net/openvswitch/vport.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (c) 2007-2014 Nicira, Inc.
   4  */
   5
   6 #include <linux/etherdevice.h>
   7 #include <linux/if.h>
   8 #include <linux/if_vlan.h>
   9 #include <linux/jhash.h>
  10 #include <linux/kernel.h>
  11 #include <linux/list.h>
  12 #include <linux/mutex.h>
  13 #include <linux/percpu.h>
  14 #include <linux/rcupdate.h>
  15 #include <linux/rtnetlink.h>
  16 #include <linux/compat.h>
  17 #include <net/net_namespace.h>
  18 #include <linux/module.h>
  19
  20 #include "datapath.h"
  21 #include "vport.h"
  22 #include "vport-internal_dev.h"
  23
  24 static LIST_HEAD(vport_ops_list);
  25
  26 /* Protected by RCU read lock for reading, ovs_mutex for writing. */
  27 static struct hlist_head *dev_table;
  28 #define VPORT_HASH_BUCKETS 1024
  29
  30 /**
  31  *      ovs_vport_init - initialize vport subsystem
  32  *
  33  * Called at module load time to initialize the vport subsystem.
  34  */
  35 int ovs_vport_init(void)
  36 {
  37         dev_table = kcalloc(VPORT_HASH_BUCKETS, sizeof(struct hlist_head),
  38                             GFP_KERNEL);
  39         if (!dev_table)
  40                 return -ENOMEM;
  41
  42         return 0;
  43 }
  44
  45 /**
  46  *      ovs_vport_exit - shutdown vport subsystem
  47  *
  48  * Called at module exit time to shutdown the vport subsystem.
  49  */
  50 void ovs_vport_exit(void)
  51 {
  52         kfree(dev_table);
  53 }
  54
  55 static struct hlist_head *hash_bucket(const struct net *net, const char *name)
  56 {
  57         unsigned int hash = jhash(name, strlen(name), (unsigned long) net);
  58         return &dev_table[hash & (VPORT_HASH_BUCKETS - 1)];
  59 }
  60
  61 int __ovs_vport_ops_register(struct vport_ops *ops)
  62 {
  63         int err = -EEXIST;
  64         struct vport_ops *o;
  65
  66         ovs_lock();
  67         list_for_each_entry(o, &vport_ops_list, list)
  68                 if (ops->type == o->type)
  69                         goto errout;
  70
  71         list_add_tail(&ops->list, &vport_ops_list);
  72         err = 0;
  73 errout:
  74         ovs_unlock();
  75         return err;
  76 }
  77 EXPORT_SYMBOL_GPL(__ovs_vport_ops_register);
  78
  79 void ovs_vport_ops_unregister(struct vport_ops *ops)
  80 {
  81         ovs_lock();
  82         list_del(&ops->list);
  83         ovs_unlock();
  84 }
  85 EXPORT_SYMBOL_GPL(ovs_vport_ops_unregister);
  86
  87 /**
  88  *      ovs_vport_locate - find a port that has already been created
  89  *
  90  * @name: name of port to find
  91  *
  92  * Must be called with ovs or RCU read lock.
  93  */
  94 struct vport *ovs_vport_locate(const struct net *net, const char *name)
  95 {
  96         struct hlist_head *bucket = hash_bucket(net, name);
  97         struct vport *vport;
  98
  99         hlist_for_each_entry_rcu(vport, bucket, hash_node)
 100                 if (!strcmp(name, ovs_vport_name(vport)) &&
 101                     net_eq(ovs_dp_get_net(vport->dp), net))
 102                         return vport;
 103
 104         return NULL;
 105 }
 106
 107 /**
 108  *      ovs_vport_alloc - allocate and initialize new vport
 109  *
 110  * @priv_size: Size of private data area to allocate.
 111  * @ops: vport device ops
 112  *
 113  * Allocate and initialize a new vport defined by @ops.  The vport will contain
 114  * a private data area of size @priv_size that can be accessed using
 115  * vport_priv().  vports that are no longer needed should be released with
 116  * vport_free().
 117  */
 118 struct vport *ovs_vport_alloc(int priv_size, const struct vport_ops *ops,
 119                           const struct vport_parms *parms)
 120 {
 121         struct vport *vport;
 122         size_t alloc_size;
 123
 124         alloc_size = sizeof(struct vport);
 125         if (priv_size) {
 126                 alloc_size = ALIGN(alloc_size, VPORT_ALIGN);
 127                 alloc_size += priv_size;
 128         }
 129
 130         vport = kzalloc(alloc_size, GFP_KERNEL);
 131         if (!vport)
 132                 return ERR_PTR(-ENOMEM);
 133
 134         vport->dp = parms->dp;
 135         vport->port_no = parms->port_no;
 136         vport->ops = ops;
 137         INIT_HLIST_NODE(&vport->dp_hash_node);
 138
 139         if (ovs_vport_set_upcall_portids(vport, parms->upcall_portids)) {
 140                 kfree(vport);
 141                 return ERR_PTR(-EINVAL);
 142         }
 143
 144         return vport;
 145 }
 146 EXPORT_SYMBOL_GPL(ovs_vport_alloc);
 147
 148 /**
 149  *      ovs_vport_free - uninitialize and free vport
 150  *
 151  * @vport: vport to free
 152  *
 153  * Frees a vport allocated with vport_alloc() when it is no longer needed.
 154  *
 155  * The caller must ensure that an RCU grace period has passed since the last
 156  * time @vport was in a datapath.
 157  */
 158 void ovs_vport_free(struct vport *vport)
 159 {
 160         /* vport is freed from RCU callback or error path, Therefore
 161          * it is safe to use raw dereference.
 162          */
 163         kfree(rcu_dereference_raw(vport->upcall_portids));
 164         kfree(vport);
 165 }
 166 EXPORT_SYMBOL_GPL(ovs_vport_free);
 167
 168 static struct vport_ops *ovs_vport_lookup(const struct vport_parms *parms)
 169 {
 170         struct vport_ops *ops;
 171
 172         list_for_each_entry(ops, &vport_ops_list, list)
 173                 if (ops->type == parms->type)
 174                         return ops;
 175
 176         return NULL;
 177 }
 178
 179 /**
 180  *      ovs_vport_add - add vport device (for kernel callers)
 181  *
 182  * @parms: Information about new vport.
 183  *
 184  * Creates a new vport with the specified configuration (which is dependent on
 185  * device type).  ovs_mutex must be held.
 186  */
 187 struct vport *ovs_vport_add(const struct vport_parms *parms)
 188 {
 189         struct vport_ops *ops;
 190         struct vport *vport;
 191
 192         ops = ovs_vport_lookup(parms);
 193         if (ops) {
 194                 struct hlist_head *bucket;
 195
 196                 if (!try_module_get(ops->owner))
 197                         return ERR_PTR(-EAFNOSUPPORT);
 198
 199                 vport = ops->create(parms);
 200                 if (IS_ERR(vport)) {
 201                         module_put(ops->owner);
 202                         return vport;
 203                 }
 204
 205                 bucket = hash_bucket(ovs_dp_get_net(vport->dp),
 206                                      ovs_vport_name(vport));
 207                 hlist_add_head_rcu(&vport->hash_node, bucket);
 208                 return vport;
 209         }
 210
 211         /* Unlock to attempt module load and return -EAGAIN if load
 212          * was successful as we need to restart the port addition
 213          * workflow.
 214          */
 215         ovs_unlock();
 216         request_module("vport-type-%d", parms->type);
 217         ovs_lock();
 218
 219         if (!ovs_vport_lookup(parms))
 220                 return ERR_PTR(-EAFNOSUPPORT);
 221         else
 222                 return ERR_PTR(-EAGAIN);
 223 }
 224
 225 /**
 226  *      ovs_vport_set_options - modify existing vport device (for kernel callers)
 227  *
 228  * @vport: vport to modify.
 229  * @options: New configuration.
 230  *
 231  * Modifies an existing device with the specified configuration (which is
 232  * dependent on device type).  ovs_mutex must be held.
 233  */
 234 int ovs_vport_set_options(struct vport *vport, struct nlattr *options)
 235 {
 236         if (!vport->ops->set_options)
 237                 return -EOPNOTSUPP;
 238         return vport->ops->set_options(vport, options);
 239 }
 240
 241 /**
 242  *      ovs_vport_del - delete existing vport device
 243  *
 244  * @vport: vport to delete.
 245  *
 246  * Detaches @vport from its datapath and destroys it.  ovs_mutex must
 247  * be held.
 248  */
 249 void ovs_vport_del(struct vport *vport)
 250 {
 251         ASSERT_OVSL();
 252
 253         hlist_del_rcu(&vport->hash_node);
 254         module_put(vport->ops->owner);
 255         vport->ops->destroy(vport);
 256 }
 257
 258 /**
 259  *      ovs_vport_get_stats - retrieve device stats
 260  *
 261  * @vport: vport from which to retrieve the stats
 262  * @stats: location to store stats
 263  *
 264  * Retrieves transmit, receive, and error stats for the given device.
 265  *
 266  * Must be called with ovs_mutex or rcu_read_lock.
 267  */
 268 void ovs_vport_get_stats(struct vport *vport, struct ovs_vport_stats *stats)
 269 {
 270         const struct rtnl_link_stats64 *dev_stats;
 271         struct rtnl_link_stats64 temp;
 272
 273         dev_stats = dev_get_stats(vport->dev, &temp);
 274         stats->rx_errors  = dev_stats->rx_errors;
 275         stats->tx_errors  = dev_stats->tx_errors;
 276         stats->tx_dropped = dev_stats->tx_dropped;
 277         stats->rx_dropped = dev_stats->rx_dropped;
 278
 279         stats->rx_bytes   = dev_stats->rx_bytes;
 280         stats->rx_packets = dev_stats->rx_packets;
 281         stats->tx_bytes   = dev_stats->tx_bytes;
 282         stats->tx_packets = dev_stats->tx_packets;
 283 }
 284
 285 /**
 286  *      ovs_vport_get_options - retrieve device options
 287  *
 288  * @vport: vport from which to retrieve the options.
 289  * @skb: sk_buff where options should be appended.
 290  *
 291  * Retrieves the configuration of the given device, appending an
 292  * %OVS_VPORT_ATTR_OPTIONS attribute that in turn contains nested
 293  * vport-specific attributes to @skb.
 294  *
 295  * Returns 0 if successful, -EMSGSIZE if @skb has insufficient room, or another
 296  * negative error code if a real error occurred.  If an error occurs, @skb is
 297  * left unmodified.
 298  *
 299  * Must be called with ovs_mutex or rcu_read_lock.
 300  */
 301 int ovs_vport_get_options(const struct vport *vport, struct sk_buff *skb)
 302 {
 303         struct nlattr *nla;
 304         int err;
 305
 306         if (!vport->ops->get_options)
 307                 return 0;
 308
 309         nla = nla_nest_start_noflag(skb, OVS_VPORT_ATTR_OPTIONS);
 310         if (!nla)
 311                 return -EMSGSIZE;
 312
 313         err = vport->ops->get_options(vport, skb);
 314         if (err) {
 315                 nla_nest_cancel(skb, nla);
 316                 return err;
 317         }
 318
 319         nla_nest_end(skb, nla);
 320         return 0;
 321 }
 322
 323 /**
 324  *      ovs_vport_set_upcall_portids - set upcall portids of @vport.
 325  *
 326  * @vport: vport to modify.
 327  * @ids: new configuration, an array of port ids.
 328  *
 329  * Sets the vport's upcall_portids to @ids.
 330  *
 331  * Returns 0 if successful, -EINVAL if @ids is zero length or cannot be parsed
 332  * as an array of U32.
 333  *
 334  * Must be called with ovs_mutex.
 335  */
 336 int ovs_vport_set_upcall_portids(struct vport *vport, const struct nlattr *ids)
 337 {
 338         struct vport_portids *old, *vport_portids;
 339
 340         if (!nla_len(ids) || nla_len(ids) % sizeof(u32))
 341                 return -EINVAL;
 342
 343         old = ovsl_dereference(vport->upcall_portids);
 344
 345         vport_portids = kmalloc(sizeof(*vport_portids) + nla_len(ids),
 346                                 GFP_KERNEL);
 347         if (!vport_portids)
 348                 return -ENOMEM;
 349
 350         vport_portids->n_ids = nla_len(ids) / sizeof(u32);
 351         vport_portids->rn_ids = reciprocal_value(vport_portids->n_ids);
 352         nla_memcpy(vport_portids->ids, ids, nla_len(ids));
 353
 354         rcu_assign_pointer(vport->upcall_portids, vport_portids);
 355
 356         if (old)
 357                 kfree_rcu(old, rcu);
 358         return 0;
 359 }
 360
 361 /**
 362  *      ovs_vport_get_upcall_portids - get the upcall_portids of @vport.
 363  *
 364  * @vport: vport from which to retrieve the portids.
 365  * @skb: sk_buff where portids should be appended.
 366  *
 367  * Retrieves the configuration of the given vport, appending the
 368  * %OVS_VPORT_ATTR_UPCALL_PID attribute which is the array of upcall
 369  * portids to @skb.
 370  *
 371  * Returns 0 if successful, -EMSGSIZE if @skb has insufficient room.
 372  * If an error occurs, @skb is left unmodified.  Must be called with
 373  * ovs_mutex or rcu_read_lock.
 374  */
 375 int ovs_vport_get_upcall_portids(const struct vport *vport,
 376                                  struct sk_buff *skb)
 377 {
 378         struct vport_portids *ids;
 379
 380         ids = rcu_dereference_ovsl(vport->upcall_portids);
 381
 382         if (vport->dp->user_features & OVS_DP_F_VPORT_PIDS)
 383                 return nla_put(skb, OVS_VPORT_ATTR_UPCALL_PID,
 384                                ids->n_ids * sizeof(u32), (void *)ids->ids);
 385         else
 386                 return nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, ids->ids[0]);
 387 }
 388
 389 /**
 390  *      ovs_vport_find_upcall_portid - find the upcall portid to send upcall.
 391  *
 392  * @vport: vport from which the missed packet is received.
 393  * @skb: skb that the missed packet was received.
 394  *
 395  * Uses the skb_get_hash() to select the upcall portid to send the
 396  * upcall.
 397  *
 398  * Returns the portid of the target socket.  Must be called with rcu_read_lock.
 399  */
 400 u32 ovs_vport_find_upcall_portid(const struct vport *vport, struct sk_buff *skb)
 401 {
 402         struct vport_portids *ids;
 403         u32 ids_index;
 404         u32 hash;
 405
 406         ids = rcu_dereference(vport->upcall_portids);
 407
 408         if (ids->n_ids == 1 && ids->ids[0] == 0)
 409                 return 0;
 410
 411         hash = skb_get_hash(skb);
 412         ids_index = hash - ids->n_ids * reciprocal_divide(hash, ids->rn_ids);
 413         return ids->ids[ids_index];
 414 }
 415
 416 /**
 417  *      ovs_vport_receive - pass up received packet to the datapath for processing
 418  *
 419  * @vport: vport that received the packet
 420  * @skb: skb that was received
 421  * @tun_key: tunnel (if any) that carried packet
 422  *
 423  * Must be called with rcu_read_lock.  The packet cannot be shared and
 424  * skb->data should point to the Ethernet header.
 425  */
 426 int ovs_vport_receive(struct vport *vport, struct sk_buff *skb,
 427                       const struct ip_tunnel_info *tun_info)
 428 {
 429         struct sw_flow_key key;
 430         int error;
 431
 432         OVS_CB(skb)->input_vport = vport;
 433         OVS_CB(skb)->mru = 0;
 434         OVS_CB(skb)->cutlen = 0;
 435         if (unlikely(dev_net(skb->dev) != ovs_dp_get_net(vport->dp))) {
 436                 u32 mark;
 437
 438                 mark = skb->mark;
 439                 skb_scrub_packet(skb, true);
 440                 skb->mark = mark;
 441                 tun_info = NULL;
 442         }
 443
 444         /* Extract flow from 'skb' into 'key'. */
 445         error = ovs_flow_key_extract(tun_info, skb, &key);
 446         if (unlikely(error)) {
 447                 kfree_skb(skb);
 448                 return error;
 449         }
 450         ovs_dp_process_packet(skb, &key);
 451         return 0;
 452 }
 453
 454 static int packet_length(const struct sk_buff *skb,
 455                          struct net_device *dev)
 456 {
 457         int length = skb->len - dev->hard_header_len;
 458
 459         if (!skb_vlan_tag_present(skb) &&
 460             eth_type_vlan(skb->protocol))
 461                 length -= VLAN_HLEN;
 462
 463         /* Don't subtract for multiple VLAN tags. Most (all?) drivers allow
 464          * (ETH_LEN + VLAN_HLEN) in addition to the mtu value, but almost none
 465          * account for 802.1ad. e.g. is_skb_forwardable().
 466          */
 467
 468         return length > 0 ? length : 0;
 469 }
 470
 471 void ovs_vport_send(struct vport *vport, struct sk_buff *skb, u8 mac_proto)
 472 {
 473         int mtu = vport->dev->mtu;
 474
 475         switch (vport->dev->type) {
 476         case ARPHRD_NONE:
 477                 if (mac_proto == MAC_PROTO_ETHERNET) {
 478                         skb_reset_network_header(skb);
 479                         skb_reset_mac_len(skb);
 480                         skb->protocol = htons(ETH_P_TEB);
 481                 } else if (mac_proto != MAC_PROTO_NONE) {
 482                         WARN_ON_ONCE(1);
 483                         goto drop;
 484                 }
 485                 break;
 486         case ARPHRD_ETHER:
 487                 if (mac_proto != MAC_PROTO_ETHERNET)
 488                         goto drop;
 489                 break;
 490         default:
 491                 goto drop;
 492         }
 493
 494         if (unlikely(packet_length(skb, vport->dev) > mtu &&
 495                      !skb_is_gso(skb))) {
 496                 net_warn_ratelimited("%s: dropped over-mtu packet: %d > %d\n",
 497                                      vport->dev->name,
 498                                      packet_length(skb, vport->dev), mtu);
 499                 vport->dev->stats.tx_errors++;
 500                 goto drop;
 501         }
 502
 503         skb->dev = vport->dev;
 504         vport->ops->send(skb);
 505         return;
 506
 507 drop:
 508         kfree_skb(skb);
 509 }