datapath/vport.c

   1 /*
   2  * Copyright (c) 2007-2015 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 #include <linux/etherdevice.h>
  20 #include <linux/if.h>
  21 #include <linux/if_vlan.h>
  22 #include <linux/jhash.h>
  23 #include <linux/kernel.h>
  24 #include <linux/list.h>
  25 #include <linux/mutex.h>
  26 #include <linux/percpu.h>
  27 #include <linux/rcupdate.h>
  28 #include <linux/rtnetlink.h>
  29 #include <linux/compat.h>
  30 #include <linux/module.h>
  31 #include <linux/if_link.h>
  32 #include <net/net_namespace.h>
  33 #include <net/lisp.h>
  34 #include <net/gre.h>
  35 #include <net/geneve.h>
  36 #include <net/route.h>
  37 #include <net/stt.h>
  38 #include <net/vxlan.h>
  39
  40 #include "datapath.h"
  41 #include "gso.h"
  42 #include "vport.h"
  43 #include "vport-internal_dev.h"
  44
  45 static LIST_HEAD(vport_ops_list);
  46
  47 /* Protected by RCU read lock for reading, ovs_mutex for writing. */
  48 static struct hlist_head *dev_table;
  49 #define VPORT_HASH_BUCKETS 1024
  50
  51 /**
  52  *      ovs_vport_init - initialize vport subsystem
  53  *
  54  * Called at module load time to initialize the vport subsystem.
  55  */
  56 int ovs_vport_init(void)
  57 {
  58         int err;
  59
  60         dev_table = kzalloc(VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
  61                             GFP_KERNEL);
  62         if (!dev_table)
  63                 return -ENOMEM;
  64
  65         err = lisp_init_module();
  66         if (err)
  67                 goto err_lisp;
  68         err = ipgre_init();
  69         if (err)
  70                 goto err_gre;
  71         err = geneve_init_module();
  72         if (err)
  73                 goto err_geneve;
  74
  75         err = vxlan_init_module();
  76         if (err)
  77                 goto err_vxlan;
  78         err = ovs_stt_init_module();
  79         if (err)
  80                 goto err_stt;
  81         return 0;
  82
  83 err_stt:
  84         vxlan_cleanup_module();
  85 err_vxlan:
  86         geneve_cleanup_module();
  87 err_geneve:
  88         ipgre_fini();
  89 err_gre:
  90         lisp_cleanup_module();
  91 err_lisp:
  92         kfree(dev_table);
  93         return err;
  94 }
  95
  96 /**
  97  *      ovs_vport_exit - shutdown vport subsystem
  98  *
  99  * Called at module exit time to shutdown the vport subsystem.
 100  */
 101 void ovs_vport_exit(void)
 102 {
 103         ovs_stt_cleanup_module();
 104         vxlan_cleanup_module();
 105         geneve_cleanup_module();
 106         ipgre_fini();
 107         lisp_cleanup_module();
 108         kfree(dev_table);
 109 }
 110
 111 static struct hlist_head *hash_bucket(const struct net *net, const char *name)
 112 {
 113         unsigned int hash = jhash(name, strlen(name), (unsigned long) net);
 114         return &dev_table[hash & (VPORT_HASH_BUCKETS - 1)];
 115 }
 116
 117 int __ovs_vport_ops_register(struct vport_ops *ops)
 118 {
 119         int err = -EEXIST;
 120         struct vport_ops *o;
 121
 122         ovs_lock();
 123         list_for_each_entry(o, &vport_ops_list, list)
 124                 if (ops->type == o->type)
 125                         goto errout;
 126
 127         list_add_tail(&ops->list, &vport_ops_list);
 128         err = 0;
 129 errout:
 130         ovs_unlock();
 131         return err;
 132 }
 133 EXPORT_SYMBOL_GPL(__ovs_vport_ops_register);
 134
 135 void ovs_vport_ops_unregister(struct vport_ops *ops)
 136 {
 137         ovs_lock();
 138         list_del(&ops->list);
 139         ovs_unlock();
 140 }
 141 EXPORT_SYMBOL_GPL(ovs_vport_ops_unregister);
 142
 143 /**
 144  *      ovs_vport_locate - find a port that has already been created
 145  *
 146  * @name: name of port to find
 147  *
 148  * Must be called with ovs or RCU read lock.
 149  */
 150 struct vport *ovs_vport_locate(const struct net *net, const char *name)
 151 {
 152         struct hlist_head *bucket = hash_bucket(net, name);
 153         struct vport *vport;
 154
 155         hlist_for_each_entry_rcu(vport, bucket, hash_node)
 156                 if (!strcmp(name, ovs_vport_name(vport)) &&
 157                     net_eq(ovs_dp_get_net(vport->dp), net))
 158                         return vport;
 159
 160         return NULL;
 161 }
 162
 163 /**
 164  *      ovs_vport_alloc - allocate and initialize new vport
 165  *
 166  * @priv_size: Size of private data area to allocate.
 167  * @ops: vport device ops
 168  *
 169  * Allocate and initialize a new vport defined by @ops.  The vport will contain
 170  * a private data area of size @priv_size that can be accessed using
 171  * vport_priv().  vports that are no longer needed should be released with
 172  * vport_free().
 173  */
 174 struct vport *ovs_vport_alloc(int priv_size, const struct vport_ops *ops,
 175                           const struct vport_parms *parms)
 176 {
 177         struct vport *vport;
 178         size_t alloc_size;
 179
 180         alloc_size = sizeof(struct vport);
 181         if (priv_size) {
 182                 alloc_size = ALIGN(alloc_size, VPORT_ALIGN);
 183                 alloc_size += priv_size;
 184         }
 185
 186         vport = kzalloc(alloc_size, GFP_KERNEL);
 187         if (!vport)
 188                 return ERR_PTR(-ENOMEM);
 189
 190         vport->dp = parms->dp;
 191         vport->port_no = parms->port_no;
 192         vport->ops = ops;
 193         INIT_HLIST_NODE(&vport->dp_hash_node);
 194
 195         if (ovs_vport_set_upcall_portids(vport, parms->upcall_portids)) {
 196                 kfree(vport);
 197                 return ERR_PTR(-EINVAL);
 198         }
 199
 200         return vport;
 201 }
 202 EXPORT_SYMBOL_GPL(ovs_vport_alloc);
 203
 204 /**
 205  *      ovs_vport_free - uninitialize and free vport
 206  *
 207  * @vport: vport to free
 208  *
 209  * Frees a vport allocated with vport_alloc() when it is no longer needed.
 210  *
 211  * The caller must ensure that an RCU grace period has passed since the last
 212  * time @vport was in a datapath.
 213  */
 214 void ovs_vport_free(struct vport *vport)
 215 {
 216         /* vport is freed from RCU callback or error path, Therefore
 217          * it is safe to use raw dereference.
 218          */
 219         kfree(rcu_dereference_raw(vport->upcall_portids));
 220         kfree(vport);
 221 }
 222 EXPORT_SYMBOL_GPL(ovs_vport_free);
 223
 224 static struct vport_ops *ovs_vport_lookup(const struct vport_parms *parms)
 225 {
 226         struct vport_ops *ops;
 227
 228         list_for_each_entry(ops, &vport_ops_list, list)
 229                 if (ops->type == parms->type)
 230                         return ops;
 231
 232         return NULL;
 233 }
 234
 235 /**
 236  *      ovs_vport_add - add vport device (for kernel callers)
 237  *
 238  * @parms: Information about new vport.
 239  *
 240  * Creates a new vport with the specified configuration (which is dependent on
 241  * device type).  ovs_mutex must be held.
 242  */
 243 struct vport *ovs_vport_add(const struct vport_parms *parms)
 244 {
 245         struct vport_ops *ops;
 246         struct vport *vport;
 247
 248         ops = ovs_vport_lookup(parms);
 249         if (ops) {
 250                 struct hlist_head *bucket;
 251
 252                 if (!try_module_get(ops->owner))
 253                         return ERR_PTR(-EAFNOSUPPORT);
 254
 255                 vport = ops->create(parms);
 256                 if (IS_ERR(vport)) {
 257                         module_put(ops->owner);
 258                         return vport;
 259                 }
 260
 261                 bucket = hash_bucket(ovs_dp_get_net(vport->dp),
 262                                      ovs_vport_name(vport));
 263                 hlist_add_head_rcu(&vport->hash_node, bucket);
 264                 return vport;
 265         }
 266
 267         /* Unlock to attempt module load and return -EAGAIN if load
 268          * was successful as we need to restart the port addition
 269          * workflow.
 270          */
 271         ovs_unlock();
 272         request_module("vport-type-%d", parms->type);
 273         ovs_lock();
 274
 275         if (!ovs_vport_lookup(parms))
 276                 return ERR_PTR(-EAFNOSUPPORT);
 277         else
 278                 return ERR_PTR(-EAGAIN);
 279 }
 280
 281 /**
 282  *      ovs_vport_set_options - modify existing vport device (for kernel callers)
 283  *
 284  * @vport: vport to modify.
 285  * @options: New configuration.
 286  *
 287  * Modifies an existing device with the specified configuration (which is
 288  * dependent on device type).  ovs_mutex must be held.
 289  */
 290 int ovs_vport_set_options(struct vport *vport, struct nlattr *options)
 291 {
 292         if (!vport->ops->set_options)
 293                 return -EOPNOTSUPP;
 294         return vport->ops->set_options(vport, options);
 295 }
 296
 297 /**
 298  *      ovs_vport_del - delete existing vport device
 299  *
 300  * @vport: vport to delete.
 301  *
 302  * Detaches @vport from its datapath and destroys it.  ovs_mutex must be
 303  * held.
 304  */
 305 void ovs_vport_del(struct vport *vport)
 306 {
 307         ASSERT_OVSL();
 308
 309         hlist_del_rcu(&vport->hash_node);
 310         module_put(vport->ops->owner);
 311         vport->ops->destroy(vport);
 312 }
 313
 314 /**
 315  *      ovs_vport_get_stats - retrieve device stats
 316  *
 317  * @vport: vport from which to retrieve the stats
 318  * @stats: location to store stats
 319  *
 320  * Retrieves transmit, receive, and error stats for the given device.
 321  *
 322  * Must be called with ovs_mutex or rcu_read_lock.
 323  */
 324 void ovs_vport_get_stats(struct vport *vport, struct ovs_vport_stats *stats)
 325 {
 326         const struct rtnl_link_stats64 *dev_stats;
 327         struct rtnl_link_stats64 temp;
 328
 329         dev_stats = dev_get_stats(vport->dev, &temp);
 330         stats->rx_errors  = dev_stats->rx_errors;
 331         stats->tx_errors  = dev_stats->tx_errors;
 332         stats->tx_dropped = dev_stats->tx_dropped;
 333         stats->rx_dropped = dev_stats->rx_dropped;
 334
 335         stats->rx_bytes   = dev_stats->rx_bytes;
 336         stats->rx_packets = dev_stats->rx_packets;
 337         stats->tx_bytes   = dev_stats->tx_bytes;
 338         stats->tx_packets = dev_stats->tx_packets;
 339 }
 340
 341 /**
 342  *      ovs_vport_get_options - retrieve device options
 343  *
 344  * @vport: vport from which to retrieve the options.
 345  * @skb: sk_buff where options should be appended.
 346  *
 347  * Retrieves the configuration of the given device, appending an
 348  * %OVS_VPORT_ATTR_OPTIONS attribute that in turn contains nested
 349  * vport-specific attributes to @skb.
 350  *
 351  * Returns 0 if successful, -EMSGSIZE if @skb has insufficient room, or another
 352  * negative error code if a real error occurred.  If an error occurs, @skb is
 353  * left unmodified.
 354  *
 355  * Must be called with ovs_mutex or rcu_read_lock.
 356  */
 357 int ovs_vport_get_options(const struct vport *vport, struct sk_buff *skb)
 358 {
 359         struct nlattr *nla;
 360         int err;
 361
 362         if (!vport->ops->get_options)
 363                 return 0;
 364
 365         nla = nla_nest_start(skb, OVS_VPORT_ATTR_OPTIONS);
 366         if (!nla)
 367                 return -EMSGSIZE;
 368
 369         err = vport->ops->get_options(vport, skb);
 370         if (err) {
 371                 nla_nest_cancel(skb, nla);
 372                 return err;
 373         }
 374
 375         nla_nest_end(skb, nla);
 376         return 0;
 377 }
 378
 379 /**
 380  *      ovs_vport_set_upcall_portids - set upcall portids of @vport.
 381  *
 382  * @vport: vport to modify.
 383  * @ids: new configuration, an array of port ids.
 384  *
 385  * Sets the vport's upcall_portids to @ids.
 386  *
 387  * Returns 0 if successful, -EINVAL if @ids is zero length or cannot be parsed
 388  * as an array of U32.
 389  *
 390  * Must be called with ovs_mutex.
 391  */
 392 int ovs_vport_set_upcall_portids(struct vport *vport, const struct nlattr *ids)
 393 {
 394         struct vport_portids *old, *vport_portids;
 395
 396         if (!nla_len(ids) || nla_len(ids) % sizeof(u32))
 397                 return -EINVAL;
 398
 399         old = ovsl_dereference(vport->upcall_portids);
 400
 401         vport_portids = kmalloc(sizeof(*vport_portids) + nla_len(ids),
 402                                 GFP_KERNEL);
 403         if (!vport_portids)
 404                 return -ENOMEM;
 405
 406         vport_portids->n_ids = nla_len(ids) / sizeof(u32);
 407         vport_portids->rn_ids = reciprocal_value(vport_portids->n_ids);
 408         nla_memcpy(vport_portids->ids, ids, nla_len(ids));
 409
 410         rcu_assign_pointer(vport->upcall_portids, vport_portids);
 411
 412         if (old)
 413                 kfree_rcu(old, rcu);
 414         return 0;
 415 }
 416
 417 /**
 418  *      ovs_vport_get_upcall_portids - get the upcall_portids of @vport.
 419  *
 420  * @vport: vport from which to retrieve the portids.
 421  * @skb: sk_buff where portids should be appended.
 422  *
 423  * Retrieves the configuration of the given vport, appending the
 424  * %OVS_VPORT_ATTR_UPCALL_PID attribute which is the array of upcall
 425  * portids to @skb.
 426  *
 427  * Returns 0 if successful, -EMSGSIZE if @skb has insufficient room.
 428  * If an error occurs, @skb is left unmodified.  Must be called with
 429  * ovs_mutex or rcu_read_lock.
 430  */
 431 int ovs_vport_get_upcall_portids(const struct vport *vport,
 432                                  struct sk_buff *skb)
 433 {
 434         struct vport_portids *ids;
 435
 436         ids = rcu_dereference_ovsl(vport->upcall_portids);
 437
 438         if (vport->dp->user_features & OVS_DP_F_VPORT_PIDS)
 439                 return nla_put(skb, OVS_VPORT_ATTR_UPCALL_PID,
 440                                ids->n_ids * sizeof(u32), (void *)ids->ids);
 441         else
 442                 return nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, ids->ids[0]);
 443 }
 444
 445 /**
 446  *      ovs_vport_find_upcall_portid - find the upcall portid to send upcall.
 447  *
 448  * @vport: vport from which the missed packet is received.
 449  * @skb: skb that the missed packet was received.
 450  *
 451  * Uses the skb_get_hash() to select the upcall portid to send the
 452  * upcall.
 453  *
 454  * Returns the portid of the target socket.  Must be called with rcu_read_lock.
 455  */
 456 u32 ovs_vport_find_upcall_portid(const struct vport *vport, struct sk_buff *skb)
 457 {
 458         struct vport_portids *ids;
 459         u32 ids_index;
 460         u32 hash;
 461
 462         ids = rcu_dereference(vport->upcall_portids);
 463
 464         if (ids->n_ids == 1 && ids->ids[0] == 0)
 465                 return 0;
 466
 467         hash = skb_get_hash(skb);
 468         ids_index = hash - ids->n_ids * reciprocal_divide(hash, ids->rn_ids);
 469         return ids->ids[ids_index];
 470 }
 471
 472 /**
 473  *      ovs_vport_receive - pass up received packet to the datapath for processing
 474  *
 475  * @vport: vport that received the packet
 476  * @skb: skb that was received
 477  * @tun_key: tunnel (if any) that carried packet
 478  *
 479  * Must be called with rcu_read_lock.  The packet cannot be shared and
 480  * skb->data should point to the Ethernet header.
 481  */
 482 int ovs_vport_receive(struct vport *vport, struct sk_buff *skb,
 483                       const struct ip_tunnel_info *tun_info)
 484 {
 485         struct sw_flow_key key;
 486         int error;
 487
 488         OVS_CB(skb)->input_vport = vport;
 489         OVS_CB(skb)->mru = 0;
 490         if (unlikely(dev_net(skb->dev) != ovs_dp_get_net(vport->dp))) {
 491                 u32 mark;
 492
 493                 mark = skb->mark;
 494                 skb_scrub_packet(skb, true);
 495                 skb->mark = mark;
 496                 tun_info = NULL;
 497         }
 498
 499         ovs_skb_init_inner_protocol(skb);
 500         skb_clear_ovs_gso_cb(skb);
 501         /* Extract flow from 'skb' into 'key'. */
 502         error = ovs_flow_key_extract(tun_info, skb, &key);
 503         if (unlikely(error)) {
 504                 kfree_skb(skb);
 505                 return error;
 506         }
 507         ovs_dp_process_packet(skb, &key);
 508         return 0;
 509 }
 510 EXPORT_SYMBOL_GPL(ovs_vport_receive);
 511
 512 static void free_vport_rcu(struct rcu_head *rcu)
 513 {
 514         struct vport *vport = container_of(rcu, struct vport, rcu);
 515
 516         ovs_vport_free(vport);
 517 }
 518
 519 void ovs_vport_deferred_free(struct vport *vport)
 520 {
 521         if (!vport)
 522                 return;
 523
 524         call_rcu(&vport->rcu, free_vport_rcu);
 525 }
 526 EXPORT_SYMBOL_GPL(ovs_vport_deferred_free);
 527
 528 static struct rtable *ovs_tunnel_route_lookup(struct net *net,
 529                                               const struct ip_tunnel_key *key,
 530                                               u32 mark,
 531                                               struct flowi4 *fl,
 532                                               u8 protocol)
 533 {
 534         struct rtable *rt;
 535
 536         memset(fl, 0, sizeof(*fl));
 537         fl->daddr = key->u.ipv4.dst;
 538         fl->saddr = key->u.ipv4.src;
 539         fl->flowi4_tos = RT_TOS(key->tos);
 540         fl->flowi4_mark = mark;
 541         fl->flowi4_proto = protocol;
 542
 543         rt = ip_route_output_key(net, fl);
 544         return rt;
 545 }
 546
 547 int ovs_tunnel_get_egress_info(struct dp_upcall_info *upcall,
 548                                struct net *net,
 549                                struct sk_buff *skb,
 550                                u8 ipproto,
 551                                __be16 tp_src,
 552                                __be16 tp_dst)
 553 {
 554         struct ip_tunnel_info *egress_tun_info = upcall->egress_tun_info;
 555         struct ip_tunnel_info *tun_info = skb_tunnel_info(skb);
 556         const struct ip_tunnel_key *tun_key;
 557         u32 skb_mark = skb->mark;
 558         struct rtable *rt;
 559         struct flowi4 fl;
 560
 561         if (unlikely(!tun_info))
 562                 return -EINVAL;
 563         if (ip_tunnel_info_af(tun_info) != AF_INET)
 564                 return -EINVAL;
 565
 566         tun_key = &tun_info->key;
 567
 568         /* Route lookup to get srouce IP address.
 569          * The process may need to be changed if the corresponding process
 570          * in vports ops changed.
 571          */
 572         rt = ovs_tunnel_route_lookup(net, tun_key, skb_mark, &fl, ipproto);
 573         if (IS_ERR(rt))
 574                 return PTR_ERR(rt);
 575
 576         ip_rt_put(rt);
 577
 578         /* Generate egress_tun_info based on tun_info,
 579          * saddr, tp_src and tp_dst
 580          */
 581         ip_tunnel_key_init(&egress_tun_info->key,
 582                            fl.saddr, tun_key->u.ipv4.dst,
 583                            tun_key->tos,
 584                            tun_key->ttl,
 585                            tp_src, tp_dst,
 586                            tun_key->tun_id,
 587                            tun_key->tun_flags);
 588         egress_tun_info->options_len = tun_info->options_len;
 589         egress_tun_info->mode = tun_info->mode;
 590         upcall->egress_tun_opts = ip_tunnel_info_opts(tun_info);
 591         return 0;
 592 }
 593 EXPORT_SYMBOL_GPL(ovs_tunnel_get_egress_info);
 594
 595 int ovs_vport_get_egress_tun_info(struct vport *vport, struct sk_buff *skb,
 596                                   struct dp_upcall_info *upcall)
 597 {
 598         /* get_egress_tun_info() is only implemented on tunnel ports. */
 599         if (unlikely(!vport->ops->get_egress_tun_info))
 600                 return -EINVAL;
 601
 602         return vport->ops->get_egress_tun_info(vport, skb, upcall);
 603 }
 604
 605 static unsigned int packet_length(const struct sk_buff *skb)
 606 {
 607         unsigned int length = skb->len - ETH_HLEN;
 608
 609         if (skb->protocol == htons(ETH_P_8021Q))
 610                 length -= VLAN_HLEN;
 611
 612         return length;
 613 }
 614
 615 void ovs_vport_send(struct vport *vport, struct sk_buff *skb)
 616 {
 617         int mtu = vport->dev->mtu;
 618
 619         if (unlikely(packet_length(skb) > mtu && !skb_is_gso(skb))) {
 620                 net_warn_ratelimited("%s: dropped over-mtu packet: %d > %d\n",
 621                                      vport->dev->name,
 622                                      packet_length(skb), mtu);
 623                 vport->dev->stats.tx_errors++;
 624                 goto drop;
 625         }
 626
 627         skb->dev = vport->dev;
 628         vport->ops->send(skb);
 629         return;
 630
 631 drop:
 632         kfree_skb(skb);
 633 }