]> git.proxmox.com Git - ovs.git/blob - datapath/datapath.c
projects / ovs.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
xenserver: xen 2.6.27 kernel doesn't need skb_checksum_setup defined
[ovs.git] / datapath / datapath.c
1 /*
2 * Distributed under the terms of the GNU GPL version 2.
3 * Copyright (c) 2007, 2008, 2009 Nicira Networks.
4 */
5
6 /* Functions for managing the dp interface/device. */
7
8 #include <linux/init.h>
9 #include <linux/module.h>
10 #include <linux/fs.h>
11 #include <linux/if_arp.h>
12 #include <linux/if_bridge.h>
13 #include <linux/if_vlan.h>
14 #include <linux/in.h>
15 #include <linux/ip.h>
16 #include <linux/delay.h>
17 #include <linux/time.h>
18 #include <linux/etherdevice.h>
19 #include <linux/kernel.h>
20 #include <linux/kthread.h>
21 #include <linux/llc.h>
22 #include <linux/mutex.h>
23 #include <linux/percpu.h>
24 #include <linux/rcupdate.h>
25 #include <linux/tcp.h>
26 #include <linux/udp.h>
27 #include <linux/version.h>
28 #include <linux/ethtool.h>
29 #include <linux/random.h>
30 #include <linux/wait.h>
31 #include <asm/system.h>
32 #include <asm/div64.h>
33 #include <asm/bug.h>
34 #include <linux/netfilter_bridge.h>
35 #include <linux/netfilter_ipv4.h>
36 #include <linux/inetdevice.h>
37 #include <linux/list.h>
38 #include <linux/rculist.h>
39 #include <linux/workqueue.h>
40 #include <linux/dmi.h>
41 #include <net/llc.h>
42
43 #include "openvswitch/datapath-protocol.h"
44 #include "datapath.h"
45 #include "actions.h"
46 #include "dp_dev.h"
47 #include "flow.h"
48
49 #include "compat.h"
50
51
52 int (*dp_ioctl_hook)(struct net_device *dev, struct ifreq *rq, int cmd);
53 EXPORT_SYMBOL(dp_ioctl_hook);
54
55 int (*dp_add_dp_hook)(struct datapath *dp);
56 EXPORT_SYMBOL(dp_add_dp_hook);
57
58 int (*dp_del_dp_hook)(struct datapath *dp);
59 EXPORT_SYMBOL(dp_del_dp_hook);
60
61 int (*dp_add_if_hook)(struct net_bridge_port *p);
62 EXPORT_SYMBOL(dp_add_if_hook);
63
64 int (*dp_del_if_hook)(struct net_bridge_port *p);
65 EXPORT_SYMBOL(dp_del_if_hook);
66
67 /* Datapaths. Protected on the read side by rcu_read_lock, on the write side
68 * by dp_mutex. dp_mutex is almost completely redundant with genl_mutex
69 * maintained by the Generic Netlink code, but the timeout path needs mutual
70 * exclusion too.
71 *
72 * dp_mutex nests inside the RTNL lock: if you need both you must take the RTNL
73 * lock first.
74 *
75 * It is safe to access the datapath and net_bridge_port structures with just
76 * dp_mutex.
77 */
78 static struct datapath *dps[ODP_MAX];
79 static DEFINE_MUTEX(dp_mutex);
80
81 /* Number of milliseconds between runs of the maintenance thread. */
82 #define MAINT_SLEEP_MSECS 1000
83
84 static int new_nbp(struct datapath *, struct net_device *, int port_no);
85
86 /* Must be called with rcu_read_lock or dp_mutex. */
87 struct datapath *get_dp(int dp_idx)
88 {
89 if (dp_idx < 0 || dp_idx >= ODP_MAX)
90 return NULL;
91 return rcu_dereference(dps[dp_idx]);
92 }
93 EXPORT_SYMBOL_GPL(get_dp);
94
95 struct datapath *get_dp_locked(int dp_idx)
96 {
97 struct datapath *dp;
98
99 mutex_lock(&dp_mutex);
100 dp = get_dp(dp_idx);
101 if (dp)
102 mutex_lock(&dp->mutex);
103 mutex_unlock(&dp_mutex);
104 return dp;
105 }
106
107 static inline size_t br_nlmsg_size(void)
108 {
109 return NLMSG_ALIGN(sizeof(struct ifinfomsg))
110 + nla_total_size(IFNAMSIZ) /* IFLA_IFNAME */
111 + nla_total_size(MAX_ADDR_LEN) /* IFLA_ADDRESS */
112 + nla_total_size(4) /* IFLA_MASTER */
113 + nla_total_size(4) /* IFLA_MTU */
114 + nla_total_size(4) /* IFLA_LINK */
115 + nla_total_size(1); /* IFLA_OPERSTATE */
116 }
117
118 static int dp_fill_ifinfo(struct sk_buff *skb,
119 const struct net_bridge_port *port,
120 int event, unsigned int flags)
121 {
122 const struct datapath *dp = port->dp;
123 const struct net_device *dev = port->dev;
124 struct ifinfomsg *hdr;
125 struct nlmsghdr *nlh;
126
127 nlh = nlmsg_put(skb, 0, 0, event, sizeof(*hdr), flags);
128 if (nlh == NULL)
129 return -EMSGSIZE;
130
131 hdr = nlmsg_data(nlh);
132 hdr->ifi_family = AF_BRIDGE;
133 hdr->__ifi_pad = 0;
134 hdr->ifi_type = dev->type;
135 hdr->ifi_index = dev->ifindex;
136 hdr->ifi_flags = dev_get_flags(dev);
137 hdr->ifi_change = 0;
138
139 NLA_PUT_STRING(skb, IFLA_IFNAME, dev->name);
140 NLA_PUT_U32(skb, IFLA_MASTER, dp->ports[ODPP_LOCAL]->dev->ifindex);
141 NLA_PUT_U32(skb, IFLA_MTU, dev->mtu);
142 #ifdef IFLA_OPERSTATE
143 NLA_PUT_U8(skb, IFLA_OPERSTATE,
144 netif_running(dev) ? dev->operstate : IF_OPER_DOWN);
145 #endif
146
147 if (dev->addr_len)
148 NLA_PUT(skb, IFLA_ADDRESS, dev->addr_len, dev->dev_addr);
149
150 if (dev->ifindex != dev->iflink)
151 NLA_PUT_U32(skb, IFLA_LINK, dev->iflink);
152
153 return nlmsg_end(skb, nlh);
154
155 nla_put_failure:
156 nlmsg_cancel(skb, nlh);
157 return -EMSGSIZE;
158 }
159
160 static void dp_ifinfo_notify(int event, struct net_bridge_port *port)
161 {
162 struct net *net = dev_net(port->dev);
163 struct sk_buff *skb;
164 int err = -ENOBUFS;
165
166 skb = nlmsg_new(br_nlmsg_size(), GFP_KERNEL);
167 if (skb == NULL)
168 goto errout;
169
170 err = dp_fill_ifinfo(skb, port, event, 0);
171 if (err < 0) {
172 /* -EMSGSIZE implies BUG in br_nlmsg_size() */
173 WARN_ON(err == -EMSGSIZE);
174 kfree_skb(skb);
175 goto errout;
176 }
177 err = rtnl_notify(skb, net, 0, RTNLGRP_LINK, NULL, GFP_KERNEL);
178 errout:
179 if (err < 0)
180 rtnl_set_sk_err(net, RTNLGRP_LINK, err);
181 }
182
183 static int create_dp(int dp_idx, const char __user *devnamep)
184 {
185 struct net_device *dp_dev;
186 char devname[IFNAMSIZ];
187 struct datapath *dp;
188 int err;
189 int i;
190
191 if (devnamep) {
192 err = -EFAULT;
193 if (strncpy_from_user(devname, devnamep, IFNAMSIZ - 1) < 0)
194 goto err;
195 devname[IFNAMSIZ - 1] = '\0';
196 } else {
197 snprintf(devname, sizeof devname, "of%d", dp_idx);
198 }
199
200 rtnl_lock();
201 mutex_lock(&dp_mutex);
202 err = -ENODEV;
203 if (!try_module_get(THIS_MODULE))
204 goto err_unlock;
205
206 /* Exit early if a datapath with that number already exists.
207 * (We don't use -EEXIST because that's ambiguous with 'devname'
208 * conflicting with an existing network device name.) */
209 err = -EBUSY;
210 if (get_dp(dp_idx))
211 goto err_put_module;
212
213 err = -ENOMEM;
214 dp = kzalloc(sizeof *dp, GFP_KERNEL);
215 if (dp == NULL)
216 goto err_put_module;
217
218 mutex_init(&dp->mutex);
219 dp->dp_idx = dp_idx;
220 for (i = 0; i < DP_N_QUEUES; i++)
221 skb_queue_head_init(&dp->queues[i]);
222 init_waitqueue_head(&dp->waitqueue);
223
224 /* Setup our datapath device */
225 dp_dev = dp_dev_create(dp, devname, ODPP_LOCAL);
226 err = PTR_ERR(dp_dev);
227 if (IS_ERR(dp_dev))
228 goto err_free_dp;
229
230 err = -ENOMEM;
231 rcu_assign_pointer(dp->table, dp_table_create(DP_L1_SIZE));
232 if (!dp->table)
233 goto err_destroy_dp_dev;
234 INIT_LIST_HEAD(&dp->port_list);
235
236 err = new_nbp(dp, dp_dev, ODPP_LOCAL);
237 if (err)
238 goto err_destroy_table;
239
240 dp->drop_frags = 0;
241 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
242 if (!dp->stats_percpu)
243 goto err_destroy_local_port;
244
245 rcu_assign_pointer(dps[dp_idx], dp);
246 mutex_unlock(&dp_mutex);
247 rtnl_unlock();
248
249 if (dp_add_dp_hook)
250 dp_add_dp_hook(dp);
251
252 return 0;
253
254 err_destroy_local_port:
255 dp_del_port(dp->ports[ODPP_LOCAL], NULL);
256 err_destroy_table:
257 dp_table_destroy(dp->table, 0);
258 err_destroy_dp_dev:
259 dp_dev_destroy(dp_dev);
260 err_free_dp:
261 kfree(dp);
262 err_put_module:
263 module_put(THIS_MODULE);
264 err_unlock:
265 mutex_unlock(&dp_mutex);
266 rtnl_unlock();
267 err:
268 return err;
269 }
270
271 static void do_destroy_dp(struct datapath *dp, struct list_head *dp_devs)
272 {
273 struct net_bridge_port *p, *n;
274 int i;
275
276 if (dp_del_dp_hook)
277 dp_del_dp_hook(dp);
278
279 /* Drop references to DP. */
280 list_for_each_entry_safe (p, n, &dp->port_list, node)
281 dp_del_port(p, dp_devs);
282
283 rcu_assign_pointer(dps[dp->dp_idx], NULL);
284 synchronize_rcu();
285
286 /* Wait until no longer in use, then destroy it. */
287 synchronize_rcu();
288 dp_table_destroy(dp->table, 1);
289 for (i = 0; i < DP_N_QUEUES; i++)
290 skb_queue_purge(&dp->queues[i]);
291 for (i = 0; i < DP_MAX_GROUPS; i++)
292 kfree(dp->groups[i]);
293 free_percpu(dp->stats_percpu);
294 kfree(dp);
295 module_put(THIS_MODULE);
296 }
297
298 static int destroy_dp(int dp_idx)
299 {
300 struct dp_dev *dp_dev, *next;
301 struct datapath *dp;
302 LIST_HEAD(dp_devs);
303 int err;
304
305 rtnl_lock();
306 mutex_lock(&dp_mutex);
307 dp = get_dp(dp_idx);
308 err = -ENODEV;
309 if (!dp)
310 goto err_unlock;
311
312 do_destroy_dp(dp, &dp_devs);
313 err = 0;
314
315 err_unlock:
316 mutex_unlock(&dp_mutex);
317 rtnl_unlock();
318 list_for_each_entry_safe (dp_dev, next, &dp_devs, list)
319 free_netdev(dp_dev->dev);
320 return err;
321 }
322
323 /* Called with RTNL lock and dp_mutex. */
324 static int new_nbp(struct datapath *dp, struct net_device *dev, int port_no)
325 {
326 struct net_bridge_port *p;
327
328 if (dev->br_port != NULL)
329 return -EBUSY;
330
331 p = kzalloc(sizeof(*p), GFP_KERNEL);
332 if (!p)
333 return -ENOMEM;
334
335 dev_set_promiscuity(dev, 1);
336 dev_hold(dev);
337 p->port_no = port_no;
338 p->dp = dp;
339 p->dev = dev;
340 if (!is_dp_dev(dev))
341 rcu_assign_pointer(dev->br_port, p);
342 else {
343 /* It would make sense to assign dev->br_port here too, but
344 * that causes packets received on internal ports to get caught
345 * in dp_frame_hook(). In turn dp_frame_hook() can reject them
346 * back to network stack, but that's a waste of time. */
347 }
348 rcu_assign_pointer(dp->ports[port_no], p);
349 list_add_rcu(&p->node, &dp->port_list);
350 dp->n_ports++;
351
352 dp_ifinfo_notify(RTM_NEWLINK, p);
353
354 return 0;
355 }
356
357 static int add_port(int dp_idx, struct odp_port __user *portp)
358 {
359 struct net_device *dev;
360 struct datapath *dp;
361 struct odp_port port;
362 int port_no;
363 int err;
364
365 err = -EFAULT;
366 if (copy_from_user(&port, portp, sizeof port))
367 goto out;
368 port.devname[IFNAMSIZ - 1] = '\0';
369 port_no = port.port;
370
371 err = -EINVAL;
372 if (port_no < 0 || port_no >= DP_MAX_PORTS)
373 goto out;
374
375 rtnl_lock();
376 dp = get_dp_locked(dp_idx);
377 err = -ENODEV;
378 if (!dp)
379 goto out_unlock_rtnl;
380
381 err = -EEXIST;
382 if (dp->ports[port_no])
383 goto out_unlock_dp;
384
385 if (!(port.flags & ODP_PORT_INTERNAL)) {
386 err = -ENODEV;
387 dev = dev_get_by_name(&init_net, port.devname);
388 if (!dev)
389 goto out_unlock_dp;
390
391 err = -EINVAL;
392 if (dev->flags & IFF_LOOPBACK || dev->type != ARPHRD_ETHER ||
393 is_dp_dev(dev))
394 goto out_put;
395 } else {
396 dev = dp_dev_create(dp, port.devname, port_no);
397 err = PTR_ERR(dev);
398 if (IS_ERR(dev))
399 goto out_unlock_dp;
400 dev_hold(dev);
401 }
402
403 err = new_nbp(dp, dev, port_no);
404 if (err)
405 goto out_put;
406
407 if (dp_add_if_hook)
408 dp_add_if_hook(dp->ports[port_no]);
409
410 out_put:
411 dev_put(dev);
412 out_unlock_dp:
413 mutex_unlock(&dp->mutex);
414 out_unlock_rtnl:
415 rtnl_unlock();
416 out:
417 return err;
418 }
419
420 int dp_del_port(struct net_bridge_port *p, struct list_head *dp_devs)
421 {
422 ASSERT_RTNL();
423
424 #ifdef SUPPORT_SYSFS
425 if (p->port_no != ODPP_LOCAL && dp_del_if_hook)
426 sysfs_remove_link(&p->dp->ifobj, p->dev->name);
427 #endif
428 dp_ifinfo_notify(RTM_DELLINK, p);
429
430 p->dp->n_ports--;
431
432 if (is_dp_dev(p->dev)) {
433 /* Make sure that no packets arrive from now on, since
434 * dp_dev_xmit() will try to find itself through
435 * p->dp->ports[], and we're about to set that to null. */
436 netif_tx_disable(p->dev);
437 }
438
439 /* First drop references to device. */
440 dev_set_promiscuity(p->dev, -1);
441 list_del_rcu(&p->node);
442 rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
443 rcu_assign_pointer(p->dev->br_port, NULL);
444
445 /* Then wait until no one is still using it, and destroy it. */
446 synchronize_rcu();
447
448 if (is_dp_dev(p->dev)) {
449 dp_dev_destroy(p->dev);
450 if (dp_devs) {
451 struct dp_dev *dp_dev = dp_dev_priv(p->dev);
452 list_add(&dp_dev->list, dp_devs);
453 }
454 }
455 if (p->port_no != ODPP_LOCAL && dp_del_if_hook) {
456 dp_del_if_hook(p);
457 } else {
458 dev_put(p->dev);
459 kfree(p);
460 }
461
462 return 0;
463 }
464
465 static int del_port(int dp_idx, int port_no)
466 {
467 struct dp_dev *dp_dev, *next;
468 struct net_bridge_port *p;
469 struct datapath *dp;
470 LIST_HEAD(dp_devs);
471 int err;
472
473 err = -EINVAL;
474 if (port_no < 0 || port_no >= DP_MAX_PORTS || port_no == ODPP_LOCAL)
475 goto out;
476
477 rtnl_lock();
478 dp = get_dp_locked(dp_idx);
479 err = -ENODEV;
480 if (!dp)
481 goto out_unlock_rtnl;
482
483 p = dp->ports[port_no];
484 err = -ENOENT;
485 if (!p)
486 goto out_unlock_dp;
487
488 err = dp_del_port(p, &dp_devs);
489
490 out_unlock_dp:
491 mutex_unlock(&dp->mutex);
492 out_unlock_rtnl:
493 rtnl_unlock();
494 out:
495 list_for_each_entry_safe (dp_dev, next, &dp_devs, list)
496 free_netdev(dp_dev->dev);
497 return err;
498 }
499
500 /* Must be called with rcu_read_lock. */
501 static void
502 do_port_input(struct net_bridge_port *p, struct sk_buff *skb)
503 {
504 /* Make our own copy of the packet. Otherwise we will mangle the
505 * packet for anyone who came before us (e.g. tcpdump via AF_PACKET).
506 * (No one comes after us, since we tell handle_bridge() that we took
507 * the packet.) */
508 skb = skb_share_check(skb, GFP_ATOMIC);
509 if (!skb)
510 return;
511
512 /* Push the Ethernet header back on. */
513 skb_push(skb, ETH_HLEN);
514 skb_reset_mac_header(skb);
515 dp_process_received_packet(skb, p);
516 }
517
518 /* Must be called with rcu_read_lock and with bottom-halves disabled. */
519 void dp_process_received_packet(struct sk_buff *skb, struct net_bridge_port *p)
520 {
521 struct datapath *dp = p->dp;
522 struct dp_stats_percpu *stats;
523 struct odp_flow_key key;
524 struct sw_flow *flow;
525
526 WARN_ON_ONCE(skb_shared(skb));
527 WARN_ON_ONCE(skb->destructor);
528
529 /* BHs are off so we don't have to use get_cpu()/put_cpu() here. */
530 stats = percpu_ptr(dp->stats_percpu, smp_processor_id());
531
532 if (flow_extract(skb, p ? p->port_no : ODPP_NONE, &key)) {
533 if (dp->drop_frags) {
534 kfree_skb(skb);
535 stats->n_frags++;
536 return;
537 }
538 }
539
540 flow = dp_table_lookup(rcu_dereference(dp->table), &key);
541 if (flow) {
542 struct sw_flow_actions *acts = rcu_dereference(flow->sf_acts);
543 flow_used(flow, skb);
544 execute_actions(dp, skb, &key, acts->actions, acts->n_actions,
545 GFP_ATOMIC);
546 stats->n_hit++;
547 } else {
548 stats->n_missed++;
549 dp_output_control(dp, skb, _ODPL_MISS_NR, 0);
550 }
551 }
552
553 /*
554 * Used as br_handle_frame_hook. (Cannot run bridge at the same time, even on
555 * different set of devices!)
556 */
557 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,22)
558 /* Called with rcu_read_lock and bottom-halves disabled. */
559 static struct sk_buff *dp_frame_hook(struct net_bridge_port *p,
560 struct sk_buff *skb)
561 {
562 do_port_input(p, skb);
563 return NULL;
564 }
565 #elif LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)
566 /* Called with rcu_read_lock and bottom-halves disabled. */
567 static int dp_frame_hook(struct net_bridge_port *p, struct sk_buff **pskb)
568 {
569 do_port_input(p, *pskb);
570 return 1;
571 }
572 #else
573 #error
574 #endif
575
576 #ifdef CONFIG_XEN
577 #if LINUX_VERSION_CODE == KERNEL_VERSION(2,6,18)
578 /* This code is copied verbatim from net/dev/core.c in Xen's
579 * linux-2.6.18-92.1.10.el5.xs5.0.0.394.644. We can't call those functions
580 * directly because they aren't exported. */
581 static int skb_pull_up_to(struct sk_buff *skb, void *ptr)
582 {
583 if (ptr < (void *)skb->tail)
584 return 1;
585 if (__pskb_pull_tail(skb,
586 ptr - (void *)skb->data - skb_headlen(skb))) {
587 return 1;
588 } else {
589 return 0;
590 }
591 }
592
593 int skb_checksum_setup(struct sk_buff *skb)
594 {
595 if (skb->proto_csum_blank) {
596 if (skb->protocol != htons(ETH_P_IP))
597 goto out;
598 if (!skb_pull_up_to(skb, skb->nh.iph + 1))
599 goto out;
600 skb->h.raw = (unsigned char *)skb->nh.iph + 4*skb->nh.iph->ihl;
601 switch (skb->nh.iph->protocol) {
602 case IPPROTO_TCP:
603 skb->csum = offsetof(struct tcphdr, check);
604 break;
605 case IPPROTO_UDP:
606 skb->csum = offsetof(struct udphdr, check);
607 break;
608 default:
609 if (net_ratelimit())
610 printk(KERN_ERR "Attempting to checksum a non-"
611 "TCP/UDP packet, dropping a protocol"
612 " %d packet", skb->nh.iph->protocol);
613 goto out;
614 }
615 if (!skb_pull_up_to(skb, skb->h.raw + skb->csum + 2))
616 goto out;
617 skb->ip_summed = CHECKSUM_HW;
618 skb->proto_csum_blank = 0;
619 }
620 return 0;
621 out:
622 return -EPROTO;
623 }
624 #endif /* linux == 2.6.18 */
625 #endif /* CONFIG_XEN */
626
627 int
628 dp_output_control(struct datapath *dp, struct sk_buff *skb, int queue_no,
629 u32 arg)
630 {
631 struct dp_stats_percpu *stats;
632 struct sk_buff_head *queue;
633 int port_no;
634 int err;
635
636 WARN_ON_ONCE(skb_shared(skb));
637 BUG_ON(queue_no != _ODPL_MISS_NR && queue_no != _ODPL_ACTION_NR);
638
639 queue = &dp->queues[queue_no];
640 err = -ENOBUFS;
641 if (skb_queue_len(queue) >= DP_MAX_QUEUE_LEN)
642 goto err_kfree_skb;
643
644 /* If a checksum-deferred packet is forwarded to the controller,
645 * correct the pointers and checksum. This happens on a regular basis
646 * only on Xen (the CHECKSUM_HW case), on which VMs can pass up packets
647 * that do not have their checksum computed. We also implement it for
648 * the non-Xen case, but it is difficult to trigger or test this case
649 * there, hence the WARN_ON_ONCE().
650 */
651 err = skb_checksum_setup(skb);
652 if (err)
653 goto err_kfree_skb;
654 #ifndef CHECKSUM_HW
655 if (skb->ip_summed == CHECKSUM_PARTIAL) {
656 WARN_ON_ONCE(1);
657 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,22)
658 /* Until 2.6.22, the start of the transport header was also the
659 * start of data to be checksummed. Linux 2.6.22 introduced
660 * the csum_start field for this purpose, but we should point
661 * the transport header to it anyway for backward
662 * compatibility, as dev_queue_xmit() does even in 2.6.28. */
663 skb_set_transport_header(skb, skb->csum_start -
664 skb_headroom(skb));
665 #endif
666 err = skb_checksum_help(skb);
667 if (err)
668 goto err_kfree_skb;
669 }
670 #else
671 if (skb->ip_summed == CHECKSUM_HW) {
672 err = skb_checksum_help(skb, 0);
673 if (err)
674 goto err_kfree_skb;
675 }
676 #endif
677
678 /* Break apart GSO packets into their component pieces. Otherwise
679 * userspace may try to stuff a 64kB packet into a 1500-byte MTU. */
680 if (skb_is_gso(skb)) {
681 struct sk_buff *nskb = skb_gso_segment(skb, 0);
682 if (nskb) {
683 kfree_skb(skb);
684 skb = nskb;
685 if (unlikely(IS_ERR(skb))) {
686 err = PTR_ERR(skb);
687 goto err;
688 }
689 } else {
690 /* XXX This case might not be possible. It's hard to
691 * tell from the skb_gso_segment() code and comment. */
692 }
693 }
694
695 /* Figure out port number. */
696 port_no = ODPP_LOCAL;
697 if (skb->dev) {
698 if (skb->dev->br_port)
699 port_no = skb->dev->br_port->port_no;
700 else if (is_dp_dev(skb->dev))
701 port_no = dp_dev_priv(skb->dev)->port_no;
702 }
703
704 /* Append each packet to queue. There will be only one packet unless
705 * we broke up a GSO packet above. */
706 do {
707 struct odp_msg *header;
708 struct sk_buff *nskb = skb->next;
709 skb->next = NULL;
710
711 err = skb_cow(skb, sizeof *header);
712 if (err) {
713 while (nskb) {
714 kfree_skb(skb);
715 skb = nskb;
716 nskb = skb->next;
717 }
718 goto err_kfree_skb;
719 }
720
721 header = (struct odp_msg*)__skb_push(skb, sizeof *header);
722 header->type = queue_no;
723 header->length = skb->len;
724 header->port = port_no;
725 header->reserved = 0;
726 header->arg = arg;
727 skb_queue_tail(queue, skb);
728
729 skb = nskb;
730 } while (skb);
731
732 wake_up_interruptible(&dp->waitqueue);
733 return 0;
734
735 err_kfree_skb:
736 kfree_skb(skb);
737 err:
738 stats = percpu_ptr(dp->stats_percpu, get_cpu());
739 stats->n_lost++;
740 put_cpu();
741
742 return err;
743 }
744
745 static int flush_flows(struct datapath *dp)
746 {
747 dp->n_flows = 0;
748 return dp_table_flush(dp);
749 }
750
751 static int validate_actions(const struct sw_flow_actions *actions)
752 {
753 unsigned int i;
754
755 for (i = 0; i < actions->n_actions; i++) {
756 const union odp_action *a = &actions->actions[i];
757 switch (a->type) {
758 case ODPAT_OUTPUT:
759 if (a->output.port >= DP_MAX_PORTS)
760 return -EINVAL;
761 break;
762
763 case ODPAT_OUTPUT_GROUP:
764 if (a->output_group.group >= DP_MAX_GROUPS)
765 return -EINVAL;
766 break;
767
768 case ODPAT_SET_VLAN_VID:
769 if (a->vlan_vid.vlan_vid & htons(~VLAN_VID_MASK))
770 return -EINVAL;
771 break;
772
773 case ODPAT_SET_VLAN_PCP:
774 if (a->vlan_pcp.vlan_pcp & ~VLAN_PCP_MASK)
775 return -EINVAL;
776 break;
777
778 default:
779 if (a->type >= ODPAT_N_ACTIONS)
780 return -EOPNOTSUPP;
781 break;
782 }
783 }
784
785 return 0;
786 }
787
788 static struct sw_flow_actions *get_actions(const struct odp_flow *flow)
789 {
790 struct sw_flow_actions *actions;
791 int error;
792
793 actions = flow_actions_alloc(flow->n_actions);
794 error = PTR_ERR(actions);
795 if (IS_ERR(actions))
796 goto error;
797
798 error = -EFAULT;
799 if (copy_from_user(actions->actions, flow->actions,
800 flow->n_actions * sizeof(union odp_action)))
801 goto error_free_actions;
802 error = validate_actions(actions);
803 if (error)
804 goto error_free_actions;
805
806 return actions;
807
808 error_free_actions:
809 kfree(actions);
810 error:
811 return ERR_PTR(error);
812 }
813
814 static void get_stats(struct sw_flow *flow, struct odp_flow_stats *stats)
815 {
816 if (flow->used.tv_sec) {
817 stats->used_sec = flow->used.tv_sec;
818 stats->used_nsec = flow->used.tv_nsec;
819 } else {
820 stats->used_sec = 0;
821 stats->used_nsec = 0;
822 }
823 stats->n_packets = flow->packet_count;
824 stats->n_bytes = flow->byte_count;
825 stats->ip_tos = flow->ip_tos;
826 stats->tcp_flags = flow->tcp_flags;
827 }
828
829 static void clear_stats(struct sw_flow *flow)
830 {
831 flow->used.tv_sec = flow->used.tv_nsec = 0;
832 flow->tcp_flags = 0;
833 flow->ip_tos = 0;
834 flow->packet_count = 0;
835 flow->byte_count = 0;
836 }
837
838 static int put_flow(struct datapath *dp, struct odp_flow_put __user *ufp)
839 {
840 struct odp_flow_put uf;
841 struct sw_flow *flow, **bucket;
842 struct dp_table *table;
843 struct odp_flow_stats stats;
844 int error;
845
846 error = -EFAULT;
847 if (copy_from_user(&uf, ufp, sizeof(struct odp_flow_put)))
848 goto error;
849 uf.flow.key.reserved = 0;
850
851 retry:
852 table = rcu_dereference(dp->table);
853 bucket = dp_table_lookup_for_insert(table, &uf.flow.key);
854 if (!bucket) {
855 /* No such flow, and the slots where it could go are full. */
856 error = uf.flags & ODPPF_CREATE ? -EXFULL : -ENOENT;
857 goto error;
858 } else if (!*bucket) {
859 /* No such flow, but we found an available slot for it. */
860 struct sw_flow_actions *acts;
861
862 error = -ENOENT;
863 if (!(uf.flags & ODPPF_CREATE))
864 goto error;
865
866 /* Expand table, if necessary, to make room. */
867 if (dp->n_flows * 4 >= table->n_buckets &&
868 table->n_buckets < DP_MAX_BUCKETS) {
869 error = dp_table_expand(dp);
870 if (error)
871 goto error;
872
873 /* The bucket's location has changed. Try again. */
874 goto retry;
875 }
876
877 /* Allocate flow. */
878 error = -ENOMEM;
879 flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
880 if (flow == NULL)
881 goto error;
882 flow->key = uf.flow.key;
883 spin_lock_init(&flow->lock);
884 clear_stats(flow);
885
886 /* Obtain actions. */
887 acts = get_actions(&uf.flow);
888 error = PTR_ERR(acts);
889 if (IS_ERR(acts))
890 goto error_free_flow;
891 rcu_assign_pointer(flow->sf_acts, acts);
892
893 /* Put flow in bucket. */
894 rcu_assign_pointer(*bucket, flow);
895 dp->n_flows++;
896 memset(&stats, 0, sizeof(struct odp_flow_stats));
897 } else {
898 /* We found a matching flow. */
899 struct sw_flow *flow = *rcu_dereference(bucket);
900 struct sw_flow_actions *old_acts, *new_acts;
901 unsigned long int flags;
902
903 /* Bail out if we're not allowed to modify an existing flow. */
904 error = -EEXIST;
905 if (!(uf.flags & ODPPF_MODIFY))
906 goto error;
907
908 /* Swap actions. */
909 new_acts = get_actions(&uf.flow);
910 error = PTR_ERR(new_acts);
911 if (IS_ERR(new_acts))
912 goto error;
913 old_acts = rcu_dereference(flow->sf_acts);
914 if (old_acts->n_actions != new_acts->n_actions ||
915 memcmp(old_acts->actions, new_acts->actions,
916 sizeof(union odp_action) * old_acts->n_actions)) {
917 rcu_assign_pointer(flow->sf_acts, new_acts);
918 flow_deferred_free_acts(old_acts);
919 } else {
920 kfree(new_acts);
921 }
922
923 /* Fetch stats, then clear them if necessary. */
924 spin_lock_irqsave(&flow->lock, flags);
925 get_stats(flow, &stats);
926 if (uf.flags & ODPPF_ZERO_STATS)
927 clear_stats(flow);
928 spin_unlock_irqrestore(&flow->lock, flags);
929 }
930
931 /* Copy stats to userspace. */
932 if (__copy_to_user(&ufp->flow.stats, &stats,
933 sizeof(struct odp_flow_stats)))
934 return -EFAULT;
935 return 0;
936
937 error_free_flow:
938 kmem_cache_free(flow_cache, flow);
939 error:
940 return error;
941 }
942
943 static int put_actions(const struct sw_flow *flow, struct odp_flow __user *ufp)
944 {
945 union odp_action __user *actions;
946 struct sw_flow_actions *sf_acts;
947 u32 n_actions;
948
949 if (__get_user(actions, &ufp->actions) ||
950 __get_user(n_actions, &ufp->n_actions))
951 return -EFAULT;
952
953 if (!n_actions)
954 return 0;
955 if (ufp->n_actions > INT_MAX / sizeof(union odp_action))
956 return -EINVAL;
957
958 sf_acts = rcu_dereference(flow->sf_acts);
959 if (__put_user(sf_acts->n_actions, &ufp->n_actions) ||
960 (actions && copy_to_user(actions, sf_acts->actions,
961 sizeof(union odp_action) *
962 min(sf_acts->n_actions, n_actions))))
963 return -EFAULT;
964
965 return 0;
966 }
967
968 static int answer_query(struct sw_flow *flow, struct odp_flow __user *ufp)
969 {
970 struct odp_flow_stats stats;
971 unsigned long int flags;
972
973 spin_lock_irqsave(&flow->lock, flags);
974 get_stats(flow, &stats);
975 spin_unlock_irqrestore(&flow->lock, flags);
976
977 if (__copy_to_user(&ufp->stats, &stats, sizeof(struct odp_flow_stats)))
978 return -EFAULT;
979 return put_actions(flow, ufp);
980 }
981
982 static int del_or_query_flow(struct datapath *dp,
983 struct odp_flow __user *ufp,
984 unsigned int cmd)
985 {
986 struct dp_table *table = rcu_dereference(dp->table);
987 struct odp_flow uf;
988 struct sw_flow *flow;
989 int error;
990
991 error = -EFAULT;
992 if (copy_from_user(&uf, ufp, sizeof uf))
993 goto error;
994 uf.key.reserved = 0;
995
996 flow = dp_table_lookup(table, &uf.key);
997 error = -ENOENT;
998 if (!flow)
999 goto error;
1000
1001 if (cmd == ODP_FLOW_DEL) {
1002 /* XXX redundant lookup */
1003 error = dp_table_delete(table, flow);
1004 if (error)
1005 goto error;
1006
1007 /* XXX These statistics might lose a few packets, since other
1008 * CPUs can be using this flow. We used to synchronize_rcu()
1009 * to make sure that we get completely accurate stats, but that
1010 * blows our performance, badly. */
1011 dp->n_flows--;
1012 error = answer_query(flow, ufp);
1013 flow_deferred_free(flow);
1014 } else {
1015 error = answer_query(flow, ufp);
1016 }
1017
1018 error:
1019 return error;
1020 }
1021
1022 static int query_multiple_flows(struct datapath *dp,
1023 const struct odp_flowvec *flowvec)
1024 {
1025 struct dp_table *table = rcu_dereference(dp->table);
1026 int i;
1027 for (i = 0; i < flowvec->n_flows; i++) {
1028 struct __user odp_flow *ufp = &flowvec->flows[i];
1029 struct odp_flow uf;
1030 struct sw_flow *flow;
1031 int error;
1032
1033 if (__copy_from_user(&uf, ufp, sizeof uf))
1034 return -EFAULT;
1035 uf.key.reserved = 0;
1036
1037 flow = dp_table_lookup(table, &uf.key);
1038 if (!flow)
1039 error = __clear_user(&ufp->stats, sizeof ufp->stats);
1040 else
1041 error = answer_query(flow, ufp);
1042 if (error)
1043 return -EFAULT;
1044 }
1045 return flowvec->n_flows;
1046 }
1047
1048 struct list_flows_cbdata {
1049 struct odp_flow __user *uflows;
1050 int n_flows;
1051 int listed_flows;
1052 };
1053
1054 static int list_flow(struct sw_flow *flow, void *cbdata_)
1055 {
1056 struct list_flows_cbdata *cbdata = cbdata_;
1057 struct odp_flow __user *ufp = &cbdata->uflows[cbdata->listed_flows++];
1058 int error;
1059
1060 if (__copy_to_user(&ufp->key, &flow->key, sizeof flow->key))
1061 return -EFAULT;
1062 error = answer_query(flow, ufp);
1063 if (error)
1064 return error;
1065
1066 if (cbdata->listed_flows >= cbdata->n_flows)
1067 return cbdata->listed_flows;
1068 return 0;
1069 }
1070
1071 static int list_flows(struct datapath *dp, const struct odp_flowvec *flowvec)
1072 {
1073 struct list_flows_cbdata cbdata;
1074 int error;
1075
1076 if (!flowvec->n_flows)
1077 return 0;
1078
1079 cbdata.uflows = flowvec->flows;
1080 cbdata.n_flows = flowvec->n_flows;
1081 cbdata.listed_flows = 0;
1082 error = dp_table_foreach(rcu_dereference(dp->table),
1083 list_flow, &cbdata);
1084 return error ? error : cbdata.listed_flows;
1085 }
1086
1087 static int do_flowvec_ioctl(struct datapath *dp, unsigned long argp,
1088 int (*function)(struct datapath *,
1089 const struct odp_flowvec *))
1090 {
1091 struct odp_flowvec __user *uflowvec;
1092 struct odp_flowvec flowvec;
1093 int retval;
1094
1095 uflowvec = (struct odp_flowvec __user *)argp;
1096 if (!access_ok(VERIFY_WRITE, uflowvec, sizeof *uflowvec) ||
1097 copy_from_user(&flowvec, uflowvec, sizeof flowvec))
1098 return -EFAULT;
1099
1100 if (flowvec.n_flows > INT_MAX / sizeof(struct odp_flow))
1101 return -EINVAL;
1102
1103 if (!access_ok(VERIFY_WRITE, flowvec.flows,
1104 flowvec.n_flows * sizeof(struct odp_flow)))
1105 return -EFAULT;
1106
1107 retval = function(dp, &flowvec);
1108 return (retval < 0 ? retval
1109 : retval == flowvec.n_flows ? 0
1110 : __put_user(retval, &uflowvec->n_flows));
1111 }
1112
1113 static int do_execute(struct datapath *dp, const struct odp_execute *executep)
1114 {
1115 struct odp_execute execute;
1116 struct odp_flow_key key;
1117 struct sk_buff *skb;
1118 struct sw_flow_actions *actions;
1119 int err;
1120
1121 err = -EFAULT;
1122 if (copy_from_user(&execute, executep, sizeof execute))
1123 goto error;
1124
1125 err = -EINVAL;
1126 if (execute.length < ETH_HLEN || execute.length > 65535)
1127 goto error;
1128
1129 err = -ENOMEM;
1130 actions = flow_actions_alloc(execute.n_actions);
1131 if (!actions)
1132 goto error;
1133
1134 err = -EFAULT;
1135 if (copy_from_user(actions->actions, execute.actions,
1136 execute.n_actions * sizeof *execute.actions))
1137 goto error_free_actions;
1138
1139 err = validate_actions(actions);
1140 if (err)
1141 goto error_free_actions;
1142
1143 err = -ENOMEM;
1144 skb = alloc_skb(execute.length, GFP_KERNEL);
1145 if (!skb)
1146 goto error_free_actions;
1147 if (execute.in_port < DP_MAX_PORTS) {
1148 struct net_bridge_port *p = dp->ports[execute.in_port];
1149 if (p)
1150 skb->dev = p->dev;
1151 }
1152
1153 err = -EFAULT;
1154 if (copy_from_user(skb_put(skb, execute.length), execute.data,
1155 execute.length))
1156 goto error_free_skb;
1157
1158 flow_extract(skb, execute.in_port, &key);
1159 err = execute_actions(dp, skb, &key, actions->actions,
1160 actions->n_actions, GFP_KERNEL);
1161 kfree(actions);
1162 return err;
1163
1164 error_free_skb:
1165 kfree_skb(skb);
1166 error_free_actions:
1167 kfree(actions);
1168 error:
1169 return err;
1170 }
1171
1172 static int
1173 get_dp_stats(struct datapath *dp, struct odp_stats __user *statsp)
1174 {
1175 struct odp_stats stats;
1176 int i;
1177
1178 stats.n_flows = dp->n_flows;
1179 stats.cur_capacity = rcu_dereference(dp->table)->n_buckets * 2;
1180 stats.max_capacity = DP_MAX_BUCKETS * 2;
1181 stats.n_ports = dp->n_ports;
1182 stats.max_ports = DP_MAX_PORTS;
1183 stats.max_groups = DP_MAX_GROUPS;
1184 stats.n_frags = stats.n_hit = stats.n_missed = stats.n_lost = 0;
1185 for_each_possible_cpu(i) {
1186 const struct dp_stats_percpu *s;
1187 s = percpu_ptr(dp->stats_percpu, i);
1188 stats.n_frags += s->n_frags;
1189 stats.n_hit += s->n_hit;
1190 stats.n_missed += s->n_missed;
1191 stats.n_lost += s->n_lost;
1192 }
1193 stats.max_miss_queue = DP_MAX_QUEUE_LEN;
1194 stats.max_action_queue = DP_MAX_QUEUE_LEN;
1195 return copy_to_user(statsp, &stats, sizeof stats) ? -EFAULT : 0;
1196 }
1197
1198 static int
1199 put_port(const struct net_bridge_port *p, struct odp_port __user *uop)
1200 {
1201 struct odp_port op;
1202 memset(&op, 0, sizeof op);
1203 strncpy(op.devname, p->dev->name, sizeof op.devname);
1204 op.port = p->port_no;
1205 op.flags = is_dp_dev(p->dev) ? ODP_PORT_INTERNAL : 0;
1206 return copy_to_user(uop, &op, sizeof op) ? -EFAULT : 0;
1207 }
1208
1209 static int
1210 query_port(struct datapath *dp, struct odp_port __user *uport)
1211 {
1212 struct odp_port port;
1213
1214 if (copy_from_user(&port, uport, sizeof port))
1215 return -EFAULT;
1216 if (port.devname[0]) {
1217 struct net_bridge_port *p;
1218 struct net_device *dev;
1219 int err;
1220
1221 port.devname[IFNAMSIZ - 1] = '\0';
1222
1223 dev = dev_get_by_name(&init_net, port.devname);
1224 if (!dev)
1225 return -ENODEV;
1226
1227 p = dev->br_port;
1228 if (!p && is_dp_dev(dev)) {
1229 struct dp_dev *dp_dev = dp_dev_priv(dev);
1230 if (dp_dev->dp == dp)
1231 p = dp->ports[dp_dev->port_no];
1232 }
1233 err = p && p->dp == dp ? put_port(p, uport) : -ENOENT;
1234 dev_put(dev);
1235
1236 return err;
1237 } else {
1238 if (port.port >= DP_MAX_PORTS)
1239 return -EINVAL;
1240 if (!dp->ports[port.port])
1241 return -ENOENT;
1242 return put_port(dp->ports[port.port], uport);
1243 }
1244 }
1245
1246 static int
1247 list_ports(struct datapath *dp, struct odp_portvec __user *pvp)
1248 {
1249 struct odp_portvec pv;
1250 struct net_bridge_port *p;
1251 int idx;
1252
1253 if (copy_from_user(&pv, pvp, sizeof pv))
1254 return -EFAULT;
1255
1256 idx = 0;
1257 if (pv.n_ports) {
1258 list_for_each_entry_rcu (p, &dp->port_list, node) {
1259 if (put_port(p, &pv.ports[idx]))
1260 return -EFAULT;
1261 if (idx++ >= pv.n_ports)
1262 break;
1263 }
1264 }
1265 return put_user(idx, &pvp->n_ports);
1266 }
1267
1268 /* RCU callback for freeing a dp_port_group */
1269 static void free_port_group(struct rcu_head *rcu)
1270 {
1271 struct dp_port_group *g = container_of(rcu, struct dp_port_group, rcu);
1272 kfree(g);
1273 }
1274
1275 static int
1276 set_port_group(struct datapath *dp, const struct odp_port_group __user *upg)
1277 {
1278 struct odp_port_group pg;
1279 struct dp_port_group *new_group, *old_group;
1280 int error;
1281
1282 error = -EFAULT;
1283 if (copy_from_user(&pg, upg, sizeof pg))
1284 goto error;
1285
1286 error = -EINVAL;
1287 if (pg.n_ports > DP_MAX_PORTS || pg.group >= DP_MAX_GROUPS)
1288 goto error;
1289
1290 error = -ENOMEM;
1291 new_group = kmalloc(sizeof *new_group + sizeof(u16) * pg.n_ports,
1292 GFP_KERNEL);
1293 if (!new_group)
1294 goto error;
1295
1296 new_group->n_ports = pg.n_ports;
1297 error = -EFAULT;
1298 if (copy_from_user(new_group->ports, pg.ports,
1299 sizeof(u16) * pg.n_ports))
1300 goto error_free;
1301
1302 old_group = rcu_dereference(dp->groups[pg.group]);
1303 rcu_assign_pointer(dp->groups[pg.group], new_group);
1304 if (old_group)
1305 call_rcu(&old_group->rcu, free_port_group);
1306 return 0;
1307
1308 error_free:
1309 kfree(new_group);
1310 error:
1311 return error;
1312 }
1313
1314 static int
1315 get_port_group(struct datapath *dp, struct odp_port_group *upg)
1316 {
1317 struct odp_port_group pg;
1318 struct dp_port_group *g;
1319 u16 n_copy;
1320
1321 if (copy_from_user(&pg, upg, sizeof pg))
1322 return -EFAULT;
1323
1324 if (pg.group >= DP_MAX_GROUPS)
1325 return -EINVAL;
1326
1327 g = dp->groups[pg.group];
1328 n_copy = g ? min_t(int, g->n_ports, pg.n_ports) : 0;
1329 if (n_copy && copy_to_user(pg.ports, g->ports, n_copy * sizeof(u16)))
1330 return -EFAULT;
1331
1332 if (put_user(g ? g->n_ports : 0, &upg->n_ports))
1333 return -EFAULT;
1334
1335 return 0;
1336 }
1337
1338 static long openvswitch_ioctl(struct file *f, unsigned int cmd,
1339 unsigned long argp)
1340 {
1341 int dp_idx = iminor(f->f_dentry->d_inode);
1342 struct datapath *dp;
1343 int drop_frags, listeners, port_no;
1344 int err;
1345
1346 /* Handle commands with special locking requirements up front. */
1347 switch (cmd) {
1348 case ODP_DP_CREATE:
1349 return create_dp(dp_idx, (char __user *)argp);
1350
1351 case ODP_DP_DESTROY:
1352 return destroy_dp(dp_idx);
1353
1354 case ODP_PORT_ADD:
1355 return add_port(dp_idx, (struct odp_port __user *)argp);
1356
1357 case ODP_PORT_DEL:
1358 err = get_user(port_no, (int __user *)argp);
1359 if (err)
1360 break;
1361 return del_port(dp_idx, port_no);
1362 }
1363
1364 dp = get_dp_locked(dp_idx);
1365 if (!dp)
1366 return -ENODEV;
1367
1368 switch (cmd) {
1369 case ODP_DP_STATS:
1370 err = get_dp_stats(dp, (struct odp_stats __user *)argp);
1371 break;
1372
1373 case ODP_GET_DROP_FRAGS:
1374 err = put_user(dp->drop_frags, (int __user *)argp);
1375 break;
1376
1377 case ODP_SET_DROP_FRAGS:
1378 err = get_user(drop_frags, (int __user *)argp);
1379 if (err)
1380 break;
1381 err = -EINVAL;
1382 if (drop_frags != 0 && drop_frags != 1)
1383 break;
1384 dp->drop_frags = drop_frags;
1385 err = 0;
1386 break;
1387
1388 case ODP_GET_LISTEN_MASK:
1389 err = put_user((int)f->private_data, (int __user *)argp);
1390 break;
1391
1392 case ODP_SET_LISTEN_MASK:
1393 err = get_user(listeners, (int __user *)argp);
1394 if (err)
1395 break;
1396 err = -EINVAL;
1397 if (listeners & ~ODPL_ALL)
1398 break;
1399 err = 0;
1400 f->private_data = (void*)listeners;
1401 break;
1402
1403 case ODP_PORT_QUERY:
1404 err = query_port(dp, (struct odp_port __user *)argp);
1405 break;
1406
1407 case ODP_PORT_LIST:
1408 err = list_ports(dp, (struct odp_portvec __user *)argp);
1409 break;
1410
1411 case ODP_PORT_GROUP_SET:
1412 err = set_port_group(dp, (struct odp_port_group __user *)argp);
1413 break;
1414
1415 case ODP_PORT_GROUP_GET:
1416 err = get_port_group(dp, (struct odp_port_group __user *)argp);
1417 break;
1418
1419 case ODP_FLOW_FLUSH:
1420 err = flush_flows(dp);
1421 break;
1422
1423 case ODP_FLOW_PUT:
1424 err = put_flow(dp, (struct odp_flow_put __user *)argp);
1425 break;
1426
1427 case ODP_FLOW_DEL:
1428 case ODP_FLOW_GET:
1429 err = del_or_query_flow(dp, (struct odp_flow __user *)argp,
1430 cmd);
1431 break;
1432
1433 case ODP_FLOW_GET_MULTIPLE:
1434 err = do_flowvec_ioctl(dp, argp, query_multiple_flows);
1435 break;
1436
1437 case ODP_FLOW_LIST:
1438 err = do_flowvec_ioctl(dp, argp, list_flows);
1439 break;
1440
1441 case ODP_EXECUTE:
1442 err = do_execute(dp, (struct odp_execute __user *)argp);
1443 break;
1444
1445 default:
1446 err = -ENOIOCTLCMD;
1447 break;
1448 }
1449 mutex_unlock(&dp->mutex);
1450 return err;
1451 }
1452
1453 static int dp_has_packet_of_interest(struct datapath *dp, int listeners)
1454 {
1455 int i;
1456 for (i = 0; i < DP_N_QUEUES; i++) {
1457 if (listeners & (1 << i) && !skb_queue_empty(&dp->queues[i]))
1458 return 1;
1459 }
1460 return 0;
1461 }
1462
1463 ssize_t openvswitch_read(struct file *f, char __user *buf, size_t nbytes,
1464 loff_t *ppos)
1465 {
1466 int listeners = (int) f->private_data;
1467 int dp_idx = iminor(f->f_dentry->d_inode);
1468 struct datapath *dp = get_dp(dp_idx);
1469 struct sk_buff *skb;
1470 struct iovec __user iov;
1471 size_t copy_bytes;
1472 int retval;
1473
1474 if (!dp)
1475 return -ENODEV;
1476
1477 if (nbytes == 0 || !listeners)
1478 return 0;
1479
1480 for (;;) {
1481 int i;
1482
1483 for (i = 0; i < DP_N_QUEUES; i++) {
1484 if (listeners & (1 << i)) {
1485 skb = skb_dequeue(&dp->queues[i]);
1486 if (skb)
1487 goto success;
1488 }
1489 }
1490
1491 if (f->f_flags & O_NONBLOCK) {
1492 retval = -EAGAIN;
1493 goto error;
1494 }
1495
1496 wait_event_interruptible(dp->waitqueue,
1497 dp_has_packet_of_interest(dp,
1498 listeners));
1499
1500 if (signal_pending(current)) {
1501 retval = -ERESTARTSYS;
1502 goto error;
1503 }
1504 }
1505 success:
1506 copy_bytes = min(skb->len, nbytes);
1507 iov.iov_base = buf;
1508 iov.iov_len = copy_bytes;
1509 retval = skb_copy_datagram_iovec(skb, 0, &iov, iov.iov_len);
1510 if (!retval)
1511 retval = copy_bytes;
1512 kfree_skb(skb);
1513
1514 error:
1515 return retval;
1516 }
1517
1518 static unsigned int openvswitch_poll(struct file *file, poll_table *wait)
1519 {
1520 int dp_idx = iminor(file->f_dentry->d_inode);
1521 struct datapath *dp = get_dp(dp_idx);
1522 unsigned int mask;
1523
1524 if (dp) {
1525 mask = 0;
1526 poll_wait(file, &dp->waitqueue, wait);
1527 if (dp_has_packet_of_interest(dp, (int)file->private_data))
1528 mask |= POLLIN | POLLRDNORM;
1529 } else {
1530 mask = POLLIN | POLLRDNORM | POLLHUP;
1531 }
1532 return mask;
1533 }
1534
1535 struct file_operations openvswitch_fops = {
1536 /* XXX .aio_read = openvswitch_aio_read, */
1537 .read = openvswitch_read,
1538 .poll = openvswitch_poll,
1539 .unlocked_ioctl = openvswitch_ioctl,
1540 /* XXX .fasync = openvswitch_fasync, */
1541 };
1542
1543 static int major;
1544 static struct llc_sap *dp_stp_sap;
1545
1546 static int dp_stp_rcv(struct sk_buff *skb, struct net_device *dev,
1547 struct packet_type *pt, struct net_device *orig_dev)
1548 {
1549 /* We don't really care about STP packets, we just listen for them for
1550 * mutual exclusion with the bridge module, so this just discards
1551 * them. */
1552 kfree_skb(skb);
1553 return 0;
1554 }
1555
1556 static int __init dp_init(void)
1557 {
1558 int err;
1559
1560 printk("Open vSwitch %s, built "__DATE__" "__TIME__"\n", VERSION BUILDNR);
1561
1562 /* Register to receive STP packets because the bridge module also
1563 * attempts to do so. Since there can only be a single listener for a
1564 * given protocol, this provides mutual exclusion against the bridge
1565 * module, preventing both of them from being loaded at the same
1566 * time. */
1567 dp_stp_sap = llc_sap_open(LLC_SAP_BSPAN, dp_stp_rcv);
1568 if (!dp_stp_sap) {
1569 printk(KERN_ERR "openvswitch: can't register sap for STP (probably the bridge module is loaded)\n");
1570 return -EADDRINUSE;
1571 }
1572
1573 err = flow_init();
1574 if (err)
1575 goto error;
1576
1577 err = register_netdevice_notifier(&dp_device_notifier);
1578 if (err)
1579 goto error_flow_exit;
1580
1581 major = register_chrdev(0, "openvswitch", &openvswitch_fops);
1582 if (err < 0)
1583 goto error_unreg_notifier;
1584
1585 /* Hook into callback used by the bridge to intercept packets.
1586 * Parasites we are. */
1587 br_handle_frame_hook = dp_frame_hook;
1588
1589 return 0;
1590
1591 error_unreg_notifier:
1592 unregister_netdevice_notifier(&dp_device_notifier);
1593 error_flow_exit:
1594 flow_exit();
1595 error:
1596 return err;
1597 }
1598
1599 static void dp_cleanup(void)
1600 {
1601 rcu_barrier();
1602 unregister_chrdev(major, "openvswitch");
1603 unregister_netdevice_notifier(&dp_device_notifier);
1604 flow_exit();
1605 br_handle_frame_hook = NULL;
1606 llc_sap_put(dp_stp_sap);
1607 }
1608
1609 module_init(dp_init);
1610 module_exit(dp_cleanup);
1611
1612 MODULE_DESCRIPTION("Open vSwitch switching datapath");
1613 MODULE_LICENSE("GPL");
openvswitch packages for PVE