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) 2007-2014 Nicira, Inc.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of version 2 of the GNU General Public
6 * License as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public License
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
16 * 02110-1301, USA
17 */
18
19#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
20
21#include <linux/init.h>
22#include <linux/module.h>
23#include <linux/if_arp.h>
24#include <linux/if_vlan.h>
25#include <linux/in.h>
26#include <linux/ip.h>
27#include <linux/jhash.h>
28#include <linux/delay.h>
29#include <linux/time.h>
30#include <linux/etherdevice.h>
31#include <linux/genetlink.h>
32#include <linux/kernel.h>
33#include <linux/kthread.h>
34#include <linux/mutex.h>
35#include <linux/percpu.h>
36#include <linux/rcupdate.h>
37#include <linux/tcp.h>
38#include <linux/udp.h>
39#include <linux/ethtool.h>
40#include <linux/wait.h>
41#include <asm/div64.h>
42#include <linux/highmem.h>
43#include <linux/netfilter_bridge.h>
44#include <linux/netfilter_ipv4.h>
45#include <linux/inetdevice.h>
46#include <linux/list.h>
47#include <linux/openvswitch.h>
48#include <linux/rculist.h>
49#include <linux/dmi.h>
50#include <net/genetlink.h>
51#include <net/net_namespace.h>
52#include <net/netns/generic.h>
53
54#include "datapath.h"
55#include "flow.h"
56#include "flow_table.h"
57#include "flow_netlink.h"
58#include "vport-internal_dev.h"
59#include "vport-netdev.h"
60
61int ovs_net_id __read_mostly;
62EXPORT_SYMBOL_GPL(ovs_net_id);
63
64static struct genl_family dp_packet_genl_family;
65static struct genl_family dp_flow_genl_family;
66static struct genl_family dp_datapath_genl_family;
67
68static const struct nla_policy flow_policy[];
69
70static const struct genl_multicast_group ovs_dp_flow_multicast_group = {
71 .name = OVS_FLOW_MCGROUP,
72};
73
74static const struct genl_multicast_group ovs_dp_datapath_multicast_group = {
75 .name = OVS_DATAPATH_MCGROUP,
76};
77
78static const struct genl_multicast_group ovs_dp_vport_multicast_group = {
79 .name = OVS_VPORT_MCGROUP,
80};
81
82/* Check if need to build a reply message.
83 * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
84static bool ovs_must_notify(struct genl_family *family, struct genl_info *info,
85 unsigned int group)
86{
87 return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
88 genl_has_listeners(family, genl_info_net(info), group);
89}
90
91static void ovs_notify(struct genl_family *family,
92 struct sk_buff *skb, struct genl_info *info)
93{
94 genl_notify(family, skb, info, 0, GFP_KERNEL);
95}
96
97/**
98 * DOC: Locking:
99 *
100 * All writes e.g. Writes to device state (add/remove datapath, port, set
101 * operations on vports, etc.), Writes to other state (flow table
102 * modifications, set miscellaneous datapath parameters, etc.) are protected
103 * by ovs_lock.
104 *
105 * Reads are protected by RCU.
106 *
107 * There are a few special cases (mostly stats) that have their own
108 * synchronization but they nest under all of above and don't interact with
109 * each other.
110 *
111 * The RTNL lock nests inside ovs_mutex.
112 */
113
114static DEFINE_MUTEX(ovs_mutex);
115
116void ovs_lock(void)
117{
118 mutex_lock(&ovs_mutex);
119}
120
121void ovs_unlock(void)
122{
123 mutex_unlock(&ovs_mutex);
124}
125
126#ifdef CONFIG_LOCKDEP
127int lockdep_ovsl_is_held(void)
128{
129 if (debug_locks)
130 return lockdep_is_held(&ovs_mutex);
131 else
132 return 1;
133}
134EXPORT_SYMBOL_GPL(lockdep_ovsl_is_held);
135#endif
136
137static struct vport *new_vport(const struct vport_parms *);
138static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
139 const struct sw_flow_key *,
140 const struct dp_upcall_info *);
141static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
142 const struct sw_flow_key *,
143 const struct dp_upcall_info *);
144
145/* Must be called with rcu_read_lock. */
146static struct datapath *get_dp_rcu(struct net *net, int dp_ifindex)
147{
148 struct net_device *dev = dev_get_by_index_rcu(net, dp_ifindex);
149
150 if (dev) {
151 struct vport *vport = ovs_internal_dev_get_vport(dev);
152 if (vport)
153 return vport->dp;
154 }
155
156 return NULL;
157}
158
159/* The caller must hold either ovs_mutex or rcu_read_lock to keep the
160 * returned dp pointer valid.
161 */
162static inline struct datapath *get_dp(struct net *net, int dp_ifindex)
163{
164 struct datapath *dp;
165
166 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_ovsl_is_held());
167 rcu_read_lock();
168 dp = get_dp_rcu(net, dp_ifindex);
169 rcu_read_unlock();
170
171 return dp;
172}
173
174/* Must be called with rcu_read_lock or ovs_mutex. */
175const char *ovs_dp_name(const struct datapath *dp)
176{
177 struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
178 return ovs_vport_name(vport);
179}
180
181static int get_dpifindex(const struct datapath *dp)
182{
183 struct vport *local;
184 int ifindex;
185
186 rcu_read_lock();
187
188 local = ovs_vport_rcu(dp, OVSP_LOCAL);
189 if (local)
190 ifindex = local->dev->ifindex;
191 else
192 ifindex = 0;
193
194 rcu_read_unlock();
195
196 return ifindex;
197}
198
199static void destroy_dp_rcu(struct rcu_head *rcu)
200{
201 struct datapath *dp = container_of(rcu, struct datapath, rcu);
202
203 ovs_flow_tbl_destroy(&dp->table);
204 free_percpu(dp->stats_percpu);
205 kfree(dp->ports);
206 kfree(dp);
207}
208
209static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
210 u16 port_no)
211{
212 return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
213}
214
215/* Called with ovs_mutex or RCU read lock. */
216struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
217{
218 struct vport *vport;
219 struct hlist_head *head;
220
221 head = vport_hash_bucket(dp, port_no);
222 hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
223 if (vport->port_no == port_no)
224 return vport;
225 }
226 return NULL;
227}
228
229/* Called with ovs_mutex. */
230static struct vport *new_vport(const struct vport_parms *parms)
231{
232 struct vport *vport;
233
234 vport = ovs_vport_add(parms);
235 if (!IS_ERR(vport)) {
236 struct datapath *dp = parms->dp;
237 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
238
239 hlist_add_head_rcu(&vport->dp_hash_node, head);
240 }
241 return vport;
242}
243
244void ovs_dp_detach_port(struct vport *p)
245{
246 ASSERT_OVSL();
247
248 /* First drop references to device. */
249 hlist_del_rcu(&p->dp_hash_node);
250
251 /* Then destroy it. */
252 ovs_vport_del(p);
253}
254
255/* Must be called with rcu_read_lock. */
256void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
257{
258 const struct vport *p = OVS_CB(skb)->input_vport;
259 struct datapath *dp = p->dp;
260 struct sw_flow *flow;
261 struct sw_flow_actions *sf_acts;
262 struct dp_stats_percpu *stats;
263 u64 *stats_counter;
264 u32 n_mask_hit;
265
266 stats = this_cpu_ptr(dp->stats_percpu);
267
268 /* Look up flow. */
269 flow = ovs_flow_tbl_lookup_stats(&dp->table, key, &n_mask_hit);
270 if (unlikely(!flow)) {
271 struct dp_upcall_info upcall;
272 int error;
273
274 memset(&upcall, 0, sizeof(upcall));
275 upcall.cmd = OVS_PACKET_CMD_MISS;
276 upcall.portid = ovs_vport_find_upcall_portid(p, skb);
277 upcall.mru = OVS_CB(skb)->mru;
278 error = ovs_dp_upcall(dp, skb, key, &upcall);
279 if (unlikely(error))
280 kfree_skb(skb);
281 else
282 consume_skb(skb);
283 stats_counter = &stats->n_missed;
284 goto out;
285 }
286
287 ovs_flow_stats_update(flow, key->tp.flags, skb);
288 sf_acts = rcu_dereference(flow->sf_acts);
289 ovs_execute_actions(dp, skb, sf_acts, key);
290
291 stats_counter = &stats->n_hit;
292
293out:
294 /* Update datapath statistics. */
295 u64_stats_update_begin(&stats->syncp);
296 (*stats_counter)++;
297 stats->n_mask_hit += n_mask_hit;
298 u64_stats_update_end(&stats->syncp);
299}
300
301int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
302 const struct sw_flow_key *key,
303 const struct dp_upcall_info *upcall_info)
304{
305 struct dp_stats_percpu *stats;
306 int err;
307
308 if (upcall_info->portid == 0) {
309 err = -ENOTCONN;
310 goto err;
311 }
312
313 if (!skb_is_gso(skb))
314 err = queue_userspace_packet(dp, skb, key, upcall_info);
315 else
316 err = queue_gso_packets(dp, skb, key, upcall_info);
317 if (err)
318 goto err;
319
320 return 0;
321
322err:
323 stats = this_cpu_ptr(dp->stats_percpu);
324
325 u64_stats_update_begin(&stats->syncp);
326 stats->n_lost++;
327 u64_stats_update_end(&stats->syncp);
328
329 return err;
330}
331
332static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
333 const struct sw_flow_key *key,
334 const struct dp_upcall_info *upcall_info)
335{
336 unsigned short gso_type = skb_shinfo(skb)->gso_type;
337 struct sw_flow_key later_key;
338 struct sk_buff *segs, *nskb;
339 int err;
340
341 BUILD_BUG_ON(sizeof(*OVS_CB(skb)) > SKB_SGO_CB_OFFSET);
342 segs = __skb_gso_segment(skb, NETIF_F_SG, false);
343 if (IS_ERR(segs))
344 return PTR_ERR(segs);
345 if (segs == NULL)
346 return -EINVAL;
347
348 if (gso_type & SKB_GSO_UDP) {
349 /* The initial flow key extracted by ovs_flow_key_extract()
350 * in this case is for a first fragment, so we need to
351 * properly mark later fragments.
352 */
353 later_key = *key;
354 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
355 }
356
357 /* Queue all of the segments. */
358 skb = segs;
359 do {
360 if (gso_type & SKB_GSO_UDP && skb != segs)
361 key = &later_key;
362
363 err = queue_userspace_packet(dp, skb, key, upcall_info);
364 if (err)
365 break;
366
367 } while ((skb = skb->next));
368
369 /* Free all of the segments. */
370 skb = segs;
371 do {
372 nskb = skb->next;
373 if (err)
374 kfree_skb(skb);
375 else
376 consume_skb(skb);
377 } while ((skb = nskb));
378 return err;
379}
380
381static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
382 unsigned int hdrlen)
383{
384 size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
385 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
386 + nla_total_size(ovs_key_attr_size()); /* OVS_PACKET_ATTR_KEY */
387
388 /* OVS_PACKET_ATTR_USERDATA */
389 if (upcall_info->userdata)
390 size += NLA_ALIGN(upcall_info->userdata->nla_len);
391
392 /* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
393 if (upcall_info->egress_tun_info)
394 size += nla_total_size(ovs_tun_key_attr_size());
395
396 /* OVS_PACKET_ATTR_ACTIONS */
397 if (upcall_info->actions_len)
398 size += nla_total_size(upcall_info->actions_len);
399
400 /* OVS_PACKET_ATTR_MRU */
401 if (upcall_info->mru)
402 size += nla_total_size(sizeof(upcall_info->mru));
403
404 return size;
405}
406
407static void pad_packet(struct datapath *dp, struct sk_buff *skb)
408{
409 if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
410 size_t plen = NLA_ALIGN(skb->len) - skb->len;
411
412 if (plen > 0)
413 memset(skb_put(skb, plen), 0, plen);
414 }
415}
416
417static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
418 const struct sw_flow_key *key,
419 const struct dp_upcall_info *upcall_info)
420{
421 struct ovs_header *upcall;
422 struct sk_buff *nskb = NULL;
423 struct sk_buff *user_skb = NULL; /* to be queued to userspace */
424 struct nlattr *nla;
425 size_t len;
426 unsigned int hlen;
427 int err, dp_ifindex;
428
429 dp_ifindex = get_dpifindex(dp);
430 if (!dp_ifindex)
431 return -ENODEV;
432
433 if (skb_vlan_tag_present(skb)) {
434 nskb = skb_clone(skb, GFP_ATOMIC);
435 if (!nskb)
436 return -ENOMEM;
437
438 nskb = __vlan_hwaccel_push_inside(nskb);
439 if (!nskb)
440 return -ENOMEM;
441
442 skb = nskb;
443 }
444
445 if (nla_attr_size(skb->len) > USHRT_MAX) {
446 err = -EFBIG;
447 goto out;
448 }
449
450 /* Complete checksum if needed */
451 if (skb->ip_summed == CHECKSUM_PARTIAL &&
452 (err = skb_checksum_help(skb)))
453 goto out;
454
455 /* Older versions of OVS user space enforce alignment of the last
456 * Netlink attribute to NLA_ALIGNTO which would require extensive
457 * padding logic. Only perform zerocopy if padding is not required.
458 */
459 if (dp->user_features & OVS_DP_F_UNALIGNED)
460 hlen = skb_zerocopy_headlen(skb);
461 else
462 hlen = skb->len;
463
464 len = upcall_msg_size(upcall_info, hlen);
465 user_skb = genlmsg_new(len, GFP_ATOMIC);
466 if (!user_skb) {
467 err = -ENOMEM;
468 goto out;
469 }
470
471 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
472 0, upcall_info->cmd);
473 upcall->dp_ifindex = dp_ifindex;
474
475 err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
476 BUG_ON(err);
477
478 if (upcall_info->userdata)
479 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
480 nla_len(upcall_info->userdata),
481 nla_data(upcall_info->userdata));
482
483 if (upcall_info->egress_tun_info) {
484 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_EGRESS_TUN_KEY);
485 err = ovs_nla_put_tunnel_info(user_skb,
486 upcall_info->egress_tun_info);
487 BUG_ON(err);
488 nla_nest_end(user_skb, nla);
489 }
490
491 if (upcall_info->actions_len) {
492 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_ACTIONS);
493 err = ovs_nla_put_actions(upcall_info->actions,
494 upcall_info->actions_len,
495 user_skb);
496 if (!err)
497 nla_nest_end(user_skb, nla);
498 else
499 nla_nest_cancel(user_skb, nla);
500 }
501
502 /* Add OVS_PACKET_ATTR_MRU */
503 if (upcall_info->mru) {
504 if (nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU,
505 upcall_info->mru)) {
506 err = -ENOBUFS;
507 goto out;
508 }
509 pad_packet(dp, user_skb);
510 }
511
512 /* Only reserve room for attribute header, packet data is added
513 * in skb_zerocopy() */
514 if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
515 err = -ENOBUFS;
516 goto out;
517 }
518 nla->nla_len = nla_attr_size(skb->len);
519
520 err = skb_zerocopy(user_skb, skb, skb->len, hlen);
521 if (err)
522 goto out;
523
524 /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
525 pad_packet(dp, user_skb);
526
527 ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
528
529 err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
530 user_skb = NULL;
531out:
532 if (err)
533 skb_tx_error(skb);
534 kfree_skb(user_skb);
535 kfree_skb(nskb);
536 return err;
537}
538
539static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
540{
541 struct ovs_header *ovs_header = info->userhdr;
542 struct net *net = sock_net(skb->sk);
543 struct nlattr **a = info->attrs;
544 struct sw_flow_actions *acts;
545 struct sk_buff *packet;
546 struct sw_flow *flow;
547 struct sw_flow_actions *sf_acts;
548 struct datapath *dp;
549 struct ethhdr *eth;
550 struct vport *input_vport;
551 u16 mru = 0;
552 int len;
553 int err;
554 bool log = !a[OVS_PACKET_ATTR_PROBE];
555
556 err = -EINVAL;
557 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
558 !a[OVS_PACKET_ATTR_ACTIONS])
559 goto err;
560
561 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
562 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
563 err = -ENOMEM;
564 if (!packet)
565 goto err;
566 skb_reserve(packet, NET_IP_ALIGN);
567
568 nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
569
570 skb_reset_mac_header(packet);
571 eth = eth_hdr(packet);
572
573 /* Normally, setting the skb 'protocol' field would be handled by a
574 * call to eth_type_trans(), but it assumes there's a sending
575 * device, which we may not have. */
576 if (eth_proto_is_802_3(eth->h_proto))
577 packet->protocol = eth->h_proto;
578 else
579 packet->protocol = htons(ETH_P_802_2);
580
581 /* Set packet's mru */
582 if (a[OVS_PACKET_ATTR_MRU]) {
583 mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
584 packet->ignore_df = 1;
585 }
586 OVS_CB(packet)->mru = mru;
587
588 /* Build an sw_flow for sending this packet. */
589 flow = ovs_flow_alloc();
590 err = PTR_ERR(flow);
591 if (IS_ERR(flow))
592 goto err_kfree_skb;
593
594 err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
595 packet, &flow->key, log);
596 if (err)
597 goto err_flow_free;
598
599 err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
600 &flow->key, &acts, log);
601 if (err)
602 goto err_flow_free;
603
604 rcu_assign_pointer(flow->sf_acts, acts);
605 packet->priority = flow->key.phy.priority;
606 packet->mark = flow->key.phy.skb_mark;
607
608 rcu_read_lock();
609 dp = get_dp_rcu(net, ovs_header->dp_ifindex);
610 err = -ENODEV;
611 if (!dp)
612 goto err_unlock;
613
614 input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
615 if (!input_vport)
616 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
617
618 if (!input_vport)
619 goto err_unlock;
620
621 packet->dev = input_vport->dev;
622 OVS_CB(packet)->input_vport = input_vport;
623 sf_acts = rcu_dereference(flow->sf_acts);
624
625 local_bh_disable();
626 err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
627 local_bh_enable();
628 rcu_read_unlock();
629
630 ovs_flow_free(flow, false);
631 return err;
632
633err_unlock:
634 rcu_read_unlock();
635err_flow_free:
636 ovs_flow_free(flow, false);
637err_kfree_skb:
638 kfree_skb(packet);
639err:
640 return err;
641}
642
643static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
644 [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
645 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
646 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
647 [OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
648 [OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
649};
650
651static const struct genl_ops dp_packet_genl_ops[] = {
652 { .cmd = OVS_PACKET_CMD_EXECUTE,
653 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
654 .policy = packet_policy,
655 .doit = ovs_packet_cmd_execute
656 }
657};
658
659static struct genl_family dp_packet_genl_family = {
660 .id = GENL_ID_GENERATE,
661 .hdrsize = sizeof(struct ovs_header),
662 .name = OVS_PACKET_FAMILY,
663 .version = OVS_PACKET_VERSION,
664 .maxattr = OVS_PACKET_ATTR_MAX,
665 .netnsok = true,
666 .parallel_ops = true,
667 .ops = dp_packet_genl_ops,
668 .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
669};
670
671static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
672 struct ovs_dp_megaflow_stats *mega_stats)
673{
674 int i;
675
676 memset(mega_stats, 0, sizeof(*mega_stats));
677
678 stats->n_flows = ovs_flow_tbl_count(&dp->table);
679 mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
680
681 stats->n_hit = stats->n_missed = stats->n_lost = 0;
682
683 for_each_possible_cpu(i) {
684 const struct dp_stats_percpu *percpu_stats;
685 struct dp_stats_percpu local_stats;
686 unsigned int start;
687
688 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
689
690 do {
691 start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
692 local_stats = *percpu_stats;
693 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
694
695 stats->n_hit += local_stats.n_hit;
696 stats->n_missed += local_stats.n_missed;
697 stats->n_lost += local_stats.n_lost;
698 mega_stats->n_mask_hit += local_stats.n_mask_hit;
699 }
700}
701
702static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
703{
704 return ovs_identifier_is_ufid(sfid) &&
705 !(ufid_flags & OVS_UFID_F_OMIT_KEY);
706}
707
708static bool should_fill_mask(uint32_t ufid_flags)
709{
710 return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
711}
712
713static bool should_fill_actions(uint32_t ufid_flags)
714{
715 return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
716}
717
718static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
719 const struct sw_flow_id *sfid,
720 uint32_t ufid_flags)
721{
722 size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
723
724 /* OVS_FLOW_ATTR_UFID */
725 if (sfid && ovs_identifier_is_ufid(sfid))
726 len += nla_total_size(sfid->ufid_len);
727
728 /* OVS_FLOW_ATTR_KEY */
729 if (!sfid || should_fill_key(sfid, ufid_flags))
730 len += nla_total_size(ovs_key_attr_size());
731
732 /* OVS_FLOW_ATTR_MASK */
733 if (should_fill_mask(ufid_flags))
734 len += nla_total_size(ovs_key_attr_size());
735
736 /* OVS_FLOW_ATTR_ACTIONS */
737 if (should_fill_actions(ufid_flags))
738 len += nla_total_size(acts->orig_len);
739
740 return len
741 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
742 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
743 + nla_total_size(8); /* OVS_FLOW_ATTR_USED */
744}
745
746/* Called with ovs_mutex or RCU read lock. */
747static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
748 struct sk_buff *skb)
749{
750 struct ovs_flow_stats stats;
751 __be16 tcp_flags;
752 unsigned long used;
753
754 ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
755
756 if (used &&
757 nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
758 return -EMSGSIZE;
759
760 if (stats.n_packets &&
761 nla_put(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats))
762 return -EMSGSIZE;
763
764 if ((u8)ntohs(tcp_flags) &&
765 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
766 return -EMSGSIZE;
767
768 return 0;
769}
770
771/* Called with ovs_mutex or RCU read lock. */
772static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
773 struct sk_buff *skb, int skb_orig_len)
774{
775 struct nlattr *start;
776 int err;
777
778 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
779 * this is the first flow to be dumped into 'skb'. This is unusual for
780 * Netlink but individual action lists can be longer than
781 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
782 * The userspace caller can always fetch the actions separately if it
783 * really wants them. (Most userspace callers in fact don't care.)
784 *
785 * This can only fail for dump operations because the skb is always
786 * properly sized for single flows.
787 */
788 start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
789 if (start) {
790 const struct sw_flow_actions *sf_acts;
791
792 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
793 err = ovs_nla_put_actions(sf_acts->actions,
794 sf_acts->actions_len, skb);
795
796 if (!err)
797 nla_nest_end(skb, start);
798 else {
799 if (skb_orig_len)
800 return err;
801
802 nla_nest_cancel(skb, start);
803 }
804 } else if (skb_orig_len) {
805 return -EMSGSIZE;
806 }
807
808 return 0;
809}
810
811/* Called with ovs_mutex or RCU read lock. */
812static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
813 struct sk_buff *skb, u32 portid,
814 u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
815{
816 const int skb_orig_len = skb->len;
817 struct ovs_header *ovs_header;
818 int err;
819
820 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
821 flags, cmd);
822 if (!ovs_header)
823 return -EMSGSIZE;
824
825 ovs_header->dp_ifindex = dp_ifindex;
826
827 err = ovs_nla_put_identifier(flow, skb);
828 if (err)
829 goto error;
830
831 if (should_fill_key(&flow->id, ufid_flags)) {
832 err = ovs_nla_put_masked_key(flow, skb);
833 if (err)
834 goto error;
835 }
836
837 if (should_fill_mask(ufid_flags)) {
838 err = ovs_nla_put_mask(flow, skb);
839 if (err)
840 goto error;
841 }
842
843 err = ovs_flow_cmd_fill_stats(flow, skb);
844 if (err)
845 goto error;
846
847 if (should_fill_actions(ufid_flags)) {
848 err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
849 if (err)
850 goto error;
851 }
852
853 genlmsg_end(skb, ovs_header);
854 return 0;
855
856error:
857 genlmsg_cancel(skb, ovs_header);
858 return err;
859}
860
861/* May not be called with RCU read lock. */
862static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
863 const struct sw_flow_id *sfid,
864 struct genl_info *info,
865 bool always,
866 uint32_t ufid_flags)
867{
868 struct sk_buff *skb;
869 size_t len;
870
871 if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
872 return NULL;
873
874 len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
875 skb = genlmsg_new(len, GFP_KERNEL);
876 if (!skb)
877 return ERR_PTR(-ENOMEM);
878
879 return skb;
880}
881
882/* Called with ovs_mutex. */
883static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
884 int dp_ifindex,
885 struct genl_info *info, u8 cmd,
886 bool always, u32 ufid_flags)
887{
888 struct sk_buff *skb;
889 int retval;
890
891 skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
892 &flow->id, info, always, ufid_flags);
893 if (IS_ERR_OR_NULL(skb))
894 return skb;
895
896 retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
897 info->snd_portid, info->snd_seq, 0,
898 cmd, ufid_flags);
899 BUG_ON(retval < 0);
900 return skb;
901}
902
903static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
904{
905 struct net *net = sock_net(skb->sk);
906 struct nlattr **a = info->attrs;
907 struct ovs_header *ovs_header = info->userhdr;
908 struct sw_flow *flow = NULL, *new_flow;
909 struct sw_flow_mask mask;
910 struct sk_buff *reply;
911 struct datapath *dp;
912 struct sw_flow_key key;
913 struct sw_flow_actions *acts;
914 struct sw_flow_match match;
915 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
916 int error;
917 bool log = !a[OVS_FLOW_ATTR_PROBE];
918
919 /* Must have key and actions. */
920 error = -EINVAL;
921 if (!a[OVS_FLOW_ATTR_KEY]) {
922 OVS_NLERR(log, "Flow key attr not present in new flow.");
923 goto error;
924 }
925 if (!a[OVS_FLOW_ATTR_ACTIONS]) {
926 OVS_NLERR(log, "Flow actions attr not present in new flow.");
927 goto error;
928 }
929
930 /* Most of the time we need to allocate a new flow, do it before
931 * locking.
932 */
933 new_flow = ovs_flow_alloc();
934 if (IS_ERR(new_flow)) {
935 error = PTR_ERR(new_flow);
936 goto error;
937 }
938
939 /* Extract key. */
940 ovs_match_init(&match, &key, &mask);
941 error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
942 a[OVS_FLOW_ATTR_MASK], log);
943 if (error)
944 goto err_kfree_flow;
945
946 ovs_flow_mask_key(&new_flow->key, &key, true, &mask);
947
948 /* Extract flow identifier. */
949 error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
950 &key, log);
951 if (error)
952 goto err_kfree_flow;
953
954 /* Validate actions. */
955 error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
956 &new_flow->key, &acts, log);
957 if (error) {
958 OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
959 goto err_kfree_flow;
960 }
961
962 reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
963 ufid_flags);
964 if (IS_ERR(reply)) {
965 error = PTR_ERR(reply);
966 goto err_kfree_acts;
967 }
968
969 ovs_lock();
970 dp = get_dp(net, ovs_header->dp_ifindex);
971 if (unlikely(!dp)) {
972 error = -ENODEV;
973 goto err_unlock_ovs;
974 }
975
976 /* Check if this is a duplicate flow */
977 if (ovs_identifier_is_ufid(&new_flow->id))
978 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
979 if (!flow)
980 flow = ovs_flow_tbl_lookup(&dp->table, &key);
981 if (likely(!flow)) {
982 rcu_assign_pointer(new_flow->sf_acts, acts);
983
984 /* Put flow in bucket. */
985 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
986 if (unlikely(error)) {
987 acts = NULL;
988 goto err_unlock_ovs;
989 }
990
991 if (unlikely(reply)) {
992 error = ovs_flow_cmd_fill_info(new_flow,
993 ovs_header->dp_ifindex,
994 reply, info->snd_portid,
995 info->snd_seq, 0,
996 OVS_FLOW_CMD_NEW,
997 ufid_flags);
998 BUG_ON(error < 0);
999 }
1000 ovs_unlock();
1001 } else {
1002 struct sw_flow_actions *old_acts;
1003
1004 /* Bail out if we're not allowed to modify an existing flow.
1005 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1006 * because Generic Netlink treats the latter as a dump
1007 * request. We also accept NLM_F_EXCL in case that bug ever
1008 * gets fixed.
1009 */
1010 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1011 | NLM_F_EXCL))) {
1012 error = -EEXIST;
1013 goto err_unlock_ovs;
1014 }
1015 /* The flow identifier has to be the same for flow updates.
1016 * Look for any overlapping flow.
1017 */
1018 if (unlikely(!ovs_flow_cmp(flow, &match))) {
1019 if (ovs_identifier_is_key(&flow->id))
1020 flow = ovs_flow_tbl_lookup_exact(&dp->table,
1021 &match);
1022 else /* UFID matches but key is different */
1023 flow = NULL;
1024 if (!flow) {
1025 error = -ENOENT;
1026 goto err_unlock_ovs;
1027 }
1028 }
1029 /* Update actions. */
1030 old_acts = ovsl_dereference(flow->sf_acts);
1031 rcu_assign_pointer(flow->sf_acts, acts);
1032
1033 if (unlikely(reply)) {
1034 error = ovs_flow_cmd_fill_info(flow,
1035 ovs_header->dp_ifindex,
1036 reply, info->snd_portid,
1037 info->snd_seq, 0,
1038 OVS_FLOW_CMD_NEW,
1039 ufid_flags);
1040 BUG_ON(error < 0);
1041 }
1042 ovs_unlock();
1043
1044 ovs_nla_free_flow_actions_rcu(old_acts);
1045 ovs_flow_free(new_flow, false);
1046 }
1047
1048 if (reply)
1049 ovs_notify(&dp_flow_genl_family, reply, info);
1050 return 0;
1051
1052err_unlock_ovs:
1053 ovs_unlock();
1054 kfree_skb(reply);
1055err_kfree_acts:
1056 ovs_nla_free_flow_actions(acts);
1057err_kfree_flow:
1058 ovs_flow_free(new_flow, false);
1059error:
1060 return error;
1061}
1062
1063/* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1064static struct sw_flow_actions *get_flow_actions(struct net *net,
1065 const struct nlattr *a,
1066 const struct sw_flow_key *key,
1067 const struct sw_flow_mask *mask,
1068 bool log)
1069{
1070 struct sw_flow_actions *acts;
1071 struct sw_flow_key masked_key;
1072 int error;
1073
1074 ovs_flow_mask_key(&masked_key, key, true, mask);
1075 error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1076 if (error) {
1077 OVS_NLERR(log,
1078 "Actions may not be safe on all matching packets");
1079 return ERR_PTR(error);
1080 }
1081
1082 return acts;
1083}
1084
1085static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1086{
1087 struct net *net = sock_net(skb->sk);
1088 struct nlattr **a = info->attrs;
1089 struct ovs_header *ovs_header = info->userhdr;
1090 struct sw_flow_key key;
1091 struct sw_flow *flow;
1092 struct sw_flow_mask mask;
1093 struct sk_buff *reply = NULL;
1094 struct datapath *dp;
1095 struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1096 struct sw_flow_match match;
1097 struct sw_flow_id sfid;
1098 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1099 int error = 0;
1100 bool log = !a[OVS_FLOW_ATTR_PROBE];
1101 bool ufid_present;
1102
1103 ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1104 if (a[OVS_FLOW_ATTR_KEY]) {
1105 ovs_match_init(&match, &key, &mask);
1106 error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1107 a[OVS_FLOW_ATTR_MASK], log);
1108 } else if (!ufid_present) {
1109 OVS_NLERR(log,
1110 "Flow set message rejected, Key attribute missing.");
1111 error = -EINVAL;
1112 }
1113 if (error)
1114 goto error;
1115
1116 /* Validate actions. */
1117 if (a[OVS_FLOW_ATTR_ACTIONS]) {
1118 if (!a[OVS_FLOW_ATTR_KEY]) {
1119 OVS_NLERR(log,
1120 "Flow key attribute not present in set flow.");
1121 error = -EINVAL;
1122 goto error;
1123 }
1124
1125 acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], &key,
1126 &mask, log);
1127 if (IS_ERR(acts)) {
1128 error = PTR_ERR(acts);
1129 goto error;
1130 }
1131
1132 /* Can allocate before locking if have acts. */
1133 reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1134 ufid_flags);
1135 if (IS_ERR(reply)) {
1136 error = PTR_ERR(reply);
1137 goto err_kfree_acts;
1138 }
1139 }
1140
1141 ovs_lock();
1142 dp = get_dp(net, ovs_header->dp_ifindex);
1143 if (unlikely(!dp)) {
1144 error = -ENODEV;
1145 goto err_unlock_ovs;
1146 }
1147 /* Check that the flow exists. */
1148 if (ufid_present)
1149 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1150 else
1151 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1152 if (unlikely(!flow)) {
1153 error = -ENOENT;
1154 goto err_unlock_ovs;
1155 }
1156
1157 /* Update actions, if present. */
1158 if (likely(acts)) {
1159 old_acts = ovsl_dereference(flow->sf_acts);
1160 rcu_assign_pointer(flow->sf_acts, acts);
1161
1162 if (unlikely(reply)) {
1163 error = ovs_flow_cmd_fill_info(flow,
1164 ovs_header->dp_ifindex,
1165 reply, info->snd_portid,
1166 info->snd_seq, 0,
1167 OVS_FLOW_CMD_NEW,
1168 ufid_flags);
1169 BUG_ON(error < 0);
1170 }
1171 } else {
1172 /* Could not alloc without acts before locking. */
1173 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1174 info, OVS_FLOW_CMD_NEW, false,
1175 ufid_flags);
1176
1177 if (IS_ERR(reply)) {
1178 error = PTR_ERR(reply);
1179 goto err_unlock_ovs;
1180 }
1181 }
1182
1183 /* Clear stats. */
1184 if (a[OVS_FLOW_ATTR_CLEAR])
1185 ovs_flow_stats_clear(flow);
1186 ovs_unlock();
1187
1188 if (reply)
1189 ovs_notify(&dp_flow_genl_family, reply, info);
1190 if (old_acts)
1191 ovs_nla_free_flow_actions_rcu(old_acts);
1192
1193 return 0;
1194
1195err_unlock_ovs:
1196 ovs_unlock();
1197 kfree_skb(reply);
1198err_kfree_acts:
1199 ovs_nla_free_flow_actions(acts);
1200error:
1201 return error;
1202}
1203
1204static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1205{
1206 struct nlattr **a = info->attrs;
1207 struct ovs_header *ovs_header = info->userhdr;
1208 struct net *net = sock_net(skb->sk);
1209 struct sw_flow_key key;
1210 struct sk_buff *reply;
1211 struct sw_flow *flow;
1212 struct datapath *dp;
1213 struct sw_flow_match match;
1214 struct sw_flow_id ufid;
1215 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1216 int err = 0;
1217 bool log = !a[OVS_FLOW_ATTR_PROBE];
1218 bool ufid_present;
1219
1220 ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1221 if (a[OVS_FLOW_ATTR_KEY]) {
1222 ovs_match_init(&match, &key, NULL);
1223 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1224 log);
1225 } else if (!ufid_present) {
1226 OVS_NLERR(log,
1227 "Flow get message rejected, Key attribute missing.");
1228 err = -EINVAL;
1229 }
1230 if (err)
1231 return err;
1232
1233 ovs_lock();
1234 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1235 if (!dp) {
1236 err = -ENODEV;
1237 goto unlock;
1238 }
1239
1240 if (ufid_present)
1241 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1242 else
1243 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1244 if (!flow) {
1245 err = -ENOENT;
1246 goto unlock;
1247 }
1248
1249 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1250 OVS_FLOW_CMD_NEW, true, ufid_flags);
1251 if (IS_ERR(reply)) {
1252 err = PTR_ERR(reply);
1253 goto unlock;
1254 }
1255
1256 ovs_unlock();
1257 return genlmsg_reply(reply, info);
1258unlock:
1259 ovs_unlock();
1260 return err;
1261}
1262
1263static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1264{
1265 struct nlattr **a = info->attrs;
1266 struct ovs_header *ovs_header = info->userhdr;
1267 struct net *net = sock_net(skb->sk);
1268 struct sw_flow_key key;
1269 struct sk_buff *reply;
1270 struct sw_flow *flow = NULL;
1271 struct datapath *dp;
1272 struct sw_flow_match match;
1273 struct sw_flow_id ufid;
1274 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1275 int err;
1276 bool log = !a[OVS_FLOW_ATTR_PROBE];
1277 bool ufid_present;
1278
1279 ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1280 if (a[OVS_FLOW_ATTR_KEY]) {
1281 ovs_match_init(&match, &key, NULL);
1282 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1283 NULL, log);
1284 if (unlikely(err))
1285 return err;
1286 }
1287
1288 ovs_lock();
1289 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1290 if (unlikely(!dp)) {
1291 err = -ENODEV;
1292 goto unlock;
1293 }
1294
1295 if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1296 err = ovs_flow_tbl_flush(&dp->table);
1297 goto unlock;
1298 }
1299
1300 if (ufid_present)
1301 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1302 else
1303 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1304 if (unlikely(!flow)) {
1305 err = -ENOENT;
1306 goto unlock;
1307 }
1308
1309 ovs_flow_tbl_remove(&dp->table, flow);
1310 ovs_unlock();
1311
1312 reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1313 &flow->id, info, false, ufid_flags);
1314 if (likely(reply)) {
1315 if (likely(!IS_ERR(reply))) {
1316 rcu_read_lock(); /*To keep RCU checker happy. */
1317 err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1318 reply, info->snd_portid,
1319 info->snd_seq, 0,
1320 OVS_FLOW_CMD_DEL,
1321 ufid_flags);
1322 rcu_read_unlock();
1323 BUG_ON(err < 0);
1324
1325 ovs_notify(&dp_flow_genl_family, reply, info);
1326 } else {
1327 netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
1328 }
1329 }
1330
1331 ovs_flow_free(flow, true);
1332 return 0;
1333unlock:
1334 ovs_unlock();
1335 return err;
1336}
1337
1338static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1339{
1340 struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1341 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1342 struct table_instance *ti;
1343 struct datapath *dp;
1344 u32 ufid_flags;
1345 int err;
1346
1347 err = genlmsg_parse(cb->nlh, &dp_flow_genl_family, a,
1348 OVS_FLOW_ATTR_MAX, flow_policy);
1349 if (err)
1350 return err;
1351 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1352
1353 rcu_read_lock();
1354 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1355 if (!dp) {
1356 rcu_read_unlock();
1357 return -ENODEV;
1358 }
1359
1360 ti = rcu_dereference(dp->table.ti);
1361 for (;;) {
1362 struct sw_flow *flow;
1363 u32 bucket, obj;
1364
1365 bucket = cb->args[0];
1366 obj = cb->args[1];
1367 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1368 if (!flow)
1369 break;
1370
1371 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1372 NETLINK_CB(cb->skb).portid,
1373 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1374 OVS_FLOW_CMD_NEW, ufid_flags) < 0)
1375 break;
1376
1377 cb->args[0] = bucket;
1378 cb->args[1] = obj;
1379 }
1380 rcu_read_unlock();
1381 return skb->len;
1382}
1383
1384static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1385 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1386 [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1387 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1388 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1389 [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1390 [OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1391 [OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1392};
1393
1394static const struct genl_ops dp_flow_genl_ops[] = {
1395 { .cmd = OVS_FLOW_CMD_NEW,
1396 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1397 .policy = flow_policy,
1398 .doit = ovs_flow_cmd_new
1399 },
1400 { .cmd = OVS_FLOW_CMD_DEL,
1401 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1402 .policy = flow_policy,
1403 .doit = ovs_flow_cmd_del
1404 },
1405 { .cmd = OVS_FLOW_CMD_GET,
1406 .flags = 0, /* OK for unprivileged users. */
1407 .policy = flow_policy,
1408 .doit = ovs_flow_cmd_get,
1409 .dumpit = ovs_flow_cmd_dump
1410 },
1411 { .cmd = OVS_FLOW_CMD_SET,
1412 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1413 .policy = flow_policy,
1414 .doit = ovs_flow_cmd_set,
1415 },
1416};
1417
1418static struct genl_family dp_flow_genl_family = {
1419 .id = GENL_ID_GENERATE,
1420 .hdrsize = sizeof(struct ovs_header),
1421 .name = OVS_FLOW_FAMILY,
1422 .version = OVS_FLOW_VERSION,
1423 .maxattr = OVS_FLOW_ATTR_MAX,
1424 .netnsok = true,
1425 .parallel_ops = true,
1426 .ops = dp_flow_genl_ops,
1427 .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1428 .mcgrps = &ovs_dp_flow_multicast_group,
1429 .n_mcgrps = 1,
1430};
1431
1432static size_t ovs_dp_cmd_msg_size(void)
1433{
1434 size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1435
1436 msgsize += nla_total_size(IFNAMSIZ);
1437 msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1438 msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1439 msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1440
1441 return msgsize;
1442}
1443
1444/* Called with ovs_mutex. */
1445static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1446 u32 portid, u32 seq, u32 flags, u8 cmd)
1447{
1448 struct ovs_header *ovs_header;
1449 struct ovs_dp_stats dp_stats;
1450 struct ovs_dp_megaflow_stats dp_megaflow_stats;
1451 int err;
1452
1453 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1454 flags, cmd);
1455 if (!ovs_header)
1456 goto error;
1457
1458 ovs_header->dp_ifindex = get_dpifindex(dp);
1459
1460 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1461 if (err)
1462 goto nla_put_failure;
1463
1464 get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1465 if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1466 &dp_stats))
1467 goto nla_put_failure;
1468
1469 if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1470 sizeof(struct ovs_dp_megaflow_stats),
1471 &dp_megaflow_stats))
1472 goto nla_put_failure;
1473
1474 if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1475 goto nla_put_failure;
1476
1477 genlmsg_end(skb, ovs_header);
1478 return 0;
1479
1480nla_put_failure:
1481 genlmsg_cancel(skb, ovs_header);
1482error:
1483 return -EMSGSIZE;
1484}
1485
1486static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1487{
1488 return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1489}
1490
1491/* Called with rcu_read_lock or ovs_mutex. */
1492static struct datapath *lookup_datapath(struct net *net,
1493 const struct ovs_header *ovs_header,
1494 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1495{
1496 struct datapath *dp;
1497
1498 if (!a[OVS_DP_ATTR_NAME])
1499 dp = get_dp(net, ovs_header->dp_ifindex);
1500 else {
1501 struct vport *vport;
1502
1503 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1504 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1505 }
1506 return dp ? dp : ERR_PTR(-ENODEV);
1507}
1508
1509static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1510{
1511 struct datapath *dp;
1512
1513 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1514 if (IS_ERR(dp))
1515 return;
1516
1517 WARN(dp->user_features, "Dropping previously announced user features\n");
1518 dp->user_features = 0;
1519}
1520
1521static void ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1522{
1523 if (a[OVS_DP_ATTR_USER_FEATURES])
1524 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1525}
1526
1527static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1528{
1529 struct nlattr **a = info->attrs;
1530 struct vport_parms parms;
1531 struct sk_buff *reply;
1532 struct datapath *dp;
1533 struct vport *vport;
1534 struct ovs_net *ovs_net;
1535 int err, i;
1536
1537 err = -EINVAL;
1538 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1539 goto err;
1540
1541 reply = ovs_dp_cmd_alloc_info();
1542 if (!reply)
1543 return -ENOMEM;
1544
1545 err = -ENOMEM;
1546 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1547 if (dp == NULL)
1548 goto err_free_reply;
1549
1550 ovs_dp_set_net(dp, sock_net(skb->sk));
1551
1552 /* Allocate table. */
1553 err = ovs_flow_tbl_init(&dp->table);
1554 if (err)
1555 goto err_free_dp;
1556
1557 dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1558 if (!dp->stats_percpu) {
1559 err = -ENOMEM;
1560 goto err_destroy_table;
1561 }
1562
1563 dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1564 GFP_KERNEL);
1565 if (!dp->ports) {
1566 err = -ENOMEM;
1567 goto err_destroy_percpu;
1568 }
1569
1570 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1571 INIT_HLIST_HEAD(&dp->ports[i]);
1572
1573 /* Set up our datapath device. */
1574 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1575 parms.type = OVS_VPORT_TYPE_INTERNAL;
1576 parms.options = NULL;
1577 parms.dp = dp;
1578 parms.port_no = OVSP_LOCAL;
1579 parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1580
1581 ovs_dp_change(dp, a);
1582
1583 /* So far only local changes have been made, now need the lock. */
1584 ovs_lock();
1585
1586 vport = new_vport(&parms);
1587 if (IS_ERR(vport)) {
1588 err = PTR_ERR(vport);
1589 if (err == -EBUSY)
1590 err = -EEXIST;
1591
1592 if (err == -EEXIST) {
1593 /* An outdated user space instance that does not understand
1594 * the concept of user_features has attempted to create a new
1595 * datapath and is likely to reuse it. Drop all user features.
1596 */
1597 if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1598 ovs_dp_reset_user_features(skb, info);
1599 }
1600
1601 goto err_destroy_ports_array;
1602 }
1603
1604 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1605 info->snd_seq, 0, OVS_DP_CMD_NEW);
1606 BUG_ON(err < 0);
1607
1608 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1609 list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1610
1611 ovs_unlock();
1612
1613 ovs_notify(&dp_datapath_genl_family, reply, info);
1614 return 0;
1615
1616err_destroy_ports_array:
1617 ovs_unlock();
1618 kfree(dp->ports);
1619err_destroy_percpu:
1620 free_percpu(dp->stats_percpu);
1621err_destroy_table:
1622 ovs_flow_tbl_destroy(&dp->table);
1623err_free_dp:
1624 kfree(dp);
1625err_free_reply:
1626 kfree_skb(reply);
1627err:
1628 return err;
1629}
1630
1631/* Called with ovs_mutex. */
1632static void __dp_destroy(struct datapath *dp)
1633{
1634 int i;
1635
1636 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1637 struct vport *vport;
1638 struct hlist_node *n;
1639
1640 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1641 if (vport->port_no != OVSP_LOCAL)
1642 ovs_dp_detach_port(vport);
1643 }
1644
1645 list_del_rcu(&dp->list_node);
1646
1647 /* OVSP_LOCAL is datapath internal port. We need to make sure that
1648 * all ports in datapath are destroyed first before freeing datapath.
1649 */
1650 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1651
1652 /* RCU destroy the flow table */
1653 call_rcu(&dp->rcu, destroy_dp_rcu);
1654}
1655
1656static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1657{
1658 struct sk_buff *reply;
1659 struct datapath *dp;
1660 int err;
1661
1662 reply = ovs_dp_cmd_alloc_info();
1663 if (!reply)
1664 return -ENOMEM;
1665
1666 ovs_lock();
1667 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1668 err = PTR_ERR(dp);
1669 if (IS_ERR(dp))
1670 goto err_unlock_free;
1671
1672 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1673 info->snd_seq, 0, OVS_DP_CMD_DEL);
1674 BUG_ON(err < 0);
1675
1676 __dp_destroy(dp);
1677 ovs_unlock();
1678
1679 ovs_notify(&dp_datapath_genl_family, reply, info);
1680
1681 return 0;
1682
1683err_unlock_free:
1684 ovs_unlock();
1685 kfree_skb(reply);
1686 return err;
1687}
1688
1689static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1690{
1691 struct sk_buff *reply;
1692 struct datapath *dp;
1693 int err;
1694
1695 reply = ovs_dp_cmd_alloc_info();
1696 if (!reply)
1697 return -ENOMEM;
1698
1699 ovs_lock();
1700 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1701 err = PTR_ERR(dp);
1702 if (IS_ERR(dp))
1703 goto err_unlock_free;
1704
1705 ovs_dp_change(dp, info->attrs);
1706
1707 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1708 info->snd_seq, 0, OVS_DP_CMD_NEW);
1709 BUG_ON(err < 0);
1710
1711 ovs_unlock();
1712 ovs_notify(&dp_datapath_genl_family, reply, info);
1713
1714 return 0;
1715
1716err_unlock_free:
1717 ovs_unlock();
1718 kfree_skb(reply);
1719 return err;
1720}
1721
1722static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1723{
1724 struct sk_buff *reply;
1725 struct datapath *dp;
1726 int err;
1727
1728 reply = ovs_dp_cmd_alloc_info();
1729 if (!reply)
1730 return -ENOMEM;
1731
1732 ovs_lock();
1733 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1734 if (IS_ERR(dp)) {
1735 err = PTR_ERR(dp);
1736 goto err_unlock_free;
1737 }
1738 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1739 info->snd_seq, 0, OVS_DP_CMD_NEW);
1740 BUG_ON(err < 0);
1741 ovs_unlock();
1742
1743 return genlmsg_reply(reply, info);
1744
1745err_unlock_free:
1746 ovs_unlock();
1747 kfree_skb(reply);
1748 return err;
1749}
1750
1751static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1752{
1753 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1754 struct datapath *dp;
1755 int skip = cb->args[0];
1756 int i = 0;
1757
1758 ovs_lock();
1759 list_for_each_entry(dp, &ovs_net->dps, list_node) {
1760 if (i >= skip &&
1761 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1762 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1763 OVS_DP_CMD_NEW) < 0)
1764 break;
1765 i++;
1766 }
1767 ovs_unlock();
1768
1769 cb->args[0] = i;
1770
1771 return skb->len;
1772}
1773
1774static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1775 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1776 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1777 [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1778};
1779
1780static const struct genl_ops dp_datapath_genl_ops[] = {
1781 { .cmd = OVS_DP_CMD_NEW,
1782 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1783 .policy = datapath_policy,
1784 .doit = ovs_dp_cmd_new
1785 },
1786 { .cmd = OVS_DP_CMD_DEL,
1787 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1788 .policy = datapath_policy,
1789 .doit = ovs_dp_cmd_del
1790 },
1791 { .cmd = OVS_DP_CMD_GET,
1792 .flags = 0, /* OK for unprivileged users. */
1793 .policy = datapath_policy,
1794 .doit = ovs_dp_cmd_get,
1795 .dumpit = ovs_dp_cmd_dump
1796 },
1797 { .cmd = OVS_DP_CMD_SET,
1798 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1799 .policy = datapath_policy,
1800 .doit = ovs_dp_cmd_set,
1801 },
1802};
1803
1804static struct genl_family dp_datapath_genl_family = {
1805 .id = GENL_ID_GENERATE,
1806 .hdrsize = sizeof(struct ovs_header),
1807 .name = OVS_DATAPATH_FAMILY,
1808 .version = OVS_DATAPATH_VERSION,
1809 .maxattr = OVS_DP_ATTR_MAX,
1810 .netnsok = true,
1811 .parallel_ops = true,
1812 .ops = dp_datapath_genl_ops,
1813 .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1814 .mcgrps = &ovs_dp_datapath_multicast_group,
1815 .n_mcgrps = 1,
1816};
1817
1818/* Called with ovs_mutex or RCU read lock. */
1819static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1820 u32 portid, u32 seq, u32 flags, u8 cmd)
1821{
1822 struct ovs_header *ovs_header;
1823 struct ovs_vport_stats vport_stats;
1824 int err;
1825
1826 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1827 flags, cmd);
1828 if (!ovs_header)
1829 return -EMSGSIZE;
1830
1831 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1832
1833 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1834 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1835 nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1836 ovs_vport_name(vport)))
1837 goto nla_put_failure;
1838
1839 ovs_vport_get_stats(vport, &vport_stats);
1840 if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1841 &vport_stats))
1842 goto nla_put_failure;
1843
1844 if (ovs_vport_get_upcall_portids(vport, skb))
1845 goto nla_put_failure;
1846
1847 err = ovs_vport_get_options(vport, skb);
1848 if (err == -EMSGSIZE)
1849 goto error;
1850
1851 genlmsg_end(skb, ovs_header);
1852 return 0;
1853
1854nla_put_failure:
1855 err = -EMSGSIZE;
1856error:
1857 genlmsg_cancel(skb, ovs_header);
1858 return err;
1859}
1860
1861static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1862{
1863 return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1864}
1865
1866/* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1867struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1868 u32 seq, u8 cmd)
1869{
1870 struct sk_buff *skb;
1871 int retval;
1872
1873 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1874 if (!skb)
1875 return ERR_PTR(-ENOMEM);
1876
1877 retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1878 BUG_ON(retval < 0);
1879
1880 return skb;
1881}
1882
1883/* Called with ovs_mutex or RCU read lock. */
1884static struct vport *lookup_vport(struct net *net,
1885 const struct ovs_header *ovs_header,
1886 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1887{
1888 struct datapath *dp;
1889 struct vport *vport;
1890
1891 if (a[OVS_VPORT_ATTR_NAME]) {
1892 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1893 if (!vport)
1894 return ERR_PTR(-ENODEV);
1895 if (ovs_header->dp_ifindex &&
1896 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1897 return ERR_PTR(-ENODEV);
1898 return vport;
1899 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1900 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1901
1902 if (port_no >= DP_MAX_PORTS)
1903 return ERR_PTR(-EFBIG);
1904
1905 dp = get_dp(net, ovs_header->dp_ifindex);
1906 if (!dp)
1907 return ERR_PTR(-ENODEV);
1908
1909 vport = ovs_vport_ovsl_rcu(dp, port_no);
1910 if (!vport)
1911 return ERR_PTR(-ENODEV);
1912 return vport;
1913 } else
1914 return ERR_PTR(-EINVAL);
1915}
1916
1917/* Called with ovs_mutex */
1918static void update_headroom(struct datapath *dp)
1919{
1920 unsigned dev_headroom, max_headroom = 0;
1921 struct net_device *dev;
1922 struct vport *vport;
1923 int i;
1924
1925 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1926 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1927 dev = vport->dev;
1928 dev_headroom = netdev_get_fwd_headroom(dev);
1929 if (dev_headroom > max_headroom)
1930 max_headroom = dev_headroom;
1931 }
1932 }
1933
1934 dp->max_headroom = max_headroom;
1935 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1936 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node)
1937 netdev_set_rx_headroom(vport->dev, max_headroom);
1938}
1939
1940static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1941{
1942 struct nlattr **a = info->attrs;
1943 struct ovs_header *ovs_header = info->userhdr;
1944 struct vport_parms parms;
1945 struct sk_buff *reply;
1946 struct vport *vport;
1947 struct datapath *dp;
1948 u32 port_no;
1949 int err;
1950
1951 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1952 !a[OVS_VPORT_ATTR_UPCALL_PID])
1953 return -EINVAL;
1954
1955 port_no = a[OVS_VPORT_ATTR_PORT_NO]
1956 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
1957 if (port_no >= DP_MAX_PORTS)
1958 return -EFBIG;
1959
1960 reply = ovs_vport_cmd_alloc_info();
1961 if (!reply)
1962 return -ENOMEM;
1963
1964 ovs_lock();
1965restart:
1966 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1967 err = -ENODEV;
1968 if (!dp)
1969 goto exit_unlock_free;
1970
1971 if (port_no) {
1972 vport = ovs_vport_ovsl(dp, port_no);
1973 err = -EBUSY;
1974 if (vport)
1975 goto exit_unlock_free;
1976 } else {
1977 for (port_no = 1; ; port_no++) {
1978 if (port_no >= DP_MAX_PORTS) {
1979 err = -EFBIG;
1980 goto exit_unlock_free;
1981 }
1982 vport = ovs_vport_ovsl(dp, port_no);
1983 if (!vport)
1984 break;
1985 }
1986 }
1987
1988 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1989 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1990 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1991 parms.dp = dp;
1992 parms.port_no = port_no;
1993 parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
1994
1995 vport = new_vport(&parms);
1996 err = PTR_ERR(vport);
1997 if (IS_ERR(vport)) {
1998 if (err == -EAGAIN)
1999 goto restart;
2000 goto exit_unlock_free;
2001 }
2002
2003 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2004 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2005
2006 if (netdev_get_fwd_headroom(vport->dev) > dp->max_headroom)
2007 update_headroom(dp);
2008 else
2009 netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2010
2011 BUG_ON(err < 0);
2012 ovs_unlock();
2013
2014 ovs_notify(&dp_vport_genl_family, reply, info);
2015 return 0;
2016
2017exit_unlock_free:
2018 ovs_unlock();
2019 kfree_skb(reply);
2020 return err;
2021}
2022
2023static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2024{
2025 struct nlattr **a = info->attrs;
2026 struct sk_buff *reply;
2027 struct vport *vport;
2028 int err;
2029
2030 reply = ovs_vport_cmd_alloc_info();
2031 if (!reply)
2032 return -ENOMEM;
2033
2034 ovs_lock();
2035 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2036 err = PTR_ERR(vport);
2037 if (IS_ERR(vport))
2038 goto exit_unlock_free;
2039
2040 if (a[OVS_VPORT_ATTR_TYPE] &&
2041 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2042 err = -EINVAL;
2043 goto exit_unlock_free;
2044 }
2045
2046 if (a[OVS_VPORT_ATTR_OPTIONS]) {
2047 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2048 if (err)
2049 goto exit_unlock_free;
2050 }
2051
2052
2053 if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2054 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2055
2056 err = ovs_vport_set_upcall_portids(vport, ids);
2057 if (err)
2058 goto exit_unlock_free;
2059 }
2060
2061 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2062 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2063 BUG_ON(err < 0);
2064
2065 ovs_unlock();
2066 ovs_notify(&dp_vport_genl_family, reply, info);
2067 return 0;
2068
2069exit_unlock_free:
2070 ovs_unlock();
2071 kfree_skb(reply);
2072 return err;
2073}
2074
2075static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2076{
2077 bool must_update_headroom = false;
2078 struct nlattr **a = info->attrs;
2079 struct sk_buff *reply;
2080 struct datapath *dp;
2081 struct vport *vport;
2082 int err;
2083
2084 reply = ovs_vport_cmd_alloc_info();
2085 if (!reply)
2086 return -ENOMEM;
2087
2088 ovs_lock();
2089 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2090 err = PTR_ERR(vport);
2091 if (IS_ERR(vport))
2092 goto exit_unlock_free;
2093
2094 if (vport->port_no == OVSP_LOCAL) {
2095 err = -EINVAL;
2096 goto exit_unlock_free;
2097 }
2098
2099 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2100 info->snd_seq, 0, OVS_VPORT_CMD_DEL);
2101 BUG_ON(err < 0);
2102
2103 /* the vport deletion may trigger dp headroom update */
2104 dp = vport->dp;
2105 if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2106 must_update_headroom = true;
2107 netdev_reset_rx_headroom(vport->dev);
2108 ovs_dp_detach_port(vport);
2109
2110 if (must_update_headroom)
2111 update_headroom(dp);
2112 ovs_unlock();
2113
2114 ovs_notify(&dp_vport_genl_family, reply, info);
2115 return 0;
2116
2117exit_unlock_free:
2118 ovs_unlock();
2119 kfree_skb(reply);
2120 return err;
2121}
2122
2123static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2124{
2125 struct nlattr **a = info->attrs;
2126 struct ovs_header *ovs_header = info->userhdr;
2127 struct sk_buff *reply;
2128 struct vport *vport;
2129 int err;
2130
2131 reply = ovs_vport_cmd_alloc_info();
2132 if (!reply)
2133 return -ENOMEM;
2134
2135 rcu_read_lock();
2136 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2137 err = PTR_ERR(vport);
2138 if (IS_ERR(vport))
2139 goto exit_unlock_free;
2140 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2141 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2142 BUG_ON(err < 0);
2143 rcu_read_unlock();
2144
2145 return genlmsg_reply(reply, info);
2146
2147exit_unlock_free:
2148 rcu_read_unlock();
2149 kfree_skb(reply);
2150 return err;
2151}
2152
2153static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2154{
2155 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2156 struct datapath *dp;
2157 int bucket = cb->args[0], skip = cb->args[1];
2158 int i, j = 0;
2159
2160 rcu_read_lock();
2161 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2162 if (!dp) {
2163 rcu_read_unlock();
2164 return -ENODEV;
2165 }
2166 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2167 struct vport *vport;
2168
2169 j = 0;
2170 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2171 if (j >= skip &&
2172 ovs_vport_cmd_fill_info(vport, skb,
2173 NETLINK_CB(cb->skb).portid,
2174 cb->nlh->nlmsg_seq,
2175 NLM_F_MULTI,
2176 OVS_VPORT_CMD_NEW) < 0)
2177 goto out;
2178
2179 j++;
2180 }
2181 skip = 0;
2182 }
2183out:
2184 rcu_read_unlock();
2185
2186 cb->args[0] = i;
2187 cb->args[1] = j;
2188
2189 return skb->len;
2190}
2191
2192static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2193 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2194 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2195 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2196 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2197 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
2198 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2199};
2200
2201static const struct genl_ops dp_vport_genl_ops[] = {
2202 { .cmd = OVS_VPORT_CMD_NEW,
2203 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2204 .policy = vport_policy,
2205 .doit = ovs_vport_cmd_new
2206 },
2207 { .cmd = OVS_VPORT_CMD_DEL,
2208 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2209 .policy = vport_policy,
2210 .doit = ovs_vport_cmd_del
2211 },
2212 { .cmd = OVS_VPORT_CMD_GET,
2213 .flags = 0, /* OK for unprivileged users. */
2214 .policy = vport_policy,
2215 .doit = ovs_vport_cmd_get,
2216 .dumpit = ovs_vport_cmd_dump
2217 },
2218 { .cmd = OVS_VPORT_CMD_SET,
2219 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2220 .policy = vport_policy,
2221 .doit = ovs_vport_cmd_set,
2222 },
2223};
2224
2225struct genl_family dp_vport_genl_family = {
2226 .id = GENL_ID_GENERATE,
2227 .hdrsize = sizeof(struct ovs_header),
2228 .name = OVS_VPORT_FAMILY,
2229 .version = OVS_VPORT_VERSION,
2230 .maxattr = OVS_VPORT_ATTR_MAX,
2231 .netnsok = true,
2232 .parallel_ops = true,
2233 .ops = dp_vport_genl_ops,
2234 .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2235 .mcgrps = &ovs_dp_vport_multicast_group,
2236 .n_mcgrps = 1,
2237};
2238
2239static struct genl_family * const dp_genl_families[] = {
2240 &dp_datapath_genl_family,
2241 &dp_vport_genl_family,
2242 &dp_flow_genl_family,
2243 &dp_packet_genl_family,
2244};
2245
2246static void dp_unregister_genl(int n_families)
2247{
2248 int i;
2249
2250 for (i = 0; i < n_families; i++)
2251 genl_unregister_family(dp_genl_families[i]);
2252}
2253
2254static int dp_register_genl(void)
2255{
2256 int err;
2257 int i;
2258
2259 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2260
2261 err = genl_register_family(dp_genl_families[i]);
2262 if (err)
2263 goto error;
2264 }
2265
2266 return 0;
2267
2268error:
2269 dp_unregister_genl(i);
2270 return err;
2271}
2272
2273static int __net_init ovs_init_net(struct net *net)
2274{
2275 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2276
2277 INIT_LIST_HEAD(&ovs_net->dps);
2278 INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2279 ovs_ct_init(net);
2280 return 0;
2281}
2282
2283static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2284 struct list_head *head)
2285{
2286 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2287 struct datapath *dp;
2288
2289 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2290 int i;
2291
2292 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2293 struct vport *vport;
2294
2295 hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2296 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2297 continue;
2298
2299 if (dev_net(vport->dev) == dnet)
2300 list_add(&vport->detach_list, head);
2301 }
2302 }
2303 }
2304}
2305
2306static void __net_exit ovs_exit_net(struct net *dnet)
2307{
2308 struct datapath *dp, *dp_next;
2309 struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2310 struct vport *vport, *vport_next;
2311 struct net *net;
2312 LIST_HEAD(head);
2313
2314 ovs_ct_exit(dnet);
2315 ovs_lock();
2316 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2317 __dp_destroy(dp);
2318
2319 rtnl_lock();
2320 for_each_net(net)
2321 list_vports_from_net(net, dnet, &head);
2322 rtnl_unlock();
2323
2324 /* Detach all vports from given namespace. */
2325 list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2326 list_del(&vport->detach_list);
2327 ovs_dp_detach_port(vport);
2328 }
2329
2330 ovs_unlock();
2331
2332 cancel_work_sync(&ovs_net->dp_notify_work);
2333}
2334
2335static struct pernet_operations ovs_net_ops = {
2336 .init = ovs_init_net,
2337 .exit = ovs_exit_net,
2338 .id = &ovs_net_id,
2339 .size = sizeof(struct ovs_net),
2340};
2341
2342static int __init dp_init(void)
2343{
2344 int err;
2345
2346 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2347
2348 pr_info("Open vSwitch switching datapath\n");
2349
2350 err = action_fifos_init();
2351 if (err)
2352 goto error;
2353
2354 err = ovs_internal_dev_rtnl_link_register();
2355 if (err)
2356 goto error_action_fifos_exit;
2357
2358 err = ovs_flow_init();
2359 if (err)
2360 goto error_unreg_rtnl_link;
2361
2362 err = ovs_vport_init();
2363 if (err)
2364 goto error_flow_exit;
2365
2366 err = register_pernet_device(&ovs_net_ops);
2367 if (err)
2368 goto error_vport_exit;
2369
2370 err = register_netdevice_notifier(&ovs_dp_device_notifier);
2371 if (err)
2372 goto error_netns_exit;
2373
2374 err = ovs_netdev_init();
2375 if (err)
2376 goto error_unreg_notifier;
2377
2378 err = dp_register_genl();
2379 if (err < 0)
2380 goto error_unreg_netdev;
2381
2382 return 0;
2383
2384error_unreg_netdev:
2385 ovs_netdev_exit();
2386error_unreg_notifier:
2387 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2388error_netns_exit:
2389 unregister_pernet_device(&ovs_net_ops);
2390error_vport_exit:
2391 ovs_vport_exit();
2392error_flow_exit:
2393 ovs_flow_exit();
2394error_unreg_rtnl_link:
2395 ovs_internal_dev_rtnl_link_unregister();
2396error_action_fifos_exit:
2397 action_fifos_exit();
2398error:
2399 return err;
2400}
2401
2402static void dp_cleanup(void)
2403{
2404 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2405 ovs_netdev_exit();
2406 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2407 unregister_pernet_device(&ovs_net_ops);
2408 rcu_barrier();
2409 ovs_vport_exit();
2410 ovs_flow_exit();
2411 ovs_internal_dev_rtnl_link_unregister();
2412 action_fifos_exit();
2413}
2414
2415module_init(dp_init);
2416module_exit(dp_cleanup);
2417
2418MODULE_DESCRIPTION("Open vSwitch switching datapath");
2419MODULE_LICENSE("GPL");