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/stt.h>
  37 #include <net/vxlan.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 = ip6gre_init();
  71         if (err)
  72                 goto err_ip6gre;
  73         err = ip6_tunnel_init();
  74         if (err)
  75                 goto err_ip6_tunnel;
  76         err = geneve_init_module();
  77         if (err)
  78                 goto err_geneve;
  79
  80         err = vxlan_init_module();
  81         if (err)
  82                 goto err_vxlan;
  83         err = ovs_stt_init_module();
  84         if (err)
  85                 goto err_stt;
  86         return 0;
  87
  88 err_stt:
  89         vxlan_cleanup_module();
  90 err_vxlan:
  91         geneve_cleanup_module();
  92 err_geneve:
  93         ip6_tunnel_cleanup();
  94 err_ip6_tunnel:
  95         ip6gre_fini();
  96 err_ip6gre:
  97         ipgre_fini();
  98 err_gre:
  99         lisp_cleanup_module();
 100 err_lisp:
 101         kfree(dev_table);
 102         return err;
 103 }
 104
 105 /**
 106  *      ovs_vport_exit - shutdown vport subsystem
 107  *
 108  * Called at module exit time to shutdown the vport subsystem.
 109  */
 110 void ovs_vport_exit(void)
 111 {
 112         ovs_stt_cleanup_module();
 113         vxlan_cleanup_module();
 114         geneve_cleanup_module();
 115         ip6_tunnel_cleanup();
 116         ip6gre_fini();
 117         ipgre_fini();
 118         lisp_cleanup_module();
 119         kfree(dev_table);
 120 }
 121
 122 static struct hlist_head *hash_bucket(const struct net *net, const char *name)
 123 {
 124         unsigned int hash = jhash(name, strlen(name), (unsigned long) net);
 125         return &dev_table[hash & (VPORT_HASH_BUCKETS - 1)];
 126 }
 127
 128 int __ovs_vport_ops_register(struct vport_ops *ops)
 129 {
 130         int err = -EEXIST;
 131         struct vport_ops *o;
 132
 133         ovs_lock();
 134         list_for_each_entry(o, &vport_ops_list, list)
 135                 if (ops->type == o->type)
 136                         goto errout;
 137
 138         list_add_tail(&ops->list, &vport_ops_list);
 139         err = 0;
 140 errout:
 141         ovs_unlock();
 142         return err;
 143 }
 144 EXPORT_SYMBOL_GPL(__ovs_vport_ops_register);
 145
 146 void ovs_vport_ops_unregister(struct vport_ops *ops)
 147 {
 148         ovs_lock();
 149         list_del(&ops->list);
 150         ovs_unlock();
 151 }
 152 EXPORT_SYMBOL_GPL(ovs_vport_ops_unregister);
 153
 154 /**
 155  *      ovs_vport_locate - find a port that has already been created
 156  *
 157  * @name: name of port to find
 158  *
 159  * Must be called with ovs or RCU read lock.
 160  */
 161 struct vport *ovs_vport_locate(const struct net *net, const char *name)
 162 {
 163         struct hlist_head *bucket = hash_bucket(net, name);
 164         struct vport *vport;
 165
 166         hlist_for_each_entry_rcu(vport, bucket, hash_node)
 167                 if (!strcmp(name, ovs_vport_name(vport)) &&
 168                     net_eq(ovs_dp_get_net(vport->dp), net))
 169                         return vport;
 170
 171         return NULL;
 172 }
 173
 174 /**
 175  *      ovs_vport_alloc - allocate and initialize new vport
 176  *
 177  * @priv_size: Size of private data area to allocate.
 178  * @ops: vport device ops
 179  *
 180  * Allocate and initialize a new vport defined by @ops.  The vport will contain
 181  * a private data area of size @priv_size that can be accessed using
 182  * vport_priv().  vports that are no longer needed should be released with
 183  * vport_free().
 184  */
 185 struct vport *ovs_vport_alloc(int priv_size, const struct vport_ops *ops,
 186                           const struct vport_parms *parms)
 187 {
 188         struct vport *vport;
 189         size_t alloc_size;
 190
 191         alloc_size = sizeof(struct vport);
 192         if (priv_size) {
 193                 alloc_size = ALIGN(alloc_size, VPORT_ALIGN);
 194                 alloc_size += priv_size;
 195         }
 196
 197         vport = kzalloc(alloc_size, GFP_KERNEL);
 198         if (!vport)
 199                 return ERR_PTR(-ENOMEM);
 200
 201         vport->dp = parms->dp;
 202         vport->port_no = parms->port_no;
 203         vport->ops = ops;
 204         INIT_HLIST_NODE(&vport->dp_hash_node);
 205
 206         if (ovs_vport_set_upcall_portids(vport, parms->upcall_portids)) {
 207                 kfree(vport);
 208                 return ERR_PTR(-EINVAL);
 209         }
 210
 211         return vport;
 212 }
 213 EXPORT_SYMBOL_GPL(ovs_vport_alloc);
 214
 215 /**
 216  *      ovs_vport_free - uninitialize and free vport
 217  *
 218  * @vport: vport to free
 219  *
 220  * Frees a vport allocated with vport_alloc() when it is no longer needed.
 221  *
 222  * The caller must ensure that an RCU grace period has passed since the last
 223  * time @vport was in a datapath.
 224  */
 225 void ovs_vport_free(struct vport *vport)
 226 {
 227         /* vport is freed from RCU callback or error path, Therefore
 228          * it is safe to use raw dereference.
 229          */
 230         kfree(rcu_dereference_raw(vport->upcall_portids));
 231         kfree(vport);
 232 }
 233 EXPORT_SYMBOL_GPL(ovs_vport_free);
 234
 235 static struct vport_ops *ovs_vport_lookup(const struct vport_parms *parms)
 236 {
 237         struct vport_ops *ops;
 238
 239         list_for_each_entry(ops, &vport_ops_list, list)
 240                 if (ops->type == parms->type)
 241                         return ops;
 242
 243         return NULL;
 244 }
 245
 246 /**
 247  *      ovs_vport_add - add vport device (for kernel callers)
 248  *
 249  * @parms: Information about new vport.
 250  *
 251  * Creates a new vport with the specified configuration (which is dependent on
 252  * device type).  ovs_mutex must be held.
 253  */
 254 struct vport *ovs_vport_add(const struct vport_parms *parms)
 255 {
 256         struct vport_ops *ops;
 257         struct vport *vport;
 258
 259         ops = ovs_vport_lookup(parms);
 260         if (ops) {
 261                 struct hlist_head *bucket;
 262
 263                 if (!try_module_get(ops->owner))
 264                         return ERR_PTR(-EAFNOSUPPORT);
 265
 266                 vport = ops->create(parms);
 267                 if (IS_ERR(vport)) {
 268                         module_put(ops->owner);
 269                         return vport;
 270                 }
 271
 272                 bucket = hash_bucket(ovs_dp_get_net(vport->dp),
 273                                      ovs_vport_name(vport));
 274                 hlist_add_head_rcu(&vport->hash_node, bucket);
 275                 return vport;
 276         }
 277
 278         /* Unlock to attempt module load and return -EAGAIN if load
 279          * was successful as we need to restart the port addition
 280          * workflow.
 281          */
 282         ovs_unlock();
 283         request_module("vport-type-%d", parms->type);
 284         ovs_lock();
 285
 286         if (!ovs_vport_lookup(parms))
 287                 return ERR_PTR(-EAFNOSUPPORT);
 288         else
 289                 return ERR_PTR(-EAGAIN);
 290 }
 291
 292 /**
 293  *      ovs_vport_set_options - modify existing vport device (for kernel callers)
 294  *
 295  * @vport: vport to modify.
 296  * @options: New configuration.
 297  *
 298  * Modifies an existing device with the specified configuration (which is
 299  * dependent on device type).  ovs_mutex must be held.
 300  */
 301 int ovs_vport_set_options(struct vport *vport, struct nlattr *options)
 302 {
 303         if (!vport->ops->set_options)
 304                 return -EOPNOTSUPP;
 305         return vport->ops->set_options(vport, options);
 306 }
 307
 308 /**
 309  *      ovs_vport_del - delete existing vport device
 310  *
 311  * @vport: vport to delete.
 312  *
 313  * Detaches @vport from its datapath and destroys it.  ovs_mutex must be
 314  * held.
 315  */
 316 void ovs_vport_del(struct vport *vport)
 317 {
 318         ASSERT_OVSL();
 319
 320         hlist_del_rcu(&vport->hash_node);
 321         module_put(vport->ops->owner);
 322         vport->ops->destroy(vport);
 323 }
 324
 325 /**
 326  *      ovs_vport_get_stats - retrieve device stats
 327  *
 328  * @vport: vport from which to retrieve the stats
 329  * @stats: location to store stats
 330  *
 331  * Retrieves transmit, receive, and error stats for the given device.
 332  *
 333  * Must be called with ovs_mutex or rcu_read_lock.
 334  */
 335 void ovs_vport_get_stats(struct vport *vport, struct ovs_vport_stats *stats)
 336 {
 337         const struct rtnl_link_stats64 *dev_stats;
 338         struct rtnl_link_stats64 temp;
 339
 340         dev_stats = dev_get_stats(vport->dev, &temp);
 341         stats->rx_errors  = dev_stats->rx_errors;
 342         stats->tx_errors  = dev_stats->tx_errors;
 343         stats->tx_dropped = dev_stats->tx_dropped;
 344         stats->rx_dropped = dev_stats->rx_dropped;
 345
 346         stats->rx_bytes   = dev_stats->rx_bytes;
 347         stats->rx_packets = dev_stats->rx_packets;
 348         stats->tx_bytes   = dev_stats->tx_bytes;
 349         stats->tx_packets = dev_stats->tx_packets;
 350 }
 351
 352 /**
 353  *      ovs_vport_get_options - retrieve device options
 354  *
 355  * @vport: vport from which to retrieve the options.
 356  * @skb: sk_buff where options should be appended.
 357  *
 358  * Retrieves the configuration of the given device, appending an
 359  * %OVS_VPORT_ATTR_OPTIONS attribute that in turn contains nested
 360  * vport-specific attributes to @skb.
 361  *
 362  * Returns 0 if successful, -EMSGSIZE if @skb has insufficient room, or another
 363  * negative error code if a real error occurred.  If an error occurs, @skb is
 364  * left unmodified.
 365  *
 366  * Must be called with ovs_mutex or rcu_read_lock.
 367  */
 368 int ovs_vport_get_options(const struct vport *vport, struct sk_buff *skb)
 369 {
 370         struct nlattr *nla;
 371         int err;
 372
 373         if (!vport->ops->get_options)
 374                 return 0;
 375
 376         nla = nla_nest_start(skb, OVS_VPORT_ATTR_OPTIONS);
 377         if (!nla)
 378                 return -EMSGSIZE;
 379
 380         err = vport->ops->get_options(vport, skb);
 381         if (err) {
 382                 nla_nest_cancel(skb, nla);
 383                 return err;
 384         }
 385
 386         nla_nest_end(skb, nla);
 387         return 0;
 388 }
 389
 390 /**
 391  *      ovs_vport_set_upcall_portids - set upcall portids of @vport.
 392  *
 393  * @vport: vport to modify.
 394  * @ids: new configuration, an array of port ids.
 395  *
 396  * Sets the vport's upcall_portids to @ids.
 397  *
 398  * Returns 0 if successful, -EINVAL if @ids is zero length or cannot be parsed
 399  * as an array of U32.
 400  *
 401  * Must be called with ovs_mutex.
 402  */
 403 int ovs_vport_set_upcall_portids(struct vport *vport, const struct nlattr *ids)
 404 {
 405         struct vport_portids *old, *vport_portids;
 406
 407         if (!nla_len(ids) || nla_len(ids) % sizeof(u32))
 408                 return -EINVAL;
 409
 410         old = ovsl_dereference(vport->upcall_portids);
 411
 412         vport_portids = kmalloc(sizeof(*vport_portids) + nla_len(ids),
 413                                 GFP_KERNEL);
 414         if (!vport_portids)
 415                 return -ENOMEM;
 416
 417         vport_portids->n_ids = nla_len(ids) / sizeof(u32);
 418         vport_portids->rn_ids = reciprocal_value(vport_portids->n_ids);
 419         nla_memcpy(vport_portids->ids, ids, nla_len(ids));
 420
 421         rcu_assign_pointer(vport->upcall_portids, vport_portids);
 422
 423         if (old)
 424                 kfree_rcu(old, rcu);
 425         return 0;
 426 }
 427
 428 /**
 429  *      ovs_vport_get_upcall_portids - get the upcall_portids of @vport.
 430  *
 431  * @vport: vport from which to retrieve the portids.
 432  * @skb: sk_buff where portids should be appended.
 433  *
 434  * Retrieves the configuration of the given vport, appending the
 435  * %OVS_VPORT_ATTR_UPCALL_PID attribute which is the array of upcall
 436  * portids to @skb.
 437  *
 438  * Returns 0 if successful, -EMSGSIZE if @skb has insufficient room.
 439  * If an error occurs, @skb is left unmodified.  Must be called with
 440  * ovs_mutex or rcu_read_lock.
 441  */
 442 int ovs_vport_get_upcall_portids(const struct vport *vport,
 443                                  struct sk_buff *skb)
 444 {
 445         struct vport_portids *ids;
 446
 447         ids = rcu_dereference_ovsl(vport->upcall_portids);
 448
 449         if (vport->dp->user_features & OVS_DP_F_VPORT_PIDS)
 450                 return nla_put(skb, OVS_VPORT_ATTR_UPCALL_PID,
 451                                ids->n_ids * sizeof(u32), (void *)ids->ids);
 452         else
 453                 return nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, ids->ids[0]);
 454 }
 455
 456 /**
 457  *      ovs_vport_find_upcall_portid - find the upcall portid to send upcall.
 458  *
 459  * @vport: vport from which the missed packet is received.
 460  * @skb: skb that the missed packet was received.
 461  *
 462  * Uses the skb_get_hash() to select the upcall portid to send the
 463  * upcall.
 464  *
 465  * Returns the portid of the target socket.  Must be called with rcu_read_lock.
 466  */
 467 u32 ovs_vport_find_upcall_portid(const struct vport *vport, struct sk_buff *skb)
 468 {
 469         struct vport_portids *ids;
 470         u32 ids_index;
 471         u32 hash;
 472
 473         ids = rcu_dereference(vport->upcall_portids);
 474
 475         if (ids->n_ids == 1 && ids->ids[0] == 0)
 476                 return 0;
 477
 478         hash = skb_get_hash(skb);
 479         ids_index = hash - ids->n_ids * reciprocal_divide(hash, ids->rn_ids);
 480         return ids->ids[ids_index];
 481 }
 482
 483 /**
 484  *      ovs_vport_receive - pass up received packet to the datapath for processing
 485  *
 486  * @vport: vport that received the packet
 487  * @skb: skb that was received
 488  * @tun_key: tunnel (if any) that carried packet
 489  *
 490  * Must be called with rcu_read_lock.  The packet cannot be shared and
 491  * skb->data should point to the Ethernet header.
 492  */
 493 int ovs_vport_receive(struct vport *vport, struct sk_buff *skb,
 494                       const struct ip_tunnel_info *tun_info)
 495 {
 496         struct sw_flow_key key;
 497         int error;
 498
 499         OVS_CB(skb)->input_vport = vport;
 500         OVS_CB(skb)->mru = 0;
 501         OVS_CB(skb)->cutlen = 0;
 502         if (unlikely(dev_net(skb->dev) != ovs_dp_get_net(vport->dp))) {
 503                 u32 mark;
 504
 505                 mark = skb->mark;
 506                 skb_scrub_packet(skb, true);
 507                 skb->mark = mark;
 508                 tun_info = NULL;
 509         }
 510
 511         ovs_skb_init_inner_protocol(skb);
 512         skb_clear_ovs_gso_cb(skb);
 513         /* Extract flow from 'skb' into 'key'. */
 514         error = ovs_flow_key_extract(tun_info, skb, &key);
 515         if (unlikely(error)) {
 516                 kfree_skb(skb);
 517                 return error;
 518         }
 519         ovs_dp_process_packet(skb, &key);
 520         return 0;
 521 }
 522
 523 static int packet_length(const struct sk_buff *skb,
 524                          struct net_device *dev)
 525 {
 526         int length = skb->len - dev->hard_header_len;
 527
 528         if (!skb_vlan_tag_present(skb) &&
 529             eth_type_vlan(skb->protocol))
 530                 length -= VLAN_HLEN;
 531
 532         /* Don't subtract for multiple VLAN tags. Most (all?) drivers allow
 533          * (ETH_LEN + VLAN_HLEN) in addition to the mtu value, but almost none
 534          * account for 802.1ad. e.g. is_skb_forwardable().
 535          */
 536
 537         return length > 0 ? length: 0;
 538 }
 539
 540 void ovs_vport_send(struct vport *vport, struct sk_buff *skb, u8 mac_proto)
 541 {
 542         int mtu = vport->dev->mtu;
 543
 544         switch (vport->dev->type) {
 545         case ARPHRD_NONE:
 546                 if (mac_proto == MAC_PROTO_ETHERNET) {
 547                         skb_reset_network_header(skb);
 548                         skb_reset_mac_len(skb);
 549                         skb->protocol = htons(ETH_P_TEB);
 550                 } else if (mac_proto != MAC_PROTO_NONE) {
 551                         WARN_ON_ONCE(1);
 552                         goto drop;
 553                 }
 554                 break;
 555         case ARPHRD_ETHER:
 556                 if (mac_proto != MAC_PROTO_ETHERNET)
 557                         goto drop;
 558                 break;
 559         default:
 560                 goto drop;
 561         }
 562
 563         if (unlikely(packet_length(skb, vport->dev) > mtu &&
 564                      !skb_is_gso(skb))) {
 565                 net_warn_ratelimited("%s: dropped over-mtu packet: %d > %d\n",
 566                                      vport->dev->name,
 567                                      packet_length(skb, vport->dev), mtu);
 568                 vport->dev->stats.tx_errors++;
 569                 goto drop;
 570         }
 571
 572         skb->dev = vport->dev;
 573         vport->ops->send(skb);
 574         return;
 575
 576 drop:
 577         kfree_skb(skb);
 578 }