Loading...
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 */
77static LIST_HEAD(dps);
78
79#define REHASH_FLOW_INTERVAL (10 * 60 * HZ)
80static void rehash_flow_table(struct work_struct *work);
81static DECLARE_DELAYED_WORK(rehash_flow_wq, rehash_flow_table);
82
83static struct vport *new_vport(const struct vport_parms *);
84static int queue_gso_packets(int dp_ifindex, struct sk_buff *,
85 const struct dp_upcall_info *);
86static 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. */
90static 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. */
108const 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
114static 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
132static 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. */
142static 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. */
158void 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. */
171void 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
211out:
212 /* Update datapath statistics. */
213 u64_stats_update_begin(&stats->sync);
214 (*stats_counter)++;
215 u64_stats_update_end(&stats->sync);
216}
217
218static 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
226int 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
253err:
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
263static 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
308static 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 (!nskb)
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
366out:
367 kfree_skb(nskb);
368 return err;
369}
370
371/* Called with genl_mutex. */
372static 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
393static int validate_actions(const struct nlattr *attr,
394 const struct sw_flow_key *key, int depth);
395
396static 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
424static int validate_tp_port(const struct sw_flow_key *flow_key)
425{
426 if (flow_key->eth.type == htons(ETH_P_IP)) {
427 if (flow_key->ipv4.tp.src && flow_key->ipv4.tp.dst)
428 return 0;
429 } else if (flow_key->eth.type == htons(ETH_P_IPV6)) {
430 if (flow_key->ipv6.tp.src && flow_key->ipv6.tp.dst)
431 return 0;
432 }
433
434 return -EINVAL;
435}
436
437static int validate_set(const struct nlattr *a,
438 const struct sw_flow_key *flow_key)
439{
440 const struct nlattr *ovs_key = nla_data(a);
441 int key_type = nla_type(ovs_key);
442
443 /* There can be only one key in a action */
444 if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
445 return -EINVAL;
446
447 if (key_type > OVS_KEY_ATTR_MAX ||
448 nla_len(ovs_key) != ovs_key_lens[key_type])
449 return -EINVAL;
450
451 switch (key_type) {
452 const struct ovs_key_ipv4 *ipv4_key;
453
454 case OVS_KEY_ATTR_PRIORITY:
455 case OVS_KEY_ATTR_ETHERNET:
456 break;
457
458 case OVS_KEY_ATTR_IPV4:
459 if (flow_key->eth.type != htons(ETH_P_IP))
460 return -EINVAL;
461
462 if (!flow_key->ipv4.addr.src || !flow_key->ipv4.addr.dst)
463 return -EINVAL;
464
465 ipv4_key = nla_data(ovs_key);
466 if (ipv4_key->ipv4_proto != flow_key->ip.proto)
467 return -EINVAL;
468
469 if (ipv4_key->ipv4_frag != flow_key->ip.frag)
470 return -EINVAL;
471
472 break;
473
474 case OVS_KEY_ATTR_TCP:
475 if (flow_key->ip.proto != IPPROTO_TCP)
476 return -EINVAL;
477
478 return validate_tp_port(flow_key);
479
480 case OVS_KEY_ATTR_UDP:
481 if (flow_key->ip.proto != IPPROTO_UDP)
482 return -EINVAL;
483
484 return validate_tp_port(flow_key);
485
486 default:
487 return -EINVAL;
488 }
489
490 return 0;
491}
492
493static int validate_userspace(const struct nlattr *attr)
494{
495 static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] = {
496 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
497 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_U64 },
498 };
499 struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
500 int error;
501
502 error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
503 attr, userspace_policy);
504 if (error)
505 return error;
506
507 if (!a[OVS_USERSPACE_ATTR_PID] ||
508 !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
509 return -EINVAL;
510
511 return 0;
512}
513
514static int validate_actions(const struct nlattr *attr,
515 const struct sw_flow_key *key, int depth)
516{
517 const struct nlattr *a;
518 int rem, err;
519
520 if (depth >= SAMPLE_ACTION_DEPTH)
521 return -EOVERFLOW;
522
523 nla_for_each_nested(a, attr, rem) {
524 /* Expected argument lengths, (u32)-1 for variable length. */
525 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
526 [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
527 [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
528 [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
529 [OVS_ACTION_ATTR_POP_VLAN] = 0,
530 [OVS_ACTION_ATTR_SET] = (u32)-1,
531 [OVS_ACTION_ATTR_SAMPLE] = (u32)-1
532 };
533 const struct ovs_action_push_vlan *vlan;
534 int type = nla_type(a);
535
536 if (type > OVS_ACTION_ATTR_MAX ||
537 (action_lens[type] != nla_len(a) &&
538 action_lens[type] != (u32)-1))
539 return -EINVAL;
540
541 switch (type) {
542 case OVS_ACTION_ATTR_UNSPEC:
543 return -EINVAL;
544
545 case OVS_ACTION_ATTR_USERSPACE:
546 err = validate_userspace(a);
547 if (err)
548 return err;
549 break;
550
551 case OVS_ACTION_ATTR_OUTPUT:
552 if (nla_get_u32(a) >= DP_MAX_PORTS)
553 return -EINVAL;
554 break;
555
556
557 case OVS_ACTION_ATTR_POP_VLAN:
558 break;
559
560 case OVS_ACTION_ATTR_PUSH_VLAN:
561 vlan = nla_data(a);
562 if (vlan->vlan_tpid != htons(ETH_P_8021Q))
563 return -EINVAL;
564 if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
565 return -EINVAL;
566 break;
567
568 case OVS_ACTION_ATTR_SET:
569 err = validate_set(a, key);
570 if (err)
571 return err;
572 break;
573
574 case OVS_ACTION_ATTR_SAMPLE:
575 err = validate_sample(a, key, depth);
576 if (err)
577 return err;
578 break;
579
580 default:
581 return -EINVAL;
582 }
583 }
584
585 if (rem > 0)
586 return -EINVAL;
587
588 return 0;
589}
590
591static void clear_stats(struct sw_flow *flow)
592{
593 flow->used = 0;
594 flow->tcp_flags = 0;
595 flow->packet_count = 0;
596 flow->byte_count = 0;
597}
598
599static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
600{
601 struct ovs_header *ovs_header = info->userhdr;
602 struct nlattr **a = info->attrs;
603 struct sw_flow_actions *acts;
604 struct sk_buff *packet;
605 struct sw_flow *flow;
606 struct datapath *dp;
607 struct ethhdr *eth;
608 int len;
609 int err;
610 int key_len;
611
612 err = -EINVAL;
613 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
614 !a[OVS_PACKET_ATTR_ACTIONS] ||
615 nla_len(a[OVS_PACKET_ATTR_PACKET]) < ETH_HLEN)
616 goto err;
617
618 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
619 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
620 err = -ENOMEM;
621 if (!packet)
622 goto err;
623 skb_reserve(packet, NET_IP_ALIGN);
624
625 memcpy(__skb_put(packet, len), nla_data(a[OVS_PACKET_ATTR_PACKET]), len);
626
627 skb_reset_mac_header(packet);
628 eth = eth_hdr(packet);
629
630 /* Normally, setting the skb 'protocol' field would be handled by a
631 * call to eth_type_trans(), but it assumes there's a sending
632 * device, which we may not have. */
633 if (ntohs(eth->h_proto) >= 1536)
634 packet->protocol = eth->h_proto;
635 else
636 packet->protocol = htons(ETH_P_802_2);
637
638 /* Build an sw_flow for sending this packet. */
639 flow = ovs_flow_alloc();
640 err = PTR_ERR(flow);
641 if (IS_ERR(flow))
642 goto err_kfree_skb;
643
644 err = ovs_flow_extract(packet, -1, &flow->key, &key_len);
645 if (err)
646 goto err_flow_free;
647
648 err = ovs_flow_metadata_from_nlattrs(&flow->key.phy.priority,
649 &flow->key.phy.in_port,
650 a[OVS_PACKET_ATTR_KEY]);
651 if (err)
652 goto err_flow_free;
653
654 err = validate_actions(a[OVS_PACKET_ATTR_ACTIONS], &flow->key, 0);
655 if (err)
656 goto err_flow_free;
657
658 flow->hash = ovs_flow_hash(&flow->key, key_len);
659
660 acts = ovs_flow_actions_alloc(a[OVS_PACKET_ATTR_ACTIONS]);
661 err = PTR_ERR(acts);
662 if (IS_ERR(acts))
663 goto err_flow_free;
664 rcu_assign_pointer(flow->sf_acts, acts);
665
666 OVS_CB(packet)->flow = flow;
667 packet->priority = flow->key.phy.priority;
668
669 rcu_read_lock();
670 dp = get_dp(ovs_header->dp_ifindex);
671 err = -ENODEV;
672 if (!dp)
673 goto err_unlock;
674
675 local_bh_disable();
676 err = ovs_execute_actions(dp, packet);
677 local_bh_enable();
678 rcu_read_unlock();
679
680 ovs_flow_free(flow);
681 return err;
682
683err_unlock:
684 rcu_read_unlock();
685err_flow_free:
686 ovs_flow_free(flow);
687err_kfree_skb:
688 kfree_skb(packet);
689err:
690 return err;
691}
692
693static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
694 [OVS_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
695 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
696 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
697};
698
699static struct genl_ops dp_packet_genl_ops[] = {
700 { .cmd = OVS_PACKET_CMD_EXECUTE,
701 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
702 .policy = packet_policy,
703 .doit = ovs_packet_cmd_execute
704 }
705};
706
707static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
708{
709 int i;
710 struct flow_table *table = genl_dereference(dp->table);
711
712 stats->n_flows = ovs_flow_tbl_count(table);
713
714 stats->n_hit = stats->n_missed = stats->n_lost = 0;
715 for_each_possible_cpu(i) {
716 const struct dp_stats_percpu *percpu_stats;
717 struct dp_stats_percpu local_stats;
718 unsigned int start;
719
720 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
721
722 do {
723 start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
724 local_stats = *percpu_stats;
725 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
726
727 stats->n_hit += local_stats.n_hit;
728 stats->n_missed += local_stats.n_missed;
729 stats->n_lost += local_stats.n_lost;
730 }
731}
732
733static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
734 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
735 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
736 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
737};
738
739static struct genl_family dp_flow_genl_family = {
740 .id = GENL_ID_GENERATE,
741 .hdrsize = sizeof(struct ovs_header),
742 .name = OVS_FLOW_FAMILY,
743 .version = OVS_FLOW_VERSION,
744 .maxattr = OVS_FLOW_ATTR_MAX
745};
746
747static struct genl_multicast_group ovs_dp_flow_multicast_group = {
748 .name = OVS_FLOW_MCGROUP
749};
750
751/* Called with genl_lock. */
752static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
753 struct sk_buff *skb, u32 pid,
754 u32 seq, u32 flags, u8 cmd)
755{
756 const int skb_orig_len = skb->len;
757 const struct sw_flow_actions *sf_acts;
758 struct ovs_flow_stats stats;
759 struct ovs_header *ovs_header;
760 struct nlattr *nla;
761 unsigned long used;
762 u8 tcp_flags;
763 int err;
764
765 sf_acts = rcu_dereference_protected(flow->sf_acts,
766 lockdep_genl_is_held());
767
768 ovs_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags, cmd);
769 if (!ovs_header)
770 return -EMSGSIZE;
771
772 ovs_header->dp_ifindex = get_dpifindex(dp);
773
774 nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
775 if (!nla)
776 goto nla_put_failure;
777 err = ovs_flow_to_nlattrs(&flow->key, skb);
778 if (err)
779 goto error;
780 nla_nest_end(skb, nla);
781
782 spin_lock_bh(&flow->lock);
783 used = flow->used;
784 stats.n_packets = flow->packet_count;
785 stats.n_bytes = flow->byte_count;
786 tcp_flags = flow->tcp_flags;
787 spin_unlock_bh(&flow->lock);
788
789 if (used &&
790 nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
791 goto nla_put_failure;
792
793 if (stats.n_packets &&
794 nla_put(skb, OVS_FLOW_ATTR_STATS,
795 sizeof(struct ovs_flow_stats), &stats))
796 goto nla_put_failure;
797
798 if (tcp_flags &&
799 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags))
800 goto nla_put_failure;
801
802 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
803 * this is the first flow to be dumped into 'skb'. This is unusual for
804 * Netlink but individual action lists can be longer than
805 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
806 * The userspace caller can always fetch the actions separately if it
807 * really wants them. (Most userspace callers in fact don't care.)
808 *
809 * This can only fail for dump operations because the skb is always
810 * properly sized for single flows.
811 */
812 err = nla_put(skb, OVS_FLOW_ATTR_ACTIONS, sf_acts->actions_len,
813 sf_acts->actions);
814 if (err < 0 && skb_orig_len)
815 goto error;
816
817 return genlmsg_end(skb, ovs_header);
818
819nla_put_failure:
820 err = -EMSGSIZE;
821error:
822 genlmsg_cancel(skb, ovs_header);
823 return err;
824}
825
826static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
827{
828 const struct sw_flow_actions *sf_acts;
829 int len;
830
831 sf_acts = rcu_dereference_protected(flow->sf_acts,
832 lockdep_genl_is_held());
833
834 /* OVS_FLOW_ATTR_KEY */
835 len = nla_total_size(FLOW_BUFSIZE);
836 /* OVS_FLOW_ATTR_ACTIONS */
837 len += nla_total_size(sf_acts->actions_len);
838 /* OVS_FLOW_ATTR_STATS */
839 len += nla_total_size(sizeof(struct ovs_flow_stats));
840 /* OVS_FLOW_ATTR_TCP_FLAGS */
841 len += nla_total_size(1);
842 /* OVS_FLOW_ATTR_USED */
843 len += nla_total_size(8);
844
845 len += NLMSG_ALIGN(sizeof(struct ovs_header));
846
847 return genlmsg_new(len, GFP_KERNEL);
848}
849
850static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
851 struct datapath *dp,
852 u32 pid, u32 seq, u8 cmd)
853{
854 struct sk_buff *skb;
855 int retval;
856
857 skb = ovs_flow_cmd_alloc_info(flow);
858 if (!skb)
859 return ERR_PTR(-ENOMEM);
860
861 retval = ovs_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd);
862 BUG_ON(retval < 0);
863 return skb;
864}
865
866static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
867{
868 struct nlattr **a = info->attrs;
869 struct ovs_header *ovs_header = info->userhdr;
870 struct sw_flow_key key;
871 struct sw_flow *flow;
872 struct sk_buff *reply;
873 struct datapath *dp;
874 struct flow_table *table;
875 int error;
876 int key_len;
877
878 /* Extract key. */
879 error = -EINVAL;
880 if (!a[OVS_FLOW_ATTR_KEY])
881 goto error;
882 error = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
883 if (error)
884 goto error;
885
886 /* Validate actions. */
887 if (a[OVS_FLOW_ATTR_ACTIONS]) {
888 error = validate_actions(a[OVS_FLOW_ATTR_ACTIONS], &key, 0);
889 if (error)
890 goto error;
891 } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
892 error = -EINVAL;
893 goto error;
894 }
895
896 dp = get_dp(ovs_header->dp_ifindex);
897 error = -ENODEV;
898 if (!dp)
899 goto error;
900
901 table = genl_dereference(dp->table);
902 flow = ovs_flow_tbl_lookup(table, &key, key_len);
903 if (!flow) {
904 struct sw_flow_actions *acts;
905
906 /* Bail out if we're not allowed to create a new flow. */
907 error = -ENOENT;
908 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
909 goto error;
910
911 /* Expand table, if necessary, to make room. */
912 if (ovs_flow_tbl_need_to_expand(table)) {
913 struct flow_table *new_table;
914
915 new_table = ovs_flow_tbl_expand(table);
916 if (!IS_ERR(new_table)) {
917 rcu_assign_pointer(dp->table, new_table);
918 ovs_flow_tbl_deferred_destroy(table);
919 table = genl_dereference(dp->table);
920 }
921 }
922
923 /* Allocate flow. */
924 flow = ovs_flow_alloc();
925 if (IS_ERR(flow)) {
926 error = PTR_ERR(flow);
927 goto error;
928 }
929 flow->key = key;
930 clear_stats(flow);
931
932 /* Obtain actions. */
933 acts = ovs_flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]);
934 error = PTR_ERR(acts);
935 if (IS_ERR(acts))
936 goto error_free_flow;
937 rcu_assign_pointer(flow->sf_acts, acts);
938
939 /* Put flow in bucket. */
940 flow->hash = ovs_flow_hash(&key, key_len);
941 ovs_flow_tbl_insert(table, flow);
942
943 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
944 info->snd_seq,
945 OVS_FLOW_CMD_NEW);
946 } else {
947 /* We found a matching flow. */
948 struct sw_flow_actions *old_acts;
949 struct nlattr *acts_attrs;
950
951 /* Bail out if we're not allowed to modify an existing flow.
952 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
953 * because Generic Netlink treats the latter as a dump
954 * request. We also accept NLM_F_EXCL in case that bug ever
955 * gets fixed.
956 */
957 error = -EEXIST;
958 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
959 info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
960 goto error;
961
962 /* Update actions. */
963 old_acts = rcu_dereference_protected(flow->sf_acts,
964 lockdep_genl_is_held());
965 acts_attrs = a[OVS_FLOW_ATTR_ACTIONS];
966 if (acts_attrs &&
967 (old_acts->actions_len != nla_len(acts_attrs) ||
968 memcmp(old_acts->actions, nla_data(acts_attrs),
969 old_acts->actions_len))) {
970 struct sw_flow_actions *new_acts;
971
972 new_acts = ovs_flow_actions_alloc(acts_attrs);
973 error = PTR_ERR(new_acts);
974 if (IS_ERR(new_acts))
975 goto error;
976
977 rcu_assign_pointer(flow->sf_acts, new_acts);
978 ovs_flow_deferred_free_acts(old_acts);
979 }
980
981 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
982 info->snd_seq, OVS_FLOW_CMD_NEW);
983
984 /* Clear stats. */
985 if (a[OVS_FLOW_ATTR_CLEAR]) {
986 spin_lock_bh(&flow->lock);
987 clear_stats(flow);
988 spin_unlock_bh(&flow->lock);
989 }
990 }
991
992 if (!IS_ERR(reply))
993 genl_notify(reply, genl_info_net(info), info->snd_pid,
994 ovs_dp_flow_multicast_group.id, info->nlhdr,
995 GFP_KERNEL);
996 else
997 netlink_set_err(init_net.genl_sock, 0,
998 ovs_dp_flow_multicast_group.id, PTR_ERR(reply));
999 return 0;
1000
1001error_free_flow:
1002 ovs_flow_free(flow);
1003error:
1004 return error;
1005}
1006
1007static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1008{
1009 struct nlattr **a = info->attrs;
1010 struct ovs_header *ovs_header = info->userhdr;
1011 struct sw_flow_key key;
1012 struct sk_buff *reply;
1013 struct sw_flow *flow;
1014 struct datapath *dp;
1015 struct flow_table *table;
1016 int err;
1017 int key_len;
1018
1019 if (!a[OVS_FLOW_ATTR_KEY])
1020 return -EINVAL;
1021 err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1022 if (err)
1023 return err;
1024
1025 dp = get_dp(ovs_header->dp_ifindex);
1026 if (!dp)
1027 return -ENODEV;
1028
1029 table = genl_dereference(dp->table);
1030 flow = ovs_flow_tbl_lookup(table, &key, key_len);
1031 if (!flow)
1032 return -ENOENT;
1033
1034 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1035 info->snd_seq, OVS_FLOW_CMD_NEW);
1036 if (IS_ERR(reply))
1037 return PTR_ERR(reply);
1038
1039 return genlmsg_reply(reply, info);
1040}
1041
1042static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1043{
1044 struct nlattr **a = info->attrs;
1045 struct ovs_header *ovs_header = info->userhdr;
1046 struct sw_flow_key key;
1047 struct sk_buff *reply;
1048 struct sw_flow *flow;
1049 struct datapath *dp;
1050 struct flow_table *table;
1051 int err;
1052 int key_len;
1053
1054 if (!a[OVS_FLOW_ATTR_KEY])
1055 return flush_flows(ovs_header->dp_ifindex);
1056 err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1057 if (err)
1058 return err;
1059
1060 dp = get_dp(ovs_header->dp_ifindex);
1061 if (!dp)
1062 return -ENODEV;
1063
1064 table = genl_dereference(dp->table);
1065 flow = ovs_flow_tbl_lookup(table, &key, key_len);
1066 if (!flow)
1067 return -ENOENT;
1068
1069 reply = ovs_flow_cmd_alloc_info(flow);
1070 if (!reply)
1071 return -ENOMEM;
1072
1073 ovs_flow_tbl_remove(table, flow);
1074
1075 err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_pid,
1076 info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1077 BUG_ON(err < 0);
1078
1079 ovs_flow_deferred_free(flow);
1080
1081 genl_notify(reply, genl_info_net(info), info->snd_pid,
1082 ovs_dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1083 return 0;
1084}
1085
1086static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1087{
1088 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1089 struct datapath *dp;
1090 struct flow_table *table;
1091
1092 dp = get_dp(ovs_header->dp_ifindex);
1093 if (!dp)
1094 return -ENODEV;
1095
1096 table = genl_dereference(dp->table);
1097
1098 for (;;) {
1099 struct sw_flow *flow;
1100 u32 bucket, obj;
1101
1102 bucket = cb->args[0];
1103 obj = cb->args[1];
1104 flow = ovs_flow_tbl_next(table, &bucket, &obj);
1105 if (!flow)
1106 break;
1107
1108 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1109 NETLINK_CB(cb->skb).pid,
1110 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1111 OVS_FLOW_CMD_NEW) < 0)
1112 break;
1113
1114 cb->args[0] = bucket;
1115 cb->args[1] = obj;
1116 }
1117 return skb->len;
1118}
1119
1120static struct genl_ops dp_flow_genl_ops[] = {
1121 { .cmd = OVS_FLOW_CMD_NEW,
1122 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1123 .policy = flow_policy,
1124 .doit = ovs_flow_cmd_new_or_set
1125 },
1126 { .cmd = OVS_FLOW_CMD_DEL,
1127 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1128 .policy = flow_policy,
1129 .doit = ovs_flow_cmd_del
1130 },
1131 { .cmd = OVS_FLOW_CMD_GET,
1132 .flags = 0, /* OK for unprivileged users. */
1133 .policy = flow_policy,
1134 .doit = ovs_flow_cmd_get,
1135 .dumpit = ovs_flow_cmd_dump
1136 },
1137 { .cmd = OVS_FLOW_CMD_SET,
1138 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1139 .policy = flow_policy,
1140 .doit = ovs_flow_cmd_new_or_set,
1141 },
1142};
1143
1144static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1145 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1146 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1147};
1148
1149static struct genl_family dp_datapath_genl_family = {
1150 .id = GENL_ID_GENERATE,
1151 .hdrsize = sizeof(struct ovs_header),
1152 .name = OVS_DATAPATH_FAMILY,
1153 .version = OVS_DATAPATH_VERSION,
1154 .maxattr = OVS_DP_ATTR_MAX
1155};
1156
1157static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1158 .name = OVS_DATAPATH_MCGROUP
1159};
1160
1161static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1162 u32 pid, u32 seq, u32 flags, u8 cmd)
1163{
1164 struct ovs_header *ovs_header;
1165 struct ovs_dp_stats dp_stats;
1166 int err;
1167
1168 ovs_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family,
1169 flags, cmd);
1170 if (!ovs_header)
1171 goto error;
1172
1173 ovs_header->dp_ifindex = get_dpifindex(dp);
1174
1175 rcu_read_lock();
1176 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1177 rcu_read_unlock();
1178 if (err)
1179 goto nla_put_failure;
1180
1181 get_dp_stats(dp, &dp_stats);
1182 if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats), &dp_stats))
1183 goto nla_put_failure;
1184
1185 return genlmsg_end(skb, ovs_header);
1186
1187nla_put_failure:
1188 genlmsg_cancel(skb, ovs_header);
1189error:
1190 return -EMSGSIZE;
1191}
1192
1193static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 pid,
1194 u32 seq, u8 cmd)
1195{
1196 struct sk_buff *skb;
1197 int retval;
1198
1199 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1200 if (!skb)
1201 return ERR_PTR(-ENOMEM);
1202
1203 retval = ovs_dp_cmd_fill_info(dp, skb, pid, seq, 0, cmd);
1204 if (retval < 0) {
1205 kfree_skb(skb);
1206 return ERR_PTR(retval);
1207 }
1208 return skb;
1209}
1210
1211/* Called with genl_mutex and optionally with RTNL lock also. */
1212static struct datapath *lookup_datapath(struct ovs_header *ovs_header,
1213 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1214{
1215 struct datapath *dp;
1216
1217 if (!a[OVS_DP_ATTR_NAME])
1218 dp = get_dp(ovs_header->dp_ifindex);
1219 else {
1220 struct vport *vport;
1221
1222 rcu_read_lock();
1223 vport = ovs_vport_locate(nla_data(a[OVS_DP_ATTR_NAME]));
1224 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1225 rcu_read_unlock();
1226 }
1227 return dp ? dp : ERR_PTR(-ENODEV);
1228}
1229
1230static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1231{
1232 struct nlattr **a = info->attrs;
1233 struct vport_parms parms;
1234 struct sk_buff *reply;
1235 struct datapath *dp;
1236 struct vport *vport;
1237 int err;
1238
1239 err = -EINVAL;
1240 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1241 goto err;
1242
1243 rtnl_lock();
1244 err = -ENODEV;
1245 if (!try_module_get(THIS_MODULE))
1246 goto err_unlock_rtnl;
1247
1248 err = -ENOMEM;
1249 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1250 if (dp == NULL)
1251 goto err_put_module;
1252 INIT_LIST_HEAD(&dp->port_list);
1253
1254 /* Allocate table. */
1255 err = -ENOMEM;
1256 rcu_assign_pointer(dp->table, ovs_flow_tbl_alloc(TBL_MIN_BUCKETS));
1257 if (!dp->table)
1258 goto err_free_dp;
1259
1260 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1261 if (!dp->stats_percpu) {
1262 err = -ENOMEM;
1263 goto err_destroy_table;
1264 }
1265
1266 /* Set up our datapath device. */
1267 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1268 parms.type = OVS_VPORT_TYPE_INTERNAL;
1269 parms.options = NULL;
1270 parms.dp = dp;
1271 parms.port_no = OVSP_LOCAL;
1272 parms.upcall_pid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1273
1274 vport = new_vport(&parms);
1275 if (IS_ERR(vport)) {
1276 err = PTR_ERR(vport);
1277 if (err == -EBUSY)
1278 err = -EEXIST;
1279
1280 goto err_destroy_percpu;
1281 }
1282
1283 reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1284 info->snd_seq, OVS_DP_CMD_NEW);
1285 err = PTR_ERR(reply);
1286 if (IS_ERR(reply))
1287 goto err_destroy_local_port;
1288
1289 list_add_tail(&dp->list_node, &dps);
1290 rtnl_unlock();
1291
1292 genl_notify(reply, genl_info_net(info), info->snd_pid,
1293 ovs_dp_datapath_multicast_group.id, info->nlhdr,
1294 GFP_KERNEL);
1295 return 0;
1296
1297err_destroy_local_port:
1298 ovs_dp_detach_port(rtnl_dereference(dp->ports[OVSP_LOCAL]));
1299err_destroy_percpu:
1300 free_percpu(dp->stats_percpu);
1301err_destroy_table:
1302 ovs_flow_tbl_destroy(genl_dereference(dp->table));
1303err_free_dp:
1304 kfree(dp);
1305err_put_module:
1306 module_put(THIS_MODULE);
1307err_unlock_rtnl:
1308 rtnl_unlock();
1309err:
1310 return err;
1311}
1312
1313static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1314{
1315 struct vport *vport, *next_vport;
1316 struct sk_buff *reply;
1317 struct datapath *dp;
1318 int err;
1319
1320 rtnl_lock();
1321 dp = lookup_datapath(info->userhdr, info->attrs);
1322 err = PTR_ERR(dp);
1323 if (IS_ERR(dp))
1324 goto exit_unlock;
1325
1326 reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1327 info->snd_seq, OVS_DP_CMD_DEL);
1328 err = PTR_ERR(reply);
1329 if (IS_ERR(reply))
1330 goto exit_unlock;
1331
1332 list_for_each_entry_safe(vport, next_vport, &dp->port_list, node)
1333 if (vport->port_no != OVSP_LOCAL)
1334 ovs_dp_detach_port(vport);
1335
1336 list_del(&dp->list_node);
1337 ovs_dp_detach_port(rtnl_dereference(dp->ports[OVSP_LOCAL]));
1338
1339 /* rtnl_unlock() will wait until all the references to devices that
1340 * are pending unregistration have been dropped. We do it here to
1341 * ensure that any internal devices (which contain DP pointers) are
1342 * fully destroyed before freeing the datapath.
1343 */
1344 rtnl_unlock();
1345
1346 call_rcu(&dp->rcu, destroy_dp_rcu);
1347 module_put(THIS_MODULE);
1348
1349 genl_notify(reply, genl_info_net(info), info->snd_pid,
1350 ovs_dp_datapath_multicast_group.id, info->nlhdr,
1351 GFP_KERNEL);
1352
1353 return 0;
1354
1355exit_unlock:
1356 rtnl_unlock();
1357 return err;
1358}
1359
1360static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1361{
1362 struct sk_buff *reply;
1363 struct datapath *dp;
1364 int err;
1365
1366 dp = lookup_datapath(info->userhdr, info->attrs);
1367 if (IS_ERR(dp))
1368 return PTR_ERR(dp);
1369
1370 reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1371 info->snd_seq, OVS_DP_CMD_NEW);
1372 if (IS_ERR(reply)) {
1373 err = PTR_ERR(reply);
1374 netlink_set_err(init_net.genl_sock, 0,
1375 ovs_dp_datapath_multicast_group.id, err);
1376 return 0;
1377 }
1378
1379 genl_notify(reply, genl_info_net(info), info->snd_pid,
1380 ovs_dp_datapath_multicast_group.id, info->nlhdr,
1381 GFP_KERNEL);
1382
1383 return 0;
1384}
1385
1386static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1387{
1388 struct sk_buff *reply;
1389 struct datapath *dp;
1390
1391 dp = lookup_datapath(info->userhdr, info->attrs);
1392 if (IS_ERR(dp))
1393 return PTR_ERR(dp);
1394
1395 reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1396 info->snd_seq, OVS_DP_CMD_NEW);
1397 if (IS_ERR(reply))
1398 return PTR_ERR(reply);
1399
1400 return genlmsg_reply(reply, info);
1401}
1402
1403static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1404{
1405 struct datapath *dp;
1406 int skip = cb->args[0];
1407 int i = 0;
1408
1409 list_for_each_entry(dp, &dps, list_node) {
1410 if (i >= skip &&
1411 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).pid,
1412 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1413 OVS_DP_CMD_NEW) < 0)
1414 break;
1415 i++;
1416 }
1417
1418 cb->args[0] = i;
1419
1420 return skb->len;
1421}
1422
1423static struct genl_ops dp_datapath_genl_ops[] = {
1424 { .cmd = OVS_DP_CMD_NEW,
1425 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1426 .policy = datapath_policy,
1427 .doit = ovs_dp_cmd_new
1428 },
1429 { .cmd = OVS_DP_CMD_DEL,
1430 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1431 .policy = datapath_policy,
1432 .doit = ovs_dp_cmd_del
1433 },
1434 { .cmd = OVS_DP_CMD_GET,
1435 .flags = 0, /* OK for unprivileged users. */
1436 .policy = datapath_policy,
1437 .doit = ovs_dp_cmd_get,
1438 .dumpit = ovs_dp_cmd_dump
1439 },
1440 { .cmd = OVS_DP_CMD_SET,
1441 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1442 .policy = datapath_policy,
1443 .doit = ovs_dp_cmd_set,
1444 },
1445};
1446
1447static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1448 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1449 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1450 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1451 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1452 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1453 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1454};
1455
1456static struct genl_family dp_vport_genl_family = {
1457 .id = GENL_ID_GENERATE,
1458 .hdrsize = sizeof(struct ovs_header),
1459 .name = OVS_VPORT_FAMILY,
1460 .version = OVS_VPORT_VERSION,
1461 .maxattr = OVS_VPORT_ATTR_MAX
1462};
1463
1464struct genl_multicast_group ovs_dp_vport_multicast_group = {
1465 .name = OVS_VPORT_MCGROUP
1466};
1467
1468/* Called with RTNL lock or RCU read lock. */
1469static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1470 u32 pid, u32 seq, u32 flags, u8 cmd)
1471{
1472 struct ovs_header *ovs_header;
1473 struct ovs_vport_stats vport_stats;
1474 int err;
1475
1476 ovs_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family,
1477 flags, cmd);
1478 if (!ovs_header)
1479 return -EMSGSIZE;
1480
1481 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1482
1483 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1484 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1485 nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1486 nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_pid))
1487 goto nla_put_failure;
1488
1489 ovs_vport_get_stats(vport, &vport_stats);
1490 if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1491 &vport_stats))
1492 goto nla_put_failure;
1493
1494 err = ovs_vport_get_options(vport, skb);
1495 if (err == -EMSGSIZE)
1496 goto error;
1497
1498 return genlmsg_end(skb, ovs_header);
1499
1500nla_put_failure:
1501 err = -EMSGSIZE;
1502error:
1503 genlmsg_cancel(skb, ovs_header);
1504 return err;
1505}
1506
1507/* Called with RTNL lock or RCU read lock. */
1508struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 pid,
1509 u32 seq, u8 cmd)
1510{
1511 struct sk_buff *skb;
1512 int retval;
1513
1514 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1515 if (!skb)
1516 return ERR_PTR(-ENOMEM);
1517
1518 retval = ovs_vport_cmd_fill_info(vport, skb, pid, seq, 0, cmd);
1519 if (retval < 0) {
1520 kfree_skb(skb);
1521 return ERR_PTR(retval);
1522 }
1523 return skb;
1524}
1525
1526/* Called with RTNL lock or RCU read lock. */
1527static struct vport *lookup_vport(struct ovs_header *ovs_header,
1528 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1529{
1530 struct datapath *dp;
1531 struct vport *vport;
1532
1533 if (a[OVS_VPORT_ATTR_NAME]) {
1534 vport = ovs_vport_locate(nla_data(a[OVS_VPORT_ATTR_NAME]));
1535 if (!vport)
1536 return ERR_PTR(-ENODEV);
1537 if (ovs_header->dp_ifindex &&
1538 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1539 return ERR_PTR(-ENODEV);
1540 return vport;
1541 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1542 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1543
1544 if (port_no >= DP_MAX_PORTS)
1545 return ERR_PTR(-EFBIG);
1546
1547 dp = get_dp(ovs_header->dp_ifindex);
1548 if (!dp)
1549 return ERR_PTR(-ENODEV);
1550
1551 vport = rcu_dereference_rtnl(dp->ports[port_no]);
1552 if (!vport)
1553 return ERR_PTR(-ENOENT);
1554 return vport;
1555 } else
1556 return ERR_PTR(-EINVAL);
1557}
1558
1559static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1560{
1561 struct nlattr **a = info->attrs;
1562 struct ovs_header *ovs_header = info->userhdr;
1563 struct vport_parms parms;
1564 struct sk_buff *reply;
1565 struct vport *vport;
1566 struct datapath *dp;
1567 u32 port_no;
1568 int err;
1569
1570 err = -EINVAL;
1571 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1572 !a[OVS_VPORT_ATTR_UPCALL_PID])
1573 goto exit;
1574
1575 rtnl_lock();
1576 dp = get_dp(ovs_header->dp_ifindex);
1577 err = -ENODEV;
1578 if (!dp)
1579 goto exit_unlock;
1580
1581 if (a[OVS_VPORT_ATTR_PORT_NO]) {
1582 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1583
1584 err = -EFBIG;
1585 if (port_no >= DP_MAX_PORTS)
1586 goto exit_unlock;
1587
1588 vport = rtnl_dereference(dp->ports[port_no]);
1589 err = -EBUSY;
1590 if (vport)
1591 goto exit_unlock;
1592 } else {
1593 for (port_no = 1; ; port_no++) {
1594 if (port_no >= DP_MAX_PORTS) {
1595 err = -EFBIG;
1596 goto exit_unlock;
1597 }
1598 vport = rtnl_dereference(dp->ports[port_no]);
1599 if (!vport)
1600 break;
1601 }
1602 }
1603
1604 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1605 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1606 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1607 parms.dp = dp;
1608 parms.port_no = port_no;
1609 parms.upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1610
1611 vport = new_vport(&parms);
1612 err = PTR_ERR(vport);
1613 if (IS_ERR(vport))
1614 goto exit_unlock;
1615
1616 reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1617 OVS_VPORT_CMD_NEW);
1618 if (IS_ERR(reply)) {
1619 err = PTR_ERR(reply);
1620 ovs_dp_detach_port(vport);
1621 goto exit_unlock;
1622 }
1623 genl_notify(reply, genl_info_net(info), info->snd_pid,
1624 ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1625
1626exit_unlock:
1627 rtnl_unlock();
1628exit:
1629 return err;
1630}
1631
1632static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1633{
1634 struct nlattr **a = info->attrs;
1635 struct sk_buff *reply;
1636 struct vport *vport;
1637 int err;
1638
1639 rtnl_lock();
1640 vport = lookup_vport(info->userhdr, a);
1641 err = PTR_ERR(vport);
1642 if (IS_ERR(vport))
1643 goto exit_unlock;
1644
1645 err = 0;
1646 if (a[OVS_VPORT_ATTR_TYPE] &&
1647 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type)
1648 err = -EINVAL;
1649
1650 if (!err && a[OVS_VPORT_ATTR_OPTIONS])
1651 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1652 if (!err && a[OVS_VPORT_ATTR_UPCALL_PID])
1653 vport->upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1654
1655 reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1656 OVS_VPORT_CMD_NEW);
1657 if (IS_ERR(reply)) {
1658 netlink_set_err(init_net.genl_sock, 0,
1659 ovs_dp_vport_multicast_group.id, PTR_ERR(reply));
1660 goto exit_unlock;
1661 }
1662
1663 genl_notify(reply, genl_info_net(info), info->snd_pid,
1664 ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1665
1666exit_unlock:
1667 rtnl_unlock();
1668 return err;
1669}
1670
1671static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1672{
1673 struct nlattr **a = info->attrs;
1674 struct sk_buff *reply;
1675 struct vport *vport;
1676 int err;
1677
1678 rtnl_lock();
1679 vport = lookup_vport(info->userhdr, a);
1680 err = PTR_ERR(vport);
1681 if (IS_ERR(vport))
1682 goto exit_unlock;
1683
1684 if (vport->port_no == OVSP_LOCAL) {
1685 err = -EINVAL;
1686 goto exit_unlock;
1687 }
1688
1689 reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1690 OVS_VPORT_CMD_DEL);
1691 err = PTR_ERR(reply);
1692 if (IS_ERR(reply))
1693 goto exit_unlock;
1694
1695 ovs_dp_detach_port(vport);
1696
1697 genl_notify(reply, genl_info_net(info), info->snd_pid,
1698 ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1699
1700exit_unlock:
1701 rtnl_unlock();
1702 return err;
1703}
1704
1705static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1706{
1707 struct nlattr **a = info->attrs;
1708 struct ovs_header *ovs_header = info->userhdr;
1709 struct sk_buff *reply;
1710 struct vport *vport;
1711 int err;
1712
1713 rcu_read_lock();
1714 vport = lookup_vport(ovs_header, a);
1715 err = PTR_ERR(vport);
1716 if (IS_ERR(vport))
1717 goto exit_unlock;
1718
1719 reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1720 OVS_VPORT_CMD_NEW);
1721 err = PTR_ERR(reply);
1722 if (IS_ERR(reply))
1723 goto exit_unlock;
1724
1725 rcu_read_unlock();
1726
1727 return genlmsg_reply(reply, info);
1728
1729exit_unlock:
1730 rcu_read_unlock();
1731 return err;
1732}
1733
1734static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1735{
1736 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1737 struct datapath *dp;
1738 u32 port_no;
1739 int retval;
1740
1741 dp = get_dp(ovs_header->dp_ifindex);
1742 if (!dp)
1743 return -ENODEV;
1744
1745 rcu_read_lock();
1746 for (port_no = cb->args[0]; port_no < DP_MAX_PORTS; port_no++) {
1747 struct vport *vport;
1748
1749 vport = rcu_dereference(dp->ports[port_no]);
1750 if (!vport)
1751 continue;
1752
1753 if (ovs_vport_cmd_fill_info(vport, skb, NETLINK_CB(cb->skb).pid,
1754 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1755 OVS_VPORT_CMD_NEW) < 0)
1756 break;
1757 }
1758 rcu_read_unlock();
1759
1760 cb->args[0] = port_no;
1761 retval = skb->len;
1762
1763 return retval;
1764}
1765
1766static void rehash_flow_table(struct work_struct *work)
1767{
1768 struct datapath *dp;
1769
1770 genl_lock();
1771
1772 list_for_each_entry(dp, &dps, list_node) {
1773 struct flow_table *old_table = genl_dereference(dp->table);
1774 struct flow_table *new_table;
1775
1776 new_table = ovs_flow_tbl_rehash(old_table);
1777 if (!IS_ERR(new_table)) {
1778 rcu_assign_pointer(dp->table, new_table);
1779 ovs_flow_tbl_deferred_destroy(old_table);
1780 }
1781 }
1782
1783 genl_unlock();
1784
1785 schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
1786}
1787
1788static struct genl_ops dp_vport_genl_ops[] = {
1789 { .cmd = OVS_VPORT_CMD_NEW,
1790 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1791 .policy = vport_policy,
1792 .doit = ovs_vport_cmd_new
1793 },
1794 { .cmd = OVS_VPORT_CMD_DEL,
1795 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1796 .policy = vport_policy,
1797 .doit = ovs_vport_cmd_del
1798 },
1799 { .cmd = OVS_VPORT_CMD_GET,
1800 .flags = 0, /* OK for unprivileged users. */
1801 .policy = vport_policy,
1802 .doit = ovs_vport_cmd_get,
1803 .dumpit = ovs_vport_cmd_dump
1804 },
1805 { .cmd = OVS_VPORT_CMD_SET,
1806 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1807 .policy = vport_policy,
1808 .doit = ovs_vport_cmd_set,
1809 },
1810};
1811
1812struct genl_family_and_ops {
1813 struct genl_family *family;
1814 struct genl_ops *ops;
1815 int n_ops;
1816 struct genl_multicast_group *group;
1817};
1818
1819static const struct genl_family_and_ops dp_genl_families[] = {
1820 { &dp_datapath_genl_family,
1821 dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1822 &ovs_dp_datapath_multicast_group },
1823 { &dp_vport_genl_family,
1824 dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1825 &ovs_dp_vport_multicast_group },
1826 { &dp_flow_genl_family,
1827 dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1828 &ovs_dp_flow_multicast_group },
1829 { &dp_packet_genl_family,
1830 dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1831 NULL },
1832};
1833
1834static void dp_unregister_genl(int n_families)
1835{
1836 int i;
1837
1838 for (i = 0; i < n_families; i++)
1839 genl_unregister_family(dp_genl_families[i].family);
1840}
1841
1842static int dp_register_genl(void)
1843{
1844 int n_registered;
1845 int err;
1846 int i;
1847
1848 n_registered = 0;
1849 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
1850 const struct genl_family_and_ops *f = &dp_genl_families[i];
1851
1852 err = genl_register_family_with_ops(f->family, f->ops,
1853 f->n_ops);
1854 if (err)
1855 goto error;
1856 n_registered++;
1857
1858 if (f->group) {
1859 err = genl_register_mc_group(f->family, f->group);
1860 if (err)
1861 goto error;
1862 }
1863 }
1864
1865 return 0;
1866
1867error:
1868 dp_unregister_genl(n_registered);
1869 return err;
1870}
1871
1872static int __init dp_init(void)
1873{
1874 struct sk_buff *dummy_skb;
1875 int err;
1876
1877 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
1878
1879 pr_info("Open vSwitch switching datapath\n");
1880
1881 err = ovs_flow_init();
1882 if (err)
1883 goto error;
1884
1885 err = ovs_vport_init();
1886 if (err)
1887 goto error_flow_exit;
1888
1889 err = register_netdevice_notifier(&ovs_dp_device_notifier);
1890 if (err)
1891 goto error_vport_exit;
1892
1893 err = dp_register_genl();
1894 if (err < 0)
1895 goto error_unreg_notifier;
1896
1897 schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
1898
1899 return 0;
1900
1901error_unreg_notifier:
1902 unregister_netdevice_notifier(&ovs_dp_device_notifier);
1903error_vport_exit:
1904 ovs_vport_exit();
1905error_flow_exit:
1906 ovs_flow_exit();
1907error:
1908 return err;
1909}
1910
1911static void dp_cleanup(void)
1912{
1913 cancel_delayed_work_sync(&rehash_flow_wq);
1914 rcu_barrier();
1915 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
1916 unregister_netdevice_notifier(&ovs_dp_device_notifier);
1917 ovs_vport_exit();
1918 ovs_flow_exit();
1919}
1920
1921module_init(dp_init);
1922module_exit(dp_cleanup);
1923
1924MODULE_DESCRIPTION("Open vSwitch switching datapath");
1925MODULE_LICENSE("GPL");
1/*
2 * Copyright (c) 2007-2013 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/lockdep.h>
48#include <linux/openvswitch.h>
49#include <linux/rculist.h>
50#include <linux/dmi.h>
51#include <linux/workqueue.h>
52#include <net/genetlink.h>
53#include <net/net_namespace.h>
54#include <net/netns/generic.h>
55
56#include "datapath.h"
57#include "flow.h"
58#include "flow_table.h"
59#include "flow_netlink.h"
60#include "vport-internal_dev.h"
61#include "vport-netdev.h"
62
63int ovs_net_id __read_mostly;
64
65static void ovs_notify(struct genl_family *family,
66 struct sk_buff *skb, struct genl_info *info)
67{
68 genl_notify(family, skb, genl_info_net(info), info->snd_portid,
69 0, info->nlhdr, GFP_KERNEL);
70}
71
72/**
73 * DOC: Locking:
74 *
75 * All writes e.g. Writes to device state (add/remove datapath, port, set
76 * operations on vports, etc.), Writes to other state (flow table
77 * modifications, set miscellaneous datapath parameters, etc.) are protected
78 * by ovs_lock.
79 *
80 * Reads are protected by RCU.
81 *
82 * There are a few special cases (mostly stats) that have their own
83 * synchronization but they nest under all of above and don't interact with
84 * each other.
85 *
86 * The RTNL lock nests inside ovs_mutex.
87 */
88
89static DEFINE_MUTEX(ovs_mutex);
90
91void ovs_lock(void)
92{
93 mutex_lock(&ovs_mutex);
94}
95
96void ovs_unlock(void)
97{
98 mutex_unlock(&ovs_mutex);
99}
100
101#ifdef CONFIG_LOCKDEP
102int lockdep_ovsl_is_held(void)
103{
104 if (debug_locks)
105 return lockdep_is_held(&ovs_mutex);
106 else
107 return 1;
108}
109#endif
110
111static struct vport *new_vport(const struct vport_parms *);
112static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
113 const struct dp_upcall_info *);
114static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
115 const struct dp_upcall_info *);
116
117/* Must be called with rcu_read_lock or ovs_mutex. */
118static struct datapath *get_dp(struct net *net, int dp_ifindex)
119{
120 struct datapath *dp = NULL;
121 struct net_device *dev;
122
123 rcu_read_lock();
124 dev = dev_get_by_index_rcu(net, dp_ifindex);
125 if (dev) {
126 struct vport *vport = ovs_internal_dev_get_vport(dev);
127 if (vport)
128 dp = vport->dp;
129 }
130 rcu_read_unlock();
131
132 return dp;
133}
134
135/* Must be called with rcu_read_lock or ovs_mutex. */
136static const char *ovs_dp_name(const struct datapath *dp)
137{
138 struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
139 return vport->ops->get_name(vport);
140}
141
142static int get_dpifindex(struct datapath *dp)
143{
144 struct vport *local;
145 int ifindex;
146
147 rcu_read_lock();
148
149 local = ovs_vport_rcu(dp, OVSP_LOCAL);
150 if (local)
151 ifindex = netdev_vport_priv(local)->dev->ifindex;
152 else
153 ifindex = 0;
154
155 rcu_read_unlock();
156
157 return ifindex;
158}
159
160static void destroy_dp_rcu(struct rcu_head *rcu)
161{
162 struct datapath *dp = container_of(rcu, struct datapath, rcu);
163
164 free_percpu(dp->stats_percpu);
165 release_net(ovs_dp_get_net(dp));
166 kfree(dp->ports);
167 kfree(dp);
168}
169
170static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
171 u16 port_no)
172{
173 return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
174}
175
176struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
177{
178 struct vport *vport;
179 struct hlist_head *head;
180
181 head = vport_hash_bucket(dp, port_no);
182 hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
183 if (vport->port_no == port_no)
184 return vport;
185 }
186 return NULL;
187}
188
189/* Called with ovs_mutex. */
190static struct vport *new_vport(const struct vport_parms *parms)
191{
192 struct vport *vport;
193
194 vport = ovs_vport_add(parms);
195 if (!IS_ERR(vport)) {
196 struct datapath *dp = parms->dp;
197 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
198
199 hlist_add_head_rcu(&vport->dp_hash_node, head);
200 }
201 return vport;
202}
203
204void ovs_dp_detach_port(struct vport *p)
205{
206 ASSERT_OVSL();
207
208 /* First drop references to device. */
209 hlist_del_rcu(&p->dp_hash_node);
210
211 /* Then destroy it. */
212 ovs_vport_del(p);
213}
214
215/* Must be called with rcu_read_lock. */
216void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
217{
218 struct datapath *dp = p->dp;
219 struct sw_flow *flow;
220 struct dp_stats_percpu *stats;
221 struct sw_flow_key key;
222 u64 *stats_counter;
223 u32 n_mask_hit;
224 int error;
225
226 stats = this_cpu_ptr(dp->stats_percpu);
227
228 /* Extract flow from 'skb' into 'key'. */
229 error = ovs_flow_extract(skb, p->port_no, &key);
230 if (unlikely(error)) {
231 kfree_skb(skb);
232 return;
233 }
234
235 /* Look up flow. */
236 flow = ovs_flow_tbl_lookup_stats(&dp->table, &key, &n_mask_hit);
237 if (unlikely(!flow)) {
238 struct dp_upcall_info upcall;
239
240 upcall.cmd = OVS_PACKET_CMD_MISS;
241 upcall.key = &key;
242 upcall.userdata = NULL;
243 upcall.portid = p->upcall_portid;
244 ovs_dp_upcall(dp, skb, &upcall);
245 consume_skb(skb);
246 stats_counter = &stats->n_missed;
247 goto out;
248 }
249
250 OVS_CB(skb)->flow = flow;
251 OVS_CB(skb)->pkt_key = &key;
252
253 ovs_flow_stats_update(OVS_CB(skb)->flow, skb);
254 ovs_execute_actions(dp, skb);
255 stats_counter = &stats->n_hit;
256
257out:
258 /* Update datapath statistics. */
259 u64_stats_update_begin(&stats->syncp);
260 (*stats_counter)++;
261 stats->n_mask_hit += n_mask_hit;
262 u64_stats_update_end(&stats->syncp);
263}
264
265static struct genl_family dp_packet_genl_family = {
266 .id = GENL_ID_GENERATE,
267 .hdrsize = sizeof(struct ovs_header),
268 .name = OVS_PACKET_FAMILY,
269 .version = OVS_PACKET_VERSION,
270 .maxattr = OVS_PACKET_ATTR_MAX,
271 .netnsok = true,
272 .parallel_ops = true,
273};
274
275int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
276 const struct dp_upcall_info *upcall_info)
277{
278 struct dp_stats_percpu *stats;
279 int err;
280
281 if (upcall_info->portid == 0) {
282 err = -ENOTCONN;
283 goto err;
284 }
285
286 if (!skb_is_gso(skb))
287 err = queue_userspace_packet(dp, skb, upcall_info);
288 else
289 err = queue_gso_packets(dp, skb, upcall_info);
290 if (err)
291 goto err;
292
293 return 0;
294
295err:
296 stats = this_cpu_ptr(dp->stats_percpu);
297
298 u64_stats_update_begin(&stats->syncp);
299 stats->n_lost++;
300 u64_stats_update_end(&stats->syncp);
301
302 return err;
303}
304
305static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
306 const struct dp_upcall_info *upcall_info)
307{
308 unsigned short gso_type = skb_shinfo(skb)->gso_type;
309 struct dp_upcall_info later_info;
310 struct sw_flow_key later_key;
311 struct sk_buff *segs, *nskb;
312 int err;
313
314 segs = __skb_gso_segment(skb, NETIF_F_SG, false);
315 if (IS_ERR(segs))
316 return PTR_ERR(segs);
317
318 /* Queue all of the segments. */
319 skb = segs;
320 do {
321 err = queue_userspace_packet(dp, skb, upcall_info);
322 if (err)
323 break;
324
325 if (skb == segs && gso_type & SKB_GSO_UDP) {
326 /* The initial flow key extracted by ovs_flow_extract()
327 * in this case is for a first fragment, so we need to
328 * properly mark later fragments.
329 */
330 later_key = *upcall_info->key;
331 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
332
333 later_info = *upcall_info;
334 later_info.key = &later_key;
335 upcall_info = &later_info;
336 }
337 } while ((skb = skb->next));
338
339 /* Free all of the segments. */
340 skb = segs;
341 do {
342 nskb = skb->next;
343 if (err)
344 kfree_skb(skb);
345 else
346 consume_skb(skb);
347 } while ((skb = nskb));
348 return err;
349}
350
351static size_t key_attr_size(void)
352{
353 return nla_total_size(4) /* OVS_KEY_ATTR_PRIORITY */
354 + nla_total_size(0) /* OVS_KEY_ATTR_TUNNEL */
355 + nla_total_size(8) /* OVS_TUNNEL_KEY_ATTR_ID */
356 + nla_total_size(4) /* OVS_TUNNEL_KEY_ATTR_IPV4_SRC */
357 + nla_total_size(4) /* OVS_TUNNEL_KEY_ATTR_IPV4_DST */
358 + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TOS */
359 + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TTL */
360 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
361 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_CSUM */
362 + nla_total_size(4) /* OVS_KEY_ATTR_IN_PORT */
363 + nla_total_size(4) /* OVS_KEY_ATTR_SKB_MARK */
364 + nla_total_size(12) /* OVS_KEY_ATTR_ETHERNET */
365 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
366 + nla_total_size(4) /* OVS_KEY_ATTR_8021Q */
367 + nla_total_size(0) /* OVS_KEY_ATTR_ENCAP */
368 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
369 + nla_total_size(40) /* OVS_KEY_ATTR_IPV6 */
370 + nla_total_size(2) /* OVS_KEY_ATTR_ICMPV6 */
371 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
372}
373
374static size_t upcall_msg_size(const struct nlattr *userdata,
375 unsigned int hdrlen)
376{
377 size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
378 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
379 + nla_total_size(key_attr_size()); /* OVS_PACKET_ATTR_KEY */
380
381 /* OVS_PACKET_ATTR_USERDATA */
382 if (userdata)
383 size += NLA_ALIGN(userdata->nla_len);
384
385 return size;
386}
387
388static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
389 const struct dp_upcall_info *upcall_info)
390{
391 struct ovs_header *upcall;
392 struct sk_buff *nskb = NULL;
393 struct sk_buff *user_skb; /* to be queued to userspace */
394 struct nlattr *nla;
395 struct genl_info info = {
396 .dst_sk = ovs_dp_get_net(dp)->genl_sock,
397 .snd_portid = upcall_info->portid,
398 };
399 size_t len;
400 unsigned int hlen;
401 int err, dp_ifindex;
402
403 dp_ifindex = get_dpifindex(dp);
404 if (!dp_ifindex)
405 return -ENODEV;
406
407 if (vlan_tx_tag_present(skb)) {
408 nskb = skb_clone(skb, GFP_ATOMIC);
409 if (!nskb)
410 return -ENOMEM;
411
412 nskb = __vlan_put_tag(nskb, nskb->vlan_proto, vlan_tx_tag_get(nskb));
413 if (!nskb)
414 return -ENOMEM;
415
416 nskb->vlan_tci = 0;
417 skb = nskb;
418 }
419
420 if (nla_attr_size(skb->len) > USHRT_MAX) {
421 err = -EFBIG;
422 goto out;
423 }
424
425 /* Complete checksum if needed */
426 if (skb->ip_summed == CHECKSUM_PARTIAL &&
427 (err = skb_checksum_help(skb)))
428 goto out;
429
430 /* Older versions of OVS user space enforce alignment of the last
431 * Netlink attribute to NLA_ALIGNTO which would require extensive
432 * padding logic. Only perform zerocopy if padding is not required.
433 */
434 if (dp->user_features & OVS_DP_F_UNALIGNED)
435 hlen = skb_zerocopy_headlen(skb);
436 else
437 hlen = skb->len;
438
439 len = upcall_msg_size(upcall_info->userdata, hlen);
440 user_skb = genlmsg_new_unicast(len, &info, GFP_ATOMIC);
441 if (!user_skb) {
442 err = -ENOMEM;
443 goto out;
444 }
445
446 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
447 0, upcall_info->cmd);
448 upcall->dp_ifindex = dp_ifindex;
449
450 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
451 ovs_nla_put_flow(upcall_info->key, upcall_info->key, user_skb);
452 nla_nest_end(user_skb, nla);
453
454 if (upcall_info->userdata)
455 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
456 nla_len(upcall_info->userdata),
457 nla_data(upcall_info->userdata));
458
459 /* Only reserve room for attribute header, packet data is added
460 * in skb_zerocopy() */
461 if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
462 err = -ENOBUFS;
463 goto out;
464 }
465 nla->nla_len = nla_attr_size(skb->len);
466
467 err = skb_zerocopy(user_skb, skb, skb->len, hlen);
468 if (err)
469 goto out;
470
471 /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
472 if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
473 size_t plen = NLA_ALIGN(user_skb->len) - user_skb->len;
474
475 if (plen > 0)
476 memset(skb_put(user_skb, plen), 0, plen);
477 }
478
479 ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
480
481 err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
482out:
483 if (err)
484 skb_tx_error(skb);
485 kfree_skb(nskb);
486 return err;
487}
488
489static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
490{
491 struct ovs_header *ovs_header = info->userhdr;
492 struct nlattr **a = info->attrs;
493 struct sw_flow_actions *acts;
494 struct sk_buff *packet;
495 struct sw_flow *flow;
496 struct datapath *dp;
497 struct ethhdr *eth;
498 int len;
499 int err;
500
501 err = -EINVAL;
502 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
503 !a[OVS_PACKET_ATTR_ACTIONS])
504 goto err;
505
506 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
507 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
508 err = -ENOMEM;
509 if (!packet)
510 goto err;
511 skb_reserve(packet, NET_IP_ALIGN);
512
513 nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
514
515 skb_reset_mac_header(packet);
516 eth = eth_hdr(packet);
517
518 /* Normally, setting the skb 'protocol' field would be handled by a
519 * call to eth_type_trans(), but it assumes there's a sending
520 * device, which we may not have. */
521 if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
522 packet->protocol = eth->h_proto;
523 else
524 packet->protocol = htons(ETH_P_802_2);
525
526 /* Build an sw_flow for sending this packet. */
527 flow = ovs_flow_alloc(false);
528 err = PTR_ERR(flow);
529 if (IS_ERR(flow))
530 goto err_kfree_skb;
531
532 err = ovs_flow_extract(packet, -1, &flow->key);
533 if (err)
534 goto err_flow_free;
535
536 err = ovs_nla_get_flow_metadata(flow, a[OVS_PACKET_ATTR_KEY]);
537 if (err)
538 goto err_flow_free;
539 acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_PACKET_ATTR_ACTIONS]));
540 err = PTR_ERR(acts);
541 if (IS_ERR(acts))
542 goto err_flow_free;
543
544 err = ovs_nla_copy_actions(a[OVS_PACKET_ATTR_ACTIONS],
545 &flow->key, 0, &acts);
546 rcu_assign_pointer(flow->sf_acts, acts);
547 if (err)
548 goto err_flow_free;
549
550 OVS_CB(packet)->flow = flow;
551 OVS_CB(packet)->pkt_key = &flow->key;
552 packet->priority = flow->key.phy.priority;
553 packet->mark = flow->key.phy.skb_mark;
554
555 rcu_read_lock();
556 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
557 err = -ENODEV;
558 if (!dp)
559 goto err_unlock;
560
561 local_bh_disable();
562 err = ovs_execute_actions(dp, packet);
563 local_bh_enable();
564 rcu_read_unlock();
565
566 ovs_flow_free(flow, false);
567 return err;
568
569err_unlock:
570 rcu_read_unlock();
571err_flow_free:
572 ovs_flow_free(flow, false);
573err_kfree_skb:
574 kfree_skb(packet);
575err:
576 return err;
577}
578
579static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
580 [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
581 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
582 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
583};
584
585static const struct genl_ops dp_packet_genl_ops[] = {
586 { .cmd = OVS_PACKET_CMD_EXECUTE,
587 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
588 .policy = packet_policy,
589 .doit = ovs_packet_cmd_execute
590 }
591};
592
593static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats,
594 struct ovs_dp_megaflow_stats *mega_stats)
595{
596 int i;
597
598 memset(mega_stats, 0, sizeof(*mega_stats));
599
600 stats->n_flows = ovs_flow_tbl_count(&dp->table);
601 mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
602
603 stats->n_hit = stats->n_missed = stats->n_lost = 0;
604
605 for_each_possible_cpu(i) {
606 const struct dp_stats_percpu *percpu_stats;
607 struct dp_stats_percpu local_stats;
608 unsigned int start;
609
610 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
611
612 do {
613 start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
614 local_stats = *percpu_stats;
615 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
616
617 stats->n_hit += local_stats.n_hit;
618 stats->n_missed += local_stats.n_missed;
619 stats->n_lost += local_stats.n_lost;
620 mega_stats->n_mask_hit += local_stats.n_mask_hit;
621 }
622}
623
624static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
625 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
626 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
627 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
628};
629
630static struct genl_family dp_flow_genl_family = {
631 .id = GENL_ID_GENERATE,
632 .hdrsize = sizeof(struct ovs_header),
633 .name = OVS_FLOW_FAMILY,
634 .version = OVS_FLOW_VERSION,
635 .maxattr = OVS_FLOW_ATTR_MAX,
636 .netnsok = true,
637 .parallel_ops = true,
638};
639
640static struct genl_multicast_group ovs_dp_flow_multicast_group = {
641 .name = OVS_FLOW_MCGROUP
642};
643
644static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
645{
646 return NLMSG_ALIGN(sizeof(struct ovs_header))
647 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
648 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_MASK */
649 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
650 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
651 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
652 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
653}
654
655/* Called with ovs_mutex. */
656static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
657 struct sk_buff *skb, u32 portid,
658 u32 seq, u32 flags, u8 cmd)
659{
660 const int skb_orig_len = skb->len;
661 struct nlattr *start;
662 struct ovs_flow_stats stats;
663 __be16 tcp_flags;
664 unsigned long used;
665 struct ovs_header *ovs_header;
666 struct nlattr *nla;
667 int err;
668
669 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
670 if (!ovs_header)
671 return -EMSGSIZE;
672
673 ovs_header->dp_ifindex = get_dpifindex(dp);
674
675 /* Fill flow key. */
676 nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
677 if (!nla)
678 goto nla_put_failure;
679
680 err = ovs_nla_put_flow(&flow->unmasked_key, &flow->unmasked_key, skb);
681 if (err)
682 goto error;
683 nla_nest_end(skb, nla);
684
685 nla = nla_nest_start(skb, OVS_FLOW_ATTR_MASK);
686 if (!nla)
687 goto nla_put_failure;
688
689 err = ovs_nla_put_flow(&flow->key, &flow->mask->key, skb);
690 if (err)
691 goto error;
692
693 nla_nest_end(skb, nla);
694
695 ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
696 if (used &&
697 nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
698 goto nla_put_failure;
699
700 if (stats.n_packets &&
701 nla_put(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats))
702 goto nla_put_failure;
703
704 if ((u8)ntohs(tcp_flags) &&
705 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
706 goto nla_put_failure;
707
708 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
709 * this is the first flow to be dumped into 'skb'. This is unusual for
710 * Netlink but individual action lists can be longer than
711 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
712 * The userspace caller can always fetch the actions separately if it
713 * really wants them. (Most userspace callers in fact don't care.)
714 *
715 * This can only fail for dump operations because the skb is always
716 * properly sized for single flows.
717 */
718 start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
719 if (start) {
720 const struct sw_flow_actions *sf_acts;
721
722 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
723
724 err = ovs_nla_put_actions(sf_acts->actions,
725 sf_acts->actions_len, skb);
726 if (!err)
727 nla_nest_end(skb, start);
728 else {
729 if (skb_orig_len)
730 goto error;
731
732 nla_nest_cancel(skb, start);
733 }
734 } else if (skb_orig_len)
735 goto nla_put_failure;
736
737 return genlmsg_end(skb, ovs_header);
738
739nla_put_failure:
740 err = -EMSGSIZE;
741error:
742 genlmsg_cancel(skb, ovs_header);
743 return err;
744}
745
746static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow,
747 struct genl_info *info)
748{
749 size_t len;
750
751 len = ovs_flow_cmd_msg_size(ovsl_dereference(flow->sf_acts));
752
753 return genlmsg_new_unicast(len, info, GFP_KERNEL);
754}
755
756static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
757 struct datapath *dp,
758 struct genl_info *info,
759 u8 cmd)
760{
761 struct sk_buff *skb;
762 int retval;
763
764 skb = ovs_flow_cmd_alloc_info(flow, info);
765 if (!skb)
766 return ERR_PTR(-ENOMEM);
767
768 retval = ovs_flow_cmd_fill_info(flow, dp, skb, info->snd_portid,
769 info->snd_seq, 0, cmd);
770 BUG_ON(retval < 0);
771 return skb;
772}
773
774static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
775{
776 struct nlattr **a = info->attrs;
777 struct ovs_header *ovs_header = info->userhdr;
778 struct sw_flow_key key, masked_key;
779 struct sw_flow *flow = NULL;
780 struct sw_flow_mask mask;
781 struct sk_buff *reply;
782 struct datapath *dp;
783 struct sw_flow_actions *acts = NULL;
784 struct sw_flow_match match;
785 bool exact_5tuple;
786 int error;
787
788 /* Extract key. */
789 error = -EINVAL;
790 if (!a[OVS_FLOW_ATTR_KEY])
791 goto error;
792
793 ovs_match_init(&match, &key, &mask);
794 error = ovs_nla_get_match(&match, &exact_5tuple,
795 a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
796 if (error)
797 goto error;
798
799 /* Validate actions. */
800 if (a[OVS_FLOW_ATTR_ACTIONS]) {
801 acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
802 error = PTR_ERR(acts);
803 if (IS_ERR(acts))
804 goto error;
805
806 ovs_flow_mask_key(&masked_key, &key, &mask);
807 error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS],
808 &masked_key, 0, &acts);
809 if (error) {
810 OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
811 goto err_kfree;
812 }
813 } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
814 error = -EINVAL;
815 goto error;
816 }
817
818 ovs_lock();
819 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
820 error = -ENODEV;
821 if (!dp)
822 goto err_unlock_ovs;
823
824 /* Check if this is a duplicate flow */
825 flow = ovs_flow_tbl_lookup(&dp->table, &key);
826 if (!flow) {
827 /* Bail out if we're not allowed to create a new flow. */
828 error = -ENOENT;
829 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
830 goto err_unlock_ovs;
831
832 /* Allocate flow. */
833 flow = ovs_flow_alloc(!exact_5tuple);
834 if (IS_ERR(flow)) {
835 error = PTR_ERR(flow);
836 goto err_unlock_ovs;
837 }
838
839 flow->key = masked_key;
840 flow->unmasked_key = key;
841 rcu_assign_pointer(flow->sf_acts, acts);
842
843 /* Put flow in bucket. */
844 error = ovs_flow_tbl_insert(&dp->table, flow, &mask);
845 if (error) {
846 acts = NULL;
847 goto err_flow_free;
848 }
849
850 reply = ovs_flow_cmd_build_info(flow, dp, info, OVS_FLOW_CMD_NEW);
851 } else {
852 /* We found a matching flow. */
853 struct sw_flow_actions *old_acts;
854
855 /* Bail out if we're not allowed to modify an existing flow.
856 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
857 * because Generic Netlink treats the latter as a dump
858 * request. We also accept NLM_F_EXCL in case that bug ever
859 * gets fixed.
860 */
861 error = -EEXIST;
862 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
863 info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
864 goto err_unlock_ovs;
865
866 /* The unmasked key has to be the same for flow updates. */
867 if (!ovs_flow_cmp_unmasked_key(flow, &match))
868 goto err_unlock_ovs;
869
870 /* Update actions. */
871 old_acts = ovsl_dereference(flow->sf_acts);
872 rcu_assign_pointer(flow->sf_acts, acts);
873 ovs_nla_free_flow_actions(old_acts);
874
875 reply = ovs_flow_cmd_build_info(flow, dp, info, OVS_FLOW_CMD_NEW);
876
877 /* Clear stats. */
878 if (a[OVS_FLOW_ATTR_CLEAR])
879 ovs_flow_stats_clear(flow);
880 }
881 ovs_unlock();
882
883 if (!IS_ERR(reply))
884 ovs_notify(&dp_flow_genl_family, reply, info);
885 else
886 genl_set_err(&dp_flow_genl_family, sock_net(skb->sk), 0,
887 0, PTR_ERR(reply));
888 return 0;
889
890err_flow_free:
891 ovs_flow_free(flow, false);
892err_unlock_ovs:
893 ovs_unlock();
894err_kfree:
895 kfree(acts);
896error:
897 return error;
898}
899
900static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
901{
902 struct nlattr **a = info->attrs;
903 struct ovs_header *ovs_header = info->userhdr;
904 struct sw_flow_key key;
905 struct sk_buff *reply;
906 struct sw_flow *flow;
907 struct datapath *dp;
908 struct sw_flow_match match;
909 int err;
910
911 if (!a[OVS_FLOW_ATTR_KEY]) {
912 OVS_NLERR("Flow get message rejected, Key attribute missing.\n");
913 return -EINVAL;
914 }
915
916 ovs_match_init(&match, &key, NULL);
917 err = ovs_nla_get_match(&match, NULL, a[OVS_FLOW_ATTR_KEY], NULL);
918 if (err)
919 return err;
920
921 ovs_lock();
922 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
923 if (!dp) {
924 err = -ENODEV;
925 goto unlock;
926 }
927
928 flow = ovs_flow_tbl_lookup(&dp->table, &key);
929 if (!flow || !ovs_flow_cmp_unmasked_key(flow, &match)) {
930 err = -ENOENT;
931 goto unlock;
932 }
933
934 reply = ovs_flow_cmd_build_info(flow, dp, info, OVS_FLOW_CMD_NEW);
935 if (IS_ERR(reply)) {
936 err = PTR_ERR(reply);
937 goto unlock;
938 }
939
940 ovs_unlock();
941 return genlmsg_reply(reply, info);
942unlock:
943 ovs_unlock();
944 return err;
945}
946
947static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
948{
949 struct nlattr **a = info->attrs;
950 struct ovs_header *ovs_header = info->userhdr;
951 struct sw_flow_key key;
952 struct sk_buff *reply;
953 struct sw_flow *flow;
954 struct datapath *dp;
955 struct sw_flow_match match;
956 int err;
957
958 ovs_lock();
959 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
960 if (!dp) {
961 err = -ENODEV;
962 goto unlock;
963 }
964
965 if (!a[OVS_FLOW_ATTR_KEY]) {
966 err = ovs_flow_tbl_flush(&dp->table);
967 goto unlock;
968 }
969
970 ovs_match_init(&match, &key, NULL);
971 err = ovs_nla_get_match(&match, NULL, a[OVS_FLOW_ATTR_KEY], NULL);
972 if (err)
973 goto unlock;
974
975 flow = ovs_flow_tbl_lookup(&dp->table, &key);
976 if (!flow || !ovs_flow_cmp_unmasked_key(flow, &match)) {
977 err = -ENOENT;
978 goto unlock;
979 }
980
981 reply = ovs_flow_cmd_alloc_info(flow, info);
982 if (!reply) {
983 err = -ENOMEM;
984 goto unlock;
985 }
986
987 ovs_flow_tbl_remove(&dp->table, flow);
988
989 err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_portid,
990 info->snd_seq, 0, OVS_FLOW_CMD_DEL);
991 BUG_ON(err < 0);
992
993 ovs_flow_free(flow, true);
994 ovs_unlock();
995
996 ovs_notify(&dp_flow_genl_family, reply, info);
997 return 0;
998unlock:
999 ovs_unlock();
1000 return err;
1001}
1002
1003static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1004{
1005 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1006 struct table_instance *ti;
1007 struct datapath *dp;
1008
1009 rcu_read_lock();
1010 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1011 if (!dp) {
1012 rcu_read_unlock();
1013 return -ENODEV;
1014 }
1015
1016 ti = rcu_dereference(dp->table.ti);
1017 for (;;) {
1018 struct sw_flow *flow;
1019 u32 bucket, obj;
1020
1021 bucket = cb->args[0];
1022 obj = cb->args[1];
1023 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1024 if (!flow)
1025 break;
1026
1027 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1028 NETLINK_CB(cb->skb).portid,
1029 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1030 OVS_FLOW_CMD_NEW) < 0)
1031 break;
1032
1033 cb->args[0] = bucket;
1034 cb->args[1] = obj;
1035 }
1036 rcu_read_unlock();
1037 return skb->len;
1038}
1039
1040static const struct genl_ops dp_flow_genl_ops[] = {
1041 { .cmd = OVS_FLOW_CMD_NEW,
1042 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1043 .policy = flow_policy,
1044 .doit = ovs_flow_cmd_new_or_set
1045 },
1046 { .cmd = OVS_FLOW_CMD_DEL,
1047 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1048 .policy = flow_policy,
1049 .doit = ovs_flow_cmd_del
1050 },
1051 { .cmd = OVS_FLOW_CMD_GET,
1052 .flags = 0, /* OK for unprivileged users. */
1053 .policy = flow_policy,
1054 .doit = ovs_flow_cmd_get,
1055 .dumpit = ovs_flow_cmd_dump
1056 },
1057 { .cmd = OVS_FLOW_CMD_SET,
1058 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1059 .policy = flow_policy,
1060 .doit = ovs_flow_cmd_new_or_set,
1061 },
1062};
1063
1064static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1065 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1066 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1067 [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1068};
1069
1070static struct genl_family dp_datapath_genl_family = {
1071 .id = GENL_ID_GENERATE,
1072 .hdrsize = sizeof(struct ovs_header),
1073 .name = OVS_DATAPATH_FAMILY,
1074 .version = OVS_DATAPATH_VERSION,
1075 .maxattr = OVS_DP_ATTR_MAX,
1076 .netnsok = true,
1077 .parallel_ops = true,
1078};
1079
1080static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1081 .name = OVS_DATAPATH_MCGROUP
1082};
1083
1084static size_t ovs_dp_cmd_msg_size(void)
1085{
1086 size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1087
1088 msgsize += nla_total_size(IFNAMSIZ);
1089 msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1090 msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1091 msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1092
1093 return msgsize;
1094}
1095
1096static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1097 u32 portid, u32 seq, u32 flags, u8 cmd)
1098{
1099 struct ovs_header *ovs_header;
1100 struct ovs_dp_stats dp_stats;
1101 struct ovs_dp_megaflow_stats dp_megaflow_stats;
1102 int err;
1103
1104 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1105 flags, cmd);
1106 if (!ovs_header)
1107 goto error;
1108
1109 ovs_header->dp_ifindex = get_dpifindex(dp);
1110
1111 rcu_read_lock();
1112 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1113 rcu_read_unlock();
1114 if (err)
1115 goto nla_put_failure;
1116
1117 get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1118 if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1119 &dp_stats))
1120 goto nla_put_failure;
1121
1122 if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1123 sizeof(struct ovs_dp_megaflow_stats),
1124 &dp_megaflow_stats))
1125 goto nla_put_failure;
1126
1127 if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1128 goto nla_put_failure;
1129
1130 return genlmsg_end(skb, ovs_header);
1131
1132nla_put_failure:
1133 genlmsg_cancel(skb, ovs_header);
1134error:
1135 return -EMSGSIZE;
1136}
1137
1138static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp,
1139 struct genl_info *info, u8 cmd)
1140{
1141 struct sk_buff *skb;
1142 int retval;
1143
1144 skb = genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info, GFP_KERNEL);
1145 if (!skb)
1146 return ERR_PTR(-ENOMEM);
1147
1148 retval = ovs_dp_cmd_fill_info(dp, skb, info->snd_portid, info->snd_seq, 0, cmd);
1149 if (retval < 0) {
1150 kfree_skb(skb);
1151 return ERR_PTR(retval);
1152 }
1153 return skb;
1154}
1155
1156/* Called with ovs_mutex. */
1157static struct datapath *lookup_datapath(struct net *net,
1158 struct ovs_header *ovs_header,
1159 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1160{
1161 struct datapath *dp;
1162
1163 if (!a[OVS_DP_ATTR_NAME])
1164 dp = get_dp(net, ovs_header->dp_ifindex);
1165 else {
1166 struct vport *vport;
1167
1168 rcu_read_lock();
1169 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1170 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1171 rcu_read_unlock();
1172 }
1173 return dp ? dp : ERR_PTR(-ENODEV);
1174}
1175
1176static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1177{
1178 struct datapath *dp;
1179
1180 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1181 if (IS_ERR(dp))
1182 return;
1183
1184 WARN(dp->user_features, "Dropping previously announced user features\n");
1185 dp->user_features = 0;
1186}
1187
1188static void ovs_dp_change(struct datapath *dp, struct nlattr **a)
1189{
1190 if (a[OVS_DP_ATTR_USER_FEATURES])
1191 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1192}
1193
1194static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1195{
1196 struct nlattr **a = info->attrs;
1197 struct vport_parms parms;
1198 struct sk_buff *reply;
1199 struct datapath *dp;
1200 struct vport *vport;
1201 struct ovs_net *ovs_net;
1202 int err, i;
1203
1204 err = -EINVAL;
1205 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1206 goto err;
1207
1208 ovs_lock();
1209
1210 err = -ENOMEM;
1211 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1212 if (dp == NULL)
1213 goto err_unlock_ovs;
1214
1215 ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1216
1217 /* Allocate table. */
1218 err = ovs_flow_tbl_init(&dp->table);
1219 if (err)
1220 goto err_free_dp;
1221
1222 dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1223 if (!dp->stats_percpu) {
1224 err = -ENOMEM;
1225 goto err_destroy_table;
1226 }
1227
1228 dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1229 GFP_KERNEL);
1230 if (!dp->ports) {
1231 err = -ENOMEM;
1232 goto err_destroy_percpu;
1233 }
1234
1235 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1236 INIT_HLIST_HEAD(&dp->ports[i]);
1237
1238 /* Set up our datapath device. */
1239 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1240 parms.type = OVS_VPORT_TYPE_INTERNAL;
1241 parms.options = NULL;
1242 parms.dp = dp;
1243 parms.port_no = OVSP_LOCAL;
1244 parms.upcall_portid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1245
1246 ovs_dp_change(dp, a);
1247
1248 vport = new_vport(&parms);
1249 if (IS_ERR(vport)) {
1250 err = PTR_ERR(vport);
1251 if (err == -EBUSY)
1252 err = -EEXIST;
1253
1254 if (err == -EEXIST) {
1255 /* An outdated user space instance that does not understand
1256 * the concept of user_features has attempted to create a new
1257 * datapath and is likely to reuse it. Drop all user features.
1258 */
1259 if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1260 ovs_dp_reset_user_features(skb, info);
1261 }
1262
1263 goto err_destroy_ports_array;
1264 }
1265
1266 reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_NEW);
1267 err = PTR_ERR(reply);
1268 if (IS_ERR(reply))
1269 goto err_destroy_local_port;
1270
1271 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1272 list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1273
1274 ovs_unlock();
1275
1276 ovs_notify(&dp_datapath_genl_family, reply, info);
1277 return 0;
1278
1279err_destroy_local_port:
1280 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1281err_destroy_ports_array:
1282 kfree(dp->ports);
1283err_destroy_percpu:
1284 free_percpu(dp->stats_percpu);
1285err_destroy_table:
1286 ovs_flow_tbl_destroy(&dp->table, false);
1287err_free_dp:
1288 release_net(ovs_dp_get_net(dp));
1289 kfree(dp);
1290err_unlock_ovs:
1291 ovs_unlock();
1292err:
1293 return err;
1294}
1295
1296/* Called with ovs_mutex. */
1297static void __dp_destroy(struct datapath *dp)
1298{
1299 int i;
1300
1301 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1302 struct vport *vport;
1303 struct hlist_node *n;
1304
1305 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1306 if (vport->port_no != OVSP_LOCAL)
1307 ovs_dp_detach_port(vport);
1308 }
1309
1310 list_del_rcu(&dp->list_node);
1311
1312 /* OVSP_LOCAL is datapath internal port. We need to make sure that
1313 * all ports in datapath are destroyed first before freeing datapath.
1314 */
1315 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1316
1317 /* RCU destroy the flow table */
1318 ovs_flow_tbl_destroy(&dp->table, true);
1319
1320 call_rcu(&dp->rcu, destroy_dp_rcu);
1321}
1322
1323static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1324{
1325 struct sk_buff *reply;
1326 struct datapath *dp;
1327 int err;
1328
1329 ovs_lock();
1330 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1331 err = PTR_ERR(dp);
1332 if (IS_ERR(dp))
1333 goto unlock;
1334
1335 reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_DEL);
1336 err = PTR_ERR(reply);
1337 if (IS_ERR(reply))
1338 goto unlock;
1339
1340 __dp_destroy(dp);
1341 ovs_unlock();
1342
1343 ovs_notify(&dp_datapath_genl_family, reply, info);
1344
1345 return 0;
1346unlock:
1347 ovs_unlock();
1348 return err;
1349}
1350
1351static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1352{
1353 struct sk_buff *reply;
1354 struct datapath *dp;
1355 int err;
1356
1357 ovs_lock();
1358 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1359 err = PTR_ERR(dp);
1360 if (IS_ERR(dp))
1361 goto unlock;
1362
1363 ovs_dp_change(dp, info->attrs);
1364
1365 reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_NEW);
1366 if (IS_ERR(reply)) {
1367 err = PTR_ERR(reply);
1368 genl_set_err(&dp_datapath_genl_family, sock_net(skb->sk), 0,
1369 0, err);
1370 err = 0;
1371 goto unlock;
1372 }
1373
1374 ovs_unlock();
1375 ovs_notify(&dp_datapath_genl_family, reply, info);
1376
1377 return 0;
1378unlock:
1379 ovs_unlock();
1380 return err;
1381}
1382
1383static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1384{
1385 struct sk_buff *reply;
1386 struct datapath *dp;
1387 int err;
1388
1389 ovs_lock();
1390 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1391 if (IS_ERR(dp)) {
1392 err = PTR_ERR(dp);
1393 goto unlock;
1394 }
1395
1396 reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_NEW);
1397 if (IS_ERR(reply)) {
1398 err = PTR_ERR(reply);
1399 goto unlock;
1400 }
1401
1402 ovs_unlock();
1403 return genlmsg_reply(reply, info);
1404
1405unlock:
1406 ovs_unlock();
1407 return err;
1408}
1409
1410static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1411{
1412 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1413 struct datapath *dp;
1414 int skip = cb->args[0];
1415 int i = 0;
1416
1417 rcu_read_lock();
1418 list_for_each_entry_rcu(dp, &ovs_net->dps, list_node) {
1419 if (i >= skip &&
1420 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1421 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1422 OVS_DP_CMD_NEW) < 0)
1423 break;
1424 i++;
1425 }
1426 rcu_read_unlock();
1427
1428 cb->args[0] = i;
1429
1430 return skb->len;
1431}
1432
1433static const struct genl_ops dp_datapath_genl_ops[] = {
1434 { .cmd = OVS_DP_CMD_NEW,
1435 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1436 .policy = datapath_policy,
1437 .doit = ovs_dp_cmd_new
1438 },
1439 { .cmd = OVS_DP_CMD_DEL,
1440 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1441 .policy = datapath_policy,
1442 .doit = ovs_dp_cmd_del
1443 },
1444 { .cmd = OVS_DP_CMD_GET,
1445 .flags = 0, /* OK for unprivileged users. */
1446 .policy = datapath_policy,
1447 .doit = ovs_dp_cmd_get,
1448 .dumpit = ovs_dp_cmd_dump
1449 },
1450 { .cmd = OVS_DP_CMD_SET,
1451 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1452 .policy = datapath_policy,
1453 .doit = ovs_dp_cmd_set,
1454 },
1455};
1456
1457static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1458 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1459 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1460 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1461 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1462 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1463 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1464};
1465
1466struct genl_family dp_vport_genl_family = {
1467 .id = GENL_ID_GENERATE,
1468 .hdrsize = sizeof(struct ovs_header),
1469 .name = OVS_VPORT_FAMILY,
1470 .version = OVS_VPORT_VERSION,
1471 .maxattr = OVS_VPORT_ATTR_MAX,
1472 .netnsok = true,
1473 .parallel_ops = true,
1474};
1475
1476static struct genl_multicast_group ovs_dp_vport_multicast_group = {
1477 .name = OVS_VPORT_MCGROUP
1478};
1479
1480/* Called with ovs_mutex or RCU read lock. */
1481static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1482 u32 portid, u32 seq, u32 flags, u8 cmd)
1483{
1484 struct ovs_header *ovs_header;
1485 struct ovs_vport_stats vport_stats;
1486 int err;
1487
1488 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1489 flags, cmd);
1490 if (!ovs_header)
1491 return -EMSGSIZE;
1492
1493 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1494
1495 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1496 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1497 nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1498 nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_portid))
1499 goto nla_put_failure;
1500
1501 ovs_vport_get_stats(vport, &vport_stats);
1502 if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1503 &vport_stats))
1504 goto nla_put_failure;
1505
1506 err = ovs_vport_get_options(vport, skb);
1507 if (err == -EMSGSIZE)
1508 goto error;
1509
1510 return genlmsg_end(skb, ovs_header);
1511
1512nla_put_failure:
1513 err = -EMSGSIZE;
1514error:
1515 genlmsg_cancel(skb, ovs_header);
1516 return err;
1517}
1518
1519/* Called with ovs_mutex or RCU read lock. */
1520struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1521 u32 seq, u8 cmd)
1522{
1523 struct sk_buff *skb;
1524 int retval;
1525
1526 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1527 if (!skb)
1528 return ERR_PTR(-ENOMEM);
1529
1530 retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1531 BUG_ON(retval < 0);
1532
1533 return skb;
1534}
1535
1536/* Called with ovs_mutex or RCU read lock. */
1537static struct vport *lookup_vport(struct net *net,
1538 struct ovs_header *ovs_header,
1539 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1540{
1541 struct datapath *dp;
1542 struct vport *vport;
1543
1544 if (a[OVS_VPORT_ATTR_NAME]) {
1545 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1546 if (!vport)
1547 return ERR_PTR(-ENODEV);
1548 if (ovs_header->dp_ifindex &&
1549 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1550 return ERR_PTR(-ENODEV);
1551 return vport;
1552 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1553 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1554
1555 if (port_no >= DP_MAX_PORTS)
1556 return ERR_PTR(-EFBIG);
1557
1558 dp = get_dp(net, ovs_header->dp_ifindex);
1559 if (!dp)
1560 return ERR_PTR(-ENODEV);
1561
1562 vport = ovs_vport_ovsl_rcu(dp, port_no);
1563 if (!vport)
1564 return ERR_PTR(-ENODEV);
1565 return vport;
1566 } else
1567 return ERR_PTR(-EINVAL);
1568}
1569
1570static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1571{
1572 struct nlattr **a = info->attrs;
1573 struct ovs_header *ovs_header = info->userhdr;
1574 struct vport_parms parms;
1575 struct sk_buff *reply;
1576 struct vport *vport;
1577 struct datapath *dp;
1578 u32 port_no;
1579 int err;
1580
1581 err = -EINVAL;
1582 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1583 !a[OVS_VPORT_ATTR_UPCALL_PID])
1584 goto exit;
1585
1586 ovs_lock();
1587 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1588 err = -ENODEV;
1589 if (!dp)
1590 goto exit_unlock;
1591
1592 if (a[OVS_VPORT_ATTR_PORT_NO]) {
1593 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1594
1595 err = -EFBIG;
1596 if (port_no >= DP_MAX_PORTS)
1597 goto exit_unlock;
1598
1599 vport = ovs_vport_ovsl(dp, port_no);
1600 err = -EBUSY;
1601 if (vport)
1602 goto exit_unlock;
1603 } else {
1604 for (port_no = 1; ; port_no++) {
1605 if (port_no >= DP_MAX_PORTS) {
1606 err = -EFBIG;
1607 goto exit_unlock;
1608 }
1609 vport = ovs_vport_ovsl(dp, port_no);
1610 if (!vport)
1611 break;
1612 }
1613 }
1614
1615 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1616 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1617 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1618 parms.dp = dp;
1619 parms.port_no = port_no;
1620 parms.upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1621
1622 vport = new_vport(&parms);
1623 err = PTR_ERR(vport);
1624 if (IS_ERR(vport))
1625 goto exit_unlock;
1626
1627 err = 0;
1628 reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
1629 OVS_VPORT_CMD_NEW);
1630 if (IS_ERR(reply)) {
1631 err = PTR_ERR(reply);
1632 ovs_dp_detach_port(vport);
1633 goto exit_unlock;
1634 }
1635
1636 ovs_notify(&dp_vport_genl_family, reply, info);
1637
1638exit_unlock:
1639 ovs_unlock();
1640exit:
1641 return err;
1642}
1643
1644static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1645{
1646 struct nlattr **a = info->attrs;
1647 struct sk_buff *reply;
1648 struct vport *vport;
1649 int err;
1650
1651 ovs_lock();
1652 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1653 err = PTR_ERR(vport);
1654 if (IS_ERR(vport))
1655 goto exit_unlock;
1656
1657 if (a[OVS_VPORT_ATTR_TYPE] &&
1658 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
1659 err = -EINVAL;
1660 goto exit_unlock;
1661 }
1662
1663 reply = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1664 if (!reply) {
1665 err = -ENOMEM;
1666 goto exit_unlock;
1667 }
1668
1669 if (a[OVS_VPORT_ATTR_OPTIONS]) {
1670 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1671 if (err)
1672 goto exit_free;
1673 }
1674
1675 if (a[OVS_VPORT_ATTR_UPCALL_PID])
1676 vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1677
1678 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1679 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1680 BUG_ON(err < 0);
1681
1682 ovs_unlock();
1683 ovs_notify(&dp_vport_genl_family, reply, info);
1684 return 0;
1685
1686exit_free:
1687 kfree_skb(reply);
1688exit_unlock:
1689 ovs_unlock();
1690 return err;
1691}
1692
1693static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1694{
1695 struct nlattr **a = info->attrs;
1696 struct sk_buff *reply;
1697 struct vport *vport;
1698 int err;
1699
1700 ovs_lock();
1701 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1702 err = PTR_ERR(vport);
1703 if (IS_ERR(vport))
1704 goto exit_unlock;
1705
1706 if (vport->port_no == OVSP_LOCAL) {
1707 err = -EINVAL;
1708 goto exit_unlock;
1709 }
1710
1711 reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
1712 info->snd_seq, OVS_VPORT_CMD_DEL);
1713 err = PTR_ERR(reply);
1714 if (IS_ERR(reply))
1715 goto exit_unlock;
1716
1717 err = 0;
1718 ovs_dp_detach_port(vport);
1719
1720 ovs_notify(&dp_vport_genl_family, reply, info);
1721
1722exit_unlock:
1723 ovs_unlock();
1724 return err;
1725}
1726
1727static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1728{
1729 struct nlattr **a = info->attrs;
1730 struct ovs_header *ovs_header = info->userhdr;
1731 struct sk_buff *reply;
1732 struct vport *vport;
1733 int err;
1734
1735 rcu_read_lock();
1736 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1737 err = PTR_ERR(vport);
1738 if (IS_ERR(vport))
1739 goto exit_unlock;
1740
1741 reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
1742 info->snd_seq, OVS_VPORT_CMD_NEW);
1743 err = PTR_ERR(reply);
1744 if (IS_ERR(reply))
1745 goto exit_unlock;
1746
1747 rcu_read_unlock();
1748
1749 return genlmsg_reply(reply, info);
1750
1751exit_unlock:
1752 rcu_read_unlock();
1753 return err;
1754}
1755
1756static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1757{
1758 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1759 struct datapath *dp;
1760 int bucket = cb->args[0], skip = cb->args[1];
1761 int i, j = 0;
1762
1763 rcu_read_lock();
1764 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1765 if (!dp) {
1766 rcu_read_unlock();
1767 return -ENODEV;
1768 }
1769 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
1770 struct vport *vport;
1771
1772 j = 0;
1773 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1774 if (j >= skip &&
1775 ovs_vport_cmd_fill_info(vport, skb,
1776 NETLINK_CB(cb->skb).portid,
1777 cb->nlh->nlmsg_seq,
1778 NLM_F_MULTI,
1779 OVS_VPORT_CMD_NEW) < 0)
1780 goto out;
1781
1782 j++;
1783 }
1784 skip = 0;
1785 }
1786out:
1787 rcu_read_unlock();
1788
1789 cb->args[0] = i;
1790 cb->args[1] = j;
1791
1792 return skb->len;
1793}
1794
1795static const struct genl_ops dp_vport_genl_ops[] = {
1796 { .cmd = OVS_VPORT_CMD_NEW,
1797 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1798 .policy = vport_policy,
1799 .doit = ovs_vport_cmd_new
1800 },
1801 { .cmd = OVS_VPORT_CMD_DEL,
1802 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1803 .policy = vport_policy,
1804 .doit = ovs_vport_cmd_del
1805 },
1806 { .cmd = OVS_VPORT_CMD_GET,
1807 .flags = 0, /* OK for unprivileged users. */
1808 .policy = vport_policy,
1809 .doit = ovs_vport_cmd_get,
1810 .dumpit = ovs_vport_cmd_dump
1811 },
1812 { .cmd = OVS_VPORT_CMD_SET,
1813 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1814 .policy = vport_policy,
1815 .doit = ovs_vport_cmd_set,
1816 },
1817};
1818
1819struct genl_family_and_ops {
1820 struct genl_family *family;
1821 const struct genl_ops *ops;
1822 int n_ops;
1823 const struct genl_multicast_group *group;
1824};
1825
1826static const struct genl_family_and_ops dp_genl_families[] = {
1827 { &dp_datapath_genl_family,
1828 dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1829 &ovs_dp_datapath_multicast_group },
1830 { &dp_vport_genl_family,
1831 dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1832 &ovs_dp_vport_multicast_group },
1833 { &dp_flow_genl_family,
1834 dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1835 &ovs_dp_flow_multicast_group },
1836 { &dp_packet_genl_family,
1837 dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1838 NULL },
1839};
1840
1841static void dp_unregister_genl(int n_families)
1842{
1843 int i;
1844
1845 for (i = 0; i < n_families; i++)
1846 genl_unregister_family(dp_genl_families[i].family);
1847}
1848
1849static int dp_register_genl(void)
1850{
1851 int n_registered;
1852 int err;
1853 int i;
1854
1855 n_registered = 0;
1856 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
1857 const struct genl_family_and_ops *f = &dp_genl_families[i];
1858
1859 f->family->ops = f->ops;
1860 f->family->n_ops = f->n_ops;
1861 f->family->mcgrps = f->group;
1862 f->family->n_mcgrps = f->group ? 1 : 0;
1863 err = genl_register_family(f->family);
1864 if (err)
1865 goto error;
1866 n_registered++;
1867 }
1868
1869 return 0;
1870
1871error:
1872 dp_unregister_genl(n_registered);
1873 return err;
1874}
1875
1876static int __net_init ovs_init_net(struct net *net)
1877{
1878 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1879
1880 INIT_LIST_HEAD(&ovs_net->dps);
1881 INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
1882 return 0;
1883}
1884
1885static void __net_exit ovs_exit_net(struct net *net)
1886{
1887 struct datapath *dp, *dp_next;
1888 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1889
1890 ovs_lock();
1891 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
1892 __dp_destroy(dp);
1893 ovs_unlock();
1894
1895 cancel_work_sync(&ovs_net->dp_notify_work);
1896}
1897
1898static struct pernet_operations ovs_net_ops = {
1899 .init = ovs_init_net,
1900 .exit = ovs_exit_net,
1901 .id = &ovs_net_id,
1902 .size = sizeof(struct ovs_net),
1903};
1904
1905static int __init dp_init(void)
1906{
1907 int err;
1908
1909 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
1910
1911 pr_info("Open vSwitch switching datapath\n");
1912
1913 err = ovs_flow_init();
1914 if (err)
1915 goto error;
1916
1917 err = ovs_vport_init();
1918 if (err)
1919 goto error_flow_exit;
1920
1921 err = register_pernet_device(&ovs_net_ops);
1922 if (err)
1923 goto error_vport_exit;
1924
1925 err = register_netdevice_notifier(&ovs_dp_device_notifier);
1926 if (err)
1927 goto error_netns_exit;
1928
1929 err = dp_register_genl();
1930 if (err < 0)
1931 goto error_unreg_notifier;
1932
1933 return 0;
1934
1935error_unreg_notifier:
1936 unregister_netdevice_notifier(&ovs_dp_device_notifier);
1937error_netns_exit:
1938 unregister_pernet_device(&ovs_net_ops);
1939error_vport_exit:
1940 ovs_vport_exit();
1941error_flow_exit:
1942 ovs_flow_exit();
1943error:
1944 return err;
1945}
1946
1947static void dp_cleanup(void)
1948{
1949 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
1950 unregister_netdevice_notifier(&ovs_dp_device_notifier);
1951 unregister_pernet_device(&ovs_net_ops);
1952 rcu_barrier();
1953 ovs_vport_exit();
1954 ovs_flow_exit();
1955}
1956
1957module_init(dp_init);
1958module_exit(dp_cleanup);
1959
1960MODULE_DESCRIPTION("Open vSwitch switching datapath");
1961MODULE_LICENSE("GPL");