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                 switch (error) {
 280                 case 0:
 281                 case -EAGAIN:
 282                 case -ERESTARTSYS:
 283                 case -EINTR:
 284                         consume_skb(skb);
 285                         break;
 286                 default:
 287                         kfree_skb(skb);
 288                         break;
 289                 }
 290                 stats_counter = &stats->n_missed;
 291                 goto out;
 292         }
 293
 294         ovs_flow_stats_update(flow, key->tp.flags, skb);
 295         sf_acts = rcu_dereference(flow->sf_acts);
 296         ovs_execute_actions(dp, skb, sf_acts, key);
 297
 298         stats_counter = &stats->n_hit;
 299
 300 out:
 301         /* Update datapath statistics. */
 302         u64_stats_update_begin(&stats->syncp);
 303         (*stats_counter)++;
 304         stats->n_mask_hit += n_mask_hit;
 305         u64_stats_update_end(&stats->syncp);
 306 }
 307
 308 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
 309                   const struct sw_flow_key *key,
 310                   const struct dp_upcall_info *upcall_info,
 311                   uint32_t cutlen)
 312 {
 313         struct dp_stats_percpu *stats;
 314         int err;
 315
 316         if (upcall_info->portid == 0) {
 317                 err = -ENOTCONN;
 318                 goto err;
 319         }
 320
 321         if (!skb_is_gso(skb))
 322                 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
 323         else
 324                 err = queue_gso_packets(dp, skb, key, upcall_info, cutlen);
 325         if (err)
 326                 goto err;
 327
 328         return 0;
 329
 330 err:
 331         stats = this_cpu_ptr(dp->stats_percpu);
 332
 333         u64_stats_update_begin(&stats->syncp);
 334         stats->n_lost++;
 335         u64_stats_update_end(&stats->syncp);
 336
 337         return err;
 338 }
 339
 340 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
 341                              const struct sw_flow_key *key,
 342                              const struct dp_upcall_info *upcall_info,
 343                                  uint32_t cutlen)
 344 {
 345         unsigned int gso_type = skb_shinfo(skb)->gso_type;
 346         struct sw_flow_key later_key;
 347         struct sk_buff *segs, *nskb;
 348         int err;
 349
 350         BUILD_BUG_ON(sizeof(*OVS_CB(skb)) > SKB_SGO_CB_OFFSET);
 351         segs = __skb_gso_segment(skb, NETIF_F_SG, false);
 352         if (IS_ERR(segs))
 353                 return PTR_ERR(segs);
 354         if (segs == NULL)
 355                 return -EINVAL;
 356
 357         if (gso_type & SKB_GSO_UDP) {
 358                 /* The initial flow key extracted by ovs_flow_key_extract()
 359                  * in this case is for a first fragment, so we need to
 360                  * properly mark later fragments.
 361                  */
 362                 later_key = *key;
 363                 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
 364         }
 365
 366         /* Queue all of the segments. */
 367         skb = segs;
 368         do {
 369                 if (gso_type & SKB_GSO_UDP && skb != segs)
 370                         key = &later_key;
 371
 372                 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
 373                 if (err)
 374                         break;
 375
 376         } while ((skb = skb->next));
 377
 378         /* Free all of the segments. */
 379         skb = segs;
 380         do {
 381                 nskb = skb->next;
 382                 if (err)
 383                         kfree_skb(skb);
 384                 else
 385                         consume_skb(skb);
 386         } while ((skb = nskb));
 387         return err;
 388 }
 389
 390 static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
 391                               unsigned int hdrlen, int actions_attrlen)
 392 {
 393         size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
 394                 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
 395                 + nla_total_size(ovs_key_attr_size()) /* OVS_PACKET_ATTR_KEY */
 396                 + nla_total_size(sizeof(unsigned int)); /* OVS_PACKET_ATTR_LEN */
 397
 398         /* OVS_PACKET_ATTR_USERDATA */
 399         if (upcall_info->userdata)
 400                 size += NLA_ALIGN(upcall_info->userdata->nla_len);
 401
 402         /* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
 403         if (upcall_info->egress_tun_info)
 404                 size += nla_total_size(ovs_tun_key_attr_size());
 405
 406         /* OVS_PACKET_ATTR_ACTIONS */
 407         if (upcall_info->actions_len)
 408                 size += nla_total_size(actions_attrlen);
 409
 410         /* OVS_PACKET_ATTR_MRU */
 411         if (upcall_info->mru)
 412                 size += nla_total_size(sizeof(upcall_info->mru));
 413
 414         return size;
 415 }
 416
 417 static void pad_packet(struct datapath *dp, struct sk_buff *skb)
 418 {
 419         if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
 420                 size_t plen = NLA_ALIGN(skb->len) - skb->len;
 421
 422                 if (plen > 0)
 423                         skb_put_zero(skb, plen);
 424         }
 425 }
 426
 427 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
 428                                   const struct sw_flow_key *key,
 429                                   const struct dp_upcall_info *upcall_info,
 430                                   uint32_t cutlen)
 431 {
 432         struct ovs_header *upcall;
 433         struct sk_buff *nskb = NULL;
 434         struct sk_buff *user_skb = NULL; /* to be queued to userspace */
 435         struct nlattr *nla;
 436         size_t len;
 437         unsigned int hlen;
 438         int err, dp_ifindex;
 439
 440         dp_ifindex = get_dpifindex(dp);
 441         if (!dp_ifindex)
 442                 return -ENODEV;
 443
 444         if (skb_vlan_tag_present(skb)) {
 445                 nskb = skb_clone(skb, GFP_ATOMIC);
 446                 if (!nskb)
 447                         return -ENOMEM;
 448
 449                 nskb = __vlan_hwaccel_push_inside(nskb);
 450                 if (!nskb)
 451                         return -ENOMEM;
 452
 453                 skb = nskb;
 454         }
 455
 456         if (nla_attr_size(skb->len) > USHRT_MAX) {
 457                 err = -EFBIG;
 458                 goto out;
 459         }
 460
 461         /* Complete checksum if needed */
 462         if (skb->ip_summed == CHECKSUM_PARTIAL &&
 463             (err = skb_csum_hwoffload_help(skb, 0)))
 464                 goto out;
 465
 466         /* Older versions of OVS user space enforce alignment of the last
 467          * Netlink attribute to NLA_ALIGNTO which would require extensive
 468          * padding logic. Only perform zerocopy if padding is not required.
 469          */
 470         if (dp->user_features & OVS_DP_F_UNALIGNED)
 471                 hlen = skb_zerocopy_headlen(skb);
 472         else
 473                 hlen = skb->len;
 474
 475         len = upcall_msg_size(upcall_info, hlen - cutlen,
 476                               OVS_CB(skb)->acts_origlen);
 477         user_skb = genlmsg_new(len, GFP_ATOMIC);
 478         if (!user_skb) {
 479                 err = -ENOMEM;
 480                 goto out;
 481         }
 482
 483         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
 484                              0, upcall_info->cmd);
 485         upcall->dp_ifindex = dp_ifindex;
 486
 487         err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
 488         BUG_ON(err);
 489
 490         if (upcall_info->userdata)
 491                 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
 492                           nla_len(upcall_info->userdata),
 493                           nla_data(upcall_info->userdata));
 494
 495         if (upcall_info->egress_tun_info) {
 496                 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_EGRESS_TUN_KEY);
 497                 err = ovs_nla_put_tunnel_info(user_skb,
 498                                               upcall_info->egress_tun_info);
 499                 BUG_ON(err);
 500                 nla_nest_end(user_skb, nla);
 501         }
 502
 503         if (upcall_info->actions_len) {
 504                 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_ACTIONS);
 505                 err = ovs_nla_put_actions(upcall_info->actions,
 506                                           upcall_info->actions_len,
 507                                           user_skb);
 508                 if (!err)
 509                         nla_nest_end(user_skb, nla);
 510                 else
 511                         nla_nest_cancel(user_skb, nla);
 512         }
 513
 514         /* Add OVS_PACKET_ATTR_MRU */
 515         if (upcall_info->mru) {
 516                 if (nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU,
 517                                 upcall_info->mru)) {
 518                         err = -ENOBUFS;
 519                         goto out;
 520                 }
 521                 pad_packet(dp, user_skb);
 522         }
 523
 524         /* Add OVS_PACKET_ATTR_LEN when packet is truncated */
 525         if (cutlen > 0) {
 526                 if (nla_put_u32(user_skb, OVS_PACKET_ATTR_LEN,
 527                                 skb->len)) {
 528                         err = -ENOBUFS;
 529                         goto out;
 530                 }
 531                 pad_packet(dp, user_skb);
 532         }
 533
 534         /* Only reserve room for attribute header, packet data is added
 535          * in skb_zerocopy() */
 536         if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
 537                 err = -ENOBUFS;
 538                 goto out;
 539         }
 540         nla->nla_len = nla_attr_size(skb->len - cutlen);
 541
 542         err = skb_zerocopy(user_skb, skb, skb->len - cutlen, hlen);
 543         if (err)
 544                 goto out;
 545
 546         /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
 547         pad_packet(dp, user_skb);
 548
 549         ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
 550
 551         err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
 552         user_skb = NULL;
 553 out:
 554         if (err)
 555                 skb_tx_error(skb);
 556         consume_skb(user_skb);
 557         consume_skb(nskb);
 558
 559         return err;
 560 }
 561
 562 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
 563 {
 564         struct ovs_header *ovs_header = info->userhdr;
 565         struct net *net = sock_net(skb->sk);
 566         struct nlattr **a = info->attrs;
 567         struct sw_flow_actions *acts;
 568         struct sk_buff *packet;
 569         struct sw_flow *flow;
 570         struct sw_flow_actions *sf_acts;
 571         struct datapath *dp;
 572         struct vport *input_vport;
 573         u16 mru = 0;
 574         int len;
 575         int err;
 576         bool log = !a[OVS_PACKET_ATTR_PROBE];
 577
 578         err = -EINVAL;
 579         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
 580             !a[OVS_PACKET_ATTR_ACTIONS])
 581                 goto err;
 582
 583         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
 584         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
 585         err = -ENOMEM;
 586         if (!packet)
 587                 goto err;
 588         skb_reserve(packet, NET_IP_ALIGN);
 589
 590         nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
 591
 592         /* Set packet's mru */
 593         if (a[OVS_PACKET_ATTR_MRU]) {
 594                 mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
 595                 packet->ignore_df = 1;
 596         }
 597         OVS_CB(packet)->mru = mru;
 598
 599         /* Build an sw_flow for sending this packet. */
 600         flow = ovs_flow_alloc();
 601         err = PTR_ERR(flow);
 602         if (IS_ERR(flow))
 603                 goto err_kfree_skb;
 604
 605         err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
 606                                              packet, &flow->key, log);
 607         if (err)
 608                 goto err_flow_free;
 609
 610         err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
 611                                    &flow->key, &acts, log);
 612         if (err)
 613                 goto err_flow_free;
 614
 615         rcu_assign_pointer(flow->sf_acts, acts);
 616         packet->priority = flow->key.phy.priority;
 617         packet->mark = flow->key.phy.skb_mark;
 618
 619         rcu_read_lock();
 620         dp = get_dp_rcu(net, ovs_header->dp_ifindex);
 621         err = -ENODEV;
 622         if (!dp)
 623                 goto err_unlock;
 624
 625         input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
 626         if (!input_vport)
 627                 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
 628
 629         if (!input_vport)
 630                 goto err_unlock;
 631
 632         packet->dev = input_vport->dev;
 633         OVS_CB(packet)->input_vport = input_vport;
 634         sf_acts = rcu_dereference(flow->sf_acts);
 635
 636         local_bh_disable();
 637         err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
 638         local_bh_enable();
 639         rcu_read_unlock();
 640
 641         ovs_flow_free(flow, false);
 642         return err;
 643
 644 err_unlock:
 645         rcu_read_unlock();
 646 err_flow_free:
 647         ovs_flow_free(flow, false);
 648 err_kfree_skb:
 649         kfree_skb(packet);
 650 err:
 651         return err;
 652 }
 653
 654 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
 655         [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
 656         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
 657         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
 658         [OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
 659         [OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
 660 };
 661
 662 static const struct genl_ops dp_packet_genl_ops[] = {
 663         { .cmd = OVS_PACKET_CMD_EXECUTE,
 664           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
 665           .policy = packet_policy,
 666           .doit = ovs_packet_cmd_execute
 667         }
 668 };
 669
 670 static struct genl_family dp_packet_genl_family __ro_after_init = {
 671         .hdrsize = sizeof(struct ovs_header),
 672         .name = OVS_PACKET_FAMILY,
 673         .version = OVS_PACKET_VERSION,
 674         .maxattr = OVS_PACKET_ATTR_MAX,
 675         .netnsok = true,
 676         .parallel_ops = true,
 677         .ops = dp_packet_genl_ops,
 678         .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
 679         .module = THIS_MODULE,
 680 };
 681
 682 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
 683                          struct ovs_dp_megaflow_stats *mega_stats)
 684 {
 685         int i;
 686
 687         memset(mega_stats, 0, sizeof(*mega_stats));
 688
 689         stats->n_flows = ovs_flow_tbl_count(&dp->table);
 690         mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
 691
 692         stats->n_hit = stats->n_missed = stats->n_lost = 0;
 693
 694         for_each_possible_cpu(i) {
 695                 const struct dp_stats_percpu *percpu_stats;
 696                 struct dp_stats_percpu local_stats;
 697                 unsigned int start;
 698
 699                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
 700
 701                 do {
 702                         start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
 703                         local_stats = *percpu_stats;
 704                 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
 705
 706                 stats->n_hit += local_stats.n_hit;
 707                 stats->n_missed += local_stats.n_missed;
 708                 stats->n_lost += local_stats.n_lost;
 709                 mega_stats->n_mask_hit += local_stats.n_mask_hit;
 710         }
 711 }
 712
 713 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
 714 {
 715         return ovs_identifier_is_ufid(sfid) &&
 716                !(ufid_flags & OVS_UFID_F_OMIT_KEY);
 717 }
 718
 719 static bool should_fill_mask(uint32_t ufid_flags)
 720 {
 721         return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
 722 }
 723
 724 static bool should_fill_actions(uint32_t ufid_flags)
 725 {
 726         return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
 727 }
 728
 729 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
 730                                     const struct sw_flow_id *sfid,
 731                                     uint32_t ufid_flags)
 732 {
 733         size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
 734
 735         /* OVS_FLOW_ATTR_UFID, or unmasked flow key as fallback
 736          * see ovs_nla_put_identifier()
 737          */
 738         if (sfid && ovs_identifier_is_ufid(sfid))
 739                 len += nla_total_size(sfid->ufid_len);
 740         else
 741                 len += nla_total_size(ovs_key_attr_size());
 742
 743         /* OVS_FLOW_ATTR_KEY */
 744         if (!sfid || should_fill_key(sfid, ufid_flags))
 745                 len += nla_total_size(ovs_key_attr_size());
 746
 747         /* OVS_FLOW_ATTR_MASK */
 748         if (should_fill_mask(ufid_flags))
 749                 len += nla_total_size(ovs_key_attr_size());
 750
 751         /* OVS_FLOW_ATTR_ACTIONS */
 752         if (should_fill_actions(ufid_flags))
 753                 len += nla_total_size(acts->orig_len);
 754
 755         return len
 756                 + nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
 757                 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
 758                 + nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */
 759 }
 760
 761 /* Called with ovs_mutex or RCU read lock. */
 762 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
 763                                    struct sk_buff *skb)
 764 {
 765         struct ovs_flow_stats stats;
 766         __be16 tcp_flags;
 767         unsigned long used;
 768
 769         ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
 770
 771         if (used &&
 772             nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
 773                               OVS_FLOW_ATTR_PAD))
 774                 return -EMSGSIZE;
 775
 776         if (stats.n_packets &&
 777             nla_put_64bit(skb, OVS_FLOW_ATTR_STATS,
 778                           sizeof(struct ovs_flow_stats), &stats,
 779                           OVS_FLOW_ATTR_PAD))
 780                 return -EMSGSIZE;
 781
 782         if ((u8)ntohs(tcp_flags) &&
 783              nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
 784                 return -EMSGSIZE;
 785
 786         return 0;
 787 }
 788
 789 /* Called with ovs_mutex or RCU read lock. */
 790 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
 791                                      struct sk_buff *skb, int skb_orig_len)
 792 {
 793         struct nlattr *start;
 794         int err;
 795
 796         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
 797          * this is the first flow to be dumped into 'skb'.  This is unusual for
 798          * Netlink but individual action lists can be longer than
 799          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
 800          * The userspace caller can always fetch the actions separately if it
 801          * really wants them.  (Most userspace callers in fact don't care.)
 802          *
 803          * This can only fail for dump operations because the skb is always
 804          * properly sized for single flows.
 805          */
 806         start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
 807         if (start) {
 808                 const struct sw_flow_actions *sf_acts;
 809
 810                 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
 811                 err = ovs_nla_put_actions(sf_acts->actions,
 812                                           sf_acts->actions_len, skb);
 813
 814                 if (!err)
 815                         nla_nest_end(skb, start);
 816                 else {
 817                         if (skb_orig_len)
 818                                 return err;
 819
 820                         nla_nest_cancel(skb, start);
 821                 }
 822         } else if (skb_orig_len) {
 823                 return -EMSGSIZE;
 824         }
 825
 826         return 0;
 827 }
 828
 829 /* Called with ovs_mutex or RCU read lock. */
 830 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
 831                                   struct sk_buff *skb, u32 portid,
 832                                   u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
 833 {
 834         const int skb_orig_len = skb->len;
 835         struct ovs_header *ovs_header;
 836         int err;
 837
 838         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
 839                                  flags, cmd);
 840         if (!ovs_header)
 841                 return -EMSGSIZE;
 842
 843         ovs_header->dp_ifindex = dp_ifindex;
 844
 845         err = ovs_nla_put_identifier(flow, skb);
 846         if (err)
 847                 goto error;
 848
 849         if (should_fill_key(&flow->id, ufid_flags)) {
 850                 err = ovs_nla_put_masked_key(flow, skb);
 851                 if (err)
 852                         goto error;
 853         }
 854
 855         if (should_fill_mask(ufid_flags)) {
 856                 err = ovs_nla_put_mask(flow, skb);
 857                 if (err)
 858                         goto error;
 859         }
 860
 861         err = ovs_flow_cmd_fill_stats(flow, skb);
 862         if (err)
 863                 goto error;
 864
 865         if (should_fill_actions(ufid_flags)) {
 866                 err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
 867                 if (err)
 868                         goto error;
 869         }
 870
 871         genlmsg_end(skb, ovs_header);
 872         return 0;
 873
 874 error:
 875         genlmsg_cancel(skb, ovs_header);
 876         return err;
 877 }
 878
 879 /* May not be called with RCU read lock. */
 880 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
 881                                                const struct sw_flow_id *sfid,
 882                                                struct genl_info *info,
 883                                                bool always,
 884                                                uint32_t ufid_flags)
 885 {
 886         struct sk_buff *skb;
 887         size_t len;
 888
 889         if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
 890                 return NULL;
 891
 892         len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
 893         skb = genlmsg_new(len, GFP_KERNEL);
 894         if (!skb)
 895                 return ERR_PTR(-ENOMEM);
 896
 897         return skb;
 898 }
 899
 900 /* Called with ovs_mutex. */
 901 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
 902                                                int dp_ifindex,
 903                                                struct genl_info *info, u8 cmd,
 904                                                bool always, u32 ufid_flags)
 905 {
 906         struct sk_buff *skb;
 907         int retval;
 908
 909         skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
 910                                       &flow->id, info, always, ufid_flags);
 911         if (IS_ERR_OR_NULL(skb))
 912                 return skb;
 913
 914         retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
 915                                         info->snd_portid, info->snd_seq, 0,
 916                                         cmd, ufid_flags);
 917         if (WARN_ON_ONCE(retval < 0)) {
 918                 kfree_skb(skb);
 919                 skb = ERR_PTR(retval);
 920         }
 921         return skb;
 922 }
 923
 924 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
 925 {
 926         struct net *net = sock_net(skb->sk);
 927         struct nlattr **a = info->attrs;
 928         struct ovs_header *ovs_header = info->userhdr;
 929         struct sw_flow *flow = NULL, *new_flow;
 930         struct sw_flow_mask mask;
 931         struct sk_buff *reply;
 932         struct datapath *dp;
 933         struct sw_flow_key *key;
 934         struct sw_flow_actions *acts;
 935         struct sw_flow_match match;
 936         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
 937         int error;
 938         bool log = !a[OVS_FLOW_ATTR_PROBE];
 939
 940         /* Must have key and actions. */
 941         error = -EINVAL;
 942         if (!a[OVS_FLOW_ATTR_KEY]) {
 943                 OVS_NLERR(log, "Flow key attr not present in new flow.");
 944                 goto error;
 945         }
 946         if (!a[OVS_FLOW_ATTR_ACTIONS]) {
 947                 OVS_NLERR(log, "Flow actions attr not present in new flow.");
 948                 goto error;
 949         }
 950
 951         /* Most of the time we need to allocate a new flow, do it before
 952          * locking.
 953          */
 954         new_flow = ovs_flow_alloc();
 955         if (IS_ERR(new_flow)) {
 956                 error = PTR_ERR(new_flow);
 957                 goto error;
 958         }
 959
 960         /* Extract key. */
 961         key = kzalloc(sizeof(*key), GFP_KERNEL);
 962         if (!key) {
 963                 error = -ENOMEM;
 964                 goto err_kfree_flow;
 965         }
 966
 967         ovs_match_init(&match, key, false, &mask);
 968         error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
 969                                   a[OVS_FLOW_ATTR_MASK], log);
 970         if (error)
 971                 goto err_kfree_key;
 972
 973         ovs_flow_mask_key(&new_flow->key, key, true, &mask);
 974
 975         /* Extract flow identifier. */
 976         error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
 977                                        key, log);
 978         if (error)
 979                 goto err_kfree_key;
 980
 981         /* Validate actions. */
 982         error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
 983                                      &new_flow->key, &acts, log);
 984         if (error) {
 985                 OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
 986                 goto err_kfree_key;
 987         }
 988
 989         reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
 990                                         ufid_flags);
 991         if (IS_ERR(reply)) {
 992                 error = PTR_ERR(reply);
 993                 goto err_kfree_acts;
 994         }
 995
 996         ovs_lock();
 997         dp = get_dp(net, ovs_header->dp_ifindex);
 998         if (unlikely(!dp)) {
 999                 error = -ENODEV;
1000                 goto err_unlock_ovs;
1001         }
1002
1003         /* Check if this is a duplicate flow */
1004         if (ovs_identifier_is_ufid(&new_flow->id))
1005                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
1006         if (!flow)
1007                 flow = ovs_flow_tbl_lookup(&dp->table, key);
1008         if (likely(!flow)) {
1009                 rcu_assign_pointer(new_flow->sf_acts, acts);
1010
1011                 /* Put flow in bucket. */
1012                 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
1013                 if (unlikely(error)) {
1014                         acts = NULL;
1015                         goto err_unlock_ovs;
1016                 }
1017
1018                 if (unlikely(reply)) {
1019                         error = ovs_flow_cmd_fill_info(new_flow,
1020                                                        ovs_header->dp_ifindex,
1021                                                        reply, info->snd_portid,
1022                                                        info->snd_seq, 0,
1023                                                        OVS_FLOW_CMD_NEW,
1024                                                        ufid_flags);
1025                         BUG_ON(error < 0);
1026                 }
1027                 ovs_unlock();
1028         } else {
1029                 struct sw_flow_actions *old_acts;
1030
1031                 /* Bail out if we're not allowed to modify an existing flow.
1032                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1033                  * because Generic Netlink treats the latter as a dump
1034                  * request.  We also accept NLM_F_EXCL in case that bug ever
1035                  * gets fixed.
1036                  */
1037                 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1038                                                          | NLM_F_EXCL))) {
1039                         error = -EEXIST;
1040                         goto err_unlock_ovs;
1041                 }
1042                 /* The flow identifier has to be the same for flow updates.
1043                  * Look for any overlapping flow.
1044                  */
1045                 if (unlikely(!ovs_flow_cmp(flow, &match))) {
1046                         if (ovs_identifier_is_key(&flow->id))
1047                                 flow = ovs_flow_tbl_lookup_exact(&dp->table,
1048                                                                  &match);
1049                         else /* UFID matches but key is different */
1050                                 flow = NULL;
1051                         if (!flow) {
1052                                 error = -ENOENT;
1053                                 goto err_unlock_ovs;
1054                         }
1055                 }
1056                 /* Update actions. */
1057                 old_acts = ovsl_dereference(flow->sf_acts);
1058                 rcu_assign_pointer(flow->sf_acts, acts);
1059
1060                 if (unlikely(reply)) {
1061                         error = ovs_flow_cmd_fill_info(flow,
1062                                                        ovs_header->dp_ifindex,
1063                                                        reply, info->snd_portid,
1064                                                        info->snd_seq, 0,
1065                                                        OVS_FLOW_CMD_NEW,
1066                                                        ufid_flags);
1067                         BUG_ON(error < 0);
1068                 }
1069                 ovs_unlock();
1070
1071                 ovs_nla_free_flow_actions_rcu(old_acts);
1072                 ovs_flow_free(new_flow, false);
1073         }
1074
1075         if (reply)
1076                 ovs_notify(&dp_flow_genl_family, reply, info);
1077
1078         kfree(key);
1079         return 0;
1080
1081 err_unlock_ovs:
1082         ovs_unlock();
1083         kfree_skb(reply);
1084 err_kfree_acts:
1085         ovs_nla_free_flow_actions(acts);
1086 err_kfree_key:
1087         kfree(key);
1088 err_kfree_flow:
1089         ovs_flow_free(new_flow, false);
1090 error:
1091         return error;
1092 }
1093
1094 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1095 static struct sw_flow_actions *get_flow_actions(struct net *net,
1096                                                 const struct nlattr *a,
1097                                                 const struct sw_flow_key *key,
1098                                                 const struct sw_flow_mask *mask,
1099                                                 bool log)
1100 {
1101         struct sw_flow_actions *acts;
1102         struct sw_flow_key masked_key;
1103         int error;
1104
1105         ovs_flow_mask_key(&masked_key, key, true, mask);
1106         error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1107         if (error) {
1108                 OVS_NLERR(log,
1109                           "Actions may not be safe on all matching packets");
1110                 return ERR_PTR(error);
1111         }
1112
1113         return acts;
1114 }
1115
1116 /* Factor out match-init and action-copy to avoid
1117  * "Wframe-larger-than=1024" warning. Because mask is only
1118  * used to get actions, we new a function to save some
1119  * stack space.
1120  *
1121  * If there are not key and action attrs, we return 0
1122  * directly. In the case, the caller will also not use the
1123  * match as before. If there is action attr, we try to get
1124  * actions and save them to *acts. Before returning from
1125  * the function, we reset the match->mask pointer. Because
1126  * we should not to return match object with dangling reference
1127  * to mask.
1128  * */
1129 static int ovs_nla_init_match_and_action(struct net *net,
1130                                          struct sw_flow_match *match,
1131                                          struct sw_flow_key *key,
1132                                          struct nlattr **a,
1133                                          struct sw_flow_actions **acts,
1134                                          bool log)
1135 {
1136         struct sw_flow_mask mask;
1137         int error = 0;
1138
1139         if (a[OVS_FLOW_ATTR_KEY]) {
1140                 ovs_match_init(match, key, true, &mask);
1141                 error = ovs_nla_get_match(net, match, a[OVS_FLOW_ATTR_KEY],
1142                                           a[OVS_FLOW_ATTR_MASK], log);
1143                 if (error)
1144                         goto error;
1145         }
1146
1147         if (a[OVS_FLOW_ATTR_ACTIONS]) {
1148                 if (!a[OVS_FLOW_ATTR_KEY]) {
1149                         OVS_NLERR(log,
1150                                   "Flow key attribute not present in set flow.");
1151                         error = -EINVAL;
1152                         goto error;
1153                 }
1154
1155                 *acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], key,
1156                                          &mask, log);
1157                 if (IS_ERR(*acts)) {
1158                         error = PTR_ERR(*acts);
1159                         goto error;
1160                 }
1161         }
1162
1163         /* On success, error is 0. */
1164 error:
1165         match->mask = NULL;
1166         return error;
1167 }
1168
1169 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1170 {
1171         struct net *net = sock_net(skb->sk);
1172         struct nlattr **a = info->attrs;
1173         struct ovs_header *ovs_header = info->userhdr;
1174         struct sw_flow_key key;
1175         struct sw_flow *flow;
1176         struct sk_buff *reply = NULL;
1177         struct datapath *dp;
1178         struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1179         struct sw_flow_match match;
1180         struct sw_flow_id sfid;
1181         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1182         int error = 0;
1183         bool log = !a[OVS_FLOW_ATTR_PROBE];
1184         bool ufid_present;
1185
1186         ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1187         if (!a[OVS_FLOW_ATTR_KEY] && !ufid_present) {
1188                 OVS_NLERR(log,
1189                           "Flow set message rejected, Key attribute missing.");
1190                 return -EINVAL;
1191         }
1192
1193         error = ovs_nla_init_match_and_action(net, &match, &key, a,
1194                                               &acts, log);
1195         if (error)
1196                 goto error;
1197
1198         if (acts) {
1199                 /* Can allocate before locking if have acts. */
1200                 reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1201                                                 ufid_flags);
1202                 if (IS_ERR(reply)) {
1203                         error = PTR_ERR(reply);
1204                         goto err_kfree_acts;
1205                 }
1206         }
1207
1208         ovs_lock();
1209         dp = get_dp(net, ovs_header->dp_ifindex);
1210         if (unlikely(!dp)) {
1211                 error = -ENODEV;
1212                 goto err_unlock_ovs;
1213         }
1214         /* Check that the flow exists. */
1215         if (ufid_present)
1216                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1217         else
1218                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1219         if (unlikely(!flow)) {
1220                 error = -ENOENT;
1221                 goto err_unlock_ovs;
1222         }
1223
1224         /* Update actions, if present. */
1225         if (likely(acts)) {
1226                 old_acts = ovsl_dereference(flow->sf_acts);
1227                 rcu_assign_pointer(flow->sf_acts, acts);
1228
1229                 if (unlikely(reply)) {
1230                         error = ovs_flow_cmd_fill_info(flow,
1231                                                        ovs_header->dp_ifindex,
1232                                                        reply, info->snd_portid,
1233                                                        info->snd_seq, 0,
1234                                                        OVS_FLOW_CMD_NEW,
1235                                                        ufid_flags);
1236                         BUG_ON(error < 0);
1237                 }
1238         } else {
1239                 /* Could not alloc without acts before locking. */
1240                 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1241                                                 info, OVS_FLOW_CMD_NEW, false,
1242                                                 ufid_flags);
1243
1244                 if (IS_ERR(reply)) {
1245                         error = PTR_ERR(reply);
1246                         goto err_unlock_ovs;
1247                 }
1248         }
1249
1250         /* Clear stats. */
1251         if (a[OVS_FLOW_ATTR_CLEAR])
1252                 ovs_flow_stats_clear(flow);
1253         ovs_unlock();
1254
1255         if (reply)
1256                 ovs_notify(&dp_flow_genl_family, reply, info);
1257         if (old_acts)
1258                 ovs_nla_free_flow_actions_rcu(old_acts);
1259
1260         return 0;
1261
1262 err_unlock_ovs:
1263         ovs_unlock();
1264         kfree_skb(reply);
1265 err_kfree_acts:
1266         ovs_nla_free_flow_actions(acts);
1267 error:
1268         return error;
1269 }
1270
1271 static int ovs_flow_cmd_get(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;
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 = 0;
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], NULL,
1291                                         log);
1292         } else if (!ufid_present) {
1293                 OVS_NLERR(log,
1294                           "Flow get message rejected, Key attribute missing.");
1295                 err = -EINVAL;
1296         }
1297         if (err)
1298                 return err;
1299
1300         ovs_lock();
1301         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1302         if (!dp) {
1303                 err = -ENODEV;
1304                 goto unlock;
1305         }
1306
1307         if (ufid_present)
1308                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1309         else
1310                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1311         if (!flow) {
1312                 err = -ENOENT;
1313                 goto unlock;
1314         }
1315
1316         reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1317                                         OVS_FLOW_CMD_NEW, true, ufid_flags);
1318         if (IS_ERR(reply)) {
1319                 err = PTR_ERR(reply);
1320                 goto unlock;
1321         }
1322
1323         ovs_unlock();
1324         return genlmsg_reply(reply, info);
1325 unlock:
1326         ovs_unlock();
1327         return err;
1328 }
1329
1330 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1331 {
1332         struct nlattr **a = info->attrs;
1333         struct ovs_header *ovs_header = info->userhdr;
1334         struct net *net = sock_net(skb->sk);
1335         struct sw_flow_key key;
1336         struct sk_buff *reply;
1337         struct sw_flow *flow = NULL;
1338         struct datapath *dp;
1339         struct sw_flow_match match;
1340         struct sw_flow_id ufid;
1341         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1342         int err;
1343         bool log = !a[OVS_FLOW_ATTR_PROBE];
1344         bool ufid_present;
1345
1346         ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1347         if (a[OVS_FLOW_ATTR_KEY]) {
1348                 ovs_match_init(&match, &key, true, NULL);
1349                 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1350                                         NULL, log);
1351                 if (unlikely(err))
1352                         return err;
1353         }
1354
1355         ovs_lock();
1356         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1357         if (unlikely(!dp)) {
1358                 err = -ENODEV;
1359                 goto unlock;
1360         }
1361
1362         if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1363                 err = ovs_flow_tbl_flush(&dp->table);
1364                 goto unlock;
1365         }
1366
1367         if (ufid_present)
1368                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1369         else
1370                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1371         if (unlikely(!flow)) {
1372                 err = -ENOENT;
1373                 goto unlock;
1374         }
1375
1376         ovs_flow_tbl_remove(&dp->table, flow);
1377         ovs_unlock();
1378
1379         reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1380                                         &flow->id, info, false, ufid_flags);
1381         if (likely(reply)) {
1382                 if (likely(!IS_ERR(reply))) {
1383                         rcu_read_lock();        /*To keep RCU checker happy. */
1384                         err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1385                                                      reply, info->snd_portid,
1386                                                      info->snd_seq, 0,
1387                                                      OVS_FLOW_CMD_DEL,
1388                                                      ufid_flags);
1389                         rcu_read_unlock();
1390                         if (WARN_ON_ONCE(err < 0)) {
1391                                 kfree_skb(reply);
1392                                 goto out_free;
1393                         }
1394
1395                         ovs_notify(&dp_flow_genl_family, reply, info);
1396                 } else {
1397                         netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
1398                 }
1399         }
1400
1401 out_free:
1402         ovs_flow_free(flow, true);
1403         return 0;
1404 unlock:
1405         ovs_unlock();
1406         return err;
1407 }
1408
1409 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1410 {
1411         struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1412         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1413         struct table_instance *ti;
1414         struct datapath *dp;
1415         u32 ufid_flags;
1416         int err;
1417
1418         err = genlmsg_parse(cb->nlh, &dp_flow_genl_family, a,
1419                             OVS_FLOW_ATTR_MAX, flow_policy, NULL);
1420         if (err)
1421                 return err;
1422         ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1423
1424         rcu_read_lock();
1425         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1426         if (!dp) {
1427                 rcu_read_unlock();
1428                 return -ENODEV;
1429         }
1430
1431         ti = rcu_dereference(dp->table.ti);
1432         for (;;) {
1433                 struct sw_flow *flow;
1434                 u32 bucket, obj;
1435
1436                 bucket = cb->args[0];
1437                 obj = cb->args[1];
1438                 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1439                 if (!flow)
1440                         break;
1441
1442                 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1443                                            NETLINK_CB(cb->skb).portid,
1444                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1445                                            OVS_FLOW_CMD_NEW, ufid_flags) < 0)
1446                         break;
1447
1448                 cb->args[0] = bucket;
1449                 cb->args[1] = obj;
1450         }
1451         rcu_read_unlock();
1452         return skb->len;
1453 }
1454
1455 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1456         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1457         [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1458         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1459         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1460         [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1461         [OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1462         [OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1463 };
1464
1465 static const struct genl_ops dp_flow_genl_ops[] = {
1466         { .cmd = OVS_FLOW_CMD_NEW,
1467           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1468           .policy = flow_policy,
1469           .doit = ovs_flow_cmd_new
1470         },
1471         { .cmd = OVS_FLOW_CMD_DEL,
1472           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1473           .policy = flow_policy,
1474           .doit = ovs_flow_cmd_del
1475         },
1476         { .cmd = OVS_FLOW_CMD_GET,
1477           .flags = 0,               /* OK for unprivileged users. */
1478           .policy = flow_policy,
1479           .doit = ovs_flow_cmd_get,
1480           .dumpit = ovs_flow_cmd_dump
1481         },
1482         { .cmd = OVS_FLOW_CMD_SET,
1483           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1484           .policy = flow_policy,
1485           .doit = ovs_flow_cmd_set,
1486         },
1487 };
1488
1489 static struct genl_family dp_flow_genl_family __ro_after_init = {
1490         .hdrsize = sizeof(struct ovs_header),
1491         .name = OVS_FLOW_FAMILY,
1492         .version = OVS_FLOW_VERSION,
1493         .maxattr = OVS_FLOW_ATTR_MAX,
1494         .netnsok = true,
1495         .parallel_ops = true,
1496         .ops = dp_flow_genl_ops,
1497         .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1498         .mcgrps = &ovs_dp_flow_multicast_group,
1499         .n_mcgrps = 1,
1500         .module = THIS_MODULE,
1501 };
1502
1503 static size_t ovs_dp_cmd_msg_size(void)
1504 {
1505         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1506
1507         msgsize += nla_total_size(IFNAMSIZ);
1508         msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1509         msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
1510         msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1511
1512         return msgsize;
1513 }
1514
1515 /* Called with ovs_mutex. */
1516 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1517                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1518 {
1519         struct ovs_header *ovs_header;
1520         struct ovs_dp_stats dp_stats;
1521         struct ovs_dp_megaflow_stats dp_megaflow_stats;
1522         int err;
1523
1524         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1525                                    flags, cmd);
1526         if (!ovs_header)
1527                 goto error;
1528
1529         ovs_header->dp_ifindex = get_dpifindex(dp);
1530
1531         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1532         if (err)
1533                 goto nla_put_failure;
1534
1535         get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1536         if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1537                           &dp_stats, OVS_DP_ATTR_PAD))
1538                 goto nla_put_failure;
1539
1540         if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1541                           sizeof(struct ovs_dp_megaflow_stats),
1542                           &dp_megaflow_stats, OVS_DP_ATTR_PAD))
1543                 goto nla_put_failure;
1544
1545         if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1546                 goto nla_put_failure;
1547
1548         genlmsg_end(skb, ovs_header);
1549         return 0;
1550
1551 nla_put_failure:
1552         genlmsg_cancel(skb, ovs_header);
1553 error:
1554         return -EMSGSIZE;
1555 }
1556
1557 static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1558 {
1559         return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1560 }
1561
1562 /* Called with rcu_read_lock or ovs_mutex. */
1563 static struct datapath *lookup_datapath(struct net *net,
1564                                         const struct ovs_header *ovs_header,
1565                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1566 {
1567         struct datapath *dp;
1568
1569         if (!a[OVS_DP_ATTR_NAME])
1570                 dp = get_dp(net, ovs_header->dp_ifindex);
1571         else {
1572                 struct vport *vport;
1573
1574                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1575                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1576         }
1577         return dp ? dp : ERR_PTR(-ENODEV);
1578 }
1579
1580 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1581 {
1582         struct datapath *dp;
1583
1584         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1585         if (IS_ERR(dp))
1586                 return;
1587
1588         pr_warn("%s: Dropping previously announced user features\n",
1589                 ovs_dp_name(dp));
1590         dp->user_features = 0;
1591 }
1592
1593 static void ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1594 {
1595         if (a[OVS_DP_ATTR_USER_FEATURES])
1596                 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1597 }
1598
1599 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1600 {
1601         struct nlattr **a = info->attrs;
1602         struct vport_parms parms;
1603         struct sk_buff *reply;
1604         struct datapath *dp;
1605         struct vport *vport;
1606         struct ovs_net *ovs_net;
1607         int err, i;
1608
1609         err = -EINVAL;
1610         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1611                 goto err;
1612
1613         reply = ovs_dp_cmd_alloc_info();
1614         if (!reply)
1615                 return -ENOMEM;
1616
1617         err = -ENOMEM;
1618         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1619         if (dp == NULL)
1620                 goto err_free_reply;
1621
1622         ovs_dp_set_net(dp, sock_net(skb->sk));
1623
1624         /* Allocate table. */
1625         err = ovs_flow_tbl_init(&dp->table);
1626         if (err)
1627                 goto err_free_dp;
1628
1629         dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1630         if (!dp->stats_percpu) {
1631                 err = -ENOMEM;
1632                 goto err_destroy_table;
1633         }
1634
1635         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1636                             GFP_KERNEL);
1637         if (!dp->ports) {
1638                 err = -ENOMEM;
1639                 goto err_destroy_percpu;
1640         }
1641
1642         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1643                 INIT_HLIST_HEAD(&dp->ports[i]);
1644
1645         /* Set up our datapath device. */
1646         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1647         parms.type = OVS_VPORT_TYPE_INTERNAL;
1648         parms.options = NULL;
1649         parms.dp = dp;
1650         parms.port_no = OVSP_LOCAL;
1651         parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1652
1653         ovs_dp_change(dp, a);
1654
1655         /* So far only local changes have been made, now need the lock. */
1656         ovs_lock();
1657
1658         vport = new_vport(&parms);
1659         if (IS_ERR(vport)) {
1660                 err = PTR_ERR(vport);
1661                 if (err == -EBUSY)
1662                         err = -EEXIST;
1663
1664                 if (err == -EEXIST) {
1665                         /* An outdated user space instance that does not understand
1666                          * the concept of user_features has attempted to create a new
1667                          * datapath and is likely to reuse it. Drop all user features.
1668                          */
1669                         if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1670                                 ovs_dp_reset_user_features(skb, info);
1671                 }
1672
1673                 goto err_destroy_ports_array;
1674         }
1675
1676         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1677                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1678         BUG_ON(err < 0);
1679
1680         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1681         list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1682
1683         ovs_unlock();
1684
1685         ovs_notify(&dp_datapath_genl_family, reply, info);
1686         return 0;
1687
1688 err_destroy_ports_array:
1689         ovs_unlock();
1690         kfree(dp->ports);
1691 err_destroy_percpu:
1692         free_percpu(dp->stats_percpu);
1693 err_destroy_table:
1694         ovs_flow_tbl_destroy(&dp->table);
1695 err_free_dp:
1696         kfree(dp);
1697 err_free_reply:
1698         kfree_skb(reply);
1699 err:
1700         return err;
1701 }
1702
1703 /* Called with ovs_mutex. */
1704 static void __dp_destroy(struct datapath *dp)
1705 {
1706         int i;
1707
1708         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1709                 struct vport *vport;
1710                 struct hlist_node *n;
1711
1712                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1713                         if (vport->port_no != OVSP_LOCAL)
1714                                 ovs_dp_detach_port(vport);
1715         }
1716
1717         list_del_rcu(&dp->list_node);
1718
1719         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1720          * all ports in datapath are destroyed first before freeing datapath.
1721          */
1722         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1723
1724         /* RCU destroy the flow table */
1725         call_rcu(&dp->rcu, destroy_dp_rcu);
1726 }
1727
1728 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1729 {
1730         struct sk_buff *reply;
1731         struct datapath *dp;
1732         int err;
1733
1734         reply = ovs_dp_cmd_alloc_info();
1735         if (!reply)
1736                 return -ENOMEM;
1737
1738         ovs_lock();
1739         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1740         err = PTR_ERR(dp);
1741         if (IS_ERR(dp))
1742                 goto err_unlock_free;
1743
1744         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1745                                    info->snd_seq, 0, OVS_DP_CMD_DEL);
1746         BUG_ON(err < 0);
1747
1748         __dp_destroy(dp);
1749         ovs_unlock();
1750
1751         ovs_notify(&dp_datapath_genl_family, reply, info);
1752
1753         return 0;
1754
1755 err_unlock_free:
1756         ovs_unlock();
1757         kfree_skb(reply);
1758         return err;
1759 }
1760
1761 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1762 {
1763         struct sk_buff *reply;
1764         struct datapath *dp;
1765         int err;
1766
1767         reply = ovs_dp_cmd_alloc_info();
1768         if (!reply)
1769                 return -ENOMEM;
1770
1771         ovs_lock();
1772         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1773         err = PTR_ERR(dp);
1774         if (IS_ERR(dp))
1775                 goto err_unlock_free;
1776
1777         ovs_dp_change(dp, info->attrs);
1778
1779         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1780                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1781         BUG_ON(err < 0);
1782
1783         ovs_unlock();
1784         ovs_notify(&dp_datapath_genl_family, reply, info);
1785
1786         return 0;
1787
1788 err_unlock_free:
1789         ovs_unlock();
1790         kfree_skb(reply);
1791         return err;
1792 }
1793
1794 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1795 {
1796         struct sk_buff *reply;
1797         struct datapath *dp;
1798         int err;
1799
1800         reply = ovs_dp_cmd_alloc_info();
1801         if (!reply)
1802                 return -ENOMEM;
1803
1804         ovs_lock();
1805         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1806         if (IS_ERR(dp)) {
1807                 err = PTR_ERR(dp);
1808                 goto err_unlock_free;
1809         }
1810         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1811                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1812         BUG_ON(err < 0);
1813         ovs_unlock();
1814
1815         return genlmsg_reply(reply, info);
1816
1817 err_unlock_free:
1818         ovs_unlock();
1819         kfree_skb(reply);
1820         return err;
1821 }
1822
1823 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1824 {
1825         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1826         struct datapath *dp;
1827         int skip = cb->args[0];
1828         int i = 0;
1829
1830         ovs_lock();
1831         list_for_each_entry(dp, &ovs_net->dps, list_node) {
1832                 if (i >= skip &&
1833                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1834                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1835                                          OVS_DP_CMD_NEW) < 0)
1836                         break;
1837                 i++;
1838         }
1839         ovs_unlock();
1840
1841         cb->args[0] = i;
1842
1843         return skb->len;
1844 }
1845
1846 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1847         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1848         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1849         [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1850 };
1851
1852 static const struct genl_ops dp_datapath_genl_ops[] = {
1853         { .cmd = OVS_DP_CMD_NEW,
1854           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1855           .policy = datapath_policy,
1856           .doit = ovs_dp_cmd_new
1857         },
1858         { .cmd = OVS_DP_CMD_DEL,
1859           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1860           .policy = datapath_policy,
1861           .doit = ovs_dp_cmd_del
1862         },
1863         { .cmd = OVS_DP_CMD_GET,
1864           .flags = 0,               /* OK for unprivileged users. */
1865           .policy = datapath_policy,
1866           .doit = ovs_dp_cmd_get,
1867           .dumpit = ovs_dp_cmd_dump
1868         },
1869         { .cmd = OVS_DP_CMD_SET,
1870           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1871           .policy = datapath_policy,
1872           .doit = ovs_dp_cmd_set,
1873         },
1874 };
1875
1876 static struct genl_family dp_datapath_genl_family __ro_after_init = {
1877         .hdrsize = sizeof(struct ovs_header),
1878         .name = OVS_DATAPATH_FAMILY,
1879         .version = OVS_DATAPATH_VERSION,
1880         .maxattr = OVS_DP_ATTR_MAX,
1881         .netnsok = true,
1882         .parallel_ops = true,
1883         .ops = dp_datapath_genl_ops,
1884         .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1885         .mcgrps = &ovs_dp_datapath_multicast_group,
1886         .n_mcgrps = 1,
1887         .module = THIS_MODULE,
1888 };
1889
1890 /* Called with ovs_mutex or RCU read lock. */
1891 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1892                                    u32 portid, u32 seq, u32 flags, u8 cmd)
1893 {
1894         struct ovs_header *ovs_header;
1895         struct ovs_vport_stats vport_stats;
1896         int err;
1897
1898         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1899                                  flags, cmd);
1900         if (!ovs_header)
1901                 return -EMSGSIZE;
1902
1903         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1904
1905         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1906             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1907             nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1908                            ovs_vport_name(vport)))
1909                 goto nla_put_failure;
1910
1911         ovs_vport_get_stats(vport, &vport_stats);
1912         if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
1913                           sizeof(struct ovs_vport_stats), &vport_stats,
1914                           OVS_VPORT_ATTR_PAD))
1915                 goto nla_put_failure;
1916
1917         if (ovs_vport_get_upcall_portids(vport, skb))
1918                 goto nla_put_failure;
1919
1920         err = ovs_vport_get_options(vport, skb);
1921         if (err == -EMSGSIZE)
1922                 goto error;
1923
1924         genlmsg_end(skb, ovs_header);
1925         return 0;
1926
1927 nla_put_failure:
1928         err = -EMSGSIZE;
1929 error:
1930         genlmsg_cancel(skb, ovs_header);
1931         return err;
1932 }
1933
1934 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1935 {
1936         return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1937 }
1938
1939 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1940 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1941                                          u32 seq, u8 cmd)
1942 {
1943         struct sk_buff *skb;
1944         int retval;
1945
1946         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1947         if (!skb)
1948                 return ERR_PTR(-ENOMEM);
1949
1950         retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1951         BUG_ON(retval < 0);
1952
1953         return skb;
1954 }
1955
1956 /* Called with ovs_mutex or RCU read lock. */
1957 static struct vport *lookup_vport(struct net *net,
1958                                   const struct ovs_header *ovs_header,
1959                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1960 {
1961         struct datapath *dp;
1962         struct vport *vport;
1963
1964         if (a[OVS_VPORT_ATTR_NAME]) {
1965                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1966                 if (!vport)
1967                         return ERR_PTR(-ENODEV);
1968                 if (ovs_header->dp_ifindex &&
1969                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1970                         return ERR_PTR(-ENODEV);
1971                 return vport;
1972         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1973                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1974
1975                 if (port_no >= DP_MAX_PORTS)
1976                         return ERR_PTR(-EFBIG);
1977
1978                 dp = get_dp(net, ovs_header->dp_ifindex);
1979                 if (!dp)
1980                         return ERR_PTR(-ENODEV);
1981
1982                 vport = ovs_vport_ovsl_rcu(dp, port_no);
1983                 if (!vport)
1984                         return ERR_PTR(-ENODEV);
1985                 return vport;
1986         } else
1987                 return ERR_PTR(-EINVAL);
1988 }
1989
1990 /* Called with ovs_mutex */
1991 static void update_headroom(struct datapath *dp)
1992 {
1993         unsigned dev_headroom, max_headroom = 0;
1994         struct net_device *dev;
1995         struct vport *vport;
1996         int i;
1997
1998         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1999                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2000                         dev = vport->dev;
2001                         dev_headroom = netdev_get_fwd_headroom(dev);
2002                         if (dev_headroom > max_headroom)
2003                                 max_headroom = dev_headroom;
2004                 }
2005         }
2006
2007         dp->max_headroom = max_headroom;
2008         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
2009                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node)
2010                         netdev_set_rx_headroom(vport->dev, max_headroom);
2011 }
2012
2013 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
2014 {
2015         struct nlattr **a = info->attrs;
2016         struct ovs_header *ovs_header = info->userhdr;
2017         struct vport_parms parms;
2018         struct sk_buff *reply;
2019         struct vport *vport;
2020         struct datapath *dp;
2021         u32 port_no;
2022         int err;
2023
2024         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
2025             !a[OVS_VPORT_ATTR_UPCALL_PID])
2026                 return -EINVAL;
2027
2028         port_no = a[OVS_VPORT_ATTR_PORT_NO]
2029                 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
2030         if (port_no >= DP_MAX_PORTS)
2031                 return -EFBIG;
2032
2033         reply = ovs_vport_cmd_alloc_info();
2034         if (!reply)
2035                 return -ENOMEM;
2036
2037         ovs_lock();
2038 restart:
2039         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2040         err = -ENODEV;
2041         if (!dp)
2042                 goto exit_unlock_free;
2043
2044         if (port_no) {
2045                 vport = ovs_vport_ovsl(dp, port_no);
2046                 err = -EBUSY;
2047                 if (vport)
2048                         goto exit_unlock_free;
2049         } else {
2050                 for (port_no = 1; ; port_no++) {
2051                         if (port_no >= DP_MAX_PORTS) {
2052                                 err = -EFBIG;
2053                                 goto exit_unlock_free;
2054                         }
2055                         vport = ovs_vport_ovsl(dp, port_no);
2056                         if (!vport)
2057                                 break;
2058                 }
2059         }
2060
2061         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2062         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2063         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2064         parms.dp = dp;
2065         parms.port_no = port_no;
2066         parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
2067
2068         vport = new_vport(&parms);
2069         err = PTR_ERR(vport);
2070         if (IS_ERR(vport)) {
2071                 if (err == -EAGAIN)
2072                         goto restart;
2073                 goto exit_unlock_free;
2074         }
2075
2076         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2077                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2078
2079         if (netdev_get_fwd_headroom(vport->dev) > dp->max_headroom)
2080                 update_headroom(dp);
2081         else
2082                 netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2083
2084         BUG_ON(err < 0);
2085         ovs_unlock();
2086
2087         ovs_notify(&dp_vport_genl_family, reply, info);
2088         return 0;
2089
2090 exit_unlock_free:
2091         ovs_unlock();
2092         kfree_skb(reply);
2093         return err;
2094 }
2095
2096 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2097 {
2098         struct nlattr **a = info->attrs;
2099         struct sk_buff *reply;
2100         struct vport *vport;
2101         int err;
2102
2103         reply = ovs_vport_cmd_alloc_info();
2104         if (!reply)
2105                 return -ENOMEM;
2106
2107         ovs_lock();
2108         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2109         err = PTR_ERR(vport);
2110         if (IS_ERR(vport))
2111                 goto exit_unlock_free;
2112
2113         if (a[OVS_VPORT_ATTR_TYPE] &&
2114             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2115                 err = -EINVAL;
2116                 goto exit_unlock_free;
2117         }
2118
2119         if (a[OVS_VPORT_ATTR_OPTIONS]) {
2120                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2121                 if (err)
2122                         goto exit_unlock_free;
2123         }
2124
2125
2126         if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2127                 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2128
2129                 err = ovs_vport_set_upcall_portids(vport, ids);
2130                 if (err)
2131                         goto exit_unlock_free;
2132         }
2133
2134         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2135                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2136         BUG_ON(err < 0);
2137
2138         ovs_unlock();
2139         ovs_notify(&dp_vport_genl_family, reply, info);
2140         return 0;
2141
2142 exit_unlock_free:
2143         ovs_unlock();
2144         kfree_skb(reply);
2145         return err;
2146 }
2147
2148 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2149 {
2150         bool must_update_headroom = false;
2151         struct nlattr **a = info->attrs;
2152         struct sk_buff *reply;
2153         struct datapath *dp;
2154         struct vport *vport;
2155         int err;
2156
2157         reply = ovs_vport_cmd_alloc_info();
2158         if (!reply)
2159                 return -ENOMEM;
2160
2161         ovs_lock();
2162         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2163         err = PTR_ERR(vport);
2164         if (IS_ERR(vport))
2165                 goto exit_unlock_free;
2166
2167         if (vport->port_no == OVSP_LOCAL) {
2168                 err = -EINVAL;
2169                 goto exit_unlock_free;
2170         }
2171
2172         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2173                                       info->snd_seq, 0, OVS_VPORT_CMD_DEL);
2174         BUG_ON(err < 0);
2175
2176         /* the vport deletion may trigger dp headroom update */
2177         dp = vport->dp;
2178         if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2179                 must_update_headroom = true;
2180         netdev_reset_rx_headroom(vport->dev);
2181         ovs_dp_detach_port(vport);
2182
2183         if (must_update_headroom)
2184                 update_headroom(dp);
2185         ovs_unlock();
2186
2187         ovs_notify(&dp_vport_genl_family, reply, info);
2188         return 0;
2189
2190 exit_unlock_free:
2191         ovs_unlock();
2192         kfree_skb(reply);
2193         return err;
2194 }
2195
2196 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2197 {
2198         struct nlattr **a = info->attrs;
2199         struct ovs_header *ovs_header = info->userhdr;
2200         struct sk_buff *reply;
2201         struct vport *vport;
2202         int err;
2203
2204         reply = ovs_vport_cmd_alloc_info();
2205         if (!reply)
2206                 return -ENOMEM;
2207
2208         rcu_read_lock();
2209         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2210         err = PTR_ERR(vport);
2211         if (IS_ERR(vport))
2212                 goto exit_unlock_free;
2213         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2214                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2215         BUG_ON(err < 0);
2216         rcu_read_unlock();
2217
2218         return genlmsg_reply(reply, info);
2219
2220 exit_unlock_free:
2221         rcu_read_unlock();
2222         kfree_skb(reply);
2223         return err;
2224 }
2225
2226 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2227 {
2228         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2229         struct datapath *dp;
2230         int bucket = cb->args[0], skip = cb->args[1];
2231         int i, j = 0;
2232
2233         rcu_read_lock();
2234         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2235         if (!dp) {
2236                 rcu_read_unlock();
2237                 return -ENODEV;
2238         }
2239         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2240                 struct vport *vport;
2241
2242                 j = 0;
2243                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2244                         if (j >= skip &&
2245                             ovs_vport_cmd_fill_info(vport, skb,
2246                                                     NETLINK_CB(cb->skb).portid,
2247                                                     cb->nlh->nlmsg_seq,
2248                                                     NLM_F_MULTI,
2249                                                     OVS_VPORT_CMD_NEW) < 0)
2250                                 goto out;
2251
2252                         j++;
2253                 }
2254                 skip = 0;
2255         }
2256 out:
2257         rcu_read_unlock();
2258
2259         cb->args[0] = i;
2260         cb->args[1] = j;
2261
2262         return skb->len;
2263 }
2264
2265 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2266         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2267         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2268         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2269         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2270         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_UNSPEC },
2271         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2272 };
2273
2274 static const struct genl_ops dp_vport_genl_ops[] = {
2275         { .cmd = OVS_VPORT_CMD_NEW,
2276           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2277           .policy = vport_policy,
2278           .doit = ovs_vport_cmd_new
2279         },
2280         { .cmd = OVS_VPORT_CMD_DEL,
2281           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2282           .policy = vport_policy,
2283           .doit = ovs_vport_cmd_del
2284         },
2285         { .cmd = OVS_VPORT_CMD_GET,
2286           .flags = 0,               /* OK for unprivileged users. */
2287           .policy = vport_policy,
2288           .doit = ovs_vport_cmd_get,
2289           .dumpit = ovs_vport_cmd_dump
2290         },
2291         { .cmd = OVS_VPORT_CMD_SET,
2292           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2293           .policy = vport_policy,
2294           .doit = ovs_vport_cmd_set,
2295         },
2296 };
2297
2298 struct genl_family dp_vport_genl_family __ro_after_init = {
2299         .hdrsize = sizeof(struct ovs_header),
2300         .name = OVS_VPORT_FAMILY,
2301         .version = OVS_VPORT_VERSION,
2302         .maxattr = OVS_VPORT_ATTR_MAX,
2303         .netnsok = true,
2304         .parallel_ops = true,
2305         .ops = dp_vport_genl_ops,
2306         .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2307         .mcgrps = &ovs_dp_vport_multicast_group,
2308         .n_mcgrps = 1,
2309         .module = THIS_MODULE,
2310 };
2311
2312 static struct genl_family * const dp_genl_families[] = {
2313         &dp_datapath_genl_family,
2314         &dp_vport_genl_family,
2315         &dp_flow_genl_family,
2316         &dp_packet_genl_family,
2317 };
2318
2319 static void dp_unregister_genl(int n_families)
2320 {
2321         int i;
2322
2323         for (i = 0; i < n_families; i++)
2324                 genl_unregister_family(dp_genl_families[i]);
2325 }
2326
2327 static int __init dp_register_genl(void)
2328 {
2329         int err;
2330         int i;
2331
2332         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2333
2334                 err = genl_register_family(dp_genl_families[i]);
2335                 if (err)
2336                         goto error;
2337         }
2338
2339         return 0;
2340
2341 error:
2342         dp_unregister_genl(i);
2343         return err;
2344 }
2345
2346 static int __net_init ovs_init_net(struct net *net)
2347 {
2348         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2349
2350         INIT_LIST_HEAD(&ovs_net->dps);
2351         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2352         ovs_ct_init(net);
2353         return 0;
2354 }
2355
2356 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2357                                             struct list_head *head)
2358 {
2359         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2360         struct datapath *dp;
2361
2362         list_for_each_entry(dp, &ovs_net->dps, list_node) {
2363                 int i;
2364
2365                 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2366                         struct vport *vport;
2367
2368                         hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2369                                 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2370                                         continue;
2371
2372                                 if (dev_net(vport->dev) == dnet)
2373                                         list_add(&vport->detach_list, head);
2374                         }
2375                 }
2376         }
2377 }
2378
2379 static void __net_exit ovs_exit_net(struct net *dnet)
2380 {
2381         struct datapath *dp, *dp_next;
2382         struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2383         struct vport *vport, *vport_next;
2384         struct net *net;
2385         LIST_HEAD(head);
2386
2387         ovs_ct_exit(dnet);
2388         ovs_lock();
2389         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2390                 __dp_destroy(dp);
2391
2392         rtnl_lock();
2393         for_each_net(net)
2394                 list_vports_from_net(net, dnet, &head);
2395         rtnl_unlock();
2396
2397         /* Detach all vports from given namespace. */
2398         list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2399                 list_del(&vport->detach_list);
2400                 ovs_dp_detach_port(vport);
2401         }
2402
2403         ovs_unlock();
2404
2405         cancel_work_sync(&ovs_net->dp_notify_work);
2406 }
2407
2408 static struct pernet_operations ovs_net_ops = {
2409         .init = ovs_init_net,
2410         .exit = ovs_exit_net,
2411         .id   = &ovs_net_id,
2412         .size = sizeof(struct ovs_net),
2413 };
2414
2415 static int __init dp_init(void)
2416 {
2417         int err;
2418
2419         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2420
2421         pr_info("Open vSwitch switching datapath\n");
2422
2423         err = action_fifos_init();
2424         if (err)
2425                 goto error;
2426
2427         err = ovs_internal_dev_rtnl_link_register();
2428         if (err)
2429                 goto error_action_fifos_exit;
2430
2431         err = ovs_flow_init();
2432         if (err)
2433                 goto error_unreg_rtnl_link;
2434
2435         err = ovs_vport_init();
2436         if (err)
2437                 goto error_flow_exit;
2438
2439         err = register_pernet_device(&ovs_net_ops);
2440         if (err)
2441                 goto error_vport_exit;
2442
2443         err = register_netdevice_notifier(&ovs_dp_device_notifier);
2444         if (err)
2445                 goto error_netns_exit;
2446
2447         err = ovs_netdev_init();
2448         if (err)
2449                 goto error_unreg_notifier;
2450
2451         err = dp_register_genl();
2452         if (err < 0)
2453                 goto error_unreg_netdev;
2454
2455         return 0;
2456
2457 error_unreg_netdev:
2458         ovs_netdev_exit();
2459 error_unreg_notifier:
2460         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2461 error_netns_exit:
2462         unregister_pernet_device(&ovs_net_ops);
2463 error_vport_exit:
2464         ovs_vport_exit();
2465 error_flow_exit:
2466         ovs_flow_exit();
2467 error_unreg_rtnl_link:
2468         ovs_internal_dev_rtnl_link_unregister();
2469 error_action_fifos_exit:
2470         action_fifos_exit();
2471 error:
2472         return err;
2473 }
2474
2475 static void dp_cleanup(void)
2476 {
2477         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2478         ovs_netdev_exit();
2479         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2480         unregister_pernet_device(&ovs_net_ops);
2481         rcu_barrier();
2482         ovs_vport_exit();
2483         ovs_flow_exit();
2484         ovs_internal_dev_rtnl_link_unregister();
2485         action_fifos_exit();
2486 }
2487
2488 module_init(dp_init);
2489 module_exit(dp_cleanup);
2490
2491 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2492 MODULE_LICENSE("GPL");
2493 MODULE_ALIAS_GENL_FAMILY(OVS_DATAPATH_FAMILY);
2494 MODULE_ALIAS_GENL_FAMILY(OVS_VPORT_FAMILY);
2495 MODULE_ALIAS_GENL_FAMILY(OVS_FLOW_FAMILY);
2496 MODULE_ALIAS_GENL_FAMILY(OVS_PACKET_FAMILY);