]> git.proxmox.com Git - mirror_ovs.git/blob - datapath/datapath.c
projects / mirror_ovs.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Initial implementation of sFlow.
[mirror_ovs.git] / datapath / datapath.c
1 /*
2 * Copyright (c) 2007, 2008, 2009 Nicira Networks.
3 * Distributed under the terms of the GNU GPL version 2.
4 *
5 * Significant portions of this file may be copied from parts of the Linux
6 * kernel, by Linus Torvalds and others.
7 */
8
9 /* Functions for managing the dp interface/device. */
10
11 #include <linux/init.h>
12 #include <linux/module.h>
13 #include <linux/fs.h>
14 #include <linux/if_arp.h>
15 #include <linux/if_bridge.h>
16 #include <linux/if_vlan.h>
17 #include <linux/in.h>
18 #include <linux/ip.h>
19 #include <linux/delay.h>
20 #include <linux/time.h>
21 #include <linux/etherdevice.h>
22 #include <linux/kernel.h>
23 #include <linux/kthread.h>
24 #include <linux/llc.h>
25 #include <linux/mutex.h>
26 #include <linux/percpu.h>
27 #include <linux/rcupdate.h>
28 #include <linux/tcp.h>
29 #include <linux/udp.h>
30 #include <linux/version.h>
31 #include <linux/ethtool.h>
32 #include <linux/random.h>
33 #include <linux/wait.h>
34 #include <asm/system.h>
35 #include <asm/div64.h>
36 #include <asm/bug.h>
37 #include <linux/netfilter_bridge.h>
38 #include <linux/netfilter_ipv4.h>
39 #include <linux/inetdevice.h>
40 #include <linux/list.h>
41 #include <linux/rculist.h>
42 #include <linux/workqueue.h>
43 #include <linux/dmi.h>
44 #include <net/llc.h>
45
46 #include "openvswitch/datapath-protocol.h"
47 #include "datapath.h"
48 #include "actions.h"
49 #include "dp_dev.h"
50 #include "flow.h"
51
52 #include "compat.h"
53
54
55 int (*dp_ioctl_hook)(struct net_device *dev, struct ifreq *rq, int cmd);
56 EXPORT_SYMBOL(dp_ioctl_hook);
57
58 /* Datapaths. Protected on the read side by rcu_read_lock, on the write side
59 * by dp_mutex.
60 *
61 * dp_mutex nests inside the RTNL lock: if you need both you must take the RTNL
62 * lock first.
63 *
64 * It is safe to access the datapath and net_bridge_port structures with just
65 * dp_mutex.
66 */
67 static struct datapath *dps[ODP_MAX];
68 static DEFINE_MUTEX(dp_mutex);
69
70 /* Number of milliseconds between runs of the maintenance thread. */
71 #define MAINT_SLEEP_MSECS 1000
72
73 static int new_nbp(struct datapath *, struct net_device *, int port_no);
74
75 /* Must be called with rcu_read_lock or dp_mutex. */
76 struct datapath *get_dp(int dp_idx)
77 {
78 if (dp_idx < 0 || dp_idx >= ODP_MAX)
79 return NULL;
80 return rcu_dereference(dps[dp_idx]);
81 }
82 EXPORT_SYMBOL_GPL(get_dp);
83
84 struct datapath *get_dp_locked(int dp_idx)
85 {
86 struct datapath *dp;
87
88 mutex_lock(&dp_mutex);
89 dp = get_dp(dp_idx);
90 if (dp)
91 mutex_lock(&dp->mutex);
92 mutex_unlock(&dp_mutex);
93 return dp;
94 }
95
96 static inline size_t br_nlmsg_size(void)
97 {
98 return NLMSG_ALIGN(sizeof(struct ifinfomsg))
99 + nla_total_size(IFNAMSIZ) /* IFLA_IFNAME */
100 + nla_total_size(MAX_ADDR_LEN) /* IFLA_ADDRESS */
101 + nla_total_size(4) /* IFLA_MASTER */
102 + nla_total_size(4) /* IFLA_MTU */
103 + nla_total_size(4) /* IFLA_LINK */
104 + nla_total_size(1); /* IFLA_OPERSTATE */
105 }
106
107 static int dp_fill_ifinfo(struct sk_buff *skb,
108 const struct net_bridge_port *port,
109 int event, unsigned int flags)
110 {
111 const struct datapath *dp = port->dp;
112 const struct net_device *dev = port->dev;
113 struct ifinfomsg *hdr;
114 struct nlmsghdr *nlh;
115
116 nlh = nlmsg_put(skb, 0, 0, event, sizeof(*hdr), flags);
117 if (nlh == NULL)
118 return -EMSGSIZE;
119
120 hdr = nlmsg_data(nlh);
121 hdr->ifi_family = AF_BRIDGE;
122 hdr->__ifi_pad = 0;
123 hdr->ifi_type = dev->type;
124 hdr->ifi_index = dev->ifindex;
125 hdr->ifi_flags = dev_get_flags(dev);
126 hdr->ifi_change = 0;
127
128 NLA_PUT_STRING(skb, IFLA_IFNAME, dev->name);
129 NLA_PUT_U32(skb, IFLA_MASTER, dp->ports[ODPP_LOCAL]->dev->ifindex);
130 NLA_PUT_U32(skb, IFLA_MTU, dev->mtu);
131 #ifdef IFLA_OPERSTATE
132 NLA_PUT_U8(skb, IFLA_OPERSTATE,
133 netif_running(dev) ? dev->operstate : IF_OPER_DOWN);
134 #endif
135
136 if (dev->addr_len)
137 NLA_PUT(skb, IFLA_ADDRESS, dev->addr_len, dev->dev_addr);
138
139 if (dev->ifindex != dev->iflink)
140 NLA_PUT_U32(skb, IFLA_LINK, dev->iflink);
141
142 return nlmsg_end(skb, nlh);
143
144 nla_put_failure:
145 nlmsg_cancel(skb, nlh);
146 return -EMSGSIZE;
147 }
148
149 static void dp_ifinfo_notify(int event, struct net_bridge_port *port)
150 {
151 struct net *net = dev_net(port->dev);
152 struct sk_buff *skb;
153 int err = -ENOBUFS;
154
155 skb = nlmsg_new(br_nlmsg_size(), GFP_KERNEL);
156 if (skb == NULL)
157 goto errout;
158
159 err = dp_fill_ifinfo(skb, port, event, 0);
160 if (err < 0) {
161 /* -EMSGSIZE implies BUG in br_nlmsg_size() */
162 WARN_ON(err == -EMSGSIZE);
163 kfree_skb(skb);
164 goto errout;
165 }
166 rtnl_notify(skb, net, 0, RTNLGRP_LINK, NULL, GFP_KERNEL);
167 return;
168 errout:
169 if (err < 0)
170 rtnl_set_sk_err(net, RTNLGRP_LINK, err);
171 }
172
173 static void release_dp(struct kobject *kobj)
174 {
175 struct datapath *dp = container_of(kobj, struct datapath, ifobj);
176 kfree(dp);
177 }
178
179 struct kobj_type dp_ktype = {
180 .release = release_dp
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 INIT_LIST_HEAD(&dp->port_list);
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 /* Initialize kobject for bridge. This will be added as
225 * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
226 dp->ifobj.kset = NULL;
227 kobject_init(&dp->ifobj, &dp_ktype);
228
229 /* Allocate table. */
230 err = -ENOMEM;
231 rcu_assign_pointer(dp->table, dp_table_create(DP_L1_SIZE));
232 if (!dp->table)
233 goto err_free_dp;
234
235 /* Set up our datapath device. */
236 dp_dev = dp_dev_create(dp, devname, ODPP_LOCAL);
237 err = PTR_ERR(dp_dev);
238 if (IS_ERR(dp_dev))
239 goto err_destroy_table;
240
241 err = new_nbp(dp, dp_dev, ODPP_LOCAL);
242 if (err) {
243 dp_dev_destroy(dp_dev);
244 goto err_destroy_table;
245 }
246
247 dp->drop_frags = 0;
248 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
249 if (!dp->stats_percpu)
250 goto err_destroy_local_port;
251
252 rcu_assign_pointer(dps[dp_idx], dp);
253 mutex_unlock(&dp_mutex);
254 rtnl_unlock();
255
256 dp_sysfs_add_dp(dp);
257
258 return 0;
259
260 err_destroy_local_port:
261 dp_del_port(dp->ports[ODPP_LOCAL]);
262 err_destroy_table:
263 dp_table_destroy(dp->table, 0);
264 err_free_dp:
265 kfree(dp);
266 err_put_module:
267 module_put(THIS_MODULE);
268 err_unlock:
269 mutex_unlock(&dp_mutex);
270 rtnl_unlock();
271 err:
272 return err;
273 }
274
275 static void do_destroy_dp(struct datapath *dp)
276 {
277 struct net_bridge_port *p, *n;
278 int i;
279
280 list_for_each_entry_safe (p, n, &dp->port_list, node)
281 if (p->port_no != ODPP_LOCAL)
282 dp_del_port(p);
283
284 dp_sysfs_del_dp(dp);
285
286 rcu_assign_pointer(dps[dp->dp_idx], NULL);
287
288 dp_del_port(dp->ports[ODPP_LOCAL]);
289
290 dp_table_destroy(dp->table, 1);
291
292 for (i = 0; i < DP_N_QUEUES; i++)
293 skb_queue_purge(&dp->queues[i]);
294 for (i = 0; i < DP_MAX_GROUPS; i++)
295 kfree(dp->groups[i]);
296 free_percpu(dp->stats_percpu);
297 kobject_put(&dp->ifobj);
298 module_put(THIS_MODULE);
299 }
300
301 static int destroy_dp(int dp_idx)
302 {
303 struct datapath *dp;
304 int err;
305
306 rtnl_lock();
307 mutex_lock(&dp_mutex);
308 dp = get_dp(dp_idx);
309 err = -ENODEV;
310 if (!dp)
311 goto err_unlock;
312
313 do_destroy_dp(dp);
314 err = 0;
315
316 err_unlock:
317 mutex_unlock(&dp_mutex);
318 rtnl_unlock();
319 return err;
320 }
321
322 static void release_nbp(struct kobject *kobj)
323 {
324 struct net_bridge_port *p = container_of(kobj, struct net_bridge_port, kobj);
325 kfree(p);
326 }
327
328 struct kobj_type brport_ktype = {
329 #ifdef CONFIG_SYSFS
330 .sysfs_ops = &brport_sysfs_ops,
331 #endif
332 .release = release_nbp
333 };
334
335 /* Called with RTNL lock and dp_mutex. */
336 static int new_nbp(struct datapath *dp, struct net_device *dev, int port_no)
337 {
338 struct net_bridge_port *p;
339
340 if (dev->br_port != NULL)
341 return -EBUSY;
342
343 p = kzalloc(sizeof(*p), GFP_KERNEL);
344 if (!p)
345 return -ENOMEM;
346
347 dev_set_promiscuity(dev, 1);
348 dev_hold(dev);
349 p->port_no = port_no;
350 p->dp = dp;
351 p->dev = dev;
352 if (!is_dp_dev(dev))
353 rcu_assign_pointer(dev->br_port, p);
354 else {
355 /* It would make sense to assign dev->br_port here too, but
356 * that causes packets received on internal ports to get caught
357 * in dp_frame_hook(). In turn dp_frame_hook() can reject them
358 * back to network stack, but that's a waste of time. */
359 }
360 rcu_assign_pointer(dp->ports[port_no], p);
361 list_add_rcu(&p->node, &dp->port_list);
362 dp->n_ports++;
363
364 /* Initialize kobject for bridge. This will be added as
365 * /sys/class/net/<devname>/brport later, if sysfs is enabled. */
366 p->kobj.kset = NULL;
367 kobject_init(&p->kobj, &brport_ktype);
368
369 dp_ifinfo_notify(RTM_NEWLINK, p);
370
371 return 0;
372 }
373
374 static int add_port(int dp_idx, struct odp_port __user *portp)
375 {
376 struct net_device *dev;
377 struct datapath *dp;
378 struct odp_port port;
379 int port_no;
380 int err;
381
382 err = -EFAULT;
383 if (copy_from_user(&port, portp, sizeof port))
384 goto out;
385 port.devname[IFNAMSIZ - 1] = '\0';
386
387 rtnl_lock();
388 dp = get_dp_locked(dp_idx);
389 err = -ENODEV;
390 if (!dp)
391 goto out_unlock_rtnl;
392
393 for (port_no = 1; port_no < DP_MAX_PORTS; port_no++)
394 if (!dp->ports[port_no])
395 goto got_port_no;
396 err = -EFBIG;
397 goto out_unlock_dp;
398
399 got_port_no:
400 if (!(port.flags & ODP_PORT_INTERNAL)) {
401 err = -ENODEV;
402 dev = dev_get_by_name(&init_net, port.devname);
403 if (!dev)
404 goto out_unlock_dp;
405
406 err = -EINVAL;
407 if (dev->flags & IFF_LOOPBACK || dev->type != ARPHRD_ETHER ||
408 is_dp_dev(dev))
409 goto out_put;
410 } else {
411 dev = dp_dev_create(dp, port.devname, port_no);
412 err = PTR_ERR(dev);
413 if (IS_ERR(dev))
414 goto out_unlock_dp;
415 dev_hold(dev);
416 }
417
418 err = new_nbp(dp, dev, port_no);
419 if (err)
420 goto out_put;
421
422 dp_sysfs_add_if(dp->ports[port_no]);
423
424 err = __put_user(port_no, &port.port);
425
426 out_put:
427 dev_put(dev);
428 out_unlock_dp:
429 mutex_unlock(&dp->mutex);
430 out_unlock_rtnl:
431 rtnl_unlock();
432 out:
433 return err;
434 }
435
436 int dp_del_port(struct net_bridge_port *p)
437 {
438 ASSERT_RTNL();
439
440 if (p->port_no != ODPP_LOCAL)
441 dp_sysfs_del_if(p);
442 dp_ifinfo_notify(RTM_DELLINK, p);
443
444 p->dp->n_ports--;
445
446 if (is_dp_dev(p->dev)) {
447 /* Make sure that no packets arrive from now on, since
448 * dp_dev_xmit() will try to find itself through
449 * p->dp->ports[], and we're about to set that to null. */
450 netif_tx_disable(p->dev);
451 }
452
453 /* First drop references to device. */
454 dev_set_promiscuity(p->dev, -1);
455 list_del_rcu(&p->node);
456 rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
457 rcu_assign_pointer(p->dev->br_port, NULL);
458
459 /* Then wait until no one is still using it, and destroy it. */
460 synchronize_rcu();
461
462 if (is_dp_dev(p->dev))
463 dp_dev_destroy(p->dev);
464 dev_put(p->dev);
465 kobject_put(&p->kobj);
466
467 return 0;
468 }
469
470 static int del_port(int dp_idx, int port_no)
471 {
472 struct net_bridge_port *p;
473 struct datapath *dp;
474 LIST_HEAD(dp_devs);
475 int err;
476
477 err = -EINVAL;
478 if (port_no < 0 || port_no >= DP_MAX_PORTS || port_no == ODPP_LOCAL)
479 goto out;
480
481 rtnl_lock();
482 dp = get_dp_locked(dp_idx);
483 err = -ENODEV;
484 if (!dp)
485 goto out_unlock_rtnl;
486
487 p = dp->ports[port_no];
488 err = -ENOENT;
489 if (!p)
490 goto out_unlock_dp;
491
492 err = dp_del_port(p);
493
494 out_unlock_dp:
495 mutex_unlock(&dp->mutex);
496 out_unlock_rtnl:
497 rtnl_unlock();
498 out:
499 return err;
500 }
501
502 /* Must be called with rcu_read_lock. */
503 static void
504 do_port_input(struct net_bridge_port *p, struct sk_buff *skb)
505 {
506 /* Make our own copy of the packet. Otherwise we will mangle the
507 * packet for anyone who came before us (e.g. tcpdump via AF_PACKET).
508 * (No one comes after us, since we tell handle_bridge() that we took
509 * the packet.) */
510 skb = skb_share_check(skb, GFP_ATOMIC);
511 if (!skb)
512 return;
513
514 /* Push the Ethernet header back on. */
515 skb_push(skb, ETH_HLEN);
516 skb_reset_mac_header(skb);
517 dp_process_received_packet(skb, p);
518 }
519
520 /* Must be called with rcu_read_lock and with bottom-halves disabled. */
521 void dp_process_received_packet(struct sk_buff *skb, struct net_bridge_port *p)
522 {
523 struct datapath *dp = p->dp;
524 struct dp_stats_percpu *stats;
525 struct odp_flow_key key;
526 struct sw_flow *flow;
527
528 WARN_ON_ONCE(skb_shared(skb));
529
530 /* BHs are off so we don't have to use get_cpu()/put_cpu() here. */
531 stats = percpu_ptr(dp->stats_percpu, smp_processor_id());
532
533 if (flow_extract(skb, p ? p->port_no : ODPP_NONE, &key)) {
534 if (dp->drop_frags) {
535 kfree_skb(skb);
536 stats->n_frags++;
537 return;
538 }
539 }
540
541 flow = dp_table_lookup(rcu_dereference(dp->table), &key);
542 if (flow) {
543 struct sw_flow_actions *acts = rcu_dereference(flow->sf_acts);
544 flow_used(flow, skb);
545 execute_actions(dp, skb, &key, acts->actions, acts->n_actions,
546 GFP_ATOMIC);
547 stats->n_hit++;
548 } else {
549 stats->n_missed++;
550 dp_output_control(dp, skb, _ODPL_MISS_NR, 0);
551 }
552 }
553
554 /*
555 * Used as br_handle_frame_hook. (Cannot run bridge at the same time, even on
556 * different set of devices!)
557 */
558 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,22)
559 /* Called with rcu_read_lock and bottom-halves disabled. */
560 static struct sk_buff *dp_frame_hook(struct net_bridge_port *p,
561 struct sk_buff *skb)
562 {
563 do_port_input(p, skb);
564 return NULL;
565 }
566 #elif LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,0)
567 /* Called with rcu_read_lock and bottom-halves disabled. */
568 static int dp_frame_hook(struct net_bridge_port *p, struct sk_buff **pskb)
569 {
570 do_port_input(p, *pskb);
571 return 1;
572 }
573 #else
574 #error
575 #endif
576
577 #if defined(CONFIG_XEN) && defined(HAVE_PROTO_DATA_VALID)
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 vswitch_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 #else
625 int vswitch_skb_checksum_setup(struct sk_buff *skb) { return 0; }
626 #endif /* CONFIG_XEN && linux == 2.6.18 */
627
628 /* Append each packet in 'skb' list to 'queue'. There will be only one packet
629 * unless we broke up a GSO packet. */
630 static int
631 queue_control_packets(struct sk_buff *skb, struct sk_buff_head *queue,
632 int queue_no, u32 arg)
633 {
634 struct sk_buff *nskb;
635 int port_no;
636 int err;
637
638 port_no = ODPP_LOCAL;
639 if (skb->dev) {
640 if (skb->dev->br_port)
641 port_no = skb->dev->br_port->port_no;
642 else if (is_dp_dev(skb->dev))
643 port_no = dp_dev_priv(skb->dev)->port_no;
644 }
645
646 do {
647 struct odp_msg *header;
648
649 nskb = skb->next;
650 skb->next = NULL;
651
652 /* If a checksum-deferred packet is forwarded to the
653 * controller, correct the pointers and checksum. This happens
654 * on a regular basis only on Xen, on which VMs can pass up
655 * packets that do not have their checksum computed.
656 */
657 err = vswitch_skb_checksum_setup(skb);
658 if (err)
659 goto err_kfree_skbs;
660 #ifndef CHECKSUM_HW
661 if (skb->ip_summed == CHECKSUM_PARTIAL) {
662 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,22)
663 /* Until 2.6.22, the start of the transport header was
664 * also the start of data to be checksummed. Linux
665 * 2.6.22 introduced the csum_start field for this
666 * purpose, but we should point the transport header to
667 * it anyway for backward compatibility, as
668 * dev_queue_xmit() does even in 2.6.28. */
669 skb_set_transport_header(skb, skb->csum_start -
670 skb_headroom(skb));
671 #endif
672 err = skb_checksum_help(skb);
673 if (err)
674 goto err_kfree_skbs;
675 }
676 #else
677 if (skb->ip_summed == CHECKSUM_HW) {
678 err = skb_checksum_help(skb, 0);
679 if (err)
680 goto err_kfree_skbs;
681 }
682 #endif
683
684 err = skb_cow(skb, sizeof *header);
685 if (err)
686 goto err_kfree_skbs;
687
688 header = (struct odp_msg*)__skb_push(skb, sizeof *header);
689 header->type = queue_no;
690 header->length = skb->len;
691 header->port = port_no;
692 header->reserved = 0;
693 header->arg = arg;
694 skb_queue_tail(queue, skb);
695
696 skb = nskb;
697 } while (skb);
698 return 0;
699
700 err_kfree_skbs:
701 kfree_skb(skb);
702 while ((skb = nskb) != NULL) {
703 nskb = skb->next;
704 kfree_skb(skb);
705 }
706 return err;
707 }
708
709 int
710 dp_output_control(struct datapath *dp, struct sk_buff *skb, int queue_no,
711 u32 arg)
712 {
713 struct dp_stats_percpu *stats;
714 struct sk_buff_head *queue;
715 int err;
716
717 WARN_ON_ONCE(skb_shared(skb));
718 BUG_ON(queue_no != _ODPL_MISS_NR && queue_no != _ODPL_ACTION_NR && queue_no != _ODPL_SFLOW_NR);
719 queue = &dp->queues[queue_no];
720 err = -ENOBUFS;
721 if (skb_queue_len(queue) >= DP_MAX_QUEUE_LEN)
722 goto err_kfree_skb;
723
724 /* Break apart GSO packets into their component pieces. Otherwise
725 * userspace may try to stuff a 64kB packet into a 1500-byte MTU. */
726 if (skb_is_gso(skb)) {
727 struct sk_buff *nskb = skb_gso_segment(skb, 0);
728 if (nskb) {
729 kfree_skb(skb);
730 skb = nskb;
731 if (unlikely(IS_ERR(skb))) {
732 err = PTR_ERR(skb);
733 goto err;
734 }
735 } else {
736 /* XXX This case might not be possible. It's hard to
737 * tell from the skb_gso_segment() code and comment. */
738 }
739 }
740
741 err = queue_control_packets(skb, queue, queue_no, arg);
742 wake_up_interruptible(&dp->waitqueue);
743 return err;
744
745 err_kfree_skb:
746 kfree_skb(skb);
747 err:
748 stats = percpu_ptr(dp->stats_percpu, get_cpu());
749 stats->n_lost++;
750 put_cpu();
751
752 return err;
753 }
754
755 static int flush_flows(struct datapath *dp)
756 {
757 dp->n_flows = 0;
758 return dp_table_flush(dp);
759 }
760
761 static int validate_actions(const struct sw_flow_actions *actions)
762 {
763 unsigned int i;
764
765 for (i = 0; i < actions->n_actions; i++) {
766 const union odp_action *a = &actions->actions[i];
767 switch (a->type) {
768 case ODPAT_OUTPUT:
769 if (a->output.port >= DP_MAX_PORTS)
770 return -EINVAL;
771 break;
772
773 case ODPAT_OUTPUT_GROUP:
774 if (a->output_group.group >= DP_MAX_GROUPS)
775 return -EINVAL;
776 break;
777
778 case ODPAT_SET_VLAN_VID:
779 if (a->vlan_vid.vlan_vid & htons(~VLAN_VID_MASK))
780 return -EINVAL;
781 break;
782
783 case ODPAT_SET_VLAN_PCP:
784 if (a->vlan_pcp.vlan_pcp
785 & ~(VLAN_PCP_MASK >> VLAN_PCP_SHIFT))
786 return -EINVAL;
787 break;
788
789 default:
790 if (a->type >= ODPAT_N_ACTIONS)
791 return -EOPNOTSUPP;
792 break;
793 }
794 }
795
796 return 0;
797 }
798
799 static struct sw_flow_actions *get_actions(const struct odp_flow *flow)
800 {
801 struct sw_flow_actions *actions;
802 int error;
803
804 actions = flow_actions_alloc(flow->n_actions);
805 error = PTR_ERR(actions);
806 if (IS_ERR(actions))
807 goto error;
808
809 error = -EFAULT;
810 if (copy_from_user(actions->actions, flow->actions,
811 flow->n_actions * sizeof(union odp_action)))
812 goto error_free_actions;
813 error = validate_actions(actions);
814 if (error)
815 goto error_free_actions;
816
817 return actions;
818
819 error_free_actions:
820 kfree(actions);
821 error:
822 return ERR_PTR(error);
823 }
824
825 static void get_stats(struct sw_flow *flow, struct odp_flow_stats *stats)
826 {
827 if (flow->used.tv_sec) {
828 stats->used_sec = flow->used.tv_sec;
829 stats->used_nsec = flow->used.tv_nsec;
830 } else {
831 stats->used_sec = 0;
832 stats->used_nsec = 0;
833 }
834 stats->n_packets = flow->packet_count;
835 stats->n_bytes = flow->byte_count;
836 stats->ip_tos = flow->ip_tos;
837 stats->tcp_flags = flow->tcp_flags;
838 stats->error = 0;
839 }
840
841 static void clear_stats(struct sw_flow *flow)
842 {
843 flow->used.tv_sec = flow->used.tv_nsec = 0;
844 flow->tcp_flags = 0;
845 flow->ip_tos = 0;
846 flow->packet_count = 0;
847 flow->byte_count = 0;
848 }
849
850 static int put_flow(struct datapath *dp, struct odp_flow_put __user *ufp)
851 {
852 struct odp_flow_put uf;
853 struct sw_flow *flow;
854 struct dp_table *table;
855 struct odp_flow_stats stats;
856 int error;
857
858 error = -EFAULT;
859 if (copy_from_user(&uf, ufp, sizeof(struct odp_flow_put)))
860 goto error;
861 uf.flow.key.reserved = 0;
862
863 table = rcu_dereference(dp->table);
864 flow = dp_table_lookup(table, &uf.flow.key);
865 if (!flow) {
866 /* No such flow. */
867 struct sw_flow_actions *acts;
868
869 error = -ENOENT;
870 if (!(uf.flags & ODPPF_CREATE))
871 goto error;
872
873 /* Expand table, if necessary, to make room. */
874 if (dp->n_flows >= table->n_buckets) {
875 error = -ENOSPC;
876 if (table->n_buckets >= DP_MAX_BUCKETS)
877 goto error;
878
879 error = dp_table_expand(dp);
880 if (error)
881 goto error;
882 table = rcu_dereference(dp->table);
883 }
884
885 /* Allocate flow. */
886 error = -ENOMEM;
887 flow = kmem_cache_alloc(flow_cache, GFP_KERNEL);
888 if (flow == NULL)
889 goto error;
890 flow->key = uf.flow.key;
891 spin_lock_init(&flow->lock);
892 clear_stats(flow);
893
894 /* Obtain actions. */
895 acts = get_actions(&uf.flow);
896 error = PTR_ERR(acts);
897 if (IS_ERR(acts))
898 goto error_free_flow;
899 rcu_assign_pointer(flow->sf_acts, acts);
900
901 /* Put flow in bucket. */
902 error = dp_table_insert(table, flow);
903 if (error)
904 goto error_free_flow_acts;
905 dp->n_flows++;
906 memset(&stats, 0, sizeof(struct odp_flow_stats));
907 } else {
908 /* We found a matching flow. */
909 struct sw_flow_actions *old_acts, *new_acts;
910 unsigned long int flags;
911
912 /* Bail out if we're not allowed to modify an existing flow. */
913 error = -EEXIST;
914 if (!(uf.flags & ODPPF_MODIFY))
915 goto error;
916
917 /* Swap actions. */
918 new_acts = get_actions(&uf.flow);
919 error = PTR_ERR(new_acts);
920 if (IS_ERR(new_acts))
921 goto error;
922 old_acts = rcu_dereference(flow->sf_acts);
923 if (old_acts->n_actions != new_acts->n_actions ||
924 memcmp(old_acts->actions, new_acts->actions,
925 sizeof(union odp_action) * old_acts->n_actions)) {
926 rcu_assign_pointer(flow->sf_acts, new_acts);
927 flow_deferred_free_acts(old_acts);
928 } else {
929 kfree(new_acts);
930 }
931
932 /* Fetch stats, then clear them if necessary. */
933 spin_lock_irqsave(&flow->lock, flags);
934 get_stats(flow, &stats);
935 if (uf.flags & ODPPF_ZERO_STATS)
936 clear_stats(flow);
937 spin_unlock_irqrestore(&flow->lock, flags);
938 }
939
940 /* Copy stats to userspace. */
941 if (__copy_to_user(&ufp->flow.stats, &stats,
942 sizeof(struct odp_flow_stats)))
943 return -EFAULT;
944 return 0;
945
946 error_free_flow_acts:
947 kfree(flow->sf_acts);
948 error_free_flow:
949 kmem_cache_free(flow_cache, flow);
950 error:
951 return error;
952 }
953
954 static int put_actions(const struct sw_flow *flow, struct odp_flow __user *ufp)
955 {
956 union odp_action __user *actions;
957 struct sw_flow_actions *sf_acts;
958 u32 n_actions;
959
960 if (__get_user(actions, &ufp->actions) ||
961 __get_user(n_actions, &ufp->n_actions))
962 return -EFAULT;
963
964 if (!n_actions)
965 return 0;
966
967 sf_acts = rcu_dereference(flow->sf_acts);
968 if (__put_user(sf_acts->n_actions, &ufp->n_actions) ||
969 (actions && copy_to_user(actions, sf_acts->actions,
970 sizeof(union odp_action) *
971 min(sf_acts->n_actions, n_actions))))
972 return -EFAULT;
973
974 return 0;
975 }
976
977 static int answer_query(struct sw_flow *flow, u32 query_flags,
978 struct odp_flow __user *ufp)
979 {
980 struct odp_flow_stats stats;
981 unsigned long int flags;
982
983 spin_lock_irqsave(&flow->lock, flags);
984 get_stats(flow, &stats);
985
986 if (query_flags & ODPFF_ZERO_TCP_FLAGS) {
987 flow->tcp_flags = 0;
988 }
989 spin_unlock_irqrestore(&flow->lock, flags);
990
991 if (__copy_to_user(&ufp->stats, &stats, sizeof(struct odp_flow_stats)))
992 return -EFAULT;
993 return put_actions(flow, ufp);
994 }
995
996 static int del_flow(struct datapath *dp, struct odp_flow __user *ufp)
997 {
998 struct dp_table *table = rcu_dereference(dp->table);
999 struct odp_flow uf;
1000 struct sw_flow *flow;
1001 int error;
1002
1003 error = -EFAULT;
1004 if (copy_from_user(&uf, ufp, sizeof uf))
1005 goto error;
1006 uf.key.reserved = 0;
1007
1008 flow = dp_table_lookup(table, &uf.key);
1009 error = -ENOENT;
1010 if (!flow)
1011 goto error;
1012
1013 /* XXX redundant lookup */
1014 error = dp_table_delete(table, flow);
1015 if (error)
1016 goto error;
1017
1018 /* XXX These statistics might lose a few packets, since other CPUs can
1019 * be using this flow. We used to synchronize_rcu() to make sure that
1020 * we get completely accurate stats, but that blows our performance,
1021 * badly. */
1022 dp->n_flows--;
1023 error = answer_query(flow, 0, ufp);
1024 flow_deferred_free(flow);
1025
1026 error:
1027 return error;
1028 }
1029
1030 static int query_flows(struct datapath *dp, const struct odp_flowvec *flowvec)
1031 {
1032 struct dp_table *table = rcu_dereference(dp->table);
1033 int i;
1034 for (i = 0; i < flowvec->n_flows; i++) {
1035 struct __user odp_flow *ufp = &flowvec->flows[i];
1036 struct odp_flow uf;
1037 struct sw_flow *flow;
1038 int error;
1039
1040 if (__copy_from_user(&uf, ufp, sizeof uf))
1041 return -EFAULT;
1042 uf.key.reserved = 0;
1043
1044 flow = dp_table_lookup(table, &uf.key);
1045 if (!flow)
1046 error = __put_user(ENOENT, &ufp->stats.error);
1047 else
1048 error = answer_query(flow, uf.flags, ufp);
1049 if (error)
1050 return -EFAULT;
1051 }
1052 return flowvec->n_flows;
1053 }
1054
1055 struct list_flows_cbdata {
1056 struct odp_flow __user *uflows;
1057 int n_flows;
1058 int listed_flows;
1059 };
1060
1061 static int list_flow(struct sw_flow *flow, void *cbdata_)
1062 {
1063 struct list_flows_cbdata *cbdata = cbdata_;
1064 struct odp_flow __user *ufp = &cbdata->uflows[cbdata->listed_flows++];
1065 int error;
1066
1067 if (__copy_to_user(&ufp->key, &flow->key, sizeof flow->key))
1068 return -EFAULT;
1069 error = answer_query(flow, 0, ufp);
1070 if (error)
1071 return error;
1072
1073 if (cbdata->listed_flows >= cbdata->n_flows)
1074 return cbdata->listed_flows;
1075 return 0;
1076 }
1077
1078 static int list_flows(struct datapath *dp, const struct odp_flowvec *flowvec)
1079 {
1080 struct list_flows_cbdata cbdata;
1081 int error;
1082
1083 if (!flowvec->n_flows)
1084 return 0;
1085
1086 cbdata.uflows = flowvec->flows;
1087 cbdata.n_flows = flowvec->n_flows;
1088 cbdata.listed_flows = 0;
1089 error = dp_table_foreach(rcu_dereference(dp->table),
1090 list_flow, &cbdata);
1091 return error ? error : cbdata.listed_flows;
1092 }
1093
1094 static int do_flowvec_ioctl(struct datapath *dp, unsigned long argp,
1095 int (*function)(struct datapath *,
1096 const struct odp_flowvec *))
1097 {
1098 struct odp_flowvec __user *uflowvec;
1099 struct odp_flowvec flowvec;
1100 int retval;
1101
1102 uflowvec = (struct odp_flowvec __user *)argp;
1103 if (!access_ok(VERIFY_WRITE, uflowvec, sizeof *uflowvec) ||
1104 copy_from_user(&flowvec, uflowvec, sizeof flowvec))
1105 return -EFAULT;
1106
1107 if (flowvec.n_flows > INT_MAX / sizeof(struct odp_flow))
1108 return -EINVAL;
1109
1110 if (!access_ok(VERIFY_WRITE, flowvec.flows,
1111 flowvec.n_flows * sizeof(struct odp_flow)))
1112 return -EFAULT;
1113
1114 retval = function(dp, &flowvec);
1115 return (retval < 0 ? retval
1116 : retval == flowvec.n_flows ? 0
1117 : __put_user(retval, &uflowvec->n_flows));
1118 }
1119
1120 static int do_execute(struct datapath *dp, const struct odp_execute *executep)
1121 {
1122 struct odp_execute execute;
1123 struct odp_flow_key key;
1124 struct sk_buff *skb;
1125 struct sw_flow_actions *actions;
1126 struct ethhdr *eth;
1127 int err;
1128
1129 err = -EFAULT;
1130 if (copy_from_user(&execute, executep, sizeof execute))
1131 goto error;
1132
1133 err = -EINVAL;
1134 if (execute.length < ETH_HLEN || execute.length > 65535)
1135 goto error;
1136
1137 err = -ENOMEM;
1138 actions = flow_actions_alloc(execute.n_actions);
1139 if (!actions)
1140 goto error;
1141
1142 err = -EFAULT;
1143 if (copy_from_user(actions->actions, execute.actions,
1144 execute.n_actions * sizeof *execute.actions))
1145 goto error_free_actions;
1146
1147 err = validate_actions(actions);
1148 if (err)
1149 goto error_free_actions;
1150
1151 err = -ENOMEM;
1152 skb = alloc_skb(execute.length, GFP_KERNEL);
1153 if (!skb)
1154 goto error_free_actions;
1155 if (execute.in_port < DP_MAX_PORTS) {
1156 struct net_bridge_port *p = dp->ports[execute.in_port];
1157 if (p)
1158 skb->dev = p->dev;
1159 }
1160
1161 err = -EFAULT;
1162 if (copy_from_user(skb_put(skb, execute.length), execute.data,
1163 execute.length))
1164 goto error_free_skb;
1165
1166 skb_reset_mac_header(skb);
1167 eth = eth_hdr(skb);
1168
1169 /* Normally, setting the skb 'protocol' field would be handled by a
1170 * call to eth_type_trans(), but it assumes there's a sending
1171 * device, which we may not have. */
1172 if (ntohs(eth->h_proto) >= 1536)
1173 skb->protocol = eth->h_proto;
1174 else
1175 skb->protocol = htons(ETH_P_802_2);
1176
1177 flow_extract(skb, execute.in_port, &key);
1178 err = execute_actions(dp, skb, &key, actions->actions,
1179 actions->n_actions, GFP_KERNEL);
1180 kfree(actions);
1181 return err;
1182
1183 error_free_skb:
1184 kfree_skb(skb);
1185 error_free_actions:
1186 kfree(actions);
1187 error:
1188 return err;
1189 }
1190
1191 static int get_dp_stats(struct datapath *dp, struct odp_stats __user *statsp)
1192 {
1193 struct odp_stats stats;
1194 int i;
1195
1196 stats.n_flows = dp->n_flows;
1197 stats.cur_capacity = rcu_dereference(dp->table)->n_buckets;
1198 stats.max_capacity = DP_MAX_BUCKETS;
1199 stats.n_ports = dp->n_ports;
1200 stats.max_ports = DP_MAX_PORTS;
1201 stats.max_groups = DP_MAX_GROUPS;
1202 stats.n_frags = stats.n_hit = stats.n_missed = stats.n_lost = 0;
1203 for_each_possible_cpu(i) {
1204 const struct dp_stats_percpu *s;
1205 s = percpu_ptr(dp->stats_percpu, i);
1206 stats.n_frags += s->n_frags;
1207 stats.n_hit += s->n_hit;
1208 stats.n_missed += s->n_missed;
1209 stats.n_lost += s->n_lost;
1210 }
1211 stats.max_miss_queue = DP_MAX_QUEUE_LEN;
1212 stats.max_action_queue = DP_MAX_QUEUE_LEN;
1213 return copy_to_user(statsp, &stats, sizeof stats) ? -EFAULT : 0;
1214 }
1215
1216 /* MTU of the dp pseudo-device: ETH_DATA_LEN or the minimum of the ports */
1217 int dp_min_mtu(const struct datapath *dp)
1218 {
1219 struct net_bridge_port *p;
1220 int mtu = 0;
1221
1222 ASSERT_RTNL();
1223
1224 list_for_each_entry_rcu (p, &dp->port_list, node) {
1225 struct net_device *dev = p->dev;
1226
1227 /* Skip any internal ports, since that's what we're trying to
1228 * set. */
1229 if (is_dp_dev(dev))
1230 continue;
1231
1232 if (!mtu || dev->mtu < mtu)
1233 mtu = dev->mtu;
1234 }
1235
1236 return mtu ? mtu : ETH_DATA_LEN;
1237 }
1238
1239 static int
1240 put_port(const struct net_bridge_port *p, struct odp_port __user *uop)
1241 {
1242 struct odp_port op;
1243 memset(&op, 0, sizeof op);
1244 strncpy(op.devname, p->dev->name, sizeof op.devname);
1245 op.port = p->port_no;
1246 op.flags = is_dp_dev(p->dev) ? ODP_PORT_INTERNAL : 0;
1247 return copy_to_user(uop, &op, sizeof op) ? -EFAULT : 0;
1248 }
1249
1250 static int
1251 query_port(struct datapath *dp, struct odp_port __user *uport)
1252 {
1253 struct odp_port port;
1254
1255 if (copy_from_user(&port, uport, sizeof port))
1256 return -EFAULT;
1257 if (port.devname[0]) {
1258 struct net_bridge_port *p;
1259 struct net_device *dev;
1260 int err;
1261
1262 port.devname[IFNAMSIZ - 1] = '\0';
1263
1264 dev = dev_get_by_name(&init_net, port.devname);
1265 if (!dev)
1266 return -ENODEV;
1267
1268 p = dev->br_port;
1269 if (!p && is_dp_dev(dev)) {
1270 struct dp_dev *dp_dev = dp_dev_priv(dev);
1271 if (dp_dev->dp == dp)
1272 p = dp->ports[dp_dev->port_no];
1273 }
1274 err = p && p->dp == dp ? put_port(p, uport) : -ENOENT;
1275 dev_put(dev);
1276
1277 return err;
1278 } else {
1279 if (port.port >= DP_MAX_PORTS)
1280 return -EINVAL;
1281 if (!dp->ports[port.port])
1282 return -ENOENT;
1283 return put_port(dp->ports[port.port], uport);
1284 }
1285 }
1286
1287 static int
1288 list_ports(struct datapath *dp, struct odp_portvec __user *pvp)
1289 {
1290 struct odp_portvec pv;
1291 struct net_bridge_port *p;
1292 int idx;
1293
1294 if (copy_from_user(&pv, pvp, sizeof pv))
1295 return -EFAULT;
1296
1297 idx = 0;
1298 if (pv.n_ports) {
1299 list_for_each_entry_rcu (p, &dp->port_list, node) {
1300 if (put_port(p, &pv.ports[idx]))
1301 return -EFAULT;
1302 if (idx++ >= pv.n_ports)
1303 break;
1304 }
1305 }
1306 return put_user(dp->n_ports, &pvp->n_ports);
1307 }
1308
1309 /* RCU callback for freeing a dp_port_group */
1310 static void free_port_group(struct rcu_head *rcu)
1311 {
1312 struct dp_port_group *g = container_of(rcu, struct dp_port_group, rcu);
1313 kfree(g);
1314 }
1315
1316 static int
1317 set_port_group(struct datapath *dp, const struct odp_port_group __user *upg)
1318 {
1319 struct odp_port_group pg;
1320 struct dp_port_group *new_group, *old_group;
1321 int error;
1322
1323 error = -EFAULT;
1324 if (copy_from_user(&pg, upg, sizeof pg))
1325 goto error;
1326
1327 error = -EINVAL;
1328 if (pg.n_ports > DP_MAX_PORTS || pg.group >= DP_MAX_GROUPS)
1329 goto error;
1330
1331 error = -ENOMEM;
1332 new_group = kmalloc(sizeof *new_group + sizeof(u16) * pg.n_ports,
1333 GFP_KERNEL);
1334 if (!new_group)
1335 goto error;
1336
1337 new_group->n_ports = pg.n_ports;
1338 error = -EFAULT;
1339 if (copy_from_user(new_group->ports, pg.ports,
1340 sizeof(u16) * pg.n_ports))
1341 goto error_free;
1342
1343 old_group = rcu_dereference(dp->groups[pg.group]);
1344 rcu_assign_pointer(dp->groups[pg.group], new_group);
1345 if (old_group)
1346 call_rcu(&old_group->rcu, free_port_group);
1347 return 0;
1348
1349 error_free:
1350 kfree(new_group);
1351 error:
1352 return error;
1353 }
1354
1355 static int
1356 get_port_group(struct datapath *dp, struct odp_port_group *upg)
1357 {
1358 struct odp_port_group pg;
1359 struct dp_port_group *g;
1360 u16 n_copy;
1361
1362 if (copy_from_user(&pg, upg, sizeof pg))
1363 return -EFAULT;
1364
1365 if (pg.group >= DP_MAX_GROUPS)
1366 return -EINVAL;
1367
1368 g = dp->groups[pg.group];
1369 n_copy = g ? min_t(int, g->n_ports, pg.n_ports) : 0;
1370 if (n_copy && copy_to_user(pg.ports, g->ports, n_copy * sizeof(u16)))
1371 return -EFAULT;
1372
1373 if (put_user(g ? g->n_ports : 0, &upg->n_ports))
1374 return -EFAULT;
1375
1376 return 0;
1377 }
1378
1379 static int get_listen_mask(const struct file *f)
1380 {
1381 return (long)f->private_data;
1382 }
1383
1384 static void set_listen_mask(struct file *f, int listen_mask)
1385 {
1386 f->private_data = (void*)(long)listen_mask;
1387 }
1388
1389 static long openvswitch_ioctl(struct file *f, unsigned int cmd,
1390 unsigned long argp)
1391 {
1392 int dp_idx = iminor(f->f_dentry->d_inode);
1393 struct datapath *dp;
1394 int drop_frags, listeners, port_no;
1395 unsigned int sflow_probability;
1396 int err;
1397
1398 /* Handle commands with special locking requirements up front. */
1399 switch (cmd) {
1400 case ODP_DP_CREATE:
1401 err = create_dp(dp_idx, (char __user *)argp);
1402 goto exit;
1403
1404 case ODP_DP_DESTROY:
1405 err = destroy_dp(dp_idx);
1406 goto exit;
1407
1408 case ODP_PORT_ADD:
1409 err = add_port(dp_idx, (struct odp_port __user *)argp);
1410 goto exit;
1411
1412 case ODP_PORT_DEL:
1413 err = get_user(port_no, (int __user *)argp);
1414 if (!err)
1415 err = del_port(dp_idx, port_no);
1416 goto exit;
1417 }
1418
1419 dp = get_dp_locked(dp_idx);
1420 err = -ENODEV;
1421 if (!dp)
1422 goto exit;
1423
1424 switch (cmd) {
1425 case ODP_DP_STATS:
1426 err = get_dp_stats(dp, (struct odp_stats __user *)argp);
1427 break;
1428
1429 case ODP_GET_DROP_FRAGS:
1430 err = put_user(dp->drop_frags, (int __user *)argp);
1431 break;
1432
1433 case ODP_SET_DROP_FRAGS:
1434 err = get_user(drop_frags, (int __user *)argp);
1435 if (err)
1436 break;
1437 err = -EINVAL;
1438 if (drop_frags != 0 && drop_frags != 1)
1439 break;
1440 dp->drop_frags = drop_frags;
1441 err = 0;
1442 break;
1443
1444 case ODP_GET_LISTEN_MASK:
1445 err = put_user(get_listen_mask(f), (int __user *)argp);
1446 break;
1447
1448 case ODP_SET_LISTEN_MASK:
1449 err = get_user(listeners, (int __user *)argp);
1450 if (err)
1451 break;
1452 err = -EINVAL;
1453 if (listeners & ~ODPL_ALL)
1454 break;
1455 err = 0;
1456 set_listen_mask(f, listeners);
1457 break;
1458
1459 case ODP_GET_SFLOW_PROBABILITY:
1460 err = put_user(dp->sflow_probability, (unsigned int __user *)argp);
1461 break;
1462
1463 case ODP_SET_SFLOW_PROBABILITY:
1464 err = get_user(sflow_probability, (unsigned int __user *)argp);
1465 if (!err)
1466 dp->sflow_probability = sflow_probability;
1467 break;
1468
1469 case ODP_PORT_QUERY:
1470 err = query_port(dp, (struct odp_port __user *)argp);
1471 break;
1472
1473 case ODP_PORT_LIST:
1474 err = list_ports(dp, (struct odp_portvec __user *)argp);
1475 break;
1476
1477 case ODP_PORT_GROUP_SET:
1478 err = set_port_group(dp, (struct odp_port_group __user *)argp);
1479 break;
1480
1481 case ODP_PORT_GROUP_GET:
1482 err = get_port_group(dp, (struct odp_port_group __user *)argp);
1483 break;
1484
1485 case ODP_FLOW_FLUSH:
1486 err = flush_flows(dp);
1487 break;
1488
1489 case ODP_FLOW_PUT:
1490 err = put_flow(dp, (struct odp_flow_put __user *)argp);
1491 break;
1492
1493 case ODP_FLOW_DEL:
1494 err = del_flow(dp, (struct odp_flow __user *)argp);
1495 break;
1496
1497 case ODP_FLOW_GET:
1498 err = do_flowvec_ioctl(dp, argp, query_flows);
1499 break;
1500
1501 case ODP_FLOW_LIST:
1502 err = do_flowvec_ioctl(dp, argp, list_flows);
1503 break;
1504
1505 case ODP_EXECUTE:
1506 err = do_execute(dp, (struct odp_execute __user *)argp);
1507 break;
1508
1509 default:
1510 err = -ENOIOCTLCMD;
1511 break;
1512 }
1513 mutex_unlock(&dp->mutex);
1514 exit:
1515 return err;
1516 }
1517
1518 static int dp_has_packet_of_interest(struct datapath *dp, int listeners)
1519 {
1520 int i;
1521 for (i = 0; i < DP_N_QUEUES; i++) {
1522 if (listeners & (1 << i) && !skb_queue_empty(&dp->queues[i]))
1523 return 1;
1524 }
1525 return 0;
1526 }
1527
1528 ssize_t openvswitch_read(struct file *f, char __user *buf, size_t nbytes,
1529 loff_t *ppos)
1530 {
1531 /* XXX is there sufficient synchronization here? */
1532 int listeners = get_listen_mask(f);
1533 int dp_idx = iminor(f->f_dentry->d_inode);
1534 struct datapath *dp = get_dp(dp_idx);
1535 struct sk_buff *skb;
1536 struct iovec __user iov;
1537 size_t copy_bytes;
1538 int retval;
1539
1540 if (!dp)
1541 return -ENODEV;
1542
1543 if (nbytes == 0 || !listeners)
1544 return 0;
1545
1546 for (;;) {
1547 int i;
1548
1549 for (i = 0; i < DP_N_QUEUES; i++) {
1550 if (listeners & (1 << i)) {
1551 skb = skb_dequeue(&dp->queues[i]);
1552 if (skb)
1553 goto success;
1554 }
1555 }
1556
1557 if (f->f_flags & O_NONBLOCK) {
1558 retval = -EAGAIN;
1559 goto error;
1560 }
1561
1562 wait_event_interruptible(dp->waitqueue,
1563 dp_has_packet_of_interest(dp,
1564 listeners));
1565
1566 if (signal_pending(current)) {
1567 retval = -ERESTARTSYS;
1568 goto error;
1569 }
1570 }
1571 success:
1572 copy_bytes = min_t(size_t, skb->len, nbytes);
1573 iov.iov_base = buf;
1574 iov.iov_len = copy_bytes;
1575 retval = skb_copy_datagram_iovec(skb, 0, &iov, iov.iov_len);
1576 if (!retval)
1577 retval = copy_bytes;
1578 kfree_skb(skb);
1579
1580 error:
1581 return retval;
1582 }
1583
1584 static unsigned int openvswitch_poll(struct file *file, poll_table *wait)
1585 {
1586 /* XXX is there sufficient synchronization here? */
1587 int dp_idx = iminor(file->f_dentry->d_inode);
1588 struct datapath *dp = get_dp(dp_idx);
1589 unsigned int mask;
1590
1591 if (dp) {
1592 mask = 0;
1593 poll_wait(file, &dp->waitqueue, wait);
1594 if (dp_has_packet_of_interest(dp, get_listen_mask(file)))
1595 mask |= POLLIN | POLLRDNORM;
1596 } else {
1597 mask = POLLIN | POLLRDNORM | POLLHUP;
1598 }
1599 return mask;
1600 }
1601
1602 struct file_operations openvswitch_fops = {
1603 /* XXX .aio_read = openvswitch_aio_read, */
1604 .read = openvswitch_read,
1605 .poll = openvswitch_poll,
1606 .unlocked_ioctl = openvswitch_ioctl,
1607 /* XXX .fasync = openvswitch_fasync, */
1608 };
1609
1610 static int major;
1611
1612 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,27)
1613 static struct llc_sap *dp_stp_sap;
1614
1615 static int dp_stp_rcv(struct sk_buff *skb, struct net_device *dev,
1616 struct packet_type *pt, struct net_device *orig_dev)
1617 {
1618 /* We don't really care about STP packets, we just listen for them for
1619 * mutual exclusion with the bridge module, so this just discards
1620 * them. */
1621 kfree_skb(skb);
1622 return 0;
1623 }
1624
1625 static int dp_avoid_bridge_init(void)
1626 {
1627 /* Register to receive STP packets because the bridge module also
1628 * attempts to do so. Since there can only be a single listener for a
1629 * given protocol, this provides mutual exclusion against the bridge
1630 * module, preventing both of them from being loaded at the same
1631 * time. */
1632 dp_stp_sap = llc_sap_open(LLC_SAP_BSPAN, dp_stp_rcv);
1633 if (!dp_stp_sap) {
1634 printk(KERN_ERR "openvswitch: can't register sap for STP (probably the bridge module is loaded)\n");
1635 return -EADDRINUSE;
1636 }
1637 return 0;
1638 }
1639
1640 static void dp_avoid_bridge_exit(void)
1641 {
1642 llc_sap_put(dp_stp_sap);
1643 }
1644 #else /* Linux 2.6.27 or later. */
1645 static int dp_avoid_bridge_init(void)
1646 {
1647 /* Linux 2.6.27 introduces a way for multiple clients to register for
1648 * STP packets, which interferes with what we try to do above.
1649 * Instead, just check whether there's a bridge hook defined. This is
1650 * not as safe--the bridge module is willing to load over the top of
1651 * us--but it provides a little bit of protection. */
1652 if (br_handle_frame_hook) {
1653 printk(KERN_ERR "openvswitch: bridge module is loaded, cannot load over it\n");
1654 return -EADDRINUSE;
1655 }
1656 return 0;
1657 }
1658
1659 static void dp_avoid_bridge_exit(void)
1660 {
1661 /* Nothing to do. */
1662 }
1663 #endif /* Linux 2.6.27 or later */
1664
1665 static int __init dp_init(void)
1666 {
1667 int err;
1668
1669 printk("Open vSwitch %s, built "__DATE__" "__TIME__"\n", VERSION BUILDNR);
1670
1671 err = dp_avoid_bridge_init();
1672 if (err)
1673 return err;
1674
1675 err = flow_init();
1676 if (err)
1677 goto error;
1678
1679 err = register_netdevice_notifier(&dp_device_notifier);
1680 if (err)
1681 goto error_flow_exit;
1682
1683 major = register_chrdev(0, "openvswitch", &openvswitch_fops);
1684 if (err < 0)
1685 goto error_unreg_notifier;
1686
1687 /* Hook into callback used by the bridge to intercept packets.
1688 * Parasites we are. */
1689 br_handle_frame_hook = dp_frame_hook;
1690
1691 return 0;
1692
1693 error_unreg_notifier:
1694 unregister_netdevice_notifier(&dp_device_notifier);
1695 error_flow_exit:
1696 flow_exit();
1697 error:
1698 return err;
1699 }
1700
1701 static void dp_cleanup(void)
1702 {
1703 rcu_barrier();
1704 unregister_chrdev(major, "openvswitch");
1705 unregister_netdevice_notifier(&dp_device_notifier);
1706 flow_exit();
1707 br_handle_frame_hook = NULL;
1708 dp_avoid_bridge_exit();
1709 }
1710
1711 module_init(dp_init);
1712 module_exit(dp_cleanup);
1713
1714 MODULE_DESCRIPTION("Open vSwitch switching datapath");
1715 MODULE_LICENSE("GPL");
openvswitch mirror