datapath/vport.c

   1 /*
   2  * Copyright (c) 2007-2012 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/version.h>
  31 #include <net/net_namespace.h>
  32
  33 #include "datapath.h"
  34 #include "vport.h"
  35 #include "vport-internal_dev.h"
  36
  37 /* List of statically compiled vport implementations.  Don't forget to also
  38  * add yours to the list at the bottom of vport.h. */
  39 static const struct vport_ops *vport_ops_list[] = {
  40         &ovs_netdev_vport_ops,
  41         &ovs_internal_vport_ops,
  42         &ovs_gre_vport_ops,
  43         &ovs_gre64_vport_ops,
  44 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,26)
  45         &ovs_vxlan_vport_ops,
  46         &ovs_lisp_vport_ops,
  47 #endif
  48 };
  49
  50 /* Protected by RCU read lock for reading, ovs_mutex for writing. */
  51 static struct hlist_head *dev_table;
  52 #define VPORT_HASH_BUCKETS 1024
  53
  54 /**
  55  *      ovs_vport_init - initialize vport subsystem
  56  *
  57  * Called at module load time to initialize the vport subsystem.
  58  */
  59 int ovs_vport_init(void)
  60 {
  61         dev_table = kzalloc(VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
  62                             GFP_KERNEL);
  63         if (!dev_table)
  64                 return -ENOMEM;
  65
  66         return 0;
  67 }
  68
  69 /**
  70  *      ovs_vport_exit - shutdown vport subsystem
  71  *
  72  * Called at module exit time to shutdown the vport subsystem.
  73  */
  74 void ovs_vport_exit(void)
  75 {
  76         kfree(dev_table);
  77 }
  78
  79 static struct hlist_head *hash_bucket(struct net *net, const char *name)
  80 {
  81         unsigned int hash = jhash(name, strlen(name), (unsigned long) net);
  82         return &dev_table[hash & (VPORT_HASH_BUCKETS - 1)];
  83 }
  84
  85 /**
  86  *      ovs_vport_locate - find a port that has already been created
  87  *
  88  * @name: name of port to find
  89  *
  90  * Must be called with ovs or RCU read lock.
  91  */
  92 struct vport *ovs_vport_locate(struct net *net, const char *name)
  93 {
  94         struct hlist_head *bucket = hash_bucket(net, name);
  95         struct vport *vport;
  96
  97         hlist_for_each_entry_rcu(vport, bucket, hash_node)
  98                 if (!strcmp(name, vport->ops->get_name(vport)) &&
  99                     net_eq(ovs_dp_get_net(vport->dp), net))
 100                         return vport;
 101
 102         return NULL;
 103 }
 104
 105 /**
 106  *      ovs_vport_alloc - allocate and initialize new vport
 107  *
 108  * @priv_size: Size of private data area to allocate.
 109  * @ops: vport device ops
 110  *
 111  * Allocate and initialize a new vport defined by @ops.  The vport will contain
 112  * a private data area of size @priv_size that can be accessed using
 113  * vport_priv().  vports that are no longer needed should be released with
 114  * ovs_vport_free().
 115  */
 116 struct vport *ovs_vport_alloc(int priv_size, const struct vport_ops *ops,
 117                               const struct vport_parms *parms)
 118 {
 119         struct vport *vport;
 120         size_t alloc_size;
 121
 122         alloc_size = sizeof(struct vport);
 123         if (priv_size) {
 124                 alloc_size = ALIGN(alloc_size, VPORT_ALIGN);
 125                 alloc_size += priv_size;
 126         }
 127
 128         vport = kzalloc(alloc_size, GFP_KERNEL);
 129         if (!vport)
 130                 return ERR_PTR(-ENOMEM);
 131
 132         vport->dp = parms->dp;
 133         vport->port_no = parms->port_no;
 134         vport->upcall_portid = parms->upcall_portid;
 135         vport->ops = ops;
 136         INIT_HLIST_NODE(&vport->dp_hash_node);
 137
 138         vport->percpu_stats = alloc_percpu(struct pcpu_tstats);
 139         if (!vport->percpu_stats) {
 140                 kfree(vport);
 141                 return ERR_PTR(-ENOMEM);
 142         }
 143
 144         spin_lock_init(&vport->stats_lock);
 145
 146         return vport;
 147 }
 148
 149 /**
 150  *      ovs_vport_free - uninitialize and free vport
 151  *
 152  * @vport: vport to free
 153  *
 154  * Frees a vport allocated with ovs_vport_alloc() when it is no longer needed.
 155  *
 156  * The caller must ensure that an RCU grace period has passed since the last
 157  * time @vport was in a datapath.
 158  */
 159 void ovs_vport_free(struct vport *vport)
 160 {
 161         free_percpu(vport->percpu_stats);
 162         kfree(vport);
 163 }
 164
 165 /**
 166  *      ovs_vport_add - add vport device (for kernel callers)
 167  *
 168  * @parms: Information about new vport.
 169  *
 170  * Creates a new vport with the specified configuration (which is dependent on
 171  * device type).  ovs_mutex must be held.
 172  */
 173 struct vport *ovs_vport_add(const struct vport_parms *parms)
 174 {
 175         struct vport *vport;
 176         int err = 0;
 177         int i;
 178
 179         for (i = 0; i < ARRAY_SIZE(vport_ops_list); i++) {
 180                 if (vport_ops_list[i]->type == parms->type) {
 181                         struct hlist_head *bucket;
 182
 183                         vport = vport_ops_list[i]->create(parms);
 184                         if (IS_ERR(vport)) {
 185                                 err = PTR_ERR(vport);
 186                                 goto out;
 187                         }
 188
 189                         bucket = hash_bucket(ovs_dp_get_net(vport->dp),
 190                                              vport->ops->get_name(vport));
 191                         hlist_add_head_rcu(&vport->hash_node, bucket);
 192                         return vport;
 193                 }
 194         }
 195
 196         err = -EAFNOSUPPORT;
 197
 198 out:
 199         return ERR_PTR(err);
 200 }
 201
 202 /**
 203  *      ovs_vport_set_options - modify existing vport device (for kernel callers)
 204  *
 205  * @vport: vport to modify.
 206  * @port: New configuration.
 207  *
 208  * Modifies an existing device with the specified configuration (which is
 209  * dependent on device type).  ovs_mutex must be held.
 210  */
 211 int ovs_vport_set_options(struct vport *vport, struct nlattr *options)
 212 {
 213         if (!vport->ops->set_options)
 214                 return -EOPNOTSUPP;
 215         return vport->ops->set_options(vport, options);
 216 }
 217
 218 /**
 219  *      ovs_vport_del - delete existing vport device
 220  *
 221  * @vport: vport to delete.
 222  *
 223  * Detaches @vport from its datapath and destroys it.  It is possible to fail
 224  * for reasons such as lack of memory.  ovs_mutex must be held.
 225  */
 226 void ovs_vport_del(struct vport *vport)
 227 {
 228         ASSERT_OVSL();
 229
 230         hlist_del_rcu(&vport->hash_node);
 231         vport->ops->destroy(vport);
 232 }
 233
 234 /**
 235  *      ovs_vport_set_stats - sets offset device stats
 236  *
 237  * @vport: vport on which to set stats
 238  * @stats: stats to set
 239  *
 240  * Provides a set of transmit, receive, and error stats to be added as an
 241  * offset to the collect data when stats are retreived.  Some devices may not
 242  * support setting the stats, in which case the result will always be
 243  * -EOPNOTSUPP.
 244  *
 245  * Must be called with ovs_mutex.
 246  */
 247 void ovs_vport_set_stats(struct vport *vport, struct ovs_vport_stats *stats)
 248 {
 249         spin_lock_bh(&vport->stats_lock);
 250         vport->offset_stats = *stats;
 251         spin_unlock_bh(&vport->stats_lock);
 252 }
 253
 254 /**
 255  *      ovs_vport_get_stats - retrieve device stats
 256  *
 257  * @vport: vport from which to retrieve the stats
 258  * @stats: location to store stats
 259  *
 260  * Retrieves transmit, receive, and error stats for the given device.
 261  *
 262  * Must be called with ovs_mutex or rcu_read_lock.
 263  */
 264 void ovs_vport_get_stats(struct vport *vport, struct ovs_vport_stats *stats)
 265 {
 266         int i;
 267
 268         /* We potentially have 3 sources of stats that need to be
 269          * combined: those we have collected (split into err_stats and
 270          * percpu_stats), offset_stats from set_stats(), and device
 271          * error stats from netdev->get_stats() (for errors that happen
 272          * downstream and therefore aren't reported through our
 273          * vport_record_error() function).
 274          * Stats from first two sources are merged and reported by ovs over
 275          * OVS_VPORT_ATTR_STATS.
 276          * netdev-stats can be directly read over netlink-ioctl.
 277          */
 278
 279         spin_lock_bh(&vport->stats_lock);
 280
 281         *stats = vport->offset_stats;
 282
 283         stats->rx_errors        += vport->err_stats.rx_errors;
 284         stats->tx_errors        += vport->err_stats.tx_errors;
 285         stats->tx_dropped       += vport->err_stats.tx_dropped;
 286         stats->rx_dropped       += vport->err_stats.rx_dropped;
 287
 288         spin_unlock_bh(&vport->stats_lock);
 289
 290         for_each_possible_cpu(i) {
 291                 const struct pcpu_tstats *percpu_stats;
 292                 struct pcpu_tstats local_stats;
 293                 unsigned int start;
 294
 295                 percpu_stats = per_cpu_ptr(vport->percpu_stats, i);
 296
 297                 do {
 298                         start = u64_stats_fetch_begin_bh(&percpu_stats->syncp);
 299                         local_stats = *percpu_stats;
 300                 } while (u64_stats_fetch_retry_bh(&percpu_stats->syncp, start));
 301
 302                 stats->rx_bytes         += local_stats.rx_bytes;
 303                 stats->rx_packets       += local_stats.rx_packets;
 304                 stats->tx_bytes         += local_stats.tx_bytes;
 305                 stats->tx_packets       += local_stats.tx_packets;
 306         }
 307 }
 308
 309 /**
 310  *      ovs_vport_get_options - retrieve device options
 311  *
 312  * @vport: vport from which to retrieve the options.
 313  * @skb: sk_buff where options should be appended.
 314  *
 315  * Retrieves the configuration of the given device, appending an
 316  * %OVS_VPORT_ATTR_OPTIONS attribute that in turn contains nested
 317  * vport-specific attributes to @skb.
 318  *
 319  * Returns 0 if successful, -EMSGSIZE if @skb has insufficient room, or another
 320  * negative error code if a real error occurred.  If an error occurs, @skb is
 321  * left unmodified.
 322  *
 323  * Must be called with ovs_mutex or rcu_read_lock.
 324  */
 325 int ovs_vport_get_options(const struct vport *vport, struct sk_buff *skb)
 326 {
 327         struct nlattr *nla;
 328         int err;
 329
 330         if (!vport->ops->get_options)
 331                 return 0;
 332
 333         nla = nla_nest_start(skb, OVS_VPORT_ATTR_OPTIONS);
 334         if (!nla)
 335                 return -EMSGSIZE;
 336
 337         err = vport->ops->get_options(vport, skb);
 338         if (err) {
 339                 nla_nest_cancel(skb, nla);
 340                 return err;
 341         }
 342
 343         nla_nest_end(skb, nla);
 344         return 0;
 345 }
 346
 347 /**
 348  *      ovs_vport_receive - pass up received packet to the datapath for processing
 349  *
 350  * @vport: vport that received the packet
 351  * @skb: skb that was received
 352  *
 353  * Must be called with rcu_read_lock.  The packet cannot be shared and
 354  * skb->data should point to the Ethernet header.  The caller must have already
 355  * called compute_ip_summed() to initialize the checksumming fields.
 356  */
 357 void ovs_vport_receive(struct vport *vport, struct sk_buff *skb,
 358                        struct ovs_key_ipv4_tunnel *tun_key)
 359 {
 360         struct pcpu_tstats *stats;
 361
 362         stats = this_cpu_ptr(vport->percpu_stats);
 363         u64_stats_update_begin(&stats->syncp);
 364         stats->rx_packets++;
 365         stats->rx_bytes += skb->len;
 366         u64_stats_update_end(&stats->syncp);
 367
 368         OVS_CB(skb)->tun_key = tun_key;
 369         ovs_dp_process_received_packet(vport, skb);
 370 }
 371
 372 /**
 373  *      ovs_vport_send - send a packet on a device
 374  *
 375  * @vport: vport on which to send the packet
 376  * @skb: skb to send
 377  *
 378  * Sends the given packet and returns the length of data sent.  Either ovs
 379  * lock or rcu_read_lock must be held.
 380  */
 381 int ovs_vport_send(struct vport *vport, struct sk_buff *skb)
 382 {
 383         int sent = vport->ops->send(vport, skb);
 384
 385         if (likely(sent > 0)) {
 386                 struct pcpu_tstats *stats;
 387
 388                 stats = this_cpu_ptr(vport->percpu_stats);
 389
 390                 u64_stats_update_begin(&stats->syncp);
 391                 stats->tx_packets++;
 392                 stats->tx_bytes += sent;
 393                 u64_stats_update_end(&stats->syncp);
 394         } else if (sent < 0) {
 395                 ovs_vport_record_error(vport, VPORT_E_TX_ERROR);
 396                 kfree_skb(skb);
 397         } else
 398                 ovs_vport_record_error(vport, VPORT_E_TX_DROPPED);
 399
 400         return sent;
 401 }
 402
 403 /**
 404  *      ovs_vport_record_error - indicate device error to generic stats layer
 405  *
 406  * @vport: vport that encountered the error
 407  * @err_type: one of enum vport_err_type types to indicate the error type
 408  *
 409  * If using the vport generic stats layer indicate that an error of the given
 410  * type has occurred.
 411  */
 412 void ovs_vport_record_error(struct vport *vport, enum vport_err_type err_type)
 413 {
 414         spin_lock(&vport->stats_lock);
 415
 416         switch (err_type) {
 417         case VPORT_E_RX_DROPPED:
 418                 vport->err_stats.rx_dropped++;
 419                 break;
 420
 421         case VPORT_E_RX_ERROR:
 422                 vport->err_stats.rx_errors++;
 423                 break;
 424
 425         case VPORT_E_TX_DROPPED:
 426                 vport->err_stats.tx_dropped++;
 427                 break;
 428
 429         case VPORT_E_TX_ERROR:
 430                 vport->err_stats.tx_errors++;
 431                 break;
 432         }
 433
 434         spin_unlock(&vport->stats_lock);
 435 }
 436
 437 static void free_vport_rcu(struct rcu_head *rcu)
 438 {
 439         struct vport *vport = container_of(rcu, struct vport, rcu);
 440
 441         ovs_vport_free(vport);
 442 }
 443
 444 void ovs_vport_deferred_free(struct vport *vport)
 445 {
 446         if (!vport)
 447                 return;
 448
 449         call_rcu(&vport->rcu, free_vport_rcu);
 450 }