datapath: Use openvswitch_handle_frame hook in >=RHEL6.4 to live side by side with...
[sliver-openvswitch.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         rtnl_lock();
164 #ifdef HAVE_RHEL_OVS_HOOK
165         rcu_assign_pointer(netdev_vport->dev->ax25_ptr, vport);
166         atomic_inc(&nr_bridges);
167         rcu_assign_pointer(openvswitch_handle_frame_hook, netdev_frame_hook);
168 #else
169         err = netdev_rx_handler_register(netdev_vport->dev, netdev_frame_hook,
170                                          vport);
171         if (err)
172                 goto error_unlock;
173 #endif
174
175         dev_set_promiscuity(netdev_vport->dev, 1);
176 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,24)
177         dev_disable_lro(netdev_vport->dev);
178 #endif
179         netdev_vport->dev->priv_flags |= IFF_OVS_DATAPATH;
180         rtnl_unlock();
181
182         return vport;
183
184 #ifndef HAVE_RHEL_OVS_HOOK
185 error_unlock:
186 #endif
187         rtnl_unlock();
188 error_put:
189         dev_put(netdev_vport->dev);
190 error_free_vport:
191         ovs_vport_free(vport);
192 error:
193         return ERR_PTR(err);
194 }
195
196 static void free_port_rcu(struct rcu_head *rcu)
197 {
198         struct netdev_vport *netdev_vport = container_of(rcu,
199                                         struct netdev_vport, rcu);
200
201 #ifdef HAVE_RHEL_OVS_HOOK
202         rcu_assign_pointer(netdev_vport->dev->ax25_ptr, NULL);
203
204         if (atomic_dec_and_test(&nr_bridges))
205                 rcu_assign_pointer(openvswitch_handle_frame_hook, NULL);
206 #endif
207         dev_put(netdev_vport->dev);
208         ovs_vport_free(vport_from_priv(netdev_vport));
209 }
210
211 static void netdev_destroy(struct vport *vport)
212 {
213         struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
214
215         rtnl_lock();
216         netdev_vport->dev->priv_flags &= ~IFF_OVS_DATAPATH;
217         netdev_rx_handler_unregister(netdev_vport->dev);
218         dev_set_promiscuity(netdev_vport->dev, -1);
219         rtnl_unlock();
220
221         call_rcu(&netdev_vport->rcu, free_port_rcu);
222 }
223
224 const char *ovs_netdev_get_name(const struct vport *vport)
225 {
226         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
227         return netdev_vport->dev->name;
228 }
229
230 int ovs_netdev_get_ifindex(const struct vport *vport)
231 {
232         const struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
233         return netdev_vport->dev->ifindex;
234 }
235
236 /* Must be called with rcu_read_lock. */
237 static void netdev_port_receive(struct vport *vport, struct sk_buff *skb)
238 {
239         if (unlikely(!vport))
240                 goto error;
241
242         if (unlikely(skb_warn_if_lro(skb)))
243                 goto error;
244
245         /* Make our own copy of the packet.  Otherwise we will mangle the
246          * packet for anyone who came before us (e.g. tcpdump via AF_PACKET).
247          * (No one comes after us, since we tell handle_bridge() that we took
248          * the packet.) */
249         skb = skb_share_check(skb, GFP_ATOMIC);
250         if (unlikely(!skb))
251                 return;
252
253         skb_push(skb, ETH_HLEN);
254
255         if (unlikely(compute_ip_summed(skb, false)))
256                 goto error;
257
258         vlan_copy_skb_tci(skb);
259
260         ovs_vport_receive(vport, skb);
261         return;
262
263 error:
264         kfree_skb(skb);
265 }
266
267 static unsigned int packet_length(const struct sk_buff *skb)
268 {
269         unsigned int length = skb->len - ETH_HLEN;
270
271         if (skb->protocol == htons(ETH_P_8021Q))
272                 length -= VLAN_HLEN;
273
274         return length;
275 }
276
277 static bool dev_supports_vlan_tx(struct net_device *dev)
278 {
279 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,37)
280         /* Software fallback means every device supports vlan_tci on TX. */
281         return true;
282 #elif defined(HAVE_VLAN_BUG_WORKAROUND)
283         return dev->features & NETIF_F_HW_VLAN_TX;
284 #else
285         /* Assume that the driver is buggy. */
286         return false;
287 #endif
288 }
289
290 static int netdev_send(struct vport *vport, struct sk_buff *skb)
291 {
292         struct netdev_vport *netdev_vport = netdev_vport_priv(vport);
293         int mtu = netdev_vport->dev->mtu;
294         int len;
295
296         if (unlikely(packet_length(skb) > mtu && !skb_is_gso(skb))) {
297                 net_warn_ratelimited("%s: dropped over-mtu packet: %d > %d\n",
298                                      netdev_vport->dev->name,
299                                      packet_length(skb), mtu);
300                 goto error;
301         }
302
303         skb->dev = netdev_vport->dev;
304         forward_ip_summed(skb, true);
305
306         if (vlan_tx_tag_present(skb) && !dev_supports_vlan_tx(skb->dev)) {
307                 int features;
308
309                 features = netif_skb_features(skb);
310
311                 if (!vlan_tso)
312                         features &= ~(NETIF_F_TSO | NETIF_F_TSO6 |
313                                       NETIF_F_UFO | NETIF_F_FSO);
314
315                 if (netif_needs_gso(skb, features)) {
316                         struct sk_buff *nskb;
317
318                         nskb = skb_gso_segment(skb, features);
319                         if (!nskb) {
320                                 if (unlikely(skb_cloned(skb) &&
321                                     pskb_expand_head(skb, 0, 0, GFP_ATOMIC))) {
322                                         kfree_skb(skb);
323                                         return 0;
324                                 }
325
326                                 skb_shinfo(skb)->gso_type &= ~SKB_GSO_DODGY;
327                                 goto tag;
328                         }
329
330                         if (IS_ERR(nskb)) {
331                                 kfree_skb(skb);
332                                 return 0;
333                         }
334                         consume_skb(skb);
335                         skb = nskb;
336
337                         len = 0;
338                         do {
339                                 nskb = skb->next;
340                                 skb->next = NULL;
341
342                                 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
343                                 if (likely(skb)) {
344                                         len += skb->len;
345                                         vlan_set_tci(skb, 0);
346                                         dev_queue_xmit(skb);
347                                 }
348
349                                 skb = nskb;
350                         } while (skb);
351
352                         return len;
353                 }
354
355 tag:
356                 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
357                 if (unlikely(!skb))
358                         return 0;
359                 vlan_set_tci(skb, 0);
360         }
361
362         len = skb->len;
363         dev_queue_xmit(skb);
364
365         return len;
366
367 error:
368         kfree_skb(skb);
369         ovs_vport_record_error(vport, VPORT_E_TX_DROPPED);
370         return 0;
371 }
372
373 /* Returns null if this device is not attached to a datapath. */
374 struct vport *ovs_netdev_get_vport(struct net_device *dev)
375 {
376 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,36) || \
377     defined HAVE_RHEL_OVS_HOOK
378 #if IFF_OVS_DATAPATH != 0
379         if (likely(dev->priv_flags & IFF_OVS_DATAPATH))
380 #else
381         if (likely(rcu_access_pointer(dev->rx_handler) == netdev_frame_hook))
382 #endif
383 #ifdef HAVE_RHEL_OVS_HOOK
384                 return (struct vport *)rcu_dereference_rtnl(dev->ax25_ptr);
385 #else
386                 return (struct vport *)rcu_dereference_rtnl(dev->rx_handler_data);
387 #endif
388         else
389                 return NULL;
390 #else
391         return (struct vport *)rcu_dereference_rtnl(dev->br_port);
392 #endif
393 }
394
395 const struct vport_ops ovs_netdev_vport_ops = {
396         .type           = OVS_VPORT_TYPE_NETDEV,
397         .flags          = VPORT_F_REQUIRED,
398         .init           = netdev_init,
399         .exit           = netdev_exit,
400         .create         = netdev_create,
401         .destroy        = netdev_destroy,
402         .get_name       = ovs_netdev_get_name,
403         .get_ifindex    = ovs_netdev_get_ifindex,
404         .send           = netdev_send,
405 };
406
407 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,36) && \
408     !defined HAVE_RHEL_OVS_HOOK
409 /*
410  * Enforces, mutual exclusion with the Linux bridge module, by declaring and
411  * exporting br_should_route_hook.  Because the bridge module also exports the
412  * same symbol, the module loader will refuse to load both modules at the same
413  * time (e.g. "bridge: exports duplicate symbol br_should_route_hook (owned by
414  * openvswitch)").
415  *
416  * Before Linux 2.6.36, Open vSwitch cannot safely coexist with the Linux
417  * bridge module, so openvswitch uses this macro in those versions.  In
418  * Linux 2.6.36 and later, Open vSwitch can coexist with the bridge module.
419  *
420  * The use of "typeof" here avoids the need to track changes in the type of
421  * br_should_route_hook over various kernel versions.
422  */
423 typeof(br_should_route_hook) br_should_route_hook;
424 EXPORT_SYMBOL(br_should_route_hook);
425 #endif