?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
dm thin metadata: add dm_thin_changed_this_transaction
[linux-2.6.git] / net / openvswitch / datapath.c
blobd8277d29e7102caf343d78b802f3371283c40de1
1 /*
2 * Copyright (c) 2007-2012 Nicira, Inc.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of version 2 of the GNU General Public
6 * License as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public License
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
16 * 02110-1301, USA
17 */
18
19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
20
21 #include <linux/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 unsigned short gso_type = skb_shinfo(skb)->gso_type;
267 struct dp_upcall_info later_info;
268 struct sw_flow_key later_key;
269 struct sk_buff *segs, *nskb;
270 int err;
271
272 segs = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
273 if (IS_ERR(segs))
274 return PTR_ERR(segs);
275
276 /* Queue all of the segments. */
277 skb = segs;
278 do {
279 err = queue_userspace_packet(dp_ifindex, skb, upcall_info);
280 if (err)
281 break;
282
283 if (skb == segs && gso_type & SKB_GSO_UDP) {
284 /* The initial flow key extracted by ovs_flow_extract()
285 * in this case is for a first fragment, so we need to
286 * properly mark later fragments.
287 */
288 later_key = *upcall_info->key;
289 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
290
291 later_info = *upcall_info;
292 later_info.key = &later_key;
293 upcall_info = &later_info;
294 }
295 } while ((skb = skb->next));
296
297 /* Free all of the segments. */
298 skb = segs;
299 do {
300 nskb = skb->next;
301 if (err)
302 kfree_skb(skb);
303 else
304 consume_skb(skb);
305 } while ((skb = nskb));
306 return err;
307 }
308
309 static int queue_userspace_packet(int dp_ifindex, struct sk_buff *skb,
310 const struct dp_upcall_info *upcall_info)
311 {
312 struct ovs_header *upcall;
313 struct sk_buff *nskb = NULL;
314 struct sk_buff *user_skb; /* to be queued to userspace */
315 struct nlattr *nla;
316 unsigned int len;
317 int err;
318
319 if (vlan_tx_tag_present(skb)) {
320 nskb = skb_clone(skb, GFP_ATOMIC);
321 if (!nskb)
322 return -ENOMEM;
323
324 nskb = __vlan_put_tag(nskb, vlan_tx_tag_get(nskb));
325 if (!nskb)
326 return -ENOMEM;
327
328 nskb->vlan_tci = 0;
329 skb = nskb;
330 }
331
332 if (nla_attr_size(skb->len) > USHRT_MAX) {
333 err = -EFBIG;
334 goto out;
335 }
336
337 len = sizeof(struct ovs_header);
338 len += nla_total_size(skb->len);
339 len += nla_total_size(FLOW_BUFSIZE);
340 if (upcall_info->cmd == OVS_PACKET_CMD_ACTION)
341 len += nla_total_size(8);
342
343 user_skb = genlmsg_new(len, GFP_ATOMIC);
344 if (!user_skb) {
345 err = -ENOMEM;
346 goto out;
347 }
348
349 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
350 0, upcall_info->cmd);
351 upcall->dp_ifindex = dp_ifindex;
352
353 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
354 ovs_flow_to_nlattrs(upcall_info->key, user_skb);
355 nla_nest_end(user_skb, nla);
356
357 if (upcall_info->userdata)
358 nla_put_u64(user_skb, OVS_PACKET_ATTR_USERDATA,
359 nla_get_u64(upcall_info->userdata));
360
361 nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
362
363 skb_copy_and_csum_dev(skb, nla_data(nla));
364
365 err = genlmsg_unicast(&init_net, user_skb, upcall_info->pid);
366
367 out:
368 kfree_skb(nskb);
369 return err;
370 }
371
372 /* Called with genl_mutex. */
373 static int flush_flows(int dp_ifindex)
374 {
375 struct flow_table *old_table;
376 struct flow_table *new_table;
377 struct datapath *dp;
378
379 dp = get_dp(dp_ifindex);
380 if (!dp)
381 return -ENODEV;
382
383 old_table = genl_dereference(dp->table);
384 new_table = ovs_flow_tbl_alloc(TBL_MIN_BUCKETS);
385 if (!new_table)
386 return -ENOMEM;
387
388 rcu_assign_pointer(dp->table, new_table);
389
390 ovs_flow_tbl_deferred_destroy(old_table);
391 return 0;
392 }
393
394 static int validate_actions(const struct nlattr *attr,
395 const struct sw_flow_key *key, int depth);
396
397 static int validate_sample(const struct nlattr *attr,
398 const struct sw_flow_key *key, int depth)
399 {
400 const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
401 const struct nlattr *probability, *actions;
402 const struct nlattr *a;
403 int rem;
404
405 memset(attrs, 0, sizeof(attrs));
406 nla_for_each_nested(a, attr, rem) {
407 int type = nla_type(a);
408 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
409 return -EINVAL;
410 attrs[type] = a;
411 }
412 if (rem)
413 return -EINVAL;
414
415 probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
416 if (!probability || nla_len(probability) != sizeof(u32))
417 return -EINVAL;
418
419 actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
420 if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
421 return -EINVAL;
422 return validate_actions(actions, key, depth + 1);
423 }
424
425 static int validate_tp_port(const struct sw_flow_key *flow_key)
426 {
427 if (flow_key->eth.type == htons(ETH_P_IP)) {
428 if (flow_key->ipv4.tp.src && flow_key->ipv4.tp.dst)
429 return 0;
430 } else if (flow_key->eth.type == htons(ETH_P_IPV6)) {
431 if (flow_key->ipv6.tp.src && flow_key->ipv6.tp.dst)
432 return 0;
433 }
434
435 return -EINVAL;
436 }
437
438 static int validate_set(const struct nlattr *a,
439 const struct sw_flow_key *flow_key)
440 {
441 const struct nlattr *ovs_key = nla_data(a);
442 int key_type = nla_type(ovs_key);
443
444 /* There can be only one key in a action */
445 if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
446 return -EINVAL;
447
448 if (key_type > OVS_KEY_ATTR_MAX ||
449 nla_len(ovs_key) != ovs_key_lens[key_type])
450 return -EINVAL;
451
452 switch (key_type) {
453 const struct ovs_key_ipv4 *ipv4_key;
454
455 case OVS_KEY_ATTR_PRIORITY:
456 case OVS_KEY_ATTR_ETHERNET:
457 break;
458
459 case OVS_KEY_ATTR_IPV4:
460 if (flow_key->eth.type != htons(ETH_P_IP))
461 return -EINVAL;
462
463 if (!flow_key->ipv4.addr.src || !flow_key->ipv4.addr.dst)
464 return -EINVAL;
465
466 ipv4_key = nla_data(ovs_key);
467 if (ipv4_key->ipv4_proto != flow_key->ip.proto)
468 return -EINVAL;
469
470 if (ipv4_key->ipv4_frag != flow_key->ip.frag)
471 return -EINVAL;
472
473 break;
474
475 case OVS_KEY_ATTR_TCP:
476 if (flow_key->ip.proto != IPPROTO_TCP)
477 return -EINVAL;
478
479 return validate_tp_port(flow_key);
480
481 case OVS_KEY_ATTR_UDP:
482 if (flow_key->ip.proto != IPPROTO_UDP)
483 return -EINVAL;
484
485 return validate_tp_port(flow_key);
486
487 default:
488 return -EINVAL;
489 }
490
491 return 0;
492 }
493
494 static int validate_userspace(const struct nlattr *attr)
495 {
496 static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] = {
497 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
498 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_U64 },
499 };
500 struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
501 int error;
502
503 error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
504 attr, userspace_policy);
505 if (error)
506 return error;
507
508 if (!a[OVS_USERSPACE_ATTR_PID] ||
509 !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
510 return -EINVAL;
511
512 return 0;
513 }
514
515 static int validate_actions(const struct nlattr *attr,
516 const struct sw_flow_key *key, int depth)
517 {
518 const struct nlattr *a;
519 int rem, err;
520
521 if (depth >= SAMPLE_ACTION_DEPTH)
522 return -EOVERFLOW;
523
524 nla_for_each_nested(a, attr, rem) {
525 /* Expected argument lengths, (u32)-1 for variable length. */
526 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
527 [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
528 [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
529 [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
530 [OVS_ACTION_ATTR_POP_VLAN] = 0,
531 [OVS_ACTION_ATTR_SET] = (u32)-1,
532 [OVS_ACTION_ATTR_SAMPLE] = (u32)-1
533 };
534 const struct ovs_action_push_vlan *vlan;
535 int type = nla_type(a);
536
537 if (type > OVS_ACTION_ATTR_MAX ||
538 (action_lens[type] != nla_len(a) &&
539 action_lens[type] != (u32)-1))
540 return -EINVAL;
541
542 switch (type) {
543 case OVS_ACTION_ATTR_UNSPEC:
544 return -EINVAL;
545
546 case OVS_ACTION_ATTR_USERSPACE:
547 err = validate_userspace(a);
548 if (err)
549 return err;
550 break;
551
552 case OVS_ACTION_ATTR_OUTPUT:
553 if (nla_get_u32(a) >= DP_MAX_PORTS)
554 return -EINVAL;
555 break;
556
557
558 case OVS_ACTION_ATTR_POP_VLAN:
559 break;
560
561 case OVS_ACTION_ATTR_PUSH_VLAN:
562 vlan = nla_data(a);
563 if (vlan->vlan_tpid != htons(ETH_P_8021Q))
564 return -EINVAL;
565 if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
566 return -EINVAL;
567 break;
568
569 case OVS_ACTION_ATTR_SET:
570 err = validate_set(a, key);
571 if (err)
572 return err;
573 break;
574
575 case OVS_ACTION_ATTR_SAMPLE:
576 err = validate_sample(a, key, depth);
577 if (err)
578 return err;
579 break;
580
581 default:
582 return -EINVAL;
583 }
584 }
585
586 if (rem > 0)
587 return -EINVAL;
588
589 return 0;
590 }
591
592 static void clear_stats(struct sw_flow *flow)
593 {
594 flow->used = 0;
595 flow->tcp_flags = 0;
596 flow->packet_count = 0;
597 flow->byte_count = 0;
598 }
599
600 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
601 {
602 struct ovs_header *ovs_header = info->userhdr;
603 struct nlattr **a = info->attrs;
604 struct sw_flow_actions *acts;
605 struct sk_buff *packet;
606 struct sw_flow *flow;
607 struct datapath *dp;
608 struct ethhdr *eth;
609 int len;
610 int err;
611 int key_len;
612
613 err = -EINVAL;
614 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
615 !a[OVS_PACKET_ATTR_ACTIONS] ||
616 nla_len(a[OVS_PACKET_ATTR_PACKET]) < ETH_HLEN)
617 goto err;
618
619 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
620 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
621 err = -ENOMEM;
622 if (!packet)
623 goto err;
624 skb_reserve(packet, NET_IP_ALIGN);
625
626 memcpy(__skb_put(packet, len), nla_data(a[OVS_PACKET_ATTR_PACKET]), len);
627
628 skb_reset_mac_header(packet);
629 eth = eth_hdr(packet);
630
631 /* Normally, setting the skb 'protocol' field would be handled by a
632 * call to eth_type_trans(), but it assumes there's a sending
633 * device, which we may not have. */
634 if (ntohs(eth->h_proto) >= 1536)
635 packet->protocol = eth->h_proto;
636 else
637 packet->protocol = htons(ETH_P_802_2);
638
639 /* Build an sw_flow for sending this packet. */
640 flow = ovs_flow_alloc();
641 err = PTR_ERR(flow);
642 if (IS_ERR(flow))
643 goto err_kfree_skb;
644
645 err = ovs_flow_extract(packet, -1, &flow->key, &key_len);
646 if (err)
647 goto err_flow_free;
648
649 err = ovs_flow_metadata_from_nlattrs(&flow->key.phy.priority,
650 &flow->key.phy.in_port,
651 a[OVS_PACKET_ATTR_KEY]);
652 if (err)
653 goto err_flow_free;
654
655 err = validate_actions(a[OVS_PACKET_ATTR_ACTIONS], &flow->key, 0);
656 if (err)
657 goto err_flow_free;
658
659 flow->hash = ovs_flow_hash(&flow->key, key_len);
660
661 acts = ovs_flow_actions_alloc(a[OVS_PACKET_ATTR_ACTIONS]);
662 err = PTR_ERR(acts);
663 if (IS_ERR(acts))
664 goto err_flow_free;
665 rcu_assign_pointer(flow->sf_acts, acts);
666
667 OVS_CB(packet)->flow = flow;
668 packet->priority = flow->key.phy.priority;
669
670 rcu_read_lock();
671 dp = get_dp(ovs_header->dp_ifindex);
672 err = -ENODEV;
673 if (!dp)
674 goto err_unlock;
675
676 local_bh_disable();
677 err = ovs_execute_actions(dp, packet);
678 local_bh_enable();
679 rcu_read_unlock();
680
681 ovs_flow_free(flow);
682 return err;
683
684 err_unlock:
685 rcu_read_unlock();
686 err_flow_free:
687 ovs_flow_free(flow);
688 err_kfree_skb:
689 kfree_skb(packet);
690 err:
691 return err;
692 }
693
694 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
695 [OVS_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
696 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
697 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
698 };
699
700 static struct genl_ops dp_packet_genl_ops[] = {
701 { .cmd = OVS_PACKET_CMD_EXECUTE,
702 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
703 .policy = packet_policy,
704 .doit = ovs_packet_cmd_execute
705 }
706 };
707
708 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
709 {
710 int i;
711 struct flow_table *table = genl_dereference(dp->table);
712
713 stats->n_flows = ovs_flow_tbl_count(table);
714
715 stats->n_hit = stats->n_missed = stats->n_lost = 0;
716 for_each_possible_cpu(i) {
717 const struct dp_stats_percpu *percpu_stats;
718 struct dp_stats_percpu local_stats;
719 unsigned int start;
720
721 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
722
723 do {
724 start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
725 local_stats = *percpu_stats;
726 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
727
728 stats->n_hit += local_stats.n_hit;
729 stats->n_missed += local_stats.n_missed;
730 stats->n_lost += local_stats.n_lost;
731 }
732 }
733
734 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
735 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
736 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
737 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
738 };
739
740 static struct genl_family dp_flow_genl_family = {
741 .id = GENL_ID_GENERATE,
742 .hdrsize = sizeof(struct ovs_header),
743 .name = OVS_FLOW_FAMILY,
744 .version = OVS_FLOW_VERSION,
745 .maxattr = OVS_FLOW_ATTR_MAX
746 };
747
748 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
749 .name = OVS_FLOW_MCGROUP
750 };
751
752 /* Called with genl_lock. */
753 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
754 struct sk_buff *skb, u32 pid,
755 u32 seq, u32 flags, u8 cmd)
756 {
757 const int skb_orig_len = skb->len;
758 const struct sw_flow_actions *sf_acts;
759 struct ovs_flow_stats stats;
760 struct ovs_header *ovs_header;
761 struct nlattr *nla;
762 unsigned long used;
763 u8 tcp_flags;
764 int err;
765
766 sf_acts = rcu_dereference_protected(flow->sf_acts,
767 lockdep_genl_is_held());
768
769 ovs_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags, cmd);
770 if (!ovs_header)
771 return -EMSGSIZE;
772
773 ovs_header->dp_ifindex = get_dpifindex(dp);
774
775 nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
776 if (!nla)
777 goto nla_put_failure;
778 err = ovs_flow_to_nlattrs(&flow->key, skb);
779 if (err)
780 goto error;
781 nla_nest_end(skb, nla);
782
783 spin_lock_bh(&flow->lock);
784 used = flow->used;
785 stats.n_packets = flow->packet_count;
786 stats.n_bytes = flow->byte_count;
787 tcp_flags = flow->tcp_flags;
788 spin_unlock_bh(&flow->lock);
789
790 if (used &&
791 nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
792 goto nla_put_failure;
793
794 if (stats.n_packets &&
795 nla_put(skb, OVS_FLOW_ATTR_STATS,
796 sizeof(struct ovs_flow_stats), &stats))
797 goto nla_put_failure;
798
799 if (tcp_flags &&
800 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags))
801 goto nla_put_failure;
802
803 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
804 * this is the first flow to be dumped into 'skb'. This is unusual for
805 * Netlink but individual action lists can be longer than
806 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
807 * The userspace caller can always fetch the actions separately if it
808 * really wants them. (Most userspace callers in fact don't care.)
809 *
810 * This can only fail for dump operations because the skb is always
811 * properly sized for single flows.
812 */
813 err = nla_put(skb, OVS_FLOW_ATTR_ACTIONS, sf_acts->actions_len,
814 sf_acts->actions);
815 if (err < 0 && skb_orig_len)
816 goto error;
817
818 return genlmsg_end(skb, ovs_header);
819
820 nla_put_failure:
821 err = -EMSGSIZE;
822 error:
823 genlmsg_cancel(skb, ovs_header);
824 return err;
825 }
826
827 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
828 {
829 const struct sw_flow_actions *sf_acts;
830 int len;
831
832 sf_acts = rcu_dereference_protected(flow->sf_acts,
833 lockdep_genl_is_held());
834
835 /* OVS_FLOW_ATTR_KEY */
836 len = nla_total_size(FLOW_BUFSIZE);
837 /* OVS_FLOW_ATTR_ACTIONS */
838 len += nla_total_size(sf_acts->actions_len);
839 /* OVS_FLOW_ATTR_STATS */
840 len += nla_total_size(sizeof(struct ovs_flow_stats));
841 /* OVS_FLOW_ATTR_TCP_FLAGS */
842 len += nla_total_size(1);
843 /* OVS_FLOW_ATTR_USED */
844 len += nla_total_size(8);
845
846 len += NLMSG_ALIGN(sizeof(struct ovs_header));
847
848 return genlmsg_new(len, GFP_KERNEL);
849 }
850
851 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
852 struct datapath *dp,
853 u32 pid, u32 seq, u8 cmd)
854 {
855 struct sk_buff *skb;
856 int retval;
857
858 skb = ovs_flow_cmd_alloc_info(flow);
859 if (!skb)
860 return ERR_PTR(-ENOMEM);
861
862 retval = ovs_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd);
863 BUG_ON(retval < 0);
864 return skb;
865 }
866
867 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
868 {
869 struct nlattr **a = info->attrs;
870 struct ovs_header *ovs_header = info->userhdr;
871 struct sw_flow_key key;
872 struct sw_flow *flow;
873 struct sk_buff *reply;
874 struct datapath *dp;
875 struct flow_table *table;
876 int error;
877 int key_len;
878
879 /* Extract key. */
880 error = -EINVAL;
881 if (!a[OVS_FLOW_ATTR_KEY])
882 goto error;
883 error = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
884 if (error)
885 goto error;
886
887 /* Validate actions. */
888 if (a[OVS_FLOW_ATTR_ACTIONS]) {
889 error = validate_actions(a[OVS_FLOW_ATTR_ACTIONS], &key, 0);
890 if (error)
891 goto error;
892 } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
893 error = -EINVAL;
894 goto error;
895 }
896
897 dp = get_dp(ovs_header->dp_ifindex);
898 error = -ENODEV;
899 if (!dp)
900 goto error;
901
902 table = genl_dereference(dp->table);
903 flow = ovs_flow_tbl_lookup(table, &key, key_len);
904 if (!flow) {
905 struct sw_flow_actions *acts;
906
907 /* Bail out if we're not allowed to create a new flow. */
908 error = -ENOENT;
909 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
910 goto error;
911
912 /* Expand table, if necessary, to make room. */
913 if (ovs_flow_tbl_need_to_expand(table)) {
914 struct flow_table *new_table;
915
916 new_table = ovs_flow_tbl_expand(table);
917 if (!IS_ERR(new_table)) {
918 rcu_assign_pointer(dp->table, new_table);
919 ovs_flow_tbl_deferred_destroy(table);
920 table = genl_dereference(dp->table);
921 }
922 }
923
924 /* Allocate flow. */
925 flow = ovs_flow_alloc();
926 if (IS_ERR(flow)) {
927 error = PTR_ERR(flow);
928 goto error;
929 }
930 flow->key = key;
931 clear_stats(flow);
932
933 /* Obtain actions. */
934 acts = ovs_flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]);
935 error = PTR_ERR(acts);
936 if (IS_ERR(acts))
937 goto error_free_flow;
938 rcu_assign_pointer(flow->sf_acts, acts);
939
940 /* Put flow in bucket. */
941 flow->hash = ovs_flow_hash(&key, key_len);
942 ovs_flow_tbl_insert(table, flow);
943
944 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
945 info->snd_seq,
946 OVS_FLOW_CMD_NEW);
947 } else {
948 /* We found a matching flow. */
949 struct sw_flow_actions *old_acts;
950 struct nlattr *acts_attrs;
951
952 /* Bail out if we're not allowed to modify an existing flow.
953 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
954 * because Generic Netlink treats the latter as a dump
955 * request. We also accept NLM_F_EXCL in case that bug ever
956 * gets fixed.
957 */
958 error = -EEXIST;
959 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
960 info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
961 goto error;
962
963 /* Update actions. */
964 old_acts = rcu_dereference_protected(flow->sf_acts,
965 lockdep_genl_is_held());
966 acts_attrs = a[OVS_FLOW_ATTR_ACTIONS];
967 if (acts_attrs &&
968 (old_acts->actions_len != nla_len(acts_attrs) ||
969 memcmp(old_acts->actions, nla_data(acts_attrs),
970 old_acts->actions_len))) {
971 struct sw_flow_actions *new_acts;
972
973 new_acts = ovs_flow_actions_alloc(acts_attrs);
974 error = PTR_ERR(new_acts);
975 if (IS_ERR(new_acts))
976 goto error;
977
978 rcu_assign_pointer(flow->sf_acts, new_acts);
979 ovs_flow_deferred_free_acts(old_acts);
980 }
981
982 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
983 info->snd_seq, OVS_FLOW_CMD_NEW);
984
985 /* Clear stats. */
986 if (a[OVS_FLOW_ATTR_CLEAR]) {
987 spin_lock_bh(&flow->lock);
988 clear_stats(flow);
989 spin_unlock_bh(&flow->lock);
990 }
991 }
992
993 if (!IS_ERR(reply))
994 genl_notify(reply, genl_info_net(info), info->snd_pid,
995 ovs_dp_flow_multicast_group.id, info->nlhdr,
996 GFP_KERNEL);
997 else
998 netlink_set_err(init_net.genl_sock, 0,
999 ovs_dp_flow_multicast_group.id, PTR_ERR(reply));
1000 return 0;
1001
1002 error_free_flow:
1003 ovs_flow_free(flow);
1004 error:
1005 return error;
1006 }
1007
1008 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1009 {
1010 struct nlattr **a = info->attrs;
1011 struct ovs_header *ovs_header = info->userhdr;
1012 struct sw_flow_key key;
1013 struct sk_buff *reply;
1014 struct sw_flow *flow;
1015 struct datapath *dp;
1016 struct flow_table *table;
1017 int err;
1018 int key_len;
1019
1020 if (!a[OVS_FLOW_ATTR_KEY])
1021 return -EINVAL;
1022 err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1023 if (err)
1024 return err;
1025
1026 dp = get_dp(ovs_header->dp_ifindex);
1027 if (!dp)
1028 return -ENODEV;
1029
1030 table = genl_dereference(dp->table);
1031 flow = ovs_flow_tbl_lookup(table, &key, key_len);
1032 if (!flow)
1033 return -ENOENT;
1034
1035 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1036 info->snd_seq, OVS_FLOW_CMD_NEW);
1037 if (IS_ERR(reply))
1038 return PTR_ERR(reply);
1039
1040 return genlmsg_reply(reply, info);
1041 }
1042
1043 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1044 {
1045 struct nlattr **a = info->attrs;
1046 struct ovs_header *ovs_header = info->userhdr;
1047 struct sw_flow_key key;
1048 struct sk_buff *reply;
1049 struct sw_flow *flow;
1050 struct datapath *dp;
1051 struct flow_table *table;
1052 int err;
1053 int key_len;
1054
1055 if (!a[OVS_FLOW_ATTR_KEY])
1056 return flush_flows(ovs_header->dp_ifindex);
1057 err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1058 if (err)
1059 return err;
1060
1061 dp = get_dp(ovs_header->dp_ifindex);
1062 if (!dp)
1063 return -ENODEV;
1064
1065 table = genl_dereference(dp->table);
1066 flow = ovs_flow_tbl_lookup(table, &key, key_len);
1067 if (!flow)
1068 return -ENOENT;
1069
1070 reply = ovs_flow_cmd_alloc_info(flow);
1071 if (!reply)
1072 return -ENOMEM;
1073
1074 ovs_flow_tbl_remove(table, flow);
1075
1076 err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_pid,
1077 info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1078 BUG_ON(err < 0);
1079
1080 ovs_flow_deferred_free(flow);
1081
1082 genl_notify(reply, genl_info_net(info), info->snd_pid,
1083 ovs_dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1084 return 0;
1085 }
1086
1087 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1088 {
1089 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1090 struct datapath *dp;
1091 struct flow_table *table;
1092
1093 dp = get_dp(ovs_header->dp_ifindex);
1094 if (!dp)
1095 return -ENODEV;
1096
1097 table = genl_dereference(dp->table);
1098
1099 for (;;) {
1100 struct sw_flow *flow;
1101 u32 bucket, obj;
1102
1103 bucket = cb->args[0];
1104 obj = cb->args[1];
1105 flow = ovs_flow_tbl_next(table, &bucket, &obj);
1106 if (!flow)
1107 break;
1108
1109 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1110 NETLINK_CB(cb->skb).pid,
1111 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1112 OVS_FLOW_CMD_NEW) < 0)
1113 break;
1114
1115 cb->args[0] = bucket;
1116 cb->args[1] = obj;
1117 }
1118 return skb->len;
1119 }
1120
1121 static struct genl_ops dp_flow_genl_ops[] = {
1122 { .cmd = OVS_FLOW_CMD_NEW,
1123 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1124 .policy = flow_policy,
1125 .doit = ovs_flow_cmd_new_or_set
1126 },
1127 { .cmd = OVS_FLOW_CMD_DEL,
1128 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1129 .policy = flow_policy,
1130 .doit = ovs_flow_cmd_del
1131 },
1132 { .cmd = OVS_FLOW_CMD_GET,
1133 .flags = 0, /* OK for unprivileged users. */
1134 .policy = flow_policy,
1135 .doit = ovs_flow_cmd_get,
1136 .dumpit = ovs_flow_cmd_dump
1137 },
1138 { .cmd = OVS_FLOW_CMD_SET,
1139 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1140 .policy = flow_policy,
1141 .doit = ovs_flow_cmd_new_or_set,
1142 },
1143 };
1144
1145 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1146 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1147 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1148 };
1149
1150 static struct genl_family dp_datapath_genl_family = {
1151 .id = GENL_ID_GENERATE,
1152 .hdrsize = sizeof(struct ovs_header),
1153 .name = OVS_DATAPATH_FAMILY,
1154 .version = OVS_DATAPATH_VERSION,
1155 .maxattr = OVS_DP_ATTR_MAX
1156 };
1157
1158 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1159 .name = OVS_DATAPATH_MCGROUP
1160 };
1161
1162 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1163 u32 pid, u32 seq, u32 flags, u8 cmd)
1164 {
1165 struct ovs_header *ovs_header;
1166 struct ovs_dp_stats dp_stats;
1167 int err;
1168
1169 ovs_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family,
1170 flags, cmd);
1171 if (!ovs_header)
1172 goto error;
1173
1174 ovs_header->dp_ifindex = get_dpifindex(dp);
1175
1176 rcu_read_lock();
1177 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1178 rcu_read_unlock();
1179 if (err)
1180 goto nla_put_failure;
1181
1182 get_dp_stats(dp, &dp_stats);
1183 if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats), &dp_stats))
1184 goto nla_put_failure;
1185
1186 return genlmsg_end(skb, ovs_header);
1187
1188 nla_put_failure:
1189 genlmsg_cancel(skb, ovs_header);
1190 error:
1191 return -EMSGSIZE;
1192 }
1193
1194 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 pid,
1195 u32 seq, u8 cmd)
1196 {
1197 struct sk_buff *skb;
1198 int retval;
1199
1200 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1201 if (!skb)
1202 return ERR_PTR(-ENOMEM);
1203
1204 retval = ovs_dp_cmd_fill_info(dp, skb, pid, seq, 0, cmd);
1205 if (retval < 0) {
1206 kfree_skb(skb);
1207 return ERR_PTR(retval);
1208 }
1209 return skb;
1210 }
1211
1212 /* Called with genl_mutex and optionally with RTNL lock also. */
1213 static struct datapath *lookup_datapath(struct ovs_header *ovs_header,
1214 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1215 {
1216 struct datapath *dp;
1217
1218 if (!a[OVS_DP_ATTR_NAME])
1219 dp = get_dp(ovs_header->dp_ifindex);
1220 else {
1221 struct vport *vport;
1222
1223 rcu_read_lock();
1224 vport = ovs_vport_locate(nla_data(a[OVS_DP_ATTR_NAME]));
1225 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1226 rcu_read_unlock();
1227 }
1228 return dp ? dp : ERR_PTR(-ENODEV);
1229 }
1230
1231 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1232 {
1233 struct nlattr **a = info->attrs;
1234 struct vport_parms parms;
1235 struct sk_buff *reply;
1236 struct datapath *dp;
1237 struct vport *vport;
1238 int err;
1239
1240 err = -EINVAL;
1241 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1242 goto err;
1243
1244 rtnl_lock();
1245 err = -ENODEV;
1246 if (!try_module_get(THIS_MODULE))
1247 goto err_unlock_rtnl;
1248
1249 err = -ENOMEM;
1250 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1251 if (dp == NULL)
1252 goto err_put_module;
1253 INIT_LIST_HEAD(&dp->port_list);
1254
1255 /* Allocate table. */
1256 err = -ENOMEM;
1257 rcu_assign_pointer(dp->table, ovs_flow_tbl_alloc(TBL_MIN_BUCKETS));
1258 if (!dp->table)
1259 goto err_free_dp;
1260
1261 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1262 if (!dp->stats_percpu) {
1263 err = -ENOMEM;
1264 goto err_destroy_table;
1265 }
1266
1267 /* Set up our datapath device. */
1268 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1269 parms.type = OVS_VPORT_TYPE_INTERNAL;
1270 parms.options = NULL;
1271 parms.dp = dp;
1272 parms.port_no = OVSP_LOCAL;
1273 parms.upcall_pid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1274
1275 vport = new_vport(&parms);
1276 if (IS_ERR(vport)) {
1277 err = PTR_ERR(vport);
1278 if (err == -EBUSY)
1279 err = -EEXIST;
1280
1281 goto err_destroy_percpu;
1282 }
1283
1284 reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1285 info->snd_seq, OVS_DP_CMD_NEW);
1286 err = PTR_ERR(reply);
1287 if (IS_ERR(reply))
1288 goto err_destroy_local_port;
1289
1290 list_add_tail(&dp->list_node, &dps);
1291 rtnl_unlock();
1292
1293 genl_notify(reply, genl_info_net(info), info->snd_pid,
1294 ovs_dp_datapath_multicast_group.id, info->nlhdr,
1295 GFP_KERNEL);
1296 return 0;
1297
1298 err_destroy_local_port:
1299 ovs_dp_detach_port(rtnl_dereference(dp->ports[OVSP_LOCAL]));
1300 err_destroy_percpu:
1301 free_percpu(dp->stats_percpu);
1302 err_destroy_table:
1303 ovs_flow_tbl_destroy(genl_dereference(dp->table));
1304 err_free_dp:
1305 kfree(dp);
1306 err_put_module:
1307 module_put(THIS_MODULE);
1308 err_unlock_rtnl:
1309 rtnl_unlock();
1310 err:
1311 return err;
1312 }
1313
1314 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1315 {
1316 struct vport *vport, *next_vport;
1317 struct sk_buff *reply;
1318 struct datapath *dp;
1319 int err;
1320
1321 rtnl_lock();
1322 dp = lookup_datapath(info->userhdr, info->attrs);
1323 err = PTR_ERR(dp);
1324 if (IS_ERR(dp))
1325 goto exit_unlock;
1326
1327 reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1328 info->snd_seq, OVS_DP_CMD_DEL);
1329 err = PTR_ERR(reply);
1330 if (IS_ERR(reply))
1331 goto exit_unlock;
1332
1333 list_for_each_entry_safe(vport, next_vport, &dp->port_list, node)
1334 if (vport->port_no != OVSP_LOCAL)
1335 ovs_dp_detach_port(vport);
1336
1337 list_del(&dp->list_node);
1338 ovs_dp_detach_port(rtnl_dereference(dp->ports[OVSP_LOCAL]));
1339
1340 /* rtnl_unlock() will wait until all the references to devices that
1341 * are pending unregistration have been dropped. We do it here to
1342 * ensure that any internal devices (which contain DP pointers) are
1343 * fully destroyed before freeing the datapath.
1344 */
1345 rtnl_unlock();
1346
1347 call_rcu(&dp->rcu, destroy_dp_rcu);
1348 module_put(THIS_MODULE);
1349
1350 genl_notify(reply, genl_info_net(info), info->snd_pid,
1351 ovs_dp_datapath_multicast_group.id, info->nlhdr,
1352 GFP_KERNEL);
1353
1354 return 0;
1355
1356 exit_unlock:
1357 rtnl_unlock();
1358 return err;
1359 }
1360
1361 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1362 {
1363 struct sk_buff *reply;
1364 struct datapath *dp;
1365 int err;
1366
1367 dp = lookup_datapath(info->userhdr, info->attrs);
1368 if (IS_ERR(dp))
1369 return PTR_ERR(dp);
1370
1371 reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1372 info->snd_seq, OVS_DP_CMD_NEW);
1373 if (IS_ERR(reply)) {
1374 err = PTR_ERR(reply);
1375 netlink_set_err(init_net.genl_sock, 0,
1376 ovs_dp_datapath_multicast_group.id, err);
1377 return 0;
1378 }
1379
1380 genl_notify(reply, genl_info_net(info), info->snd_pid,
1381 ovs_dp_datapath_multicast_group.id, info->nlhdr,
1382 GFP_KERNEL);
1383
1384 return 0;
1385 }
1386
1387 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1388 {
1389 struct sk_buff *reply;
1390 struct datapath *dp;
1391
1392 dp = lookup_datapath(info->userhdr, info->attrs);
1393 if (IS_ERR(dp))
1394 return PTR_ERR(dp);
1395
1396 reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1397 info->snd_seq, OVS_DP_CMD_NEW);
1398 if (IS_ERR(reply))
1399 return PTR_ERR(reply);
1400
1401 return genlmsg_reply(reply, info);
1402 }
1403
1404 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1405 {
1406 struct datapath *dp;
1407 int skip = cb->args[0];
1408 int i = 0;
1409
1410 list_for_each_entry(dp, &dps, list_node) {
1411 if (i >= skip &&
1412 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).pid,
1413 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1414 OVS_DP_CMD_NEW) < 0)
1415 break;
1416 i++;
1417 }
1418
1419 cb->args[0] = i;
1420
1421 return skb->len;
1422 }
1423
1424 static struct genl_ops dp_datapath_genl_ops[] = {
1425 { .cmd = OVS_DP_CMD_NEW,
1426 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1427 .policy = datapath_policy,
1428 .doit = ovs_dp_cmd_new
1429 },
1430 { .cmd = OVS_DP_CMD_DEL,
1431 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1432 .policy = datapath_policy,
1433 .doit = ovs_dp_cmd_del
1434 },
1435 { .cmd = OVS_DP_CMD_GET,
1436 .flags = 0, /* OK for unprivileged users. */
1437 .policy = datapath_policy,
1438 .doit = ovs_dp_cmd_get,
1439 .dumpit = ovs_dp_cmd_dump
1440 },
1441 { .cmd = OVS_DP_CMD_SET,
1442 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1443 .policy = datapath_policy,
1444 .doit = ovs_dp_cmd_set,
1445 },
1446 };
1447
1448 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1449 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1450 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1451 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1452 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1453 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1454 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1455 };
1456
1457 static struct genl_family dp_vport_genl_family = {
1458 .id = GENL_ID_GENERATE,
1459 .hdrsize = sizeof(struct ovs_header),
1460 .name = OVS_VPORT_FAMILY,
1461 .version = OVS_VPORT_VERSION,
1462 .maxattr = OVS_VPORT_ATTR_MAX
1463 };
1464
1465 struct genl_multicast_group ovs_dp_vport_multicast_group = {
1466 .name = OVS_VPORT_MCGROUP
1467 };
1468
1469 /* Called with RTNL lock or RCU read lock. */
1470 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1471 u32 pid, u32 seq, u32 flags, u8 cmd)
1472 {
1473 struct ovs_header *ovs_header;
1474 struct ovs_vport_stats vport_stats;
1475 int err;
1476
1477 ovs_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family,
1478 flags, cmd);
1479 if (!ovs_header)
1480 return -EMSGSIZE;
1481
1482 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1483
1484 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1485 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1486 nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1487 nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_pid))
1488 goto nla_put_failure;
1489
1490 ovs_vport_get_stats(vport, &vport_stats);
1491 if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1492 &vport_stats))
1493 goto nla_put_failure;
1494
1495 err = ovs_vport_get_options(vport, skb);
1496 if (err == -EMSGSIZE)
1497 goto error;
1498
1499 return genlmsg_end(skb, ovs_header);
1500
1501 nla_put_failure:
1502 err = -EMSGSIZE;
1503 error:
1504 genlmsg_cancel(skb, ovs_header);
1505 return err;
1506 }
1507
1508 /* Called with RTNL lock or RCU read lock. */
1509 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 pid,
1510 u32 seq, u8 cmd)
1511 {
1512 struct sk_buff *skb;
1513 int retval;
1514
1515 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1516 if (!skb)
1517 return ERR_PTR(-ENOMEM);
1518
1519 retval = ovs_vport_cmd_fill_info(vport, skb, pid, seq, 0, cmd);
1520 if (retval < 0) {
1521 kfree_skb(skb);
1522 return ERR_PTR(retval);
1523 }
1524 return skb;
1525 }
1526
1527 /* Called with RTNL lock or RCU read lock. */
1528 static struct vport *lookup_vport(struct ovs_header *ovs_header,
1529 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1530 {
1531 struct datapath *dp;
1532 struct vport *vport;
1533
1534 if (a[OVS_VPORT_ATTR_NAME]) {
1535 vport = ovs_vport_locate(nla_data(a[OVS_VPORT_ATTR_NAME]));
1536 if (!vport)
1537 return ERR_PTR(-ENODEV);
1538 if (ovs_header->dp_ifindex &&
1539 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1540 return ERR_PTR(-ENODEV);
1541 return vport;
1542 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1543 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1544
1545 if (port_no >= DP_MAX_PORTS)
1546 return ERR_PTR(-EFBIG);
1547
1548 dp = get_dp(ovs_header->dp_ifindex);
1549 if (!dp)
1550 return ERR_PTR(-ENODEV);
1551
1552 vport = rcu_dereference_rtnl(dp->ports[port_no]);
1553 if (!vport)
1554 return ERR_PTR(-ENOENT);
1555 return vport;
1556 } else
1557 return ERR_PTR(-EINVAL);
1558 }
1559
1560 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1561 {
1562 struct nlattr **a = info->attrs;
1563 struct ovs_header *ovs_header = info->userhdr;
1564 struct vport_parms parms;
1565 struct sk_buff *reply;
1566 struct vport *vport;
1567 struct datapath *dp;
1568 u32 port_no;
1569 int err;
1570
1571 err = -EINVAL;
1572 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1573 !a[OVS_VPORT_ATTR_UPCALL_PID])
1574 goto exit;
1575
1576 rtnl_lock();
1577 dp = get_dp(ovs_header->dp_ifindex);
1578 err = -ENODEV;
1579 if (!dp)
1580 goto exit_unlock;
1581
1582 if (a[OVS_VPORT_ATTR_PORT_NO]) {
1583 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1584
1585 err = -EFBIG;
1586 if (port_no >= DP_MAX_PORTS)
1587 goto exit_unlock;
1588
1589 vport = rtnl_dereference(dp->ports[port_no]);
1590 err = -EBUSY;
1591 if (vport)
1592 goto exit_unlock;
1593 } else {
1594 for (port_no = 1; ; port_no++) {
1595 if (port_no >= DP_MAX_PORTS) {
1596 err = -EFBIG;
1597 goto exit_unlock;
1598 }
1599 vport = rtnl_dereference(dp->ports[port_no]);
1600 if (!vport)
1601 break;
1602 }
1603 }
1604
1605 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1606 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1607 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1608 parms.dp = dp;
1609 parms.port_no = port_no;
1610 parms.upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1611
1612 vport = new_vport(&parms);
1613 err = PTR_ERR(vport);
1614 if (IS_ERR(vport))
1615 goto exit_unlock;
1616
1617 reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1618 OVS_VPORT_CMD_NEW);
1619 if (IS_ERR(reply)) {
1620 err = PTR_ERR(reply);
1621 ovs_dp_detach_port(vport);
1622 goto exit_unlock;
1623 }
1624 genl_notify(reply, genl_info_net(info), info->snd_pid,
1625 ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1626
1627 exit_unlock:
1628 rtnl_unlock();
1629 exit:
1630 return err;
1631 }
1632
1633 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1634 {
1635 struct nlattr **a = info->attrs;
1636 struct sk_buff *reply;
1637 struct vport *vport;
1638 int err;
1639
1640 rtnl_lock();
1641 vport = lookup_vport(info->userhdr, a);
1642 err = PTR_ERR(vport);
1643 if (IS_ERR(vport))
1644 goto exit_unlock;
1645
1646 err = 0;
1647 if (a[OVS_VPORT_ATTR_TYPE] &&
1648 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type)
1649 err = -EINVAL;
1650
1651 if (!err && a[OVS_VPORT_ATTR_OPTIONS])
1652 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1653 if (err)
1654 goto exit_unlock;
1655 if (a[OVS_VPORT_ATTR_UPCALL_PID])
1656 vport->upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1657
1658 reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1659 OVS_VPORT_CMD_NEW);
1660 if (IS_ERR(reply)) {
1661 netlink_set_err(init_net.genl_sock, 0,
1662 ovs_dp_vport_multicast_group.id, PTR_ERR(reply));
1663 goto exit_unlock;
1664 }
1665
1666 genl_notify(reply, genl_info_net(info), info->snd_pid,
1667 ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1668
1669 exit_unlock:
1670 rtnl_unlock();
1671 return err;
1672 }
1673
1674 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1675 {
1676 struct nlattr **a = info->attrs;
1677 struct sk_buff *reply;
1678 struct vport *vport;
1679 int err;
1680
1681 rtnl_lock();
1682 vport = lookup_vport(info->userhdr, a);
1683 err = PTR_ERR(vport);
1684 if (IS_ERR(vport))
1685 goto exit_unlock;
1686
1687 if (vport->port_no == OVSP_LOCAL) {
1688 err = -EINVAL;
1689 goto exit_unlock;
1690 }
1691
1692 reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1693 OVS_VPORT_CMD_DEL);
1694 err = PTR_ERR(reply);
1695 if (IS_ERR(reply))
1696 goto exit_unlock;
1697
1698 ovs_dp_detach_port(vport);
1699
1700 genl_notify(reply, genl_info_net(info), info->snd_pid,
1701 ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1702
1703 exit_unlock:
1704 rtnl_unlock();
1705 return err;
1706 }
1707
1708 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1709 {
1710 struct nlattr **a = info->attrs;
1711 struct ovs_header *ovs_header = info->userhdr;
1712 struct sk_buff *reply;
1713 struct vport *vport;
1714 int err;
1715
1716 rcu_read_lock();
1717 vport = lookup_vport(ovs_header, a);
1718 err = PTR_ERR(vport);
1719 if (IS_ERR(vport))
1720 goto exit_unlock;
1721
1722 reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1723 OVS_VPORT_CMD_NEW);
1724 err = PTR_ERR(reply);
1725 if (IS_ERR(reply))
1726 goto exit_unlock;
1727
1728 rcu_read_unlock();
1729
1730 return genlmsg_reply(reply, info);
1731
1732 exit_unlock:
1733 rcu_read_unlock();
1734 return err;
1735 }
1736
1737 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1738 {
1739 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1740 struct datapath *dp;
1741 u32 port_no;
1742 int retval;
1743
1744 dp = get_dp(ovs_header->dp_ifindex);
1745 if (!dp)
1746 return -ENODEV;
1747
1748 rcu_read_lock();
1749 for (port_no = cb->args[0]; port_no < DP_MAX_PORTS; port_no++) {
1750 struct vport *vport;
1751
1752 vport = rcu_dereference(dp->ports[port_no]);
1753 if (!vport)
1754 continue;
1755
1756 if (ovs_vport_cmd_fill_info(vport, skb, NETLINK_CB(cb->skb).pid,
1757 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1758 OVS_VPORT_CMD_NEW) < 0)
1759 break;
1760 }
1761 rcu_read_unlock();
1762
1763 cb->args[0] = port_no;
1764 retval = skb->len;
1765
1766 return retval;
1767 }
1768
1769 static void rehash_flow_table(struct work_struct *work)
1770 {
1771 struct datapath *dp;
1772
1773 genl_lock();
1774
1775 list_for_each_entry(dp, &dps, list_node) {
1776 struct flow_table *old_table = genl_dereference(dp->table);
1777 struct flow_table *new_table;
1778
1779 new_table = ovs_flow_tbl_rehash(old_table);
1780 if (!IS_ERR(new_table)) {
1781 rcu_assign_pointer(dp->table, new_table);
1782 ovs_flow_tbl_deferred_destroy(old_table);
1783 }
1784 }
1785
1786 genl_unlock();
1787
1788 schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
1789 }
1790
1791 static struct genl_ops dp_vport_genl_ops[] = {
1792 { .cmd = OVS_VPORT_CMD_NEW,
1793 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1794 .policy = vport_policy,
1795 .doit = ovs_vport_cmd_new
1796 },
1797 { .cmd = OVS_VPORT_CMD_DEL,
1798 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1799 .policy = vport_policy,
1800 .doit = ovs_vport_cmd_del
1801 },
1802 { .cmd = OVS_VPORT_CMD_GET,
1803 .flags = 0, /* OK for unprivileged users. */
1804 .policy = vport_policy,
1805 .doit = ovs_vport_cmd_get,
1806 .dumpit = ovs_vport_cmd_dump
1807 },
1808 { .cmd = OVS_VPORT_CMD_SET,
1809 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1810 .policy = vport_policy,
1811 .doit = ovs_vport_cmd_set,
1812 },
1813 };
1814
1815 struct genl_family_and_ops {
1816 struct genl_family *family;
1817 struct genl_ops *ops;
1818 int n_ops;
1819 struct genl_multicast_group *group;
1820 };
1821
1822 static const struct genl_family_and_ops dp_genl_families[] = {
1823 { &dp_datapath_genl_family,
1824 dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1825 &ovs_dp_datapath_multicast_group },
1826 { &dp_vport_genl_family,
1827 dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1828 &ovs_dp_vport_multicast_group },
1829 { &dp_flow_genl_family,
1830 dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1831 &ovs_dp_flow_multicast_group },
1832 { &dp_packet_genl_family,
1833 dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1834 NULL },
1835 };
1836
1837 static void dp_unregister_genl(int n_families)
1838 {
1839 int i;
1840
1841 for (i = 0; i < n_families; i++)
1842 genl_unregister_family(dp_genl_families[i].family);
1843 }
1844
1845 static int dp_register_genl(void)
1846 {
1847 int n_registered;
1848 int err;
1849 int i;
1850
1851 n_registered = 0;
1852 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
1853 const struct genl_family_and_ops *f = &dp_genl_families[i];
1854
1855 err = genl_register_family_with_ops(f->family, f->ops,
1856 f->n_ops);
1857 if (err)
1858 goto error;
1859 n_registered++;
1860
1861 if (f->group) {
1862 err = genl_register_mc_group(f->family, f->group);
1863 if (err)
1864 goto error;
1865 }
1866 }
1867
1868 return 0;
1869
1870 error:
1871 dp_unregister_genl(n_registered);
1872 return err;
1873 }
1874
1875 static int __init dp_init(void)
1876 {
1877 struct sk_buff *dummy_skb;
1878 int err;
1879
1880 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
1881
1882 pr_info("Open vSwitch switching datapath\n");
1883
1884 err = ovs_flow_init();
1885 if (err)
1886 goto error;
1887
1888 err = ovs_vport_init();
1889 if (err)
1890 goto error_flow_exit;
1891
1892 err = register_netdevice_notifier(&ovs_dp_device_notifier);
1893 if (err)
1894 goto error_vport_exit;
1895
1896 err = dp_register_genl();
1897 if (err < 0)
1898 goto error_unreg_notifier;
1899
1900 schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
1901
1902 return 0;
1903
1904 error_unreg_notifier:
1905 unregister_netdevice_notifier(&ovs_dp_device_notifier);
1906 error_vport_exit:
1907 ovs_vport_exit();
1908 error_flow_exit:
1909 ovs_flow_exit();
1910 error:
1911 return err;
1912 }
1913
1914 static void dp_cleanup(void)
1915 {
1916 cancel_delayed_work_sync(&rehash_flow_wq);
1917 rcu_barrier();
1918 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
1919 unregister_netdevice_notifier(&ovs_dp_device_notifier);
1920 ovs_vport_exit();
1921 ovs_flow_exit();
1922 }
1923
1924 module_init(dp_init);
1925 module_exit(dp_cleanup);
1926
1927 MODULE_DESCRIPTION("Open vSwitch switching datapath");
1928 MODULE_LICENSE("GPL");
Linus' kernel tree