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