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