?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
NFC: SHDLC implementation
[linux-2.6.git] / net / openvswitch / datapath.c
blobe44e631ea952d3aadd1770f1e85b69b38b004325
1 /*
2 * Copyright (c) 2007-2012 Nicira Networks.
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/init.h>
22 #include <linux/module.h>
23 #include <linux/if_arp.h>
24 #include <linux/if_vlan.h>
25 #include <linux/in.h>
26 #include <linux/ip.h>
27 #include <linux/jhash.h>
28 #include <linux/delay.h>
29 #include <linux/time.h>
30 #include <linux/etherdevice.h>
31 #include <linux/genetlink.h>
32 #include <linux/kernel.h>
33 #include <linux/kthread.h>
34 #include <linux/mutex.h>
35 #include <linux/percpu.h>
36 #include <linux/rcupdate.h>
37 #include <linux/tcp.h>
38 #include <linux/udp.h>
39 #include <linux/ethtool.h>
40 #include <linux/wait.h>
41 #include <asm/div64.h>
42 #include <linux/highmem.h>
43 #include <linux/netfilter_bridge.h>
44 #include <linux/netfilter_ipv4.h>
45 #include <linux/inetdevice.h>
46 #include <linux/list.h>
47 #include <linux/openvswitch.h>
48 #include <linux/rculist.h>
49 #include <linux/dmi.h>
50 #include <linux/workqueue.h>
51 #include <net/genetlink.h>
52
53 #include "datapath.h"
54 #include "flow.h"
55 #include "vport-internal_dev.h"
56
57 /**
58 * DOC: Locking:
59 *
60 * Writes to device state (add/remove datapath, port, set operations on vports,
61 * etc.) are protected by RTNL.
62 *
63 * Writes to other state (flow table modifications, set miscellaneous datapath
64 * parameters, etc.) are protected by genl_mutex. The RTNL lock nests inside
65 * genl_mutex.
66 *
67 * Reads are protected by RCU.
68 *
69 * There are a few special cases (mostly stats) that have their own
70 * synchronization but they nest under all of above and don't interact with
71 * each other.
72 */
73
74 /* Global list of datapaths to enable dumping them all out.
75 * Protected by genl_mutex.
76 */
77 static LIST_HEAD(dps);
78
79 #define REHASH_FLOW_INTERVAL (10 * 60 * HZ)
80 static void rehash_flow_table(struct work_struct *work);
81 static DECLARE_DELAYED_WORK(rehash_flow_wq, rehash_flow_table);
82
83 static struct vport *new_vport(const struct vport_parms *);
84 static int queue_gso_packets(int dp_ifindex, struct sk_buff *,
85 const struct dp_upcall_info *);
86 static int queue_userspace_packet(int dp_ifindex, struct sk_buff *,
87 const struct dp_upcall_info *);
88
89 /* Must be called with rcu_read_lock, genl_mutex, or RTNL lock. */
90 static struct datapath *get_dp(int dp_ifindex)
91 {
92 struct datapath *dp = NULL;
93 struct net_device *dev;
94
95 rcu_read_lock();
96 dev = dev_get_by_index_rcu(&init_net, dp_ifindex);
97 if (dev) {
98 struct vport *vport = ovs_internal_dev_get_vport(dev);
99 if (vport)
100 dp = vport->dp;
101 }
102 rcu_read_unlock();
103
104 return dp;
105 }
106
107 /* Must be called with rcu_read_lock or RTNL lock. */
108 const char *ovs_dp_name(const struct datapath *dp)
109 {
110 struct vport *vport = rcu_dereference_rtnl(dp->ports[OVSP_LOCAL]);
111 return vport->ops->get_name(vport);
112 }
113
114 static int get_dpifindex(struct datapath *dp)
115 {
116 struct vport *local;
117 int ifindex;
118
119 rcu_read_lock();
120
121 local = rcu_dereference(dp->ports[OVSP_LOCAL]);
122 if (local)
123 ifindex = local->ops->get_ifindex(local);
124 else
125 ifindex = 0;
126
127 rcu_read_unlock();
128
129 return ifindex;
130 }
131
132 static void destroy_dp_rcu(struct rcu_head *rcu)
133 {
134 struct datapath *dp = container_of(rcu, struct datapath, rcu);
135
136 ovs_flow_tbl_destroy((__force struct flow_table *)dp->table);
137 free_percpu(dp->stats_percpu);
138 kfree(dp);
139 }
140
141 /* Called with RTNL lock and genl_lock. */
142 static struct vport *new_vport(const struct vport_parms *parms)
143 {
144 struct vport *vport;
145
146 vport = ovs_vport_add(parms);
147 if (!IS_ERR(vport)) {
148 struct datapath *dp = parms->dp;
149
150 rcu_assign_pointer(dp->ports[parms->port_no], vport);
151 list_add(&vport->node, &dp->port_list);
152 }
153
154 return vport;
155 }
156
157 /* Called with RTNL lock. */
158 void ovs_dp_detach_port(struct vport *p)
159 {
160 ASSERT_RTNL();
161
162 /* First drop references to device. */
163 list_del(&p->node);
164 rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
165
166 /* Then destroy it. */
167 ovs_vport_del(p);
168 }
169
170 /* Must be called with rcu_read_lock. */
171 void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
172 {
173 struct datapath *dp = p->dp;
174 struct sw_flow *flow;
175 struct dp_stats_percpu *stats;
176 struct sw_flow_key key;
177 u64 *stats_counter;
178 int error;
179 int key_len;
180
181 stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
182
183 /* Extract flow from 'skb' into 'key'. */
184 error = ovs_flow_extract(skb, p->port_no, &key, &key_len);
185 if (unlikely(error)) {
186 kfree_skb(skb);
187 return;
188 }
189
190 /* Look up flow. */
191 flow = ovs_flow_tbl_lookup(rcu_dereference(dp->table), &key, key_len);
192 if (unlikely(!flow)) {
193 struct dp_upcall_info upcall;
194
195 upcall.cmd = OVS_PACKET_CMD_MISS;
196 upcall.key = &key;
197 upcall.userdata = NULL;
198 upcall.pid = p->upcall_pid;
199 ovs_dp_upcall(dp, skb, &upcall);
200 consume_skb(skb);
201 stats_counter = &stats->n_missed;
202 goto out;
203 }
204
205 OVS_CB(skb)->flow = flow;
206
207 stats_counter = &stats->n_hit;
208 ovs_flow_used(OVS_CB(skb)->flow, skb);
209 ovs_execute_actions(dp, skb);
210
211 out:
212 /* Update datapath statistics. */
213 u64_stats_update_begin(&stats->sync);
214 (*stats_counter)++;
215 u64_stats_update_end(&stats->sync);
216 }
217
218 static struct genl_family dp_packet_genl_family = {
219 .id = GENL_ID_GENERATE,
220 .hdrsize = sizeof(struct ovs_header),
221 .name = OVS_PACKET_FAMILY,
222 .version = OVS_PACKET_VERSION,
223 .maxattr = OVS_PACKET_ATTR_MAX
224 };
225
226 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
227 const struct dp_upcall_info *upcall_info)
228 {
229 struct dp_stats_percpu *stats;
230 int dp_ifindex;
231 int err;
232
233 if (upcall_info->pid == 0) {
234 err = -ENOTCONN;
235 goto err;
236 }
237
238 dp_ifindex = get_dpifindex(dp);
239 if (!dp_ifindex) {
240 err = -ENODEV;
241 goto err;
242 }
243
244 if (!skb_is_gso(skb))
245 err = queue_userspace_packet(dp_ifindex, skb, upcall_info);
246 else
247 err = queue_gso_packets(dp_ifindex, skb, upcall_info);
248 if (err)
249 goto err;
250
251 return 0;
252
253 err:
254 stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
255
256 u64_stats_update_begin(&stats->sync);
257 stats->n_lost++;
258 u64_stats_update_end(&stats->sync);
259
260 return err;
261 }
262
263 static int queue_gso_packets(int dp_ifindex, struct sk_buff *skb,
264 const struct dp_upcall_info *upcall_info)
265 {
266 struct dp_upcall_info later_info;
267 struct sw_flow_key later_key;
268 struct sk_buff *segs, *nskb;
269 int err;
270
271 segs = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
272 if (IS_ERR(skb))
273 return PTR_ERR(skb);
274
275 /* Queue all of the segments. */
276 skb = segs;
277 do {
278 err = queue_userspace_packet(dp_ifindex, skb, upcall_info);
279 if (err)
280 break;
281
282 if (skb == segs && skb_shinfo(skb)->gso_type & SKB_GSO_UDP) {
283 /* The initial flow key extracted by ovs_flow_extract()
284 * in this case is for a first fragment, so we need to
285 * properly mark later fragments.
286 */
287 later_key = *upcall_info->key;
288 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
289
290 later_info = *upcall_info;
291 later_info.key = &later_key;
292 upcall_info = &later_info;
293 }
294 } while ((skb = skb->next));
295
296 /* Free all of the segments. */
297 skb = segs;
298 do {
299 nskb = skb->next;
300 if (err)
301 kfree_skb(skb);
302 else
303 consume_skb(skb);
304 } while ((skb = nskb));
305 return err;
306 }
307
308 static int queue_userspace_packet(int dp_ifindex, struct sk_buff *skb,
309 const struct dp_upcall_info *upcall_info)
310 {
311 struct ovs_header *upcall;
312 struct sk_buff *nskb = NULL;
313 struct sk_buff *user_skb; /* to be queued to userspace */
314 struct nlattr *nla;
315 unsigned int len;
316 int err;
317
318 if (vlan_tx_tag_present(skb)) {
319 nskb = skb_clone(skb, GFP_ATOMIC);
320 if (!nskb)
321 return -ENOMEM;
322
323 nskb = __vlan_put_tag(nskb, vlan_tx_tag_get(nskb));
324 if (!skb)
325 return -ENOMEM;
326
327 nskb->vlan_tci = 0;
328 skb = nskb;
329 }
330
331 if (nla_attr_size(skb->len) > USHRT_MAX) {
332 err = -EFBIG;
333 goto out;
334 }
335
336 len = sizeof(struct ovs_header);
337 len += nla_total_size(skb->len);
338 len += nla_total_size(FLOW_BUFSIZE);
339 if (upcall_info->cmd == OVS_PACKET_CMD_ACTION)
340 len += nla_total_size(8);
341
342 user_skb = genlmsg_new(len, GFP_ATOMIC);
343 if (!user_skb) {
344 err = -ENOMEM;
345 goto out;
346 }
347
348 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
349 0, upcall_info->cmd);
350 upcall->dp_ifindex = dp_ifindex;
351
352 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
353 ovs_flow_to_nlattrs(upcall_info->key, user_skb);
354 nla_nest_end(user_skb, nla);
355
356 if (upcall_info->userdata)
357 nla_put_u64(user_skb, OVS_PACKET_ATTR_USERDATA,
358 nla_get_u64(upcall_info->userdata));
359
360 nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
361
362 skb_copy_and_csum_dev(skb, nla_data(nla));
363
364 err = genlmsg_unicast(&init_net, user_skb, upcall_info->pid);
365
366 out:
367 kfree_skb(nskb);
368 return err;
369 }
370
371 /* Called with genl_mutex. */
372 static int flush_flows(int dp_ifindex)
373 {
374 struct flow_table *old_table;
375 struct flow_table *new_table;
376 struct datapath *dp;
377
378 dp = get_dp(dp_ifindex);
379 if (!dp)
380 return -ENODEV;
381
382 old_table = genl_dereference(dp->table);
383 new_table = ovs_flow_tbl_alloc(TBL_MIN_BUCKETS);
384 if (!new_table)
385 return -ENOMEM;
386
387 rcu_assign_pointer(dp->table, new_table);
388
389 ovs_flow_tbl_deferred_destroy(old_table);
390 return 0;
391 }
392
393 static int validate_actions(const struct nlattr *attr,
394 const struct sw_flow_key *key, int depth);
395
396 static int validate_sample(const struct nlattr *attr,
397 const struct sw_flow_key *key, int depth)
398 {
399 const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
400 const struct nlattr *probability, *actions;
401 const struct nlattr *a;
402 int rem;
403
404 memset(attrs, 0, sizeof(attrs));
405 nla_for_each_nested(a, attr, rem) {
406 int type = nla_type(a);
407 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
408 return -EINVAL;
409 attrs[type] = a;
410 }
411 if (rem)
412 return -EINVAL;
413
414 probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
415 if (!probability || nla_len(probability) != sizeof(u32))
416 return -EINVAL;
417
418 actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
419 if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
420 return -EINVAL;
421 return validate_actions(actions, key, depth + 1);
422 }
423
424 static int validate_set(const struct nlattr *a,
425 const struct sw_flow_key *flow_key)
426 {
427 const struct nlattr *ovs_key = nla_data(a);
428 int key_type = nla_type(ovs_key);
429
430 /* There can be only one key in a action */
431 if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
432 return -EINVAL;
433
434 if (key_type > OVS_KEY_ATTR_MAX ||
435 nla_len(ovs_key) != ovs_key_lens[key_type])
436 return -EINVAL;
437
438 switch (key_type) {
439 const struct ovs_key_ipv4 *ipv4_key;
440
441 case OVS_KEY_ATTR_PRIORITY:
442 case OVS_KEY_ATTR_ETHERNET:
443 break;
444
445 case OVS_KEY_ATTR_IPV4:
446 if (flow_key->eth.type != htons(ETH_P_IP))
447 return -EINVAL;
448
449 if (!flow_key->ipv4.addr.src || !flow_key->ipv4.addr.dst)
450 return -EINVAL;
451
452 ipv4_key = nla_data(ovs_key);
453 if (ipv4_key->ipv4_proto != flow_key->ip.proto)
454 return -EINVAL;
455
456 if (ipv4_key->ipv4_frag != flow_key->ip.frag)
457 return -EINVAL;
458
459 break;
460
461 case OVS_KEY_ATTR_TCP:
462 if (flow_key->ip.proto != IPPROTO_TCP)
463 return -EINVAL;
464
465 if (!flow_key->ipv4.tp.src || !flow_key->ipv4.tp.dst)
466 return -EINVAL;
467
468 break;
469
470 case OVS_KEY_ATTR_UDP:
471 if (flow_key->ip.proto != IPPROTO_UDP)
472 return -EINVAL;
473
474 if (!flow_key->ipv4.tp.src || !flow_key->ipv4.tp.dst)
475 return -EINVAL;
476 break;
477
478 default:
479 return -EINVAL;
480 }
481
482 return 0;
483 }
484
485 static int validate_userspace(const struct nlattr *attr)
486 {
487 static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] = {
488 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
489 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_U64 },
490 };
491 struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
492 int error;
493
494 error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
495 attr, userspace_policy);
496 if (error)
497 return error;
498
499 if (!a[OVS_USERSPACE_ATTR_PID] ||
500 !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
501 return -EINVAL;
502
503 return 0;
504 }
505
506 static int validate_actions(const struct nlattr *attr,
507 const struct sw_flow_key *key, int depth)
508 {
509 const struct nlattr *a;
510 int rem, err;
511
512 if (depth >= SAMPLE_ACTION_DEPTH)
513 return -EOVERFLOW;
514
515 nla_for_each_nested(a, attr, rem) {
516 /* Expected argument lengths, (u32)-1 for variable length. */
517 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
518 [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
519 [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
520 [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
521 [OVS_ACTION_ATTR_POP_VLAN] = 0,
522 [OVS_ACTION_ATTR_SET] = (u32)-1,
523 [OVS_ACTION_ATTR_SAMPLE] = (u32)-1
524 };
525 const struct ovs_action_push_vlan *vlan;
526 int type = nla_type(a);
527
528 if (type > OVS_ACTION_ATTR_MAX ||
529 (action_lens[type] != nla_len(a) &&
530 action_lens[type] != (u32)-1))
531 return -EINVAL;
532
533 switch (type) {
534 case OVS_ACTION_ATTR_UNSPEC:
535 return -EINVAL;
536
537 case OVS_ACTION_ATTR_USERSPACE:
538 err = validate_userspace(a);
539 if (err)
540 return err;
541 break;
542
543 case OVS_ACTION_ATTR_OUTPUT:
544 if (nla_get_u32(a) >= DP_MAX_PORTS)
545 return -EINVAL;
546 break;
547
548
549 case OVS_ACTION_ATTR_POP_VLAN:
550 break;
551
552 case OVS_ACTION_ATTR_PUSH_VLAN:
553 vlan = nla_data(a);
554 if (vlan->vlan_tpid != htons(ETH_P_8021Q))
555 return -EINVAL;
556 if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
557 return -EINVAL;
558 break;
559
560 case OVS_ACTION_ATTR_SET:
561 err = validate_set(a, key);
562 if (err)
563 return err;
564 break;
565
566 case OVS_ACTION_ATTR_SAMPLE:
567 err = validate_sample(a, key, depth);
568 if (err)
569 return err;
570 break;
571
572 default:
573 return -EINVAL;
574 }
575 }
576
577 if (rem > 0)
578 return -EINVAL;
579
580 return 0;
581 }
582
583 static void clear_stats(struct sw_flow *flow)
584 {
585 flow->used = 0;
586 flow->tcp_flags = 0;
587 flow->packet_count = 0;
588 flow->byte_count = 0;
589 }
590
591 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
592 {
593 struct ovs_header *ovs_header = info->userhdr;
594 struct nlattr **a = info->attrs;
595 struct sw_flow_actions *acts;
596 struct sk_buff *packet;
597 struct sw_flow *flow;
598 struct datapath *dp;
599 struct ethhdr *eth;
600 int len;
601 int err;
602 int key_len;
603
604 err = -EINVAL;
605 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
606 !a[OVS_PACKET_ATTR_ACTIONS] ||
607 nla_len(a[OVS_PACKET_ATTR_PACKET]) < ETH_HLEN)
608 goto err;
609
610 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
611 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
612 err = -ENOMEM;
613 if (!packet)
614 goto err;
615 skb_reserve(packet, NET_IP_ALIGN);
616
617 memcpy(__skb_put(packet, len), nla_data(a[OVS_PACKET_ATTR_PACKET]), len);
618
619 skb_reset_mac_header(packet);
620 eth = eth_hdr(packet);
621
622 /* Normally, setting the skb 'protocol' field would be handled by a
623 * call to eth_type_trans(), but it assumes there's a sending
624 * device, which we may not have. */
625 if (ntohs(eth->h_proto) >= 1536)
626 packet->protocol = eth->h_proto;
627 else
628 packet->protocol = htons(ETH_P_802_2);
629
630 /* Build an sw_flow for sending this packet. */
631 flow = ovs_flow_alloc();
632 err = PTR_ERR(flow);
633 if (IS_ERR(flow))
634 goto err_kfree_skb;
635
636 err = ovs_flow_extract(packet, -1, &flow->key, &key_len);
637 if (err)
638 goto err_flow_free;
639
640 err = ovs_flow_metadata_from_nlattrs(&flow->key.phy.priority,
641 &flow->key.phy.in_port,
642 a[OVS_PACKET_ATTR_KEY]);
643 if (err)
644 goto err_flow_free;
645
646 err = validate_actions(a[OVS_PACKET_ATTR_ACTIONS], &flow->key, 0);
647 if (err)
648 goto err_flow_free;
649
650 flow->hash = ovs_flow_hash(&flow->key, key_len);
651
652 acts = ovs_flow_actions_alloc(a[OVS_PACKET_ATTR_ACTIONS]);
653 err = PTR_ERR(acts);
654 if (IS_ERR(acts))
655 goto err_flow_free;
656 rcu_assign_pointer(flow->sf_acts, acts);
657
658 OVS_CB(packet)->flow = flow;
659 packet->priority = flow->key.phy.priority;
660
661 rcu_read_lock();
662 dp = get_dp(ovs_header->dp_ifindex);
663 err = -ENODEV;
664 if (!dp)
665 goto err_unlock;
666
667 local_bh_disable();
668 err = ovs_execute_actions(dp, packet);
669 local_bh_enable();
670 rcu_read_unlock();
671
672 ovs_flow_free(flow);
673 return err;
674
675 err_unlock:
676 rcu_read_unlock();
677 err_flow_free:
678 ovs_flow_free(flow);
679 err_kfree_skb:
680 kfree_skb(packet);
681 err:
682 return err;
683 }
684
685 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
686 [OVS_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
687 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
688 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
689 };
690
691 static struct genl_ops dp_packet_genl_ops[] = {
692 { .cmd = OVS_PACKET_CMD_EXECUTE,
693 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
694 .policy = packet_policy,
695 .doit = ovs_packet_cmd_execute
696 }
697 };
698
699 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
700 {
701 int i;
702 struct flow_table *table = genl_dereference(dp->table);
703
704 stats->n_flows = ovs_flow_tbl_count(table);
705
706 stats->n_hit = stats->n_missed = stats->n_lost = 0;
707 for_each_possible_cpu(i) {
708 const struct dp_stats_percpu *percpu_stats;
709 struct dp_stats_percpu local_stats;
710 unsigned int start;
711
712 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
713
714 do {
715 start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
716 local_stats = *percpu_stats;
717 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
718
719 stats->n_hit += local_stats.n_hit;
720 stats->n_missed += local_stats.n_missed;
721 stats->n_lost += local_stats.n_lost;
722 }
723 }
724
725 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
726 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
727 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
728 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
729 };
730
731 static struct genl_family dp_flow_genl_family = {
732 .id = GENL_ID_GENERATE,
733 .hdrsize = sizeof(struct ovs_header),
734 .name = OVS_FLOW_FAMILY,
735 .version = OVS_FLOW_VERSION,
736 .maxattr = OVS_FLOW_ATTR_MAX
737 };
738
739 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
740 .name = OVS_FLOW_MCGROUP
741 };
742
743 /* Called with genl_lock. */
744 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
745 struct sk_buff *skb, u32 pid,
746 u32 seq, u32 flags, u8 cmd)
747 {
748 const int skb_orig_len = skb->len;
749 const struct sw_flow_actions *sf_acts;
750 struct ovs_flow_stats stats;
751 struct ovs_header *ovs_header;
752 struct nlattr *nla;
753 unsigned long used;
754 u8 tcp_flags;
755 int err;
756
757 sf_acts = rcu_dereference_protected(flow->sf_acts,
758 lockdep_genl_is_held());
759
760 ovs_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags, cmd);
761 if (!ovs_header)
762 return -EMSGSIZE;
763
764 ovs_header->dp_ifindex = get_dpifindex(dp);
765
766 nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
767 if (!nla)
768 goto nla_put_failure;
769 err = ovs_flow_to_nlattrs(&flow->key, skb);
770 if (err)
771 goto error;
772 nla_nest_end(skb, nla);
773
774 spin_lock_bh(&flow->lock);
775 used = flow->used;
776 stats.n_packets = flow->packet_count;
777 stats.n_bytes = flow->byte_count;
778 tcp_flags = flow->tcp_flags;
779 spin_unlock_bh(&flow->lock);
780
781 if (used)
782 NLA_PUT_U64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used));
783
784 if (stats.n_packets)
785 NLA_PUT(skb, OVS_FLOW_ATTR_STATS,
786 sizeof(struct ovs_flow_stats), &stats);
787
788 if (tcp_flags)
789 NLA_PUT_U8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags);
790
791 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
792 * this is the first flow to be dumped into 'skb'. This is unusual for
793 * Netlink but individual action lists can be longer than
794 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
795 * The userspace caller can always fetch the actions separately if it
796 * really wants them. (Most userspace callers in fact don't care.)
797 *
798 * This can only fail for dump operations because the skb is always
799 * properly sized for single flows.
800 */
801 err = nla_put(skb, OVS_FLOW_ATTR_ACTIONS, sf_acts->actions_len,
802 sf_acts->actions);
803 if (err < 0 && skb_orig_len)
804 goto error;
805
806 return genlmsg_end(skb, ovs_header);
807
808 nla_put_failure:
809 err = -EMSGSIZE;
810 error:
811 genlmsg_cancel(skb, ovs_header);
812 return err;
813 }
814
815 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
816 {
817 const struct sw_flow_actions *sf_acts;
818 int len;
819
820 sf_acts = rcu_dereference_protected(flow->sf_acts,
821 lockdep_genl_is_held());
822
823 /* OVS_FLOW_ATTR_KEY */
824 len = nla_total_size(FLOW_BUFSIZE);
825 /* OVS_FLOW_ATTR_ACTIONS */
826 len += nla_total_size(sf_acts->actions_len);
827 /* OVS_FLOW_ATTR_STATS */
828 len += nla_total_size(sizeof(struct ovs_flow_stats));
829 /* OVS_FLOW_ATTR_TCP_FLAGS */
830 len += nla_total_size(1);
831 /* OVS_FLOW_ATTR_USED */
832 len += nla_total_size(8);
833
834 len += NLMSG_ALIGN(sizeof(struct ovs_header));
835
836 return genlmsg_new(len, GFP_KERNEL);
837 }
838
839 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
840 struct datapath *dp,
841 u32 pid, u32 seq, u8 cmd)
842 {
843 struct sk_buff *skb;
844 int retval;
845
846 skb = ovs_flow_cmd_alloc_info(flow);
847 if (!skb)
848 return ERR_PTR(-ENOMEM);
849
850 retval = ovs_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd);
851 BUG_ON(retval < 0);
852 return skb;
853 }
854
855 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
856 {
857 struct nlattr **a = info->attrs;
858 struct ovs_header *ovs_header = info->userhdr;
859 struct sw_flow_key key;
860 struct sw_flow *flow;
861 struct sk_buff *reply;
862 struct datapath *dp;
863 struct flow_table *table;
864 int error;
865 int key_len;
866
867 /* Extract key. */
868 error = -EINVAL;
869 if (!a[OVS_FLOW_ATTR_KEY])
870 goto error;
871 error = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
872 if (error)
873 goto error;
874
875 /* Validate actions. */
876 if (a[OVS_FLOW_ATTR_ACTIONS]) {
877 error = validate_actions(a[OVS_FLOW_ATTR_ACTIONS], &key, 0);
878 if (error)
879 goto error;
880 } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
881 error = -EINVAL;
882 goto error;
883 }
884
885 dp = get_dp(ovs_header->dp_ifindex);
886 error = -ENODEV;
887 if (!dp)
888 goto error;
889
890 table = genl_dereference(dp->table);
891 flow = ovs_flow_tbl_lookup(table, &key, key_len);
892 if (!flow) {
893 struct sw_flow_actions *acts;
894
895 /* Bail out if we're not allowed to create a new flow. */
896 error = -ENOENT;
897 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
898 goto error;
899
900 /* Expand table, if necessary, to make room. */
901 if (ovs_flow_tbl_need_to_expand(table)) {
902 struct flow_table *new_table;
903
904 new_table = ovs_flow_tbl_expand(table);
905 if (!IS_ERR(new_table)) {
906 rcu_assign_pointer(dp->table, new_table);
907 ovs_flow_tbl_deferred_destroy(table);
908 table = genl_dereference(dp->table);
909 }
910 }
911
912 /* Allocate flow. */
913 flow = ovs_flow_alloc();
914 if (IS_ERR(flow)) {
915 error = PTR_ERR(flow);
916 goto error;
917 }
918 flow->key = key;
919 clear_stats(flow);
920
921 /* Obtain actions. */
922 acts = ovs_flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]);
923 error = PTR_ERR(acts);
924 if (IS_ERR(acts))
925 goto error_free_flow;
926 rcu_assign_pointer(flow->sf_acts, acts);
927
928 /* Put flow in bucket. */
929 flow->hash = ovs_flow_hash(&key, key_len);
930 ovs_flow_tbl_insert(table, flow);
931
932 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
933 info->snd_seq,
934 OVS_FLOW_CMD_NEW);
935 } else {
936 /* We found a matching flow. */
937 struct sw_flow_actions *old_acts;
938 struct nlattr *acts_attrs;
939
940 /* Bail out if we're not allowed to modify an existing flow.
941 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
942 * because Generic Netlink treats the latter as a dump
943 * request. We also accept NLM_F_EXCL in case that bug ever
944 * gets fixed.
945 */
946 error = -EEXIST;
947 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
948 info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
949 goto error;
950
951 /* Update actions. */
952 old_acts = rcu_dereference_protected(flow->sf_acts,
953 lockdep_genl_is_held());
954 acts_attrs = a[OVS_FLOW_ATTR_ACTIONS];
955 if (acts_attrs &&
956 (old_acts->actions_len != nla_len(acts_attrs) ||
957 memcmp(old_acts->actions, nla_data(acts_attrs),
958 old_acts->actions_len))) {
959 struct sw_flow_actions *new_acts;
960
961 new_acts = ovs_flow_actions_alloc(acts_attrs);
962 error = PTR_ERR(new_acts);
963 if (IS_ERR(new_acts))
964 goto error;
965
966 rcu_assign_pointer(flow->sf_acts, new_acts);
967 ovs_flow_deferred_free_acts(old_acts);
968 }
969
970 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
971 info->snd_seq, OVS_FLOW_CMD_NEW);
972
973 /* Clear stats. */
974 if (a[OVS_FLOW_ATTR_CLEAR]) {
975 spin_lock_bh(&flow->lock);
976 clear_stats(flow);
977 spin_unlock_bh(&flow->lock);
978 }
979 }
980
981 if (!IS_ERR(reply))
982 genl_notify(reply, genl_info_net(info), info->snd_pid,
983 ovs_dp_flow_multicast_group.id, info->nlhdr,
984 GFP_KERNEL);
985 else
986 netlink_set_err(init_net.genl_sock, 0,
987 ovs_dp_flow_multicast_group.id, PTR_ERR(reply));
988 return 0;
989
990 error_free_flow:
991 ovs_flow_free(flow);
992 error:
993 return error;
994 }
995
996 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
997 {
998 struct nlattr **a = info->attrs;
999 struct ovs_header *ovs_header = info->userhdr;
1000 struct sw_flow_key key;
1001 struct sk_buff *reply;
1002 struct sw_flow *flow;
1003 struct datapath *dp;
1004 struct flow_table *table;
1005 int err;
1006 int key_len;
1007
1008 if (!a[OVS_FLOW_ATTR_KEY])
1009 return -EINVAL;
1010 err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1011 if (err)
1012 return err;
1013
1014 dp = get_dp(ovs_header->dp_ifindex);
1015 if (!dp)
1016 return -ENODEV;
1017
1018 table = genl_dereference(dp->table);
1019 flow = ovs_flow_tbl_lookup(table, &key, key_len);
1020 if (!flow)
1021 return -ENOENT;
1022
1023 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1024 info->snd_seq, OVS_FLOW_CMD_NEW);
1025 if (IS_ERR(reply))
1026 return PTR_ERR(reply);
1027
1028 return genlmsg_reply(reply, info);
1029 }
1030
1031 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1032 {
1033 struct nlattr **a = info->attrs;
1034 struct ovs_header *ovs_header = info->userhdr;
1035 struct sw_flow_key key;
1036 struct sk_buff *reply;
1037 struct sw_flow *flow;
1038 struct datapath *dp;
1039 struct flow_table *table;
1040 int err;
1041 int key_len;
1042
1043 if (!a[OVS_FLOW_ATTR_KEY])
1044 return flush_flows(ovs_header->dp_ifindex);
1045 err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1046 if (err)
1047 return err;
1048
1049 dp = get_dp(ovs_header->dp_ifindex);
1050 if (!dp)
1051 return -ENODEV;
1052
1053 table = genl_dereference(dp->table);
1054 flow = ovs_flow_tbl_lookup(table, &key, key_len);
1055 if (!flow)
1056 return -ENOENT;
1057
1058 reply = ovs_flow_cmd_alloc_info(flow);
1059 if (!reply)
1060 return -ENOMEM;
1061
1062 ovs_flow_tbl_remove(table, flow);
1063
1064 err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_pid,
1065 info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1066 BUG_ON(err < 0);
1067
1068 ovs_flow_deferred_free(flow);
1069
1070 genl_notify(reply, genl_info_net(info), info->snd_pid,
1071 ovs_dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1072 return 0;
1073 }
1074
1075 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1076 {
1077 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1078 struct datapath *dp;
1079 struct flow_table *table;
1080
1081 dp = get_dp(ovs_header->dp_ifindex);
1082 if (!dp)
1083 return -ENODEV;
1084
1085 table = genl_dereference(dp->table);
1086
1087 for (;;) {
1088 struct sw_flow *flow;
1089 u32 bucket, obj;
1090
1091 bucket = cb->args[0];
1092 obj = cb->args[1];
1093 flow = ovs_flow_tbl_next(table, &bucket, &obj);
1094 if (!flow)
1095 break;
1096
1097 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1098 NETLINK_CB(cb->skb).pid,
1099 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1100 OVS_FLOW_CMD_NEW) < 0)
1101 break;
1102
1103 cb->args[0] = bucket;
1104 cb->args[1] = obj;
1105 }
1106 return skb->len;
1107 }
1108
1109 static struct genl_ops dp_flow_genl_ops[] = {
1110 { .cmd = OVS_FLOW_CMD_NEW,
1111 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1112 .policy = flow_policy,
1113 .doit = ovs_flow_cmd_new_or_set
1114 },
1115 { .cmd = OVS_FLOW_CMD_DEL,
1116 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1117 .policy = flow_policy,
1118 .doit = ovs_flow_cmd_del
1119 },
1120 { .cmd = OVS_FLOW_CMD_GET,
1121 .flags = 0, /* OK for unprivileged users. */
1122 .policy = flow_policy,
1123 .doit = ovs_flow_cmd_get,
1124 .dumpit = ovs_flow_cmd_dump
1125 },
1126 { .cmd = OVS_FLOW_CMD_SET,
1127 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1128 .policy = flow_policy,
1129 .doit = ovs_flow_cmd_new_or_set,
1130 },
1131 };
1132
1133 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1134 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1135 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1136 };
1137
1138 static struct genl_family dp_datapath_genl_family = {
1139 .id = GENL_ID_GENERATE,
1140 .hdrsize = sizeof(struct ovs_header),
1141 .name = OVS_DATAPATH_FAMILY,
1142 .version = OVS_DATAPATH_VERSION,
1143 .maxattr = OVS_DP_ATTR_MAX
1144 };
1145
1146 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1147 .name = OVS_DATAPATH_MCGROUP
1148 };
1149
1150 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1151 u32 pid, u32 seq, u32 flags, u8 cmd)
1152 {
1153 struct ovs_header *ovs_header;
1154 struct ovs_dp_stats dp_stats;
1155 int err;
1156
1157 ovs_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family,
1158 flags, cmd);
1159 if (!ovs_header)
1160 goto error;
1161
1162 ovs_header->dp_ifindex = get_dpifindex(dp);
1163
1164 rcu_read_lock();
1165 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1166 rcu_read_unlock();
1167 if (err)
1168 goto nla_put_failure;
1169
1170 get_dp_stats(dp, &dp_stats);
1171 NLA_PUT(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats), &dp_stats);
1172
1173 return genlmsg_end(skb, ovs_header);
1174
1175 nla_put_failure:
1176 genlmsg_cancel(skb, ovs_header);
1177 error:
1178 return -EMSGSIZE;
1179 }
1180
1181 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 pid,
1182 u32 seq, u8 cmd)
1183 {
1184 struct sk_buff *skb;
1185 int retval;
1186
1187 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1188 if (!skb)
1189 return ERR_PTR(-ENOMEM);
1190
1191 retval = ovs_dp_cmd_fill_info(dp, skb, pid, seq, 0, cmd);
1192 if (retval < 0) {
1193 kfree_skb(skb);
1194 return ERR_PTR(retval);
1195 }
1196 return skb;
1197 }
1198
1199 /* Called with genl_mutex and optionally with RTNL lock also. */
1200 static struct datapath *lookup_datapath(struct ovs_header *ovs_header,
1201 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1202 {
1203 struct datapath *dp;
1204
1205 if (!a[OVS_DP_ATTR_NAME])
1206 dp = get_dp(ovs_header->dp_ifindex);
1207 else {
1208 struct vport *vport;
1209
1210 rcu_read_lock();
1211 vport = ovs_vport_locate(nla_data(a[OVS_DP_ATTR_NAME]));
1212 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1213 rcu_read_unlock();
1214 }
1215 return dp ? dp : ERR_PTR(-ENODEV);
1216 }
1217
1218 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1219 {
1220 struct nlattr **a = info->attrs;
1221 struct vport_parms parms;
1222 struct sk_buff *reply;
1223 struct datapath *dp;
1224 struct vport *vport;
1225 int err;
1226
1227 err = -EINVAL;
1228 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1229 goto err;
1230
1231 rtnl_lock();
1232 err = -ENODEV;
1233 if (!try_module_get(THIS_MODULE))
1234 goto err_unlock_rtnl;
1235
1236 err = -ENOMEM;
1237 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1238 if (dp == NULL)
1239 goto err_put_module;
1240 INIT_LIST_HEAD(&dp->port_list);
1241
1242 /* Allocate table. */
1243 err = -ENOMEM;
1244 rcu_assign_pointer(dp->table, ovs_flow_tbl_alloc(TBL_MIN_BUCKETS));
1245 if (!dp->table)
1246 goto err_free_dp;
1247
1248 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1249 if (!dp->stats_percpu) {
1250 err = -ENOMEM;
1251 goto err_destroy_table;
1252 }
1253
1254 /* Set up our datapath device. */
1255 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1256 parms.type = OVS_VPORT_TYPE_INTERNAL;
1257 parms.options = NULL;
1258 parms.dp = dp;
1259 parms.port_no = OVSP_LOCAL;
1260 parms.upcall_pid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1261
1262 vport = new_vport(&parms);
1263 if (IS_ERR(vport)) {
1264 err = PTR_ERR(vport);
1265 if (err == -EBUSY)
1266 err = -EEXIST;
1267
1268 goto err_destroy_percpu;
1269 }
1270
1271 reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1272 info->snd_seq, OVS_DP_CMD_NEW);
1273 err = PTR_ERR(reply);
1274 if (IS_ERR(reply))
1275 goto err_destroy_local_port;
1276
1277 list_add_tail(&dp->list_node, &dps);
1278 rtnl_unlock();
1279
1280 genl_notify(reply, genl_info_net(info), info->snd_pid,
1281 ovs_dp_datapath_multicast_group.id, info->nlhdr,
1282 GFP_KERNEL);
1283 return 0;
1284
1285 err_destroy_local_port:
1286 ovs_dp_detach_port(rtnl_dereference(dp->ports[OVSP_LOCAL]));
1287 err_destroy_percpu:
1288 free_percpu(dp->stats_percpu);
1289 err_destroy_table:
1290 ovs_flow_tbl_destroy(genl_dereference(dp->table));
1291 err_free_dp:
1292 kfree(dp);
1293 err_put_module:
1294 module_put(THIS_MODULE);
1295 err_unlock_rtnl:
1296 rtnl_unlock();
1297 err:
1298 return err;
1299 }
1300
1301 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1302 {
1303 struct vport *vport, *next_vport;
1304 struct sk_buff *reply;
1305 struct datapath *dp;
1306 int err;
1307
1308 rtnl_lock();
1309 dp = lookup_datapath(info->userhdr, info->attrs);
1310 err = PTR_ERR(dp);
1311 if (IS_ERR(dp))
1312 goto exit_unlock;
1313
1314 reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1315 info->snd_seq, OVS_DP_CMD_DEL);
1316 err = PTR_ERR(reply);
1317 if (IS_ERR(reply))
1318 goto exit_unlock;
1319
1320 list_for_each_entry_safe(vport, next_vport, &dp->port_list, node)
1321 if (vport->port_no != OVSP_LOCAL)
1322 ovs_dp_detach_port(vport);
1323
1324 list_del(&dp->list_node);
1325 ovs_dp_detach_port(rtnl_dereference(dp->ports[OVSP_LOCAL]));
1326
1327 /* rtnl_unlock() will wait until all the references to devices that
1328 * are pending unregistration have been dropped. We do it here to
1329 * ensure that any internal devices (which contain DP pointers) are
1330 * fully destroyed before freeing the datapath.
1331 */
1332 rtnl_unlock();
1333
1334 call_rcu(&dp->rcu, destroy_dp_rcu);
1335 module_put(THIS_MODULE);
1336
1337 genl_notify(reply, genl_info_net(info), info->snd_pid,
1338 ovs_dp_datapath_multicast_group.id, info->nlhdr,
1339 GFP_KERNEL);
1340
1341 return 0;
1342
1343 exit_unlock:
1344 rtnl_unlock();
1345 return err;
1346 }
1347
1348 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1349 {
1350 struct sk_buff *reply;
1351 struct datapath *dp;
1352 int err;
1353
1354 dp = lookup_datapath(info->userhdr, info->attrs);
1355 if (IS_ERR(dp))
1356 return PTR_ERR(dp);
1357
1358 reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1359 info->snd_seq, OVS_DP_CMD_NEW);
1360 if (IS_ERR(reply)) {
1361 err = PTR_ERR(reply);
1362 netlink_set_err(init_net.genl_sock, 0,
1363 ovs_dp_datapath_multicast_group.id, err);
1364 return 0;
1365 }
1366
1367 genl_notify(reply, genl_info_net(info), info->snd_pid,
1368 ovs_dp_datapath_multicast_group.id, info->nlhdr,
1369 GFP_KERNEL);
1370
1371 return 0;
1372 }
1373
1374 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1375 {
1376 struct sk_buff *reply;
1377 struct datapath *dp;
1378
1379 dp = lookup_datapath(info->userhdr, info->attrs);
1380 if (IS_ERR(dp))
1381 return PTR_ERR(dp);
1382
1383 reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1384 info->snd_seq, OVS_DP_CMD_NEW);
1385 if (IS_ERR(reply))
1386 return PTR_ERR(reply);
1387
1388 return genlmsg_reply(reply, info);
1389 }
1390
1391 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1392 {
1393 struct datapath *dp;
1394 int skip = cb->args[0];
1395 int i = 0;
1396
1397 list_for_each_entry(dp, &dps, list_node) {
1398 if (i >= skip &&
1399 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).pid,
1400 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1401 OVS_DP_CMD_NEW) < 0)
1402 break;
1403 i++;
1404 }
1405
1406 cb->args[0] = i;
1407
1408 return skb->len;
1409 }
1410
1411 static struct genl_ops dp_datapath_genl_ops[] = {
1412 { .cmd = OVS_DP_CMD_NEW,
1413 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1414 .policy = datapath_policy,
1415 .doit = ovs_dp_cmd_new
1416 },
1417 { .cmd = OVS_DP_CMD_DEL,
1418 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1419 .policy = datapath_policy,
1420 .doit = ovs_dp_cmd_del
1421 },
1422 { .cmd = OVS_DP_CMD_GET,
1423 .flags = 0, /* OK for unprivileged users. */
1424 .policy = datapath_policy,
1425 .doit = ovs_dp_cmd_get,
1426 .dumpit = ovs_dp_cmd_dump
1427 },
1428 { .cmd = OVS_DP_CMD_SET,
1429 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1430 .policy = datapath_policy,
1431 .doit = ovs_dp_cmd_set,
1432 },
1433 };
1434
1435 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1436 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1437 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1438 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1439 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1440 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1441 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1442 };
1443
1444 static struct genl_family dp_vport_genl_family = {
1445 .id = GENL_ID_GENERATE,
1446 .hdrsize = sizeof(struct ovs_header),
1447 .name = OVS_VPORT_FAMILY,
1448 .version = OVS_VPORT_VERSION,
1449 .maxattr = OVS_VPORT_ATTR_MAX
1450 };
1451
1452 struct genl_multicast_group ovs_dp_vport_multicast_group = {
1453 .name = OVS_VPORT_MCGROUP
1454 };
1455
1456 /* Called with RTNL lock or RCU read lock. */
1457 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1458 u32 pid, u32 seq, u32 flags, u8 cmd)
1459 {
1460 struct ovs_header *ovs_header;
1461 struct ovs_vport_stats vport_stats;
1462 int err;
1463
1464 ovs_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family,
1465 flags, cmd);
1466 if (!ovs_header)
1467 return -EMSGSIZE;
1468
1469 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1470
1471 NLA_PUT_U32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no);
1472 NLA_PUT_U32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type);
1473 NLA_PUT_STRING(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport));
1474 NLA_PUT_U32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_pid);
1475
1476 ovs_vport_get_stats(vport, &vport_stats);
1477 NLA_PUT(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1478 &vport_stats);
1479
1480 err = ovs_vport_get_options(vport, skb);
1481 if (err == -EMSGSIZE)
1482 goto error;
1483
1484 return genlmsg_end(skb, ovs_header);
1485
1486 nla_put_failure:
1487 err = -EMSGSIZE;
1488 error:
1489 genlmsg_cancel(skb, ovs_header);
1490 return err;
1491 }
1492
1493 /* Called with RTNL lock or RCU read lock. */
1494 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 pid,
1495 u32 seq, u8 cmd)
1496 {
1497 struct sk_buff *skb;
1498 int retval;
1499
1500 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1501 if (!skb)
1502 return ERR_PTR(-ENOMEM);
1503
1504 retval = ovs_vport_cmd_fill_info(vport, skb, pid, seq, 0, cmd);
1505 if (retval < 0) {
1506 kfree_skb(skb);
1507 return ERR_PTR(retval);
1508 }
1509 return skb;
1510 }
1511
1512 /* Called with RTNL lock or RCU read lock. */
1513 static struct vport *lookup_vport(struct ovs_header *ovs_header,
1514 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1515 {
1516 struct datapath *dp;
1517 struct vport *vport;
1518
1519 if (a[OVS_VPORT_ATTR_NAME]) {
1520 vport = ovs_vport_locate(nla_data(a[OVS_VPORT_ATTR_NAME]));
1521 if (!vport)
1522 return ERR_PTR(-ENODEV);
1523 if (ovs_header->dp_ifindex &&
1524 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1525 return ERR_PTR(-ENODEV);
1526 return vport;
1527 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1528 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1529
1530 if (port_no >= DP_MAX_PORTS)
1531 return ERR_PTR(-EFBIG);
1532
1533 dp = get_dp(ovs_header->dp_ifindex);
1534 if (!dp)
1535 return ERR_PTR(-ENODEV);
1536
1537 vport = rcu_dereference_rtnl(dp->ports[port_no]);
1538 if (!vport)
1539 return ERR_PTR(-ENOENT);
1540 return vport;
1541 } else
1542 return ERR_PTR(-EINVAL);
1543 }
1544
1545 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1546 {
1547 struct nlattr **a = info->attrs;
1548 struct ovs_header *ovs_header = info->userhdr;
1549 struct vport_parms parms;
1550 struct sk_buff *reply;
1551 struct vport *vport;
1552 struct datapath *dp;
1553 u32 port_no;
1554 int err;
1555
1556 err = -EINVAL;
1557 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1558 !a[OVS_VPORT_ATTR_UPCALL_PID])
1559 goto exit;
1560
1561 rtnl_lock();
1562 dp = get_dp(ovs_header->dp_ifindex);
1563 err = -ENODEV;
1564 if (!dp)
1565 goto exit_unlock;
1566
1567 if (a[OVS_VPORT_ATTR_PORT_NO]) {
1568 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1569
1570 err = -EFBIG;
1571 if (port_no >= DP_MAX_PORTS)
1572 goto exit_unlock;
1573
1574 vport = rtnl_dereference(dp->ports[port_no]);
1575 err = -EBUSY;
1576 if (vport)
1577 goto exit_unlock;
1578 } else {
1579 for (port_no = 1; ; port_no++) {
1580 if (port_no >= DP_MAX_PORTS) {
1581 err = -EFBIG;
1582 goto exit_unlock;
1583 }
1584 vport = rtnl_dereference(dp->ports[port_no]);
1585 if (!vport)
1586 break;
1587 }
1588 }
1589
1590 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1591 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1592 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1593 parms.dp = dp;
1594 parms.port_no = port_no;
1595 parms.upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1596
1597 vport = new_vport(&parms);
1598 err = PTR_ERR(vport);
1599 if (IS_ERR(vport))
1600 goto exit_unlock;
1601
1602 reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1603 OVS_VPORT_CMD_NEW);
1604 if (IS_ERR(reply)) {
1605 err = PTR_ERR(reply);
1606 ovs_dp_detach_port(vport);
1607 goto exit_unlock;
1608 }
1609 genl_notify(reply, genl_info_net(info), info->snd_pid,
1610 ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1611
1612 exit_unlock:
1613 rtnl_unlock();
1614 exit:
1615 return err;
1616 }
1617
1618 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1619 {
1620 struct nlattr **a = info->attrs;
1621 struct sk_buff *reply;
1622 struct vport *vport;
1623 int err;
1624
1625 rtnl_lock();
1626 vport = lookup_vport(info->userhdr, a);
1627 err = PTR_ERR(vport);
1628 if (IS_ERR(vport))
1629 goto exit_unlock;
1630
1631 err = 0;
1632 if (a[OVS_VPORT_ATTR_TYPE] &&
1633 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type)
1634 err = -EINVAL;
1635
1636 if (!err && a[OVS_VPORT_ATTR_OPTIONS])
1637 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1638 if (!err && a[OVS_VPORT_ATTR_UPCALL_PID])
1639 vport->upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1640
1641 reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1642 OVS_VPORT_CMD_NEW);
1643 if (IS_ERR(reply)) {
1644 err = PTR_ERR(reply);
1645 netlink_set_err(init_net.genl_sock, 0,
1646 ovs_dp_vport_multicast_group.id, err);
1647 return 0;
1648 }
1649
1650 genl_notify(reply, genl_info_net(info), info->snd_pid,
1651 ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1652
1653 exit_unlock:
1654 rtnl_unlock();
1655 return err;
1656 }
1657
1658 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1659 {
1660 struct nlattr **a = info->attrs;
1661 struct sk_buff *reply;
1662 struct vport *vport;
1663 int err;
1664
1665 rtnl_lock();
1666 vport = lookup_vport(info->userhdr, a);
1667 err = PTR_ERR(vport);
1668 if (IS_ERR(vport))
1669 goto exit_unlock;
1670
1671 if (vport->port_no == OVSP_LOCAL) {
1672 err = -EINVAL;
1673 goto exit_unlock;
1674 }
1675
1676 reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1677 OVS_VPORT_CMD_DEL);
1678 err = PTR_ERR(reply);
1679 if (IS_ERR(reply))
1680 goto exit_unlock;
1681
1682 ovs_dp_detach_port(vport);
1683
1684 genl_notify(reply, genl_info_net(info), info->snd_pid,
1685 ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1686
1687 exit_unlock:
1688 rtnl_unlock();
1689 return err;
1690 }
1691
1692 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1693 {
1694 struct nlattr **a = info->attrs;
1695 struct ovs_header *ovs_header = info->userhdr;
1696 struct sk_buff *reply;
1697 struct vport *vport;
1698 int err;
1699
1700 rcu_read_lock();
1701 vport = lookup_vport(ovs_header, a);
1702 err = PTR_ERR(vport);
1703 if (IS_ERR(vport))
1704 goto exit_unlock;
1705
1706 reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1707 OVS_VPORT_CMD_NEW);
1708 err = PTR_ERR(reply);
1709 if (IS_ERR(reply))
1710 goto exit_unlock;
1711
1712 rcu_read_unlock();
1713
1714 return genlmsg_reply(reply, info);
1715
1716 exit_unlock:
1717 rcu_read_unlock();
1718 return err;
1719 }
1720
1721 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1722 {
1723 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1724 struct datapath *dp;
1725 u32 port_no;
1726 int retval;
1727
1728 dp = get_dp(ovs_header->dp_ifindex);
1729 if (!dp)
1730 return -ENODEV;
1731
1732 rcu_read_lock();
1733 for (port_no = cb->args[0]; port_no < DP_MAX_PORTS; port_no++) {
1734 struct vport *vport;
1735
1736 vport = rcu_dereference(dp->ports[port_no]);
1737 if (!vport)
1738 continue;
1739
1740 if (ovs_vport_cmd_fill_info(vport, skb, NETLINK_CB(cb->skb).pid,
1741 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1742 OVS_VPORT_CMD_NEW) < 0)
1743 break;
1744 }
1745 rcu_read_unlock();
1746
1747 cb->args[0] = port_no;
1748 retval = skb->len;
1749
1750 return retval;
1751 }
1752
1753 static void rehash_flow_table(struct work_struct *work)
1754 {
1755 struct datapath *dp;
1756
1757 genl_lock();
1758
1759 list_for_each_entry(dp, &dps, list_node) {
1760 struct flow_table *old_table = genl_dereference(dp->table);
1761 struct flow_table *new_table;
1762
1763 new_table = ovs_flow_tbl_rehash(old_table);
1764 if (!IS_ERR(new_table)) {
1765 rcu_assign_pointer(dp->table, new_table);
1766 ovs_flow_tbl_deferred_destroy(old_table);
1767 }
1768 }
1769
1770 genl_unlock();
1771
1772 schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
1773 }
1774
1775 static struct genl_ops dp_vport_genl_ops[] = {
1776 { .cmd = OVS_VPORT_CMD_NEW,
1777 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1778 .policy = vport_policy,
1779 .doit = ovs_vport_cmd_new
1780 },
1781 { .cmd = OVS_VPORT_CMD_DEL,
1782 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1783 .policy = vport_policy,
1784 .doit = ovs_vport_cmd_del
1785 },
1786 { .cmd = OVS_VPORT_CMD_GET,
1787 .flags = 0, /* OK for unprivileged users. */
1788 .policy = vport_policy,
1789 .doit = ovs_vport_cmd_get,
1790 .dumpit = ovs_vport_cmd_dump
1791 },
1792 { .cmd = OVS_VPORT_CMD_SET,
1793 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1794 .policy = vport_policy,
1795 .doit = ovs_vport_cmd_set,
1796 },
1797 };
1798
1799 struct genl_family_and_ops {
1800 struct genl_family *family;
1801 struct genl_ops *ops;
1802 int n_ops;
1803 struct genl_multicast_group *group;
1804 };
1805
1806 static const struct genl_family_and_ops dp_genl_families[] = {
1807 { &dp_datapath_genl_family,
1808 dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1809 &ovs_dp_datapath_multicast_group },
1810 { &dp_vport_genl_family,
1811 dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1812 &ovs_dp_vport_multicast_group },
1813 { &dp_flow_genl_family,
1814 dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1815 &ovs_dp_flow_multicast_group },
1816 { &dp_packet_genl_family,
1817 dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1818 NULL },
1819 };
1820
1821 static void dp_unregister_genl(int n_families)
1822 {
1823 int i;
1824
1825 for (i = 0; i < n_families; i++)
1826 genl_unregister_family(dp_genl_families[i].family);
1827 }
1828
1829 static int dp_register_genl(void)
1830 {
1831 int n_registered;
1832 int err;
1833 int i;
1834
1835 n_registered = 0;
1836 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
1837 const struct genl_family_and_ops *f = &dp_genl_families[i];
1838
1839 err = genl_register_family_with_ops(f->family, f->ops,
1840 f->n_ops);
1841 if (err)
1842 goto error;
1843 n_registered++;
1844
1845 if (f->group) {
1846 err = genl_register_mc_group(f->family, f->group);
1847 if (err)
1848 goto error;
1849 }
1850 }
1851
1852 return 0;
1853
1854 error:
1855 dp_unregister_genl(n_registered);
1856 return err;
1857 }
1858
1859 static int __init dp_init(void)
1860 {
1861 struct sk_buff *dummy_skb;
1862 int err;
1863
1864 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
1865
1866 pr_info("Open vSwitch switching datapath\n");
1867
1868 err = ovs_flow_init();
1869 if (err)
1870 goto error;
1871
1872 err = ovs_vport_init();
1873 if (err)
1874 goto error_flow_exit;
1875
1876 err = register_netdevice_notifier(&ovs_dp_device_notifier);
1877 if (err)
1878 goto error_vport_exit;
1879
1880 err = dp_register_genl();
1881 if (err < 0)
1882 goto error_unreg_notifier;
1883
1884 schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
1885
1886 return 0;
1887
1888 error_unreg_notifier:
1889 unregister_netdevice_notifier(&ovs_dp_device_notifier);
1890 error_vport_exit:
1891 ovs_vport_exit();
1892 error_flow_exit:
1893 ovs_flow_exit();
1894 error:
1895 return err;
1896 }
1897
1898 static void dp_cleanup(void)
1899 {
1900 cancel_delayed_work_sync(&rehash_flow_wq);
1901 rcu_barrier();
1902 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
1903 unregister_netdevice_notifier(&ovs_dp_device_notifier);
1904 ovs_vport_exit();
1905 ovs_flow_exit();
1906 }
1907
1908 module_init(dp_init);
1909 module_exit(dp_cleanup);
1910
1911 MODULE_DESCRIPTION("Open vSwitch switching datapath");
1912 MODULE_LICENSE("GPL");
Linus' kernel tree