ofproto-dpif: Destroy bundle after moving its last port out.
[cascardo/ovs.git] / datapath / vport-netdev.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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
20
21 #include <linux/if_arp.h>
22 #include <linux/if_bridge.h>
23 #include <linux/if_vlan.h>
24 #include <linux/kernel.h>
25 #include <linux/llc.h>
26 #include <linux/rtnetlink.h>
27 #include <linux/skbuff.h>
28
29 #include <net/llc.h>
30
31 #include "checksum.h"
32 #include "datapath.h"
33 #include "vlan.h"
34 #include "vport-internal_dev.h"
35 #include "vport-netdev.h"
36
37 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,37) && \
38         !defined(HAVE_VLAN_BUG_WORKAROUND)
39 #include <linux/module.h>
40
41 static int vlan_tso __read_mostly;
42 module_param(vlan_tso, int, 0644);
43 MODULE_PARM_DESC(vlan_tso, "Enable TSO for VLAN packets");
44 #else
45 #define vlan_tso true
46 #endif
47
48 #ifdef HAVE_RHEL_OVS_HOOK
49 static atomic_t nr_bridges = ATOMIC_INIT(0);
50
51 extern struct sk_buff *(*openvswitch_handle_frame_hook)(struct sk_buff *skb);
52 #endif
53
54 static void netdev_port_receive(struct vport *vport, struct sk_buff *skb);
55
56 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,39)
57 /* Called with rcu_read_lock and bottom-halves disabled. */
58 static rx_handler_result_t netdev_frame_hook(struct sk_buff **pskb)
59 {
60         struct sk_buff *skb = *pskb;
61         struct vport *vport;
62
63         if (unlikely(skb->pkt_type == PACKET_LOOPBACK))
64                 return RX_HANDLER_PASS;
65
66         vport = ovs_netdev_get_vport(skb->dev);
67
68         netdev_port_receive(vport, skb);
69
70         return RX_HANDLER_CONSUMED;
71 }
72 #elif LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,36) || \
73       defined HAVE_RHEL_OVS_HOOK
74 /* Called with rcu_read_lock and bottom-halves disabled. */
75 static struct sk_buff *netdev_frame_hook(struct sk_buff *skb)
76 {
77         struct vport *vport;
78
79         if (unlikely(skb->pkt_type == PACKET_LOOPBACK))
80                 return skb;
81
82         vport = ovs_netdev_get_vport(skb->dev);
83
84         netdev_port_receive(vport, skb);
85
86         return NULL;
87 }
88 #elif LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,22)
89 /*
90  * Used as br_handle_frame_hook.  (Cannot run bridge at the same time, even on
91  * different set of devices!)
92  */
93 /* Called with rcu_read_lock and bottom-halves disabled. */
94 static struct sk_buff *netdev_frame_hook(struct net_bridge_port *p,
95                                          struct sk_buff *skb)
96 {
97         netdev_port_receive((struct vport *)p, skb);
98         return NULL;
99 }
100 #elif LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)
101 /*
102  * Used as br_handle_frame_hook.  (Cannot run bridge at the same time, even on
103  * different set of devices!)
104  */
105 /* Called with rcu_read_lock and bottom-halves disabled. */
106 static int netdev_frame_hook(struct net_bridge_port *p, struct sk_buff **pskb)
107 {
108         netdev_port_receive((struct vport *)p, *pskb);
109         return 1;
110 }
111 #else
112 #error
113 #endif
114
115 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,36) || \
116     defined HAVE_RHEL_OVS_HOOK
117 static int netdev_init(void) { return 0; }
118 static void netdev_exit(void) { }
119 #else
120 static int netdev_init(void)
121 {
122         /* Hook into callback used by the bridge to intercept packets.
123          * Parasites we are. */
124         br_handle_frame_hook = netdev_frame_hook;
125
126         return 0;
127 }
128
129 static void netdev_exit(void)
130 {
131         br_handle_frame_hook = NULL;
132 }
133 #endif
134
135 static struct vport *netdev_create(const struct vport_parms *parms)
136 {
137         struct vport *vport;
138         struct netdev_vport *netdev_vport;
139         int err;
140
141         vport = ovs_vport_alloc(sizeof(struct netdev_vport),
142                                 &ovs_netdev_vport_ops, parms);
143         if (IS_ERR(vport)) {
144                 err = PTR_ERR(vport);
145                 goto error;
146         }
147
148         netdev_vport = netdev_vport_priv(vport);
149
150         netdev_vport->dev = dev_get_by_name(ovs_dp_get_net(vport->dp), parms->name);
151         if (!netdev_vport->dev) {
152                 err = -ENODEV;
153                 goto error_free_vport;
154         }
155
156         if (netdev_vport->dev->flags & IFF_LOOPBACK ||
157             netdev_vport->dev->type != ARPHRD_ETHER ||
158             ovs_is_internal_dev(netdev_vport->dev)) {
159                 err = -EINVAL;
160                 goto error_put;
161         }
162
163 #ifdef HAVE_RHEL_OVS_HOOK
164         rcu_assign_pointer(netdev_vport->dev->ax25_ptr, vport);
165         atomic_inc(&nr_bridges);
166         rcu_assign_pointer(openvswitch_handle_frame_hook, netdev_frame_hook);
167 #else
168         err = netdev_rx_handler_register(netdev_vport->dev, netdev_frame_hook,
169                                          vport);
170         if (err)
171                 goto error_put;
172 #endif
173
174         dev_set_promiscuity(netdev_vport->dev, 1);
175 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,24)
176         dev_disable_lro(netdev_vport->dev);
177 #endif
178         netdev_vport->dev->priv_flags |= IFF_OVS_DATAPATH;
179
180         return vport;
181
182 error_put:
183         dev_put(netdev_vport->dev);
184 error_free_vport:
185         ovs_vport_free(vport);
186 error:
187         return ERR_PTR(err);
188 }
189
190 static void free_port_rcu(struct rcu_head *rcu)
191 {
192         struct netdev_vport *netdev_vport = container_of(rcu,
193                                         struct netdev_vport, rcu);
194
195 #ifdef HAVE_RHEL_OVS_HOOK
196         rcu_assign_pointer(netdev_vport->dev->ax25_ptr, NULL);
197
198         if (atomic_dec_and_test(&nr_bridges))
199                 rcu_assign_pointer(openvswitch_handle_frame_hook, NULL);
200 #endif
201         dev_put(netdev_vport->dev);
202         ovs_vport_free(vport_from_priv(netdev_vport));
203 }
204
205 static void netdev_destroy(struct vport *vport)
206 {
207         struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
208
209         netdev_vport->dev->priv_flags &= ~IFF_OVS_DATAPATH;
210         netdev_rx_handler_unregister(netdev_vport->dev);
211         dev_set_promiscuity(netdev_vport->dev, -1);
212
213         call_rcu(&netdev_vport->rcu, free_port_rcu);
214 }
215
216 const char *ovs_netdev_get_name(const struct vport *vport)
217 {
218         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
219         return netdev_vport->dev->name;
220 }
221
222 int ovs_netdev_get_ifindex(const struct vport *vport)
223 {
224         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
225         return netdev_vport->dev->ifindex;
226 }
227
228 /* Must be called with rcu_read_lock. */
229 static void netdev_port_receive(struct vport *vport, struct sk_buff *skb)
230 {
231         if (unlikely(!vport))
232                 goto error;
233
234         if (unlikely(skb_warn_if_lro(skb)))
235                 goto error;
236
237         /* Make our own copy of the packet.  Otherwise we will mangle the
238          * packet for anyone who came before us (e.g. tcpdump via AF_PACKET).
239          * (No one comes after us, since we tell handle_bridge() that we took
240          * the packet.) */
241         skb = skb_share_check(skb, GFP_ATOMIC);
242         if (unlikely(!skb))
243                 return;
244
245         skb_push(skb, ETH_HLEN);
246
247         if (unlikely(compute_ip_summed(skb, false)))
248                 goto error;
249
250         vlan_copy_skb_tci(skb);
251
252         ovs_vport_receive(vport, skb);
253         return;
254
255 error:
256         kfree_skb(skb);
257 }
258
259 static unsigned int packet_length(const struct sk_buff *skb)
260 {
261         unsigned int length = skb->len - ETH_HLEN;
262
263         if (skb->protocol == htons(ETH_P_8021Q))
264                 length -= VLAN_HLEN;
265
266         return length;
267 }
268
269 static bool dev_supports_vlan_tx(struct net_device *dev)
270 {
271 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,37)
272         /* Software fallback means every device supports vlan_tci on TX. */
273         return true;
274 #elif defined(HAVE_VLAN_BUG_WORKAROUND)
275         return dev->features & NETIF_F_HW_VLAN_TX;
276 #else
277         /* Assume that the driver is buggy. */
278         return false;
279 #endif
280 }
281
282 static int netdev_send(struct vport *vport, struct sk_buff *skb)
283 {
284         struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
285         int mtu = netdev_vport->dev->mtu;
286         int len;
287
288         if (unlikely(packet_length(skb) > mtu && !skb_is_gso(skb))) {
289                 net_warn_ratelimited("%s: dropped over-mtu packet: %d > %d\n",
290                                      netdev_vport->dev->name,
291                                      packet_length(skb), mtu);
292                 goto error;
293         }
294
295         skb->dev = netdev_vport->dev;
296         forward_ip_summed(skb, true);
297
298         if (vlan_tx_tag_present(skb) && !dev_supports_vlan_tx(skb->dev)) {
299                 int features;
300
301                 features = netif_skb_features(skb);
302
303                 if (!vlan_tso)
304                         features &= ~(NETIF_F_TSO | NETIF_F_TSO6 |
305                                       NETIF_F_UFO | NETIF_F_FSO);
306
307                 if (netif_needs_gso(skb, features)) {
308                         struct sk_buff *nskb;
309
310                         nskb = skb_gso_segment(skb, features);
311                         if (!nskb) {
312                                 if (unlikely(skb_cloned(skb) &&
313                                     pskb_expand_head(skb, 0, 0, GFP_ATOMIC))) {
314                                         kfree_skb(skb);
315                                         return 0;
316                                 }
317
318                                 skb_shinfo(skb)->gso_type &= ~SKB_GSO_DODGY;
319                                 goto tag;
320                         }
321
322                         if (IS_ERR(nskb)) {
323                                 kfree_skb(skb);
324                                 return 0;
325                         }
326                         consume_skb(skb);
327                         skb = nskb;
328
329                         len = 0;
330                         do {
331                                 nskb = skb->next;
332                                 skb->next = NULL;
333
334                                 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
335                                 if (likely(skb)) {
336                                         len += skb->len;
337                                         vlan_set_tci(skb, 0);
338                                         dev_queue_xmit(skb);
339                                 }
340
341                                 skb = nskb;
342                         } while (skb);
343
344                         return len;
345                 }
346
347 tag:
348                 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
349                 if (unlikely(!skb))
350                         return 0;
351                 vlan_set_tci(skb, 0);
352         }
353
354         len = skb->len;
355         dev_queue_xmit(skb);
356
357         return len;
358
359 error:
360         kfree_skb(skb);
361         ovs_vport_record_error(vport, VPORT_E_TX_DROPPED);
362         return 0;
363 }
364
365 /* Returns null if this device is not attached to a datapath. */
366 struct vport *ovs_netdev_get_vport(struct net_device *dev)
367 {
368 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,36) || \
369     defined HAVE_RHEL_OVS_HOOK
370 #if IFF_OVS_DATAPATH != 0
371         if (likely(dev->priv_flags & IFF_OVS_DATAPATH))
372 #else
373         if (likely(rcu_access_pointer(dev->rx_handler) == netdev_frame_hook))
374 #endif
375 #ifdef HAVE_RHEL_OVS_HOOK
376                 return (struct vport *)rcu_dereference_rtnl(dev->ax25_ptr);
377 #else
378                 return (struct vport *)rcu_dereference_rtnl(dev->rx_handler_data);
379 #endif
380         else
381                 return NULL;
382 #else
383         return (struct vport *)rcu_dereference_rtnl(dev->br_port);
384 #endif
385 }
386
387 const struct vport_ops ovs_netdev_vport_ops = {
388         .type           = OVS_VPORT_TYPE_NETDEV,
389         .flags          = VPORT_F_REQUIRED,
390         .init           = netdev_init,
391         .exit           = netdev_exit,
392         .create         = netdev_create,
393         .destroy        = netdev_destroy,
394         .get_name       = ovs_netdev_get_name,
395         .get_ifindex    = ovs_netdev_get_ifindex,
396         .send           = netdev_send,
397 };
398
399 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,36) && \
400     !defined HAVE_RHEL_OVS_HOOK
401 /*
402  * Enforces, mutual exclusion with the Linux bridge module, by declaring and
403  * exporting br_should_route_hook.  Because the bridge module also exports the
404  * same symbol, the module loader will refuse to load both modules at the same
405  * time (e.g. "bridge: exports duplicate symbol br_should_route_hook (owned by
406  * openvswitch)").
407  *
408  * Before Linux 2.6.36, Open vSwitch cannot safely coexist with the Linux
409  * bridge module, so openvswitch uses this macro in those versions.  In
410  * Linux 2.6.36 and later, Open vSwitch can coexist with the bridge module.
411  *
412  * The use of "typeof" here avoids the need to track changes in the type of
413  * br_should_route_hook over various kernel versions.
414  */
415 typeof(br_should_route_hook) br_should_route_hook;
416 EXPORT_SYMBOL(br_should_route_hook);
417 #endif