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