                 /* Could not alloc without acts before locking. */
1181                 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1182                                                 info, OVS_FLOW_CMD_NEW, false,
1183                                                 ufid_flags);
1184
1185                 if (IS_ERR(reply)) {
1186                         error = PTR_ERR(reply);
1187                         goto err_unlock_ovs;
1188                 }
1189         }
1190
1191         /* Clear stats. */
1192         if (a[OVS_FLOW_ATTR_CLEAR])
1193                 ovs_flow_stats_clear(flow);
1194         ovs_unlock();
1195
1196         if (reply)
1197                 ovs_notify(&dp_flow_genl_family, reply, info);
1198         if (old_acts)
1199                 ovs_nla_free_flow_actions_rcu(old_acts);
1200
1201         return 0;
1202
1203 err_unlock_ovs:
1204         ovs_unlock();
1205         kfree_skb(reply);
1206 err_kfree_acts:
1207         ovs_nla_free_flow_actions(acts);
1208 error:
1209         return error;
1210 }
1211
1212 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1213 {
1214         struct nlattr **a = info->attrs;
1215         struct ovs_header *ovs_header = info->userhdr;
1216         struct net *net = sock_net(skb->sk);
1217         struct sw_flow_key key;
1218         struct sk_buff *reply;
1219         struct sw_flow *flow;
1220         struct datapath *dp;
1221         struct sw_flow_match match;
1222         struct sw_flow_id ufid;
1223         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1224         int err = 0;
1225         bool log = !a[OVS_FLOW_ATTR_PROBE];
1226         bool ufid_present;
1227
1228         ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1229         if (a[OVS_FLOW_ATTR_KEY]) {
1230                 ovs_match_init(&match, &key, true, NULL);
1231                 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1232                                         log);
1233         } else if (!ufid_present) {
1234                 OVS_NLERR(log,
1235                           "Flow get message rejected, Key attribute missing.");
1236                 err = -EINVAL;
1237         }
1238         if (err)
1239                 return err;
1240
1241         ovs_lock();
1242         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1243         if (!dp) {
1244                 err = -ENODEV;
1245                 goto unlock;
1246         }
1247
1248         if (ufid_present)
1249                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1250         else
1251                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1252         if (!flow) {
1253                 err = -ENOENT;
1254                 goto unlock;
1255         }
1256
1257         reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1258                                         OVS_FLOW_CMD_NEW, true, ufid_flags);
1259         if (IS_ERR(reply)) {
1260                 err = PTR_ERR(reply);
1261                 goto unlock;
1262         }
1263
1264         ovs_unlock();
1265         return genlmsg_reply(reply, info);
1266 unlock:
1267         ovs_unlock();
1268         return err;
1269 }
1270
1271 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1272 {
1273         struct nlattr **a = info->attrs;
1274         struct ovs_header *ovs_header = info->userhdr;
1275         struct net *net = sock_net(skb->sk);
1276         struct sw_flow_key key;
1277         struct sk_buff *reply;
1278         struct sw_flow *flow = NULL;
1279         struct datapath *dp;
1280         struct sw_flow_match match;
1281         struct sw_flow_id ufid;
1282         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1283         int err;
1284         bool log = !a[OVS_FLOW_ATTR_PROBE];
1285         bool ufid_present;
1286
1287         ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1288         if (a[OVS_FLOW_ATTR_KEY]) {
1289                 ovs_match_init(&match, &key, true, NULL);
1290                 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1291                                         NULL, log);
1292                 if (unlikely(err))
1293                         return err;
1294         }
1295
1296         ovs_lock();
1297         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1298         if (unlikely(!dp)) {
1299                 err = -ENODEV;
1300                 goto unlock;
1301         }
1302
1303         if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1304                 err = ovs_flow_tbl_flush(&dp->table);
1305                 goto unlock;
1306         }
1307
1308         if (ufid_present)
1309                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1310         else
1311                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1312         if (unlikely(!flow)) {
1313                 err = -ENOENT;
1314                 goto unlock;
1315         }
1316
1317         ovs_flow_tbl_remove(&dp->table, flow);
1318         ovs_unlock();
1319
1320         reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1321                                         &flow->id, info, false, ufid_flags);
1322         if (likely(reply)) {
1323                 if (likely(!IS_ERR(reply))) {
1324                         rcu_read_lock();        /*To keep RCU checker happy. */
1325                         err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1326                                                      reply, info->snd_portid,
1327                                                      info->snd_seq, 0,
1328                                                      OVS_FLOW_CMD_DEL,
1329                                                      ufid_flags);
1330                         rcu_read_unlock();
1331                         BUG_ON(err < 0);
1332
1333                         ovs_notify(&dp_flow_genl_family, reply, info);
1334                 } else {
1335                         netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
1336                 }
1337         }
1338
1339         ovs_flow_free(flow, true);
1340         return 0;
1341 unlock:
1342         ovs_unlock();
1343         return err;
1344 }
1345
1346 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1347 {
1348         struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1349         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1350         struct table_instance *ti;
1351         struct datapath *dp;
1352         u32 ufid_flags;
1353         int err;
1354
1355         err = genlmsg_parse(cb->nlh, &dp_flow_genl_family, a,
1356                             OVS_FLOW_ATTR_MAX, flow_policy, NULL);
1357         if (err)
1358                 return err;
1359         ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1360
1361         rcu_read_lock();
1362         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1363         if (!dp) {
1364                 rcu_read_unlock();
1365                 return -ENODEV;
1366         }
1367
1368         ti = rcu_dereference(dp->table.ti);
1369         for (;;) {
1370                 struct sw_flow *flow;
1371                 u32 bucket, obj;
1372
1373                 bucket = cb->args[0];
1374                 obj = cb->args[1];
1375                 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1376                 if (!flow)
1377                         break;
1378
1379                 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1380                                            NETLINK_CB(cb->skb).portid,
1381                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1382                                            OVS_FLOW_CMD_NEW, ufid_flags) < 0)
1383                         break;
1384
1385                 cb->args[0] = bucket;
1386                 cb->args[1] = obj;
1387         }
1388         rcu_read_unlock();
1389         return skb->len;
1390 }
1391
1392 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1393         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1394         [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1395         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1396         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1397         [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1398         [OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1399         [OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1400 };
1401
1402 static const struct genl_ops dp_flow_genl_ops[] = {
1403         { .cmd = OVS_FLOW_CMD_NEW,
1404           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1405           .policy = flow_policy,
1406           .doit = ovs_flow_cmd_new
1407         },
1408         { .cmd = OVS_FLOW_CMD_DEL,
1409           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1410           .policy = flow_policy,
1411           .doit = ovs_flow_cmd_del
1412         },
1413         { .cmd = OVS_FLOW_CMD_GET,
1414           .flags = 0,               /* OK for unprivileged users. */
1415           .policy = flow_policy,
1416           .doit = ovs_flow_cmd_get,
1417           .dumpit = ovs_flow_cmd_dump
1418         },
1419         { .cmd = OVS_FLOW_CMD_SET,
1420           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1421           .policy = flow_policy,
1422           .doit = ovs_flow_cmd_set,
1423         },
1424 };
1425
1426 static struct genl_family dp_flow_genl_family __ro_after_init = {
1427         .hdrsize = sizeof(struct ovs_header),
1428         .name = OVS_FLOW_FAMILY,
1429         .version = OVS_FLOW_VERSION,
1430         .maxattr = OVS_FLOW_ATTR_MAX,
1431         .netnsok = true,
1432         .parallel_ops = true,
1433         .ops = dp_flow_genl_ops,
1434         .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1435         .mcgrps = &ovs_dp_flow_multicast_group,
1436         .n_mcgrps = 1,
1437         .module = THIS_MODULE,
1438 };
1439
1440 static size_t ovs_dp_cmd_msg_size(void)
1441 {
1442         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1443
1444         msgsize += nla_total_size(IFNAMSIZ);
1445         msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1446         msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
1447         msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1448
1449         return msgsize;
1450 }
1451
1452 /* Called with ovs_mutex. */
1453 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1454                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1455 {
1456         struct ovs_header *ovs_header;
1457         struct ovs_dp_stats dp_stats;
1458         struct ovs_dp_megaflow_stats dp_megaflow_stats;
1459         int err;
1460
1461         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1462                                    flags, cmd);
1463         if (!ovs_header)
1464                 goto error;
1465
1466         ovs_header->dp_ifindex = get_dpifindex(dp);
1467
1468         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1469         if (err)
1470                 goto nla_put_failure;
1471
1472         get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1473         if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1474                           &dp_stats, OVS_DP_ATTR_PAD))
1475                 goto nla_put_failure;
1476
1477         if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1478                           sizeof(struct ovs_dp_megaflow_stats),
1479                           &dp_megaflow_stats, OVS_DP_ATTR_PAD))
1480                 goto nla_put_failure;
1481
1482         if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1483                 goto nla_put_failure;
1484
1485         genlmsg_end(skb, ovs_header);
1486         return 0;
1487
1488 nla_put_failure:
1489         genlmsg_cancel(skb, ovs_header);
1490 error:
1491         return -EMSGSIZE;
1492 }
1493
1494 static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1495 {
1496         return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1497 }
1498
1499 /* Called with rcu_read_lock or ovs_mutex. */
1500 static struct datapath *lookup_datapath(struct net *net,
1501                                         const struct ovs_header *ovs_header,
1502                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1503 {
1504         struct datapath *dp;
1505
1506         if (!a[OVS_DP_ATTR_NAME])
1507                 dp = get_dp(net, ovs_header->dp_ifindex);
1508         else {
1509                 struct vport *vport;
1510
1511                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1512                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1513         }
1514         return dp ? dp : ERR_PTR(-ENODEV);
1515 }
1516
1517 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1518 {
1519         struct datapath *dp;
1520
1521         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1522         if (IS_ERR(dp))
1523                 return;
1524
1525         WARN(dp->user_features, "Dropping previously announced user features\n");
1526         dp->user_features = 0;
1527 }
1528
1529 static void ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1530 {
1531         if (a[OVS_DP_ATTR_USER_FEATURES])
1532                 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1533 }
1534
1535 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1536 {
1537         struct nlattr **a = info->attrs;
1538         struct vport_parms parms;
1539         struct sk_buff *reply;
1540         struct datapath *dp;
1541         struct vport *vport;
1542         struct ovs_net *ovs_net;
1543         int err, i;
1544
1545         err = -EINVAL;
1546         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1547                 goto err;
1548
1549         reply = ovs_dp_cmd_alloc_info();
1550         if (!reply)
1551                 return -ENOMEM;
1552
1553         err = -ENOMEM;
1554         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1555         if (dp == NULL)
1556                 goto err_free_reply;
1557
1558         ovs_dp_set_net(dp, sock_net(skb->sk));
1559
1560         /* Allocate table. */
1561         err = ovs_flow_tbl_init(&dp->table);
1562         if (err)
1563                 goto err_free_dp;
1564
1565         dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1566         if (!dp->stats_percpu) {
1567                 err = -ENOMEM;
1568                 goto err_destroy_table;
1569         }
1570
1571         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1572                             GFP_KERNEL);
1573         if (!dp->ports) {
1574                 err = -ENOMEM;
1575                 goto err_destroy_percpu;
1576         }
1577
1578         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1579                 INIT_HLIST_HEAD(&dp->ports[i]);
1580
1581         /* Set up our datapath device. */
1582         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1583         parms.type = OVS_VPORT_TYPE_INTERNAL;
1584         parms.options = NULL;
1585         parms.dp = dp;
1586         parms.port_no = OVSP_LOCAL;
1587         parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1588
1589         ovs_dp_change(dp, a);
1590
1591         /* So far only local changes have been made, now need the lock. */
1592         ovs_lock();
1593
1594         vport = new_vport(&parms);
1595         if (IS_ERR(vport)) {
1596                 err = PTR_ERR(vport);
1597                 if (err == -EBUSY)
1598                         err = -EEXIST;
1599
1600                 if (err == -EEXIST) {
1601                         /* An outdated user space instance that does not understand
1602                          * the concept of user_features has attempted to create a new
1603                          * datapath and is likely to reuse it. Drop all user features.
1604                          */
1605                         if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1606                                 ovs_dp_reset_user_features(skb, info);
1607                 }
1608
1609                 goto err_destroy_ports_array;
1610         }
1611
1612         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1613                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1614         BUG_ON(err < 0);
1615
1616         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1617         list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1618
1619         ovs_unlock();
1620
1621         ovs_notify(&dp_datapath_genl_family, reply, info);
1622         return 0;
1623
1624 err_destroy_ports_array:
1625         ovs_unlock();
1626         kfree(dp->ports);
1627 err_destroy_percpu:
1628         free_percpu(dp->stats_percpu);
1629 err_destroy_table:
1630         ovs_flow_tbl_destroy(&dp->table);
1631 err_free_dp:
1632         kfree(dp);
1633 err_free_reply:
1634         kfree_skb(reply);
1635 err:
1636         return err;
1637 }
1638
1639 /* Called with ovs_mutex. */
1640 static void __dp_destroy(struct datapath *dp)
1641 {
1642         int i;
1643
1644         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1645                 struct vport *vport;
1646                 struct hlist_node *n;
1647
1648                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1649                         if (vport->port_no != OVSP_LOCAL)
1650                                 ovs_dp_detach_port(vport);
1651         }
1652
1653         list_del_rcu(&dp->list_node);
1654
1655         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1656          * all ports in datapath are destroyed first before freeing datapath.
1657          */
1658         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1659
1660         /* RCU destroy the flow table */
1661         call_rcu(&dp->rcu, destroy_dp_rcu);
1662 }
1663
1664 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1665 {
1666         struct sk_buff *reply;
1667         struct datapath *dp;
1668         int err;
1669
1670         reply = ovs_dp_cmd_alloc_info();
1671         if (!reply)
1672                 return -ENOMEM;
1673
1674         ovs_lock();
1675         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1676         err = PTR_ERR(dp);
1677         if (IS_ERR(dp))
1678                 goto err_unlock_free;
1679
1680         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1681                                    info->snd_seq, 0, OVS_DP_CMD_DEL);
1682         BUG_ON(err < 0);
1683
1684         __dp_destroy(dp);
1685         ovs_unlock();
1686
1687         ovs_notify(&dp_datapath_genl_family, reply, info);
1688
1689         return 0;
1690
1691 err_unlock_free:
1692         ovs_unlock();
1693         kfree_skb(reply);
1694         return err;
1695 }
1696
1697 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1698 {
1699         struct sk_buff *reply;
1700         struct datapath *dp;
1701         int err;
1702
1703         reply = ovs_dp_cmd_alloc_info();
1704         if (!reply)
1705                 return -ENOMEM;
1706
1707         ovs_lock();
1708         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1709         err = PTR_ERR(dp);
1710         if (IS_ERR(dp))
1711                 goto err_unlock_free;
1712
1713         ovs_dp_change(dp, info->attrs);
1714
1715         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1716                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1717         BUG_ON(err < 0);
1718
1719         ovs_unlock();
1720         ovs_notify(&dp_datapath_genl_family, reply, info);
1721
1722         return 0;
1723
1724 err_unlock_free:
1725         ovs_unlock();
1726         kfree_skb(reply);
1727         return err;
1728 }
1729
1730 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1731 {
1732         struct sk_buff *reply;
1733         struct datapath *dp;
1734         int err;
1735
1736         reply = ovs_dp_cmd_alloc_info();
1737         if (!reply)
1738                 return -ENOMEM;
1739
1740         ovs_lock();
1741         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1742         if (IS_ERR(dp)) {
1743                 err = PTR_ERR(dp);
1744                 goto err_unlock_free;
1745         }
1746         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1747                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1748         BUG_ON(err < 0);
1749         ovs_unlock();
1750
1751         return genlmsg_reply(reply, info);
1752
1753 err_unlock_free:
1754         ovs_unlock();
1755         kfree_skb(reply);
1756         return err;
1757 }
1758
1759 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1760 {
1761         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1762         struct datapath *dp;
1763         int skip = cb->args[0];
1764         int i = 0;
1765
1766         ovs_lock();
1767         list_for_each_entry(dp, &ovs_net->dps, list_node) {
1768                 if (i >= skip &&
1769                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1770                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1771                                          OVS_DP_CMD_NEW) < 0)
1772                         break;
1773                 i++;
1774         }
1775         ovs_unlock();
1776
1777         cb->args[0] = i;
1778
1779         return skb->len;
1780 }
1781
1782 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1783         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1784         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1785         [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1786 };
1787
1788 static const struct genl_ops dp_datapath_genl_ops[] = {
1789         { .cmd = OVS_DP_CMD_NEW,
1790           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1791           .policy = datapath_policy,
1792           .doit = ovs_dp_cmd_new
1793         },
1794         { .cmd = OVS_DP_CMD_DEL,
1795           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1796           .policy = datapath_policy,
1797           .doit = ovs_dp_cmd_del
1798         },
1799         { .cmd = OVS_DP_CMD_GET,
1800           .flags = 0,               /* OK for unprivileged users. */
1801           .policy = datapath_policy,
1802           .doit = ovs_dp_cmd_get,
1803           .dumpit = ovs_dp_cmd_dump
1804         },
1805         { .cmd = OVS_DP_CMD_SET,
1806           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1807           .policy = datapath_policy,
1808           .doit = ovs_dp_cmd_set,
1809         },
1810 };
1811
1812 static struct genl_family dp_datapath_genl_family __ro_after_init = {
1813         .hdrsize = sizeof(struct ovs_header),
1814         .name = OVS_DATAPATH_FAMILY,
1815         .version = OVS_DATAPATH_VERSION,
1816         .maxattr = OVS_DP_ATTR_MAX,
1817         .netnsok = true,
1818         .parallel_ops = true,
1819         .ops = dp_datapath_genl_ops,
1820         .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1821         .mcgrps = &ovs_dp_datapath_multicast_group,
1822         .n_mcgrps = 1,
1823         .module = THIS_MODULE,
1824 };
1825
1826 /* Called with ovs_mutex or RCU read lock. */
1827 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1828                                    u32 portid, u32 seq, u32 flags, u8 cmd)
1829 {
1830         struct ovs_header *ovs_header;
1831         struct ovs_vport_stats vport_stats;
1832         int err;
1833
1834         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1835                                  flags, cmd);
1836         if (!ovs_header)
1837                 return -EMSGSIZE;
1838
1839         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1840
1841         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1842             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1843             nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1844                            ovs_vport_name(vport)))
1845                 goto nla_put_failure;
1846
1847         ovs_vport_get_stats(vport, &vport_stats);
1848         if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
1849                           sizeof(struct ovs_vport_stats), &vport_stats,
1850                           OVS_VPORT_ATTR_PAD))
1851                 goto nla_put_failure;
1852
1853         if (ovs_vport_get_upcall_portids(vport, skb))
1854                 goto nla_put_failure;
1855
1856         err = ovs_vport_get_options(vport, skb);
1857         if (err == -EMSGSIZE)
1858                 goto error;
1859
1860         genlmsg_end(skb, ovs_header);
1861         return 0;
1862
1863 nla_put_failure:
1864         err = -EMSGSIZE;
1865 error:
1866         genlmsg_cancel(skb, ovs_header);
1867         return err;
1868 }
1869
1870 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1871 {
1872         return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1873 }
1874
1875 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1876 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1877                                          u32 seq, u8 cmd)
1878 {
1879         struct sk_buff *skb;
1880         int retval;
1881
1882         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1883         if (!skb)
1884                 return ERR_PTR(-ENOMEM);
1885
1886         retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1887         BUG_ON(retval < 0);
1888
1889         return skb;
1890 }
1891
1892 /* Called with ovs_mutex or RCU read lock. */
1893 static struct vport *lookup_vport(struct net *net,
1894                                   const struct ovs_header *ovs_header,
1895                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1896 {
1897         struct datapath *dp;
1898         struct vport *vport;
1899
1900         if (a[OVS_VPORT_ATTR_NAME]) {
1901                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1902                 if (!vport)
1903                         return ERR_PTR(-ENODEV);
1904                 if (ovs_header->dp_ifindex &&
1905                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1906                         return ERR_PTR(-ENODEV);
1907                 return vport;
1908         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1909                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1910
1911                 if (port_no >= DP_MAX_PORTS)
1912                         return ERR_PTR(-EFBIG);
1913
1914                 dp = get_dp(net, ovs_header->dp_ifindex);
1915                 if (!dp)
1916                         return ERR_PTR(-ENODEV);
1917
1918                 vport = ovs_vport_ovsl_rcu(dp, port_no);
1919                 if (!vport)
1920                         return ERR_PTR(-ENODEV);
1921                 return vport;
1922         } else
1923                 return ERR_PTR(-EINVAL);
1924 }
1925
1926 /* Called with ovs_mutex */
1927 static void update_headroom(struct datapath *dp)
1928 {
1929         unsigned dev_headroom, max_headroom = 0;
1930         struct net_device *dev;
1931         struct vport *vport;
1932         int i;
1933
1934         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1935                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1936                         dev = vport->dev;
1937                         dev_headroom = netdev_get_fwd_headroom(dev);
1938                         if (dev_headroom > max_headroom)
1939                                 max_headroom = dev_headroom;
1940                 }
1941         }
1942
1943         dp->max_headroom = max_headroom;
1944         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1945                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node)
1946                         netdev_set_rx_headroom(vport->dev, max_headroom);
1947 }
1948
1949 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1950 {
1951         struct nlattr **a = info->attrs;
1952         struct ovs_header *ovs_header = info->userhdr;
1953         struct vport_parms parms;
1954         struct sk_buff *reply;
1955         struct vport *vport;
1956         struct datapath *dp;
1957         u32 port_no;
1958         int err;
1959
1960         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1961             !a[OVS_VPORT_ATTR_UPCALL_PID])
1962                 return -EINVAL;
1963
1964         port_no = a[OVS_VPORT_ATTR_PORT_NO]
1965                 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
1966         if (port_no >= DP_MAX_PORTS)
1967                 return -EFBIG;
1968
1969         reply = ovs_vport_cmd_alloc_info();
1970         if (!reply)
1971                 return -ENOMEM;
1972
1973         ovs_lock();
1974 restart:
1975         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1976         err = -ENODEV;
1977         if (!dp)
1978                 goto exit_unlock_free;
1979
1980         if (port_no) {
1981                 vport = ovs_vport_ovsl(dp, port_no);
1982                 err = -EBUSY;
1983                 if (vport)
1984                         goto exit_unlock_free;
1985         } else {
1986                 for (port_no = 1; ; port_no++) {
1987                         if (port_no >= DP_MAX_PORTS) {
1988                                 err = -EFBIG;
1989                                 goto exit_unlock_free;
1990                         }
1991                         vport = ovs_vport_ovsl(dp, port_no);
1992                         if (!vport)
1993                                 break;
1994                 }
1995         }
1996
1997         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1998         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1999         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2000         parms.dp = dp;
2001         parms.port_no = port_no;
2002         parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
2003
2004         vport = new_vport(&parms);
2005         err = PTR_ERR(vport);
2006         if (IS_ERR(vport)) {
2007                 if (err == -EAGAIN)
2008                         goto restart;
2009                 goto exit_unlock_free;
2010         }
2011
2012         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2013                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2014
2015         if (netdev_get_fwd_headroom(vport->dev) > dp->max_headroom)
2016                 update_headroom(dp);
2017         else
2018                 netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2019
2020         BUG_ON(err < 0);
2021         ovs_unlock();
2022
2023         ovs_notify(&dp_vport_genl_family, reply, info);
2024         return 0;
2025
2026 exit_unlock_free:
2027         ovs_unlock();
2028         kfree_skb(reply);
2029         return err;
2030 }
2031
2032 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2033 {
2034         struct nlattr **a = info->attrs;
2035         struct sk_buff *reply;
2036         struct vport *vport;
2037         int err;
2038
2039         reply = ovs_vport_cmd_alloc_info();
2040         if (!reply)
2041                 return -ENOMEM;
2042
2043         ovs_lock();
2044         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2045         err = PTR_ERR(vport);
2046         if (IS_ERR(vport))
2047                 goto exit_unlock_free;
2048
2049         if (a[OVS_VPORT_ATTR_TYPE] &&
2050             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2051                 err = -EINVAL;
2052                 goto exit_unlock_free;
2053         }
2054
2055         if (a[OVS_VPORT_ATTR_OPTIONS]) {
2056                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2057                 if (err)
2058                         goto exit_unlock_free;
2059         }
2060
2061
2062         if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2063                 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2064
2065                 err = ovs_vport_set_upcall_portids(vport, ids);
2066                 if (err)
2067                         goto exit_unlock_free;
2068         }
2069
2070         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2071                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2072         BUG_ON(err < 0);
2073
2074         ovs_unlock();
2075         ovs_notify(&dp_vport_genl_family, reply, info);
2076         return 0;
2077
2078 exit_unlock_free:
2079         ovs_unlock();
2080         kfree_skb(reply);
2081         return err;
2082 }
2083
2084 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2085 {
2086         bool must_update_headroom = false;
2087         struct nlattr **a = info->attrs;
2088         struct sk_buff *reply;
2089         struct datapath *dp;
2090         struct vport *vport;
2091         int err;
2092
2093         reply = ovs_vport_cmd_alloc_info();
2094         if (!reply)
2095                 return -ENOMEM;
2096
2097         ovs_lock();
2098         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2099         err = PTR_ERR(vport);
2100         if (IS_ERR(vport))
2101                 goto exit_unlock_free;
2102
2103         if (vport->port_no == OVSP_LOCAL) {
2104                 err = -EINVAL;
2105                 goto exit_unlock_free;
2106         }
2107
2108         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2109                                       info->snd_seq, 0, OVS_VPORT_CMD_DEL);
2110         BUG_ON(err < 0);
2111
2112         /* the vport deletion may trigger dp headroom update */
2113         dp = vport->dp;
2114         if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2115                 must_update_headroom = true;
2116         netdev_reset_rx_headroom(vport->dev);
2117         ovs_dp_detach_port(vport);
2118
2119         if (must_update_headroom)
2120                 update_headroom(dp);
2121         ovs_unlock();
2122
2123         ovs_notify(&dp_vport_genl_family, reply, info);
2124         return 0;
2125
2126 exit_unlock_free:
2127         ovs_unlock();
2128         kfree_skb(reply);
2129         return err;
2130 }
2131
2132 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2133 {
2134         struct nlattr **a = info->attrs;
2135         struct ovs_header *ovs_header = info->userhdr;
2136         struct sk_buff *reply;
2137         struct vport *vport;
2138         int err;
2139
2140         reply = ovs_vport_cmd_alloc_info();
2141         if (!reply)
2142                 return -ENOMEM;
2143
2144         rcu_read_lock();
2145         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2146         err = PTR_ERR(vport);
2147         if (IS_ERR(vport))
2148                 goto exit_unlock_free;
2149         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2150                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2151         BUG_ON(err < 0);
2152         rcu_read_unlock();
2153
2154         return genlmsg_reply(reply, info);
2155
2156 exit_unlock_free:
2157         rcu_read_unlock();
2158         kfree_skb(reply);
2159         return err;
2160 }
2161
2162 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2163 {
2164         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2165         struct datapath *dp;
2166         int bucket = cb->args[0], skip = cb->args[1];
2167         int i, j = 0;
2168
2169         rcu_read_lock();
2170         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2171         if (!dp) {
2172                 rcu_read_unlock();
2173                 return -ENODEV;
2174         }
2175         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2176                 struct vport *vport;
2177
2178                 j = 0;
2179                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2180                         if (j >= skip &&
2181                             ovs_vport_cmd_fill_info(vport, skb,
2182                                                     NETLINK_CB(cb->skb).portid,
2183                                                     cb->nlh->nlmsg_seq,
2184                                                     NLM_F_MULTI,
2185                                                     OVS_VPORT_CMD_NEW) < 0)
2186                                 goto out;
2187
2188                         j++;
2189                 }
2190                 skip = 0;
2191         }
2192 out:
2193         rcu_read_unlock();
2194
2195         cb->args[0] = i;
2196         cb->args[1] = j;
2197
2198         return skb->len;
2199 }
2200
2201 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2202         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2203         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2204         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2205         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2206         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
2207         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2208 };
2209
2210 static const struct genl_ops dp_vport_genl_ops[] = {
2211         { .cmd = OVS_VPORT_CMD_NEW,
2212           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2213           .policy = vport_policy,
2214           .doit = ovs_vport_cmd_new
2215         },
2216         { .cmd = OVS_VPORT_CMD_DEL,
2217           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2218           .policy = vport_policy,
2219           .doit = ovs_vport_cmd_del
2220         },
2221         { .cmd = OVS_VPORT_CMD_GET,
2222           .flags = 0,               /* OK for unprivileged users. */
2223           .policy = vport_policy,
2224           .doit = ovs_vport_cmd_get,
2225           .dumpit = ovs_vport_cmd_dump
2226         },
2227         { .cmd = OVS_VPORT_CMD_SET,
2228           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2229           .policy = vport_policy,
2230           .doit = ovs_vport_cmd_set,
2231         },
2232 };
2233
2234 struct genl_family dp_vport_genl_family __ro_after_init = {
2235         .hdrsize = sizeof(struct ovs_header),
2236         .name = OVS_VPORT_FAMILY,
2237         .version = OVS_VPORT_VERSION,
2238         .maxattr = OVS_VPORT_ATTR_MAX,
2239         .netnsok = true,
2240         .parallel_ops = true,
2241         .ops = dp_vport_genl_ops,
2242         .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2243         .mcgrps = &ovs_dp_vport_multicast_group,
2244         .n_mcgrps = 1,
2245         .module = THIS_MODULE,
2246 };
2247
2248 static struct genl_family * const dp_genl_families[] = {
2249         &dp_datapath_genl_family,
2250         &dp_vport_genl_family,
2251         &dp_flow_genl_family,
2252         &dp_packet_genl_family,
2253 };
2254
2255 static void dp_unregister_genl(int n_families)
2256 {
2257         int i;
2258
2259         for (i = 0; i < n_families; i++)
2260                 genl_unregister_family(dp_genl_families[i]);
2261 }
2262
2263 static int __init dp_register_genl(void)
2264 {
2265         int err;
2266         int i;
2267
2268         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2269
2270                 err = genl_register_family(dp_genl_families[i]);
2271                 if (err)
2272                         goto error;
2273         }
2274
2275         return 0;
2276
2277 error:
2278         dp_unregister_genl(i);
2279         return err;
2280 }
2281
2282 static int __net_init ovs_init_net(struct net *net)
2283 {
2284         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2285
2286         INIT_LIST_HEAD(&ovs_net->dps);
2287         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2288         ovs_ct_init(net);
2289         return 0;
2290 }
2291
2292 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2293                                             struct list_head *head)
2294 {
2295         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2296         struct datapath *dp;
2297
2298         list_for_each_entry(dp, &ovs_net->dps, list_node) {
2299                 int i;
2300
2301                 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2302                         struct vport *vport;
2303
2304                         hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2305                                 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2306                                         continue;
2307
2308                                 if (dev_net(vport->dev) == dnet)
2309                                         list_add(&vport->detach_list, head);
2310                         }
2311                 }
2312         }
2313 }
2314
2315 static void __net_exit ovs_exit_net(struct net *dnet)
2316 {
2317         struct datapath *dp, *dp_next;
2318         struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2319         struct vport *vport, *vport_next;
2320         struct net *net;
2321         LIST_HEAD(head);
2322
2323         ovs_ct_exit(dnet);
2324         ovs_lock();
2325         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2326                 __dp_destroy(dp);
2327
2328         rtnl_lock();
2329         for_each_net(net)
2330                 list_vports_from_net(net, dnet, &head);
2331         rtnl_unlock();
2332
2333         /* Detach all vports from given namespace. */
2334         list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2335                 list_del(&vport->detach_list);
2336                 ovs_dp_detach_port(vport);
2337         }
2338
2339         ovs_unlock();
2340
2341         cancel_work_sync(&ovs_net->dp_notify_work);
2342 }
2343
2344 static struct pernet_operations ovs_net_ops = {
2345         .init = ovs_init_net,
2346         .exit = ovs_exit_net,
2347         .id   = &ovs_net_id,
2348         .size = sizeof(struct ovs_net),
2349 };
2350
2351 static int __init dp_init(void)
2352 {
2353         int err;
2354
2355         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2356
2357         pr_info("Open vSwitch switching datapath\n");
2358
2359         err = action_fifos_init();
2360         if (err)
2361                 goto error;
2362
2363         err = ovs_internal_dev_rtnl_link_register();
2364         if (err)
2365                 goto error_action_fifos_exit;
2366
2367         err = ovs_flow_init();
2368         if (err)
2369                 goto error_unreg_rtnl_link;
2370
2371         err = ovs_vport_init();
2372         if (err)
2373                 goto error_flow_exit;
2374
2375         err = register_pernet_device(&ovs_net_ops);
2376         if (err)
2377                 goto error_vport_exit;
2378
2379         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2380         if (err)
2381                 goto error_netns_exit;
2382
2383         err = ovs_netdev_init();
2384         if (err)
2385                 goto error_unreg_notifier;
2386
2387         err = dp_register_genl();
2388         if (err < 0)
2389                 goto error_unreg_netdev;
2390
2391         return 0;
2392
2393 error_unreg_netdev:
2394         ovs_netdev_exit();
2395 error_unreg_notifier:
2396         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2397 error_netns_exit:
2398         unregister_pernet_device(&ovs_net_ops);
2399 error_vport_exit:
2400         ovs_vport_exit();
2401 error_flow_exit:
2402         ovs_flow_exit();
2403 error_unreg_rtnl_link:
2404         ovs_internal_dev_rtnl_link_unregister();
2405 error_action_fifos_exit:
2406         action_fifos_exit();
2407 error:
2408         return err;
2409 }
2410
2411 static void dp_cleanup(void)
2412 {
2413         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2414         ovs_netdev_exit();
2415         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2416         unregister_pernet_device(&ovs_net_ops);
2417         rcu_barrier();
2418         ovs_vport_exit();
2419         ovs_flow_exit();
2420         ovs_internal_dev_rtnl_link_unregister();
2421         action_fifos_exit();
2422 }
2423
2424 module_init(dp_init);
2425 module_exit(dp_cleanup);
2426
2427 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2428 MODULE_LICENSE("GPL");
2429 MODULE_ALIAS_GENL_FAMILY(OVS_DATAPATH_FAMILY);
2430 MODULE_ALIAS_GENL_FAMILY(OVS_VPORT_FAMILY);
2431 MODULE_ALIAS_GENL_FAMILY(OVS_FLOW_FAMILY);
2432 MODULE_ALIAS_GENL_FAMILY(OVS_PACKET_FAMILY);