datapath/vport.c

   1 /*
   2  * Copyright (c) 2010, 2011 Nicira Networks.
   3  * Distributed under the terms of the GNU GPL version 2.
   4  *
   5  * Significant portions of this file may be copied from parts of the Linux
   6  * kernel, by Linus Torvalds and others.
   7  */
   8
   9 #include <linux/dcache.h>
  10 #include <linux/etherdevice.h>
  11 #include <linux/if.h>
  12 #include <linux/if_vlan.h>
  13 #include <linux/kernel.h>
  14 #include <linux/list.h>
  15 #include <linux/mutex.h>
  16 #include <linux/percpu.h>
  17 #include <linux/rcupdate.h>
  18 #include <linux/rtnetlink.h>
  19 #include <linux/compat.h>
  20 #include <linux/version.h>
  21
  22 #include "vport.h"
  23 #include "vport-internal_dev.h"
  24
  25 /* List of statically compiled vport implementations.  Don't forget to also
  26  * add yours to the list at the bottom of vport.h. */
  27 static const struct vport_ops *base_vport_ops_list[] = {
  28         &netdev_vport_ops,
  29         &internal_vport_ops,
  30         &patch_vport_ops,
  31         &gre_vport_ops,
  32 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,26)
  33         &capwap_vport_ops,
  34 #endif
  35 };
  36
  37 static const struct vport_ops **vport_ops_list;
  38 static int n_vport_types;
  39
  40 /* Protected by RCU read lock for reading, RTNL lock for writing. */
  41 static struct hlist_head *dev_table;
  42 #define VPORT_HASH_BUCKETS 1024
  43
  44 /**
  45  *      vport_init - initialize vport subsystem
  46  *
  47  * Called at module load time to initialize the vport subsystem and any
  48  * compiled in vport types.
  49  */
  50 int vport_init(void)
  51 {
  52         int err;
  53         int i;
  54
  55         dev_table = kzalloc(VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
  56                             GFP_KERNEL);
  57         if (!dev_table) {
  58                 err = -ENOMEM;
  59                 goto error;
  60         }
  61
  62         vport_ops_list = kmalloc(ARRAY_SIZE(base_vport_ops_list) *
  63                                  sizeof(struct vport_ops *), GFP_KERNEL);
  64         if (!vport_ops_list) {
  65                 err = -ENOMEM;
  66                 goto error_dev_table;
  67         }
  68
  69         for (i = 0; i < ARRAY_SIZE(base_vport_ops_list); i++) {
  70                 const struct vport_ops *new_ops = base_vport_ops_list[i];
  71
  72                 if (new_ops->init)
  73                         err = new_ops->init();
  74                 else
  75                         err = 0;
  76
  77                 if (!err)
  78                         vport_ops_list[n_vport_types++] = new_ops;
  79                 else if (new_ops->flags & VPORT_F_REQUIRED) {
  80                         vport_exit();
  81                         goto error;
  82                 }
  83         }
  84
  85         return 0;
  86
  87 error_dev_table:
  88         kfree(dev_table);
  89 error:
  90         return err;
  91 }
  92
  93 /**
  94  *      vport_exit - shutdown vport subsystem
  95  *
  96  * Called at module exit time to shutdown the vport subsystem and any
  97  * initialized vport types.
  98  */
  99 void vport_exit(void)
 100 {
 101         int i;
 102
 103         for (i = 0; i < n_vport_types; i++) {
 104                 if (vport_ops_list[i]->exit)
 105                         vport_ops_list[i]->exit();
 106         }
 107
 108         kfree(vport_ops_list);
 109         kfree(dev_table);
 110 }
 111
 112 static struct hlist_head *hash_bucket(const char *name)
 113 {
 114         unsigned int hash = full_name_hash(name, strlen(name));
 115         return &dev_table[hash & (VPORT_HASH_BUCKETS - 1)];
 116 }
 117
 118 /**
 119  *      vport_locate - find a port that has already been created
 120  *
 121  * @name: name of port to find
 122  *
 123  * Must be called with RTNL or RCU read lock.
 124  */
 125 struct vport *vport_locate(const char *name)
 126 {
 127         struct hlist_head *bucket = hash_bucket(name);
 128         struct vport *vport;
 129         struct hlist_node *node;
 130
 131         hlist_for_each_entry_rcu(vport, node, bucket, hash_node)
 132                 if (!strcmp(name, vport_get_name(vport)))
 133                         return vport;
 134
 135         return NULL;
 136 }
 137
 138 static void release_vport(struct kobject *kobj)
 139 {
 140         struct vport *p = container_of(kobj, struct vport, kobj);
 141         kfree(p);
 142 }
 143
 144 static struct kobj_type brport_ktype = {
 145 #ifdef CONFIG_SYSFS
 146         .sysfs_ops = &brport_sysfs_ops,
 147 #endif
 148         .release = release_vport
 149 };
 150
 151 /**
 152  *      vport_alloc - allocate and initialize new vport
 153  *
 154  * @priv_size: Size of private data area to allocate.
 155  * @ops: vport device ops
 156  *
 157  * Allocate and initialize a new vport defined by @ops.  The vport will contain
 158  * a private data area of size @priv_size that can be accessed using
 159  * vport_priv().  vports that are no longer needed should be released with
 160  * vport_free().
 161  */
 162 struct vport *vport_alloc(int priv_size, const struct vport_ops *ops, const struct vport_parms *parms)
 163 {
 164         struct vport *vport;
 165         size_t alloc_size;
 166
 167         alloc_size = sizeof(struct vport);
 168         if (priv_size) {
 169                 alloc_size = ALIGN(alloc_size, VPORT_ALIGN);
 170                 alloc_size += priv_size;
 171         }
 172
 173         vport = kzalloc(alloc_size, GFP_KERNEL);
 174         if (!vport)
 175                 return ERR_PTR(-ENOMEM);
 176
 177         vport->dp = parms->dp;
 178         vport->port_no = parms->port_no;
 179         atomic_set(&vport->sflow_pool, 0);
 180         vport->ops = ops;
 181
 182         /* Initialize kobject for bridge.  This will be added as
 183          * /sys/class/net/<devname>/brport later, if sysfs is enabled. */
 184         vport->kobj.kset = NULL;
 185         kobject_init(&vport->kobj, &brport_ktype);
 186
 187         vport->percpu_stats = alloc_percpu(struct vport_percpu_stats);
 188         if (!vport->percpu_stats)
 189                 return ERR_PTR(-ENOMEM);
 190
 191         spin_lock_init(&vport->stats_lock);
 192
 193         return vport;
 194 }
 195
 196 /**
 197  *      vport_free - uninitialize and free vport
 198  *
 199  * @vport: vport to free
 200  *
 201  * Frees a vport allocated with vport_alloc() when it is no longer needed.
 202  *
 203  * The caller must ensure that an RCU grace period has passed since the last
 204  * time @vport was in a datapath.
 205  */
 206 void vport_free(struct vport *vport)
 207 {
 208         free_percpu(vport->percpu_stats);
 209
 210         kobject_put(&vport->kobj);
 211 }
 212
 213 /**
 214  *      vport_add - add vport device (for kernel callers)
 215  *
 216  * @parms: Information about new vport.
 217  *
 218  * Creates a new vport with the specified configuration (which is dependent on
 219  * device type).  RTNL lock must be held.
 220  */
 221 struct vport *vport_add(const struct vport_parms *parms)
 222 {
 223         struct vport *vport;
 224         int err = 0;
 225         int i;
 226
 227         ASSERT_RTNL();
 228
 229         for (i = 0; i < n_vport_types; i++) {
 230                 if (vport_ops_list[i]->type == parms->type) {
 231                         vport = vport_ops_list[i]->create(parms);
 232                         if (IS_ERR(vport)) {
 233                                 err = PTR_ERR(vport);
 234                                 goto out;
 235                         }
 236
 237                         hlist_add_head_rcu(&vport->hash_node,
 238                                            hash_bucket(vport_get_name(vport)));
 239                         return vport;
 240                 }
 241         }
 242
 243         err = -EAFNOSUPPORT;
 244
 245 out:
 246         return ERR_PTR(err);
 247 }
 248
 249 /**
 250  *      vport_set_options - modify existing vport device (for kernel callers)
 251  *
 252  * @vport: vport to modify.
 253  * @port: New configuration.
 254  *
 255  * Modifies an existing device with the specified configuration (which is
 256  * dependent on device type).  RTNL lock must be held.
 257  */
 258 int vport_set_options(struct vport *vport, struct nlattr *options)
 259 {
 260         ASSERT_RTNL();
 261
 262         if (!vport->ops->set_options)
 263                 return -EOPNOTSUPP;
 264         return vport->ops->set_options(vport, options);
 265 }
 266
 267 /**
 268  *      vport_del - delete existing vport device
 269  *
 270  * @vport: vport to delete.
 271  *
 272  * Detaches @vport from its datapath and destroys it.  It is possible to fail
 273  * for reasons such as lack of memory.  RTNL lock must be held.
 274  */
 275 void vport_del(struct vport *vport)
 276 {
 277         ASSERT_RTNL();
 278
 279         hlist_del_rcu(&vport->hash_node);
 280
 281         vport->ops->destroy(vport);
 282 }
 283
 284 /**
 285  *      vport_set_addr - set device Ethernet address (for kernel callers)
 286  *
 287  * @vport: vport on which to set Ethernet address.
 288  * @addr: New address.
 289  *
 290  * Sets the Ethernet address of the given device.  Some devices may not support
 291  * setting the Ethernet address, in which case the result will always be
 292  * -EOPNOTSUPP.  RTNL lock must be held.
 293  */
 294 int vport_set_addr(struct vport *vport, const unsigned char *addr)
 295 {
 296         ASSERT_RTNL();
 297
 298         if (!is_valid_ether_addr(addr))
 299                 return -EADDRNOTAVAIL;
 300
 301         if (vport->ops->set_addr)
 302                 return vport->ops->set_addr(vport, addr);
 303         else
 304                 return -EOPNOTSUPP;
 305 }
 306
 307 /**
 308  *      vport_set_stats - sets offset device stats
 309  *
 310  * @vport: vport on which to set stats
 311  * @stats: stats to set
 312  *
 313  * Provides a set of transmit, receive, and error stats to be added as an
 314  * offset to the collect data when stats are retreived.  Some devices may not
 315  * support setting the stats, in which case the result will always be
 316  * -EOPNOTSUPP.
 317  *
 318  * Must be called with RTNL lock.
 319  */
 320 void vport_set_stats(struct vport *vport, struct ovs_vport_stats *stats)
 321 {
 322         ASSERT_RTNL();
 323
 324         spin_lock_bh(&vport->stats_lock);
 325         vport->offset_stats = *stats;
 326         spin_unlock_bh(&vport->stats_lock);
 327 }
 328
 329 /**
 330  *      vport_get_name - retrieve device name
 331  *
 332  * @vport: vport from which to retrieve the name.
 333  *
 334  * Retrieves the name of the given device.  Either RTNL lock or rcu_read_lock
 335  * must be held for the entire duration that the name is in use.
 336  */
 337 const char *vport_get_name(const struct vport *vport)
 338 {
 339         return vport->ops->get_name(vport);
 340 }
 341
 342 /**
 343  *      vport_get_type - retrieve device type
 344  *
 345  * @vport: vport from which to retrieve the type.
 346  *
 347  * Retrieves the type of the given device.
 348  */
 349 enum ovs_vport_type vport_get_type(const struct vport *vport)
 350 {
 351         return vport->ops->type;
 352 }
 353
 354 /**
 355  *      vport_get_addr - retrieve device Ethernet address (for kernel callers)
 356  *
 357  * @vport: vport from which to retrieve the Ethernet address.
 358  *
 359  * Retrieves the Ethernet address of the given device.  Either RTNL lock or
 360  * rcu_read_lock must be held for the entire duration that the Ethernet address
 361  * is in use.
 362  */
 363 const unsigned char *vport_get_addr(const struct vport *vport)
 364 {
 365         return vport->ops->get_addr(vport);
 366 }
 367
 368 /**
 369  *      vport_get_kobj - retrieve associated kobj
 370  *
 371  * @vport: vport from which to retrieve the associated kobj
 372  *
 373  * Retrieves the associated kobj or null if no kobj.  The returned kobj is
 374  * valid for as long as the vport exists.
 375  */
 376 struct kobject *vport_get_kobj(const struct vport *vport)
 377 {
 378         if (vport->ops->get_kobj)
 379                 return vport->ops->get_kobj(vport);
 380         else
 381                 return NULL;
 382 }
 383
 384 /**
 385  *      vport_get_stats - retrieve device stats
 386  *
 387  * @vport: vport from which to retrieve the stats
 388  * @stats: location to store stats
 389  *
 390  * Retrieves transmit, receive, and error stats for the given device.
 391  *
 392  * Must be called with RTNL lock or rcu_read_lock.
 393  */
 394 void vport_get_stats(struct vport *vport, struct ovs_vport_stats *stats)
 395 {
 396         int i;
 397
 398         /* We potentially have 3 sources of stats that need to be
 399          * combined: those we have collected (split into err_stats and
 400          * percpu_stats), offset_stats from set_stats(), and device
 401          * error stats from netdev->get_stats() (for errors that happen
 402          * downstream and therefore aren't reported through our
 403          * vport_record_error() function).
 404          * Stats from first two sources are merged and reported by ovs over
 405          * OVS_VPORT_ATTR_STATS.
 406          * netdev-stats can be directly read over netlink-ioctl.
 407          */
 408
 409         spin_lock_bh(&vport->stats_lock);
 410
 411         *stats = vport->offset_stats;
 412
 413         stats->rx_errors        += vport->err_stats.rx_errors;
 414         stats->tx_errors        += vport->err_stats.tx_errors;
 415         stats->tx_dropped       += vport->err_stats.tx_dropped;
 416         stats->rx_dropped       += vport->err_stats.rx_dropped;
 417
 418         spin_unlock_bh(&vport->stats_lock);
 419
 420         for_each_possible_cpu(i) {
 421                 const struct vport_percpu_stats *percpu_stats;
 422                 struct vport_percpu_stats local_stats;
 423                 unsigned seqcount;
 424
 425                 percpu_stats = per_cpu_ptr(vport->percpu_stats, i);
 426
 427                 do {
 428                         seqcount = read_seqcount_begin(&percpu_stats->seqlock);
 429                         local_stats = *percpu_stats;
 430                 } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount));
 431
 432                 stats->rx_bytes         += local_stats.rx_bytes;
 433                 stats->rx_packets       += local_stats.rx_packets;
 434                 stats->tx_bytes         += local_stats.tx_bytes;
 435                 stats->tx_packets       += local_stats.tx_packets;
 436         }
 437 }
 438
 439 /**
 440  *      vport_get_flags - retrieve device flags
 441  *
 442  * @vport: vport from which to retrieve the flags
 443  *
 444  * Retrieves the flags of the given device.
 445  *
 446  * Must be called with RTNL lock or rcu_read_lock.
 447  */
 448 unsigned vport_get_flags(const struct vport *vport)
 449 {
 450         return vport->ops->get_dev_flags(vport);
 451 }
 452
 453 /**
 454  *      vport_get_flags - check whether device is running
 455  *
 456  * @vport: vport on which to check status.
 457  *
 458  * Checks whether the given device is running.
 459  *
 460  * Must be called with RTNL lock or rcu_read_lock.
 461  */
 462 int vport_is_running(const struct vport *vport)
 463 {
 464         return vport->ops->is_running(vport);
 465 }
 466
 467 /**
 468  *      vport_get_flags - retrieve device operating state
 469  *
 470  * @vport: vport from which to check status
 471  *
 472  * Retrieves the RFC2863 operstate of the given device.
 473  *
 474  * Must be called with RTNL lock or rcu_read_lock.
 475  */
 476 unsigned char vport_get_operstate(const struct vport *vport)
 477 {
 478         return vport->ops->get_operstate(vport);
 479 }
 480
 481 /**
 482  *      vport_get_ifindex - retrieve device system interface index
 483  *
 484  * @vport: vport from which to retrieve index
 485  *
 486  * Retrieves the system interface index of the given device or 0 if
 487  * the device does not have one (in the case of virtual ports).
 488  * Returns a negative index on error.
 489  *
 490  * Must be called with RTNL lock or rcu_read_lock.
 491  */
 492 int vport_get_ifindex(const struct vport *vport)
 493 {
 494         if (vport->ops->get_ifindex)
 495                 return vport->ops->get_ifindex(vport);
 496         else
 497                 return 0;
 498 }
 499
 500 /**
 501  *      vport_get_mtu - retrieve device MTU
 502  *
 503  * @vport: vport from which to retrieve MTU
 504  *
 505  * Retrieves the MTU of the given device.  Returns 0 if @vport does not have an
 506  * MTU (as e.g. some tunnels do not).  Either RTNL lock or rcu_read_lock must
 507  * be held.
 508  */
 509 int vport_get_mtu(const struct vport *vport)
 510 {
 511         if (!vport->ops->get_mtu)
 512                 return 0;
 513         return vport->ops->get_mtu(vport);
 514 }
 515
 516 /**
 517  *      vport_get_options - retrieve device options
 518  *
 519  * @vport: vport from which to retrieve the options.
 520  * @skb: sk_buff where options should be appended.
 521  *
 522  * Retrieves the configuration of the given device, appending an
 523  * %OVS_VPORT_ATTR_OPTIONS attribute that in turn contains nested
 524  * vport-specific attributes to @skb.
 525  *
 526  * Returns 0 if successful, -EMSGSIZE if @skb has insufficient room, or another
 527  * negative error code if a real error occurred.  If an error occurs, @skb is
 528  * left unmodified.
 529  *
 530  * Must be called with RTNL lock or rcu_read_lock.
 531  */
 532 int vport_get_options(const struct vport *vport, struct sk_buff *skb)
 533 {
 534         struct nlattr *nla;
 535
 536         nla = nla_nest_start(skb, OVS_VPORT_ATTR_OPTIONS);
 537         if (!nla)
 538                 return -EMSGSIZE;
 539
 540         if (vport->ops->get_options) {
 541                 int err = vport->ops->get_options(vport, skb);
 542                 if (err) {
 543                         nla_nest_cancel(skb, nla);
 544                         return err;
 545                 }
 546         }
 547
 548         nla_nest_end(skb, nla);
 549         return 0;
 550 }
 551
 552 /**
 553  *      vport_receive - pass up received packet to the datapath for processing
 554  *
 555  * @vport: vport that received the packet
 556  * @skb: skb that was received
 557  *
 558  * Must be called with rcu_read_lock.  The packet cannot be shared and
 559  * skb->data should point to the Ethernet header.  The caller must have already
 560  * called compute_ip_summed() to initialize the checksumming fields.
 561  */
 562 void vport_receive(struct vport *vport, struct sk_buff *skb)
 563 {
 564         struct vport_percpu_stats *stats;
 565
 566         local_bh_disable();
 567         stats = per_cpu_ptr(vport->percpu_stats, smp_processor_id());
 568
 569         write_seqcount_begin(&stats->seqlock);
 570         stats->rx_packets++;
 571         stats->rx_bytes += skb->len;
 572         write_seqcount_end(&stats->seqlock);
 573
 574         local_bh_enable();
 575
 576         if (!(vport->ops->flags & VPORT_F_FLOW))
 577                 OVS_CB(skb)->flow = NULL;
 578
 579         if (!(vport->ops->flags & VPORT_F_TUN_ID))
 580                 OVS_CB(skb)->tun_id = 0;
 581
 582         dp_process_received_packet(vport, skb);
 583 }
 584
 585 /**
 586  *      vport_send - send a packet on a device
 587  *
 588  * @vport: vport on which to send the packet
 589  * @skb: skb to send
 590  *
 591  * Sends the given packet and returns the length of data sent.  Either RTNL
 592  * lock or rcu_read_lock must be held.
 593  */
 594 int vport_send(struct vport *vport, struct sk_buff *skb)
 595 {
 596         struct vport_percpu_stats *stats;
 597         int sent = vport->ops->send(vport, skb);
 598
 599         local_bh_disable();
 600         stats = per_cpu_ptr(vport->percpu_stats, smp_processor_id());
 601
 602         write_seqcount_begin(&stats->seqlock);
 603         stats->tx_packets++;
 604         stats->tx_bytes += sent;
 605         write_seqcount_end(&stats->seqlock);
 606
 607         local_bh_enable();
 608
 609         return sent;
 610 }
 611
 612 /**
 613  *      vport_record_error - indicate device error to generic stats layer
 614  *
 615  * @vport: vport that encountered the error
 616  * @err_type: one of enum vport_err_type types to indicate the error type
 617  *
 618  * If using the vport generic stats layer indicate that an error of the given
 619  * type has occured.
 620  */
 621 void vport_record_error(struct vport *vport, enum vport_err_type err_type)
 622 {
 623         spin_lock_bh(&vport->stats_lock);
 624
 625         switch (err_type) {
 626         case VPORT_E_RX_DROPPED:
 627                 vport->err_stats.rx_dropped++;
 628                 break;
 629
 630         case VPORT_E_RX_ERROR:
 631                 vport->err_stats.rx_errors++;
 632                 break;
 633
 634         case VPORT_E_TX_DROPPED:
 635                 vport->err_stats.tx_dropped++;
 636                 break;
 637
 638         case VPORT_E_TX_ERROR:
 639                 vport->err_stats.tx_errors++;
 640                 break;
 641         };
 642
 643         spin_unlock_bh(&vport->stats_lock);
 644 }