IPv6 NAT: Do not drop DNATed 6to4/6rd packets
[linux-block.git] / net / openvswitch / vport.c
CommitLineData
ccb1352e 1/*
caf2ee14 2 * Copyright (c) 2007-2012 Nicira, Inc.
ccb1352e
JG
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
ccb1352e
JG
19#include <linux/etherdevice.h>
20#include <linux/if.h>
21#include <linux/if_vlan.h>
46df7b81 22#include <linux/jhash.h>
ccb1352e
JG
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>
46df7b81 30#include <net/net_namespace.h>
ccb1352e 31
46df7b81 32#include "datapath.h"
ccb1352e
JG
33#include "vport.h"
34#include "vport-internal_dev.h"
35
36/* List of statically compiled vport implementations. Don't forget to also
37 * add yours to the list at the bottom of vport.h. */
38static const struct vport_ops *vport_ops_list[] = {
39 &ovs_netdev_vport_ops,
40 &ovs_internal_vport_ops,
aa310701 41
fb825a55 42#ifdef CONFIG_OPENVSWITCH_GRE
aa310701
PS
43 &ovs_gre_vport_ops,
44#endif
58264848
PS
45#ifdef CONFIG_OPENVSWITCH_VXLAN
46 &ovs_vxlan_vport_ops,
47#endif
ccb1352e
JG
48};
49
8e4e1713 50/* Protected by RCU read lock for reading, ovs_mutex for writing. */
ccb1352e
JG
51static 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 */
59int 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 */
74void ovs_vport_exit(void)
75{
76 kfree(dev_table);
77}
78
46df7b81 79static struct hlist_head *hash_bucket(struct net *net, const char *name)
ccb1352e 80{
46df7b81 81 unsigned int hash = jhash(name, strlen(name), (unsigned long) net);
ccb1352e
JG
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 *
8e4e1713 90 * Must be called with ovs or RCU read lock.
ccb1352e 91 */
46df7b81 92struct vport *ovs_vport_locate(struct net *net, const char *name)
ccb1352e 93{
46df7b81 94 struct hlist_head *bucket = hash_bucket(net, name);
ccb1352e 95 struct vport *vport;
ccb1352e 96
b67bfe0d 97 hlist_for_each_entry_rcu(vport, bucket, hash_node)
46df7b81
PS
98 if (!strcmp(name, vport->ops->get_name(vport)) &&
99 net_eq(ovs_dp_get_net(vport->dp), net))
ccb1352e
JG
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 * vport_free().
115 */
116struct 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;
15e47304 134 vport->upcall_portid = parms->upcall_portid;
ccb1352e 135 vport->ops = ops;
15eac2a7 136 INIT_HLIST_NODE(&vport->dp_hash_node);
ccb1352e 137
e0f0ecf3 138 vport->percpu_stats = alloc_percpu(struct pcpu_tstats);
f0a98ae8
DC
139 if (!vport->percpu_stats) {
140 kfree(vport);
ccb1352e 141 return ERR_PTR(-ENOMEM);
f0a98ae8 142 }
ccb1352e
JG
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 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 */
159void 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
8e4e1713 171 * device type). ovs_mutex must be held.
ccb1352e
JG
172 */
173struct vport *ovs_vport_add(const struct vport_parms *parms)
174{
175 struct vport *vport;
176 int err = 0;
177 int i;
178
ccb1352e
JG
179 for (i = 0; i < ARRAY_SIZE(vport_ops_list); i++) {
180 if (vport_ops_list[i]->type == parms->type) {
46df7b81
PS
181 struct hlist_head *bucket;
182
ccb1352e
JG
183 vport = vport_ops_list[i]->create(parms);
184 if (IS_ERR(vport)) {
185 err = PTR_ERR(vport);
186 goto out;
187 }
188
46df7b81
PS
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);
ccb1352e
JG
192 return vport;
193 }
194 }
195
196 err = -EAFNOSUPPORT;
197
198out:
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.
2694838d 206 * @options: New configuration.
ccb1352e
JG
207 *
208 * Modifies an existing device with the specified configuration (which is
8e4e1713 209 * dependent on device type). ovs_mutex must be held.
ccb1352e
JG
210 */
211int ovs_vport_set_options(struct vport *vport, struct nlattr *options)
212{
ccb1352e
JG
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
8e4e1713 224 * for reasons such as lack of memory. ovs_mutex must be held.
ccb1352e
JG
225 */
226void ovs_vport_del(struct vport *vport)
227{
8e4e1713 228 ASSERT_OVSL();
ccb1352e
JG
229
230 hlist_del_rcu(&vport->hash_node);
231
232 vport->ops->destroy(vport);
233}
234
235/**
236 * ovs_vport_get_stats - retrieve device stats
237 *
238 * @vport: vport from which to retrieve the stats
239 * @stats: location to store stats
240 *
241 * Retrieves transmit, receive, and error stats for the given device.
242 *
8e4e1713 243 * Must be called with ovs_mutex or rcu_read_lock.
ccb1352e
JG
244 */
245void ovs_vport_get_stats(struct vport *vport, struct ovs_vport_stats *stats)
246{
247 int i;
248
249 memset(stats, 0, sizeof(*stats));
250
251 /* We potentially have 2 sources of stats that need to be combined:
252 * those we have collected (split into err_stats and percpu_stats) from
253 * set_stats() and device error stats from netdev->get_stats() (for
254 * errors that happen downstream and therefore aren't reported through
255 * our vport_record_error() function).
256 * Stats from first source are reported by ovs (OVS_VPORT_ATTR_STATS).
257 * netdev-stats can be directly read over netlink-ioctl.
258 */
259
260 spin_lock_bh(&vport->stats_lock);
261
262 stats->rx_errors = vport->err_stats.rx_errors;
263 stats->tx_errors = vport->err_stats.tx_errors;
264 stats->tx_dropped = vport->err_stats.tx_dropped;
265 stats->rx_dropped = vport->err_stats.rx_dropped;
266
267 spin_unlock_bh(&vport->stats_lock);
268
269 for_each_possible_cpu(i) {
e0f0ecf3
PS
270 const struct pcpu_tstats *percpu_stats;
271 struct pcpu_tstats local_stats;
ccb1352e
JG
272 unsigned int start;
273
274 percpu_stats = per_cpu_ptr(vport->percpu_stats, i);
275
276 do {
e0f0ecf3 277 start = u64_stats_fetch_begin_bh(&percpu_stats->syncp);
ccb1352e 278 local_stats = *percpu_stats;
e0f0ecf3 279 } while (u64_stats_fetch_retry_bh(&percpu_stats->syncp, start));
ccb1352e
JG
280
281 stats->rx_bytes += local_stats.rx_bytes;
282 stats->rx_packets += local_stats.rx_packets;
283 stats->tx_bytes += local_stats.tx_bytes;
284 stats->tx_packets += local_stats.tx_packets;
285 }
286}
287
288/**
289 * ovs_vport_get_options - retrieve device options
290 *
291 * @vport: vport from which to retrieve the options.
292 * @skb: sk_buff where options should be appended.
293 *
294 * Retrieves the configuration of the given device, appending an
295 * %OVS_VPORT_ATTR_OPTIONS attribute that in turn contains nested
296 * vport-specific attributes to @skb.
297 *
298 * Returns 0 if successful, -EMSGSIZE if @skb has insufficient room, or another
299 * negative error code if a real error occurred. If an error occurs, @skb is
300 * left unmodified.
301 *
8e4e1713 302 * Must be called with ovs_mutex or rcu_read_lock.
ccb1352e
JG
303 */
304int ovs_vport_get_options(const struct vport *vport, struct sk_buff *skb)
305{
306 struct nlattr *nla;
5d963352
TG
307 int err;
308
309 if (!vport->ops->get_options)
310 return 0;
ccb1352e
JG
311
312 nla = nla_nest_start(skb, OVS_VPORT_ATTR_OPTIONS);
313 if (!nla)
314 return -EMSGSIZE;
315
5d963352
TG
316 err = vport->ops->get_options(vport, skb);
317 if (err) {
318 nla_nest_cancel(skb, nla);
319 return err;
ccb1352e
JG
320 }
321
322 nla_nest_end(skb, nla);
323 return 0;
324}
325
326/**
327 * ovs_vport_receive - pass up received packet to the datapath for processing
328 *
329 * @vport: vport that received the packet
330 * @skb: skb that was received
2694838d 331 * @tun_key: tunnel (if any) that carried packet
ccb1352e
JG
332 *
333 * Must be called with rcu_read_lock. The packet cannot be shared and
d176ca2a 334 * skb->data should point to the Ethernet header.
ccb1352e 335 */
7d5437c7
PS
336void ovs_vport_receive(struct vport *vport, struct sk_buff *skb,
337 struct ovs_key_ipv4_tunnel *tun_key)
ccb1352e 338{
e0f0ecf3 339 struct pcpu_tstats *stats;
ccb1352e 340
404f2f10 341 stats = this_cpu_ptr(vport->percpu_stats);
e0f0ecf3 342 u64_stats_update_begin(&stats->syncp);
ccb1352e
JG
343 stats->rx_packets++;
344 stats->rx_bytes += skb->len;
e0f0ecf3 345 u64_stats_update_end(&stats->syncp);
ccb1352e 346
7d5437c7 347 OVS_CB(skb)->tun_key = tun_key;
ccb1352e
JG
348 ovs_dp_process_received_packet(vport, skb);
349}
350
351/**
352 * ovs_vport_send - send a packet on a device
353 *
354 * @vport: vport on which to send the packet
355 * @skb: skb to send
356 *
8e4e1713 357 * Sends the given packet and returns the length of data sent. Either ovs
ccb1352e
JG
358 * lock or rcu_read_lock must be held.
359 */
360int ovs_vport_send(struct vport *vport, struct sk_buff *skb)
361{
362 int sent = vport->ops->send(vport, skb);
363
91b7514c 364 if (likely(sent > 0)) {
e0f0ecf3 365 struct pcpu_tstats *stats;
ccb1352e 366
404f2f10 367 stats = this_cpu_ptr(vport->percpu_stats);
ccb1352e 368
e0f0ecf3 369 u64_stats_update_begin(&stats->syncp);
ccb1352e
JG
370 stats->tx_packets++;
371 stats->tx_bytes += sent;
e0f0ecf3 372 u64_stats_update_end(&stats->syncp);
91b7514c
PS
373 } else if (sent < 0) {
374 ovs_vport_record_error(vport, VPORT_E_TX_ERROR);
375 kfree_skb(skb);
376 } else
377 ovs_vport_record_error(vport, VPORT_E_TX_DROPPED);
378
ccb1352e
JG
379 return sent;
380}
381
382/**
383 * ovs_vport_record_error - indicate device error to generic stats layer
384 *
385 * @vport: vport that encountered the error
386 * @err_type: one of enum vport_err_type types to indicate the error type
387 *
388 * If using the vport generic stats layer indicate that an error of the given
af784163 389 * type has occurred.
ccb1352e
JG
390 */
391void ovs_vport_record_error(struct vport *vport, enum vport_err_type err_type)
392{
393 spin_lock(&vport->stats_lock);
394
395 switch (err_type) {
396 case VPORT_E_RX_DROPPED:
397 vport->err_stats.rx_dropped++;
398 break;
399
400 case VPORT_E_RX_ERROR:
401 vport->err_stats.rx_errors++;
402 break;
403
404 case VPORT_E_TX_DROPPED:
405 vport->err_stats.tx_dropped++;
406 break;
407
408 case VPORT_E_TX_ERROR:
409 vport->err_stats.tx_errors++;
410 break;
a2bf91b5 411 }
ccb1352e
JG
412
413 spin_unlock(&vport->stats_lock);
414}
aa310701
PS
415
416static void free_vport_rcu(struct rcu_head *rcu)
417{
418 struct vport *vport = container_of(rcu, struct vport, rcu);
419
420 ovs_vport_free(vport);
421}
422
423void ovs_vport_deferred_free(struct vport *vport)
424{
425 if (!vport)
426 return;
427
428 call_rcu(&vport->rcu, free_vport_rcu);
429}