Setting tag sliver-openvswitch-2.1.90-1
[sliver-openvswitch.git] / datapath / datapath.c
1 /*
2  * Copyright (c) 2007-2013 Nicira, Inc.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of version 2 of the GNU General Public
6  * License as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful, but
9  * WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public License
14  * along with this program; if not, write to the Free Software
15  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
16  * 02110-1301, USA
17  */
18
19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
20
21 #include <linux/init.h>
22 #include <linux/module.h>
23 #include <linux/if_arp.h>
24 #include <linux/if_vlan.h>
25 #include <linux/in.h>
26 #include <linux/ip.h>
27 #include <linux/jhash.h>
28 #include <linux/delay.h>
29 #include <linux/time.h>
30 #include <linux/etherdevice.h>
31 #include <linux/genetlink.h>
32 #include <linux/kernel.h>
33 #include <linux/kthread.h>
34 #include <linux/mutex.h>
35 #include <linux/percpu.h>
36 #include <linux/rcupdate.h>
37 #include <linux/tcp.h>
38 #include <linux/udp.h>
39 #include <linux/version.h>
40 #include <linux/ethtool.h>
41 #include <linux/wait.h>
42 #include <asm/div64.h>
43 #include <linux/highmem.h>
44 #include <linux/netfilter_bridge.h>
45 #include <linux/netfilter_ipv4.h>
46 #include <linux/inetdevice.h>
47 #include <linux/list.h>
48 #include <linux/openvswitch.h>
49 #include <linux/rculist.h>
50 #include <linux/dmi.h>
51 #include <linux/genetlink.h>
52 #include <net/genetlink.h>
53 #include <net/genetlink.h>
54 #include <net/net_namespace.h>
55 #include <net/netns/generic.h>
56
57 #include "datapath.h"
58 #include "flow.h"
59 #include "flow_table.h"
60 #include "flow_netlink.h"
61 #include "vlan.h"
62 #include "vport-internal_dev.h"
63 #include "vport-netdev.h"
64
65 int ovs_net_id __read_mostly;
66
67 static void ovs_notify(struct sk_buff *skb, struct genl_info *info,
68                        struct genl_multicast_group *grp)
69 {
70         genl_notify(skb, genl_info_net(info), info->snd_portid,
71                     grp->id, info->nlhdr, GFP_KERNEL);
72 }
73
74 /**
75  * DOC: Locking:
76  *
77  * All writes e.g. Writes to device state (add/remove datapath, port, set
78  * operations on vports, etc.), Writes to other state (flow table
79  * modifications, set miscellaneous datapath parameters, etc.) are protected
80  * by ovs_lock.
81  *
82  * Reads are protected by RCU.
83  *
84  * There are a few special cases (mostly stats) that have their own
85  * synchronization but they nest under all of above and don't interact with
86  * each other.
87  *
88  * The RTNL lock nests inside ovs_mutex.
89  */
90
91 static DEFINE_MUTEX(ovs_mutex);
92
93 void ovs_lock(void)
94 {
95         mutex_lock(&ovs_mutex);
96 }
97
98 void ovs_unlock(void)
99 {
100         mutex_unlock(&ovs_mutex);
101 }
102
103 #ifdef CONFIG_LOCKDEP
104 int lockdep_ovsl_is_held(void)
105 {
106         if (debug_locks)
107                 return lockdep_is_held(&ovs_mutex);
108         else
109                 return 1;
110 }
111 #endif
112
113 static struct vport *new_vport(const struct vport_parms *);
114 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
115                              const struct dp_upcall_info *);
116 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
117                                   const struct dp_upcall_info *);
118
119 /* Must be called with rcu_read_lock or ovs_mutex. */
120 static struct datapath *get_dp(struct net *net, int dp_ifindex)
121 {
122         struct datapath *dp = NULL;
123         struct net_device *dev;
124
125         rcu_read_lock();
126         dev = dev_get_by_index_rcu(net, dp_ifindex);
127         if (dev) {
128                 struct vport *vport = ovs_internal_dev_get_vport(dev);
129                 if (vport)
130                         dp = vport->dp;
131         }
132         rcu_read_unlock();
133
134         return dp;
135 }
136
137 /* Must be called with rcu_read_lock or ovs_mutex. */
138 const char *ovs_dp_name(const struct datapath *dp)
139 {
140         struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
141         return vport->ops->get_name(vport);
142 }
143
144 static int get_dpifindex(struct datapath *dp)
145 {
146         struct vport *local;
147         int ifindex;
148
149         rcu_read_lock();
150
151         local = ovs_vport_rcu(dp, OVSP_LOCAL);
152         if (local)
153                 ifindex = netdev_vport_priv(local)->dev->ifindex;
154         else
155                 ifindex = 0;
156
157         rcu_read_unlock();
158
159         return ifindex;
160 }
161
162 static void destroy_dp_rcu(struct rcu_head *rcu)
163 {
164         struct datapath *dp = container_of(rcu, struct datapath, rcu);
165
166         free_percpu(dp->stats_percpu);
167         release_net(ovs_dp_get_net(dp));
168         kfree(dp->ports);
169         kfree(dp);
170 }
171
172 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
173                                             u16 port_no)
174 {
175         return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
176 }
177
178 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
179 {
180         struct vport *vport;
181         struct hlist_head *head;
182
183         head = vport_hash_bucket(dp, port_no);
184         hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
185                 if (vport->port_no == port_no)
186                         return vport;
187         }
188         return NULL;
189 }
190
191 /* Called with ovs_mutex. */
192 static struct vport *new_vport(const struct vport_parms *parms)
193 {
194         struct vport *vport;
195
196         vport = ovs_vport_add(parms);
197         if (!IS_ERR(vport)) {
198                 struct datapath *dp = parms->dp;
199                 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
200
201                 hlist_add_head_rcu(&vport->dp_hash_node, head);
202         }
203         return vport;
204 }
205
206 void ovs_dp_detach_port(struct vport *p)
207 {
208         ASSERT_OVSL();
209
210         /* First drop references to device. */
211         hlist_del_rcu(&p->dp_hash_node);
212
213         /* Then destroy it. */
214         ovs_vport_del(p);
215 }
216
217 /* Must be called with rcu_read_lock. */
218 void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
219 {
220         struct datapath *dp = p->dp;
221         struct sw_flow *flow;
222         struct dp_stats_percpu *stats;
223         struct sw_flow_key key;
224         u64 *stats_counter;
225         u32 n_mask_hit;
226         int error;
227
228         stats = this_cpu_ptr(dp->stats_percpu);
229
230         /* Extract flow from 'skb' into 'key'. */
231         error = ovs_flow_extract(skb, p->port_no, &key);
232         if (unlikely(error)) {
233                 kfree_skb(skb);
234                 return;
235         }
236
237         /* Look up flow. */
238         flow = ovs_flow_tbl_lookup_stats(&dp->table, &key, &n_mask_hit);
239         if (unlikely(!flow)) {
240                 struct dp_upcall_info upcall;
241
242                 upcall.cmd = OVS_PACKET_CMD_MISS;
243                 upcall.key = &key;
244                 upcall.userdata = NULL;
245                 upcall.portid = p->upcall_portid;
246                 ovs_dp_upcall(dp, skb, &upcall);
247                 consume_skb(skb);
248                 stats_counter = &stats->n_missed;
249                 goto out;
250         }
251
252         OVS_CB(skb)->flow = flow;
253         OVS_CB(skb)->pkt_key = &key;
254
255         ovs_flow_stats_update(OVS_CB(skb)->flow, skb);
256         ovs_execute_actions(dp, skb);
257         stats_counter = &stats->n_hit;
258
259 out:
260         /* Update datapath statistics. */
261         u64_stats_update_begin(&stats->sync);
262         (*stats_counter)++;
263         stats->n_mask_hit += n_mask_hit;
264         u64_stats_update_end(&stats->sync);
265 }
266
267 static struct genl_family dp_packet_genl_family = {
268         .id = GENL_ID_GENERATE,
269         .hdrsize = sizeof(struct ovs_header),
270         .name = OVS_PACKET_FAMILY,
271         .version = OVS_PACKET_VERSION,
272         .maxattr = OVS_PACKET_ATTR_MAX,
273         .netnsok = true,
274          SET_PARALLEL_OPS
275 };
276
277 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
278                   const struct dp_upcall_info *upcall_info)
279 {
280         struct dp_stats_percpu *stats;
281         int err;
282
283         if (upcall_info->portid == 0) {
284                 err = -ENOTCONN;
285                 goto err;
286         }
287
288         if (!skb_is_gso(skb))
289                 err = queue_userspace_packet(dp, skb, upcall_info);
290         else
291                 err = queue_gso_packets(dp, skb, upcall_info);
292         if (err)
293                 goto err;
294
295         return 0;
296
297 err:
298         stats = this_cpu_ptr(dp->stats_percpu);
299
300         u64_stats_update_begin(&stats->sync);
301         stats->n_lost++;
302         u64_stats_update_end(&stats->sync);
303
304         return err;
305 }
306
307 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
308                              const struct dp_upcall_info *upcall_info)
309 {
310         unsigned short gso_type = skb_shinfo(skb)->gso_type;
311         struct dp_upcall_info later_info;
312         struct sw_flow_key later_key;
313         struct sk_buff *segs, *nskb;
314         int err;
315
316         segs = __skb_gso_segment(skb, NETIF_F_SG, false);
317         if (IS_ERR(segs))
318                 return PTR_ERR(segs);
319
320         /* Queue all of the segments. */
321         skb = segs;
322         do {
323                 err = queue_userspace_packet(dp, skb, upcall_info);
324                 if (err)
325                         break;
326
327                 if (skb == segs && gso_type & SKB_GSO_UDP) {
328                         /* The initial flow key extracted by ovs_flow_extract()
329                          * in this case is for a first fragment, so we need to
330                          * properly mark later fragments.
331                          */
332                         later_key = *upcall_info->key;
333                         later_key.ip.frag = OVS_FRAG_TYPE_LATER;
334
335                         later_info = *upcall_info;
336                         later_info.key = &later_key;
337                         upcall_info = &later_info;
338                 }
339         } while ((skb = skb->next));
340
341         /* Free all of the segments. */
342         skb = segs;
343         do {
344                 nskb = skb->next;
345                 if (err)
346                         kfree_skb(skb);
347                 else
348                         consume_skb(skb);
349         } while ((skb = nskb));
350         return err;
351 }
352
353 static size_t key_attr_size(void)
354 {
355         return    nla_total_size(4)   /* OVS_KEY_ATTR_PRIORITY */
356                 + nla_total_size(0)   /* OVS_KEY_ATTR_TUNNEL */
357                   + nla_total_size(8)   /* OVS_TUNNEL_KEY_ATTR_ID */
358                   + nla_total_size(4)   /* OVS_TUNNEL_KEY_ATTR_IPV4_SRC */
359                   + nla_total_size(4)   /* OVS_TUNNEL_KEY_ATTR_IPV4_DST */
360                   + nla_total_size(1)   /* OVS_TUNNEL_KEY_ATTR_TOS */
361                   + nla_total_size(1)   /* OVS_TUNNEL_KEY_ATTR_TTL */
362                   + nla_total_size(0)   /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
363                   + nla_total_size(0)   /* OVS_TUNNEL_KEY_ATTR_CSUM */
364                 + nla_total_size(4)   /* OVS_KEY_ATTR_IN_PORT */
365                 + nla_total_size(4)   /* OVS_KEY_ATTR_SKB_MARK */
366                 + nla_total_size(12)  /* OVS_KEY_ATTR_ETHERNET */
367                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
368                 + nla_total_size(4)   /* OVS_KEY_ATTR_8021Q */
369                 + nla_total_size(0)   /* OVS_KEY_ATTR_ENCAP */
370                 + nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
371                 + nla_total_size(40)  /* OVS_KEY_ATTR_IPV6 */
372                 + nla_total_size(2)   /* OVS_KEY_ATTR_ICMPV6 */
373                 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
374 }
375
376 static size_t upcall_msg_size(const struct nlattr *userdata,
377                               unsigned int hdrlen)
378 {
379         size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
380                 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
381                 + nla_total_size(key_attr_size()); /* OVS_PACKET_ATTR_KEY */
382
383         /* OVS_PACKET_ATTR_USERDATA */
384         if (userdata)
385                 size += NLA_ALIGN(userdata->nla_len);
386
387         return size;
388 }
389
390 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
391                                   const struct dp_upcall_info *upcall_info)
392 {
393         struct ovs_header *upcall;
394         struct sk_buff *nskb = NULL;
395         struct sk_buff *user_skb; /* to be queued to userspace */
396         struct nlattr *nla;
397         struct genl_info info = {
398 #if LINUX_VERSION_CODE >= KERNEL_VERSION(3,14,0)
399                 .dst_sk = ovs_dp_get_net(dp)->genl_sock,
400 #endif
401                 .snd_portid = upcall_info->portid,
402         };
403         size_t len;
404         unsigned int hlen;
405         int err, dp_ifindex;
406
407         dp_ifindex = get_dpifindex(dp);
408         if (!dp_ifindex)
409                 return -ENODEV;
410
411         if (vlan_tx_tag_present(skb)) {
412                 nskb = skb_clone(skb, GFP_ATOMIC);
413                 if (!nskb)
414                         return -ENOMEM;
415
416                 nskb = __vlan_put_tag(nskb, nskb->vlan_proto, vlan_tx_tag_get(nskb));
417                 if (!nskb)
418                         return -ENOMEM;
419
420                 vlan_set_tci(nskb, 0);
421
422                 skb = nskb;
423         }
424
425         if (nla_attr_size(skb->len) > USHRT_MAX) {
426                 err = -EFBIG;
427                 goto out;
428         }
429
430         /* Complete checksum if needed */
431         if (skb->ip_summed == CHECKSUM_PARTIAL &&
432             (err = skb_checksum_help(skb)))
433                 goto out;
434
435         /* Older versions of OVS user space enforce alignment of the last
436          * Netlink attribute to NLA_ALIGNTO which would require extensive
437          * padding logic. Only perform zerocopy if padding is not required.
438          */
439         if (dp->user_features & OVS_DP_F_UNALIGNED)
440                 hlen = skb_zerocopy_headlen(skb);
441         else
442                 hlen = skb->len;
443
444         len = upcall_msg_size(upcall_info->userdata, hlen);
445         user_skb = genlmsg_new_unicast(len, &info, GFP_ATOMIC);
446         if (!user_skb) {
447                 err = -ENOMEM;
448                 goto out;
449         }
450
451         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
452                              0, upcall_info->cmd);
453         upcall->dp_ifindex = dp_ifindex;
454
455         nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
456         ovs_nla_put_flow(upcall_info->key, upcall_info->key, user_skb);
457         nla_nest_end(user_skb, nla);
458
459         if (upcall_info->userdata)
460                 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
461                           nla_len(upcall_info->userdata),
462                           nla_data(upcall_info->userdata));
463
464         /* Only reserve room for attribute header, packet data is added
465          * in skb_zerocopy() */
466         if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
467                 err = -ENOBUFS;
468                 goto out;
469         }
470         nla->nla_len = nla_attr_size(skb->len);
471
472         skb_zerocopy(user_skb, skb, skb->len, hlen);
473
474         /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
475         if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
476                 size_t plen = NLA_ALIGN(user_skb->len) - user_skb->len;
477
478                 if (plen > 0)
479                         memset(skb_put(user_skb, plen), 0, plen);
480         }
481
482         ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
483
484         err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
485 out:
486         kfree_skb(nskb);
487         return err;
488 }
489
490 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
491 {
492         struct ovs_header *ovs_header = info->userhdr;
493         struct nlattr **a = info->attrs;
494         struct sw_flow_actions *acts;
495         struct sk_buff *packet;
496         struct sw_flow *flow;
497         struct datapath *dp;
498         struct ethhdr *eth;
499         int len;
500         int err;
501
502         err = -EINVAL;
503         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
504             !a[OVS_PACKET_ATTR_ACTIONS])
505                 goto err;
506
507         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
508         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
509         err = -ENOMEM;
510         if (!packet)
511                 goto err;
512         skb_reserve(packet, NET_IP_ALIGN);
513
514         nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
515
516         skb_reset_mac_header(packet);
517         eth = eth_hdr(packet);
518
519         /* Normally, setting the skb 'protocol' field would be handled by a
520          * call to eth_type_trans(), but it assumes there's a sending
521          * device, which we may not have. */
522         if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
523                 packet->protocol = eth->h_proto;
524         else
525                 packet->protocol = htons(ETH_P_802_2);
526
527         /* Build an sw_flow for sending this packet. */
528         flow = ovs_flow_alloc();
529         err = PTR_ERR(flow);
530         if (IS_ERR(flow))
531                 goto err_kfree_skb;
532
533         err = ovs_flow_extract(packet, -1, &flow->key);
534         if (err)
535                 goto err_flow_free;
536
537         err = ovs_nla_get_flow_metadata(flow, a[OVS_PACKET_ATTR_KEY]);
538         if (err)
539                 goto err_flow_free;
540         acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_PACKET_ATTR_ACTIONS]));
541         err = PTR_ERR(acts);
542         if (IS_ERR(acts))
543                 goto err_flow_free;
544
545         err = ovs_nla_copy_actions(a[OVS_PACKET_ATTR_ACTIONS],
546                                    &flow->key, 0, &acts);
547         rcu_assign_pointer(flow->sf_acts, acts);
548         if (err)
549                 goto err_flow_free;
550
551         OVS_CB(packet)->flow = flow;
552         OVS_CB(packet)->pkt_key = &flow->key;
553         packet->priority = flow->key.phy.priority;
554         packet->mark = flow->key.phy.skb_mark;
555
556         rcu_read_lock();
557         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
558         err = -ENODEV;
559         if (!dp)
560                 goto err_unlock;
561
562         local_bh_disable();
563         err = ovs_execute_actions(dp, packet);
564         local_bh_enable();
565         rcu_read_unlock();
566
567         ovs_flow_free(flow, false);
568         return err;
569
570 err_unlock:
571         rcu_read_unlock();
572 err_flow_free:
573         ovs_flow_free(flow, false);
574 err_kfree_skb:
575         kfree_skb(packet);
576 err:
577         return err;
578 }
579
580 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
581         [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
582         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
583         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
584 };
585
586 static struct genl_ops dp_packet_genl_ops[] = {
587         { .cmd = OVS_PACKET_CMD_EXECUTE,
588           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
589           .policy = packet_policy,
590           .doit = ovs_packet_cmd_execute
591         }
592 };
593
594 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats,
595                          struct ovs_dp_megaflow_stats *mega_stats)
596 {
597         int i;
598
599         memset(mega_stats, 0, sizeof(*mega_stats));
600
601         stats->n_flows = ovs_flow_tbl_count(&dp->table);
602         mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
603
604         stats->n_hit = stats->n_missed = stats->n_lost = 0;
605
606         for_each_possible_cpu(i) {
607                 const struct dp_stats_percpu *percpu_stats;
608                 struct dp_stats_percpu local_stats;
609                 unsigned int start;
610
611                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
612
613                 do {
614                         start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
615                         local_stats = *percpu_stats;
616                 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
617
618                 stats->n_hit += local_stats.n_hit;
619                 stats->n_missed += local_stats.n_missed;
620                 stats->n_lost += local_stats.n_lost;
621                 mega_stats->n_mask_hit += local_stats.n_mask_hit;
622         }
623 }
624
625 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
626         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
627         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
628         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
629 };
630
631 static struct genl_family dp_flow_genl_family = {
632         .id = GENL_ID_GENERATE,
633         .hdrsize = sizeof(struct ovs_header),
634         .name = OVS_FLOW_FAMILY,
635         .version = OVS_FLOW_VERSION,
636         .maxattr = OVS_FLOW_ATTR_MAX,
637         .netnsok = true,
638          SET_PARALLEL_OPS
639 };
640
641 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
642         .name = OVS_FLOW_MCGROUP
643 };
644
645 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
646 {
647         return NLMSG_ALIGN(sizeof(struct ovs_header))
648                 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
649                 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_MASK */
650                 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
651                 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
652                 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
653                 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
654 }
655
656 /* Called with ovs_mutex. */
657 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
658                                   struct sk_buff *skb, u32 portid,
659                                   u32 seq, u32 flags, u8 cmd)
660 {
661         const int skb_orig_len = skb->len;
662         struct nlattr *start;
663         struct ovs_flow_stats stats;
664         __be16 tcp_flags;
665         unsigned long used;
666         struct ovs_header *ovs_header;
667         struct nlattr *nla;
668         int err;
669
670         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
671         if (!ovs_header)
672                 return -EMSGSIZE;
673
674         ovs_header->dp_ifindex = get_dpifindex(dp);
675
676         /* Fill flow key. */
677         nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
678         if (!nla)
679                 goto nla_put_failure;
680
681         err = ovs_nla_put_flow(&flow->unmasked_key, &flow->unmasked_key, skb);
682         if (err)
683                 goto error;
684         nla_nest_end(skb, nla);
685
686         nla = nla_nest_start(skb, OVS_FLOW_ATTR_MASK);
687         if (!nla)
688                 goto nla_put_failure;
689
690         err = ovs_nla_put_flow(&flow->key, &flow->mask->key, skb);
691         if (err)
692                 goto error;
693
694         nla_nest_end(skb, nla);
695
696         ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
697         if (used &&
698             nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
699                 goto nla_put_failure;
700
701         if (stats.n_packets &&
702             nla_put(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats))
703                 goto nla_put_failure;
704
705         if ((u8)ntohs(tcp_flags) &&
706              nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
707                 goto nla_put_failure;
708
709         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
710          * this is the first flow to be dumped into 'skb'.  This is unusual for
711          * Netlink but individual action lists can be longer than
712          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
713          * The userspace caller can always fetch the actions separately if it
714          * really wants them.  (Most userspace callers in fact don't care.)
715          *
716          * This can only fail for dump operations because the skb is always
717          * properly sized for single flows.
718          */
719         start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
720         if (start) {
721                 const struct sw_flow_actions *sf_acts;
722
723                 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
724
725                 err = ovs_nla_put_actions(sf_acts->actions,
726                                           sf_acts->actions_len, skb);
727                 if (!err)
728                         nla_nest_end(skb, start);
729                 else {
730                         if (skb_orig_len)
731                                 goto error;
732
733                         nla_nest_cancel(skb, start);
734                 }
735         } else if (skb_orig_len)
736                 goto nla_put_failure;
737
738         return genlmsg_end(skb, ovs_header);
739
740 nla_put_failure:
741         err = -EMSGSIZE;
742 error:
743         genlmsg_cancel(skb, ovs_header);
744         return err;
745 }
746
747 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow,
748                                                struct genl_info *info)
749 {
750         size_t len;
751
752         len = ovs_flow_cmd_msg_size(ovsl_dereference(flow->sf_acts));
753
754         return genlmsg_new_unicast(len, info, GFP_KERNEL);
755 }
756
757 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
758                                                struct datapath *dp,
759                                                struct genl_info *info,
760                                                u8 cmd)
761 {
762         struct sk_buff *skb;
763         int retval;
764
765         skb = ovs_flow_cmd_alloc_info(flow, info);
766         if (!skb)
767                 return ERR_PTR(-ENOMEM);
768
769         retval = ovs_flow_cmd_fill_info(flow, dp, skb, info->snd_portid,
770                                         info->snd_seq, 0, cmd);
771         BUG_ON(retval < 0);
772         return skb;
773 }
774
775 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
776 {
777         struct nlattr **a = info->attrs;
778         struct ovs_header *ovs_header = info->userhdr;
779         struct sw_flow_key key, masked_key;
780         struct sw_flow *flow = NULL;
781         struct sw_flow_mask mask;
782         struct sk_buff *reply;
783         struct datapath *dp;
784         struct sw_flow_actions *acts = NULL;
785         struct sw_flow_match match;
786         int error;
787
788         /* Extract key. */
789         error = -EINVAL;
790         if (!a[OVS_FLOW_ATTR_KEY])
791                 goto error;
792
793         ovs_match_init(&match, &key, &mask);
794         error = ovs_nla_get_match(&match,
795                                   a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
796         if (error)
797                 goto error;
798
799         /* Validate actions. */
800         if (a[OVS_FLOW_ATTR_ACTIONS]) {
801                 acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
802                 error = PTR_ERR(acts);
803                 if (IS_ERR(acts))
804                         goto error;
805
806                 ovs_flow_mask_key(&masked_key, &key, &mask);
807                 error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS],
808                                              &masked_key, 0, &acts);
809                 if (error) {
810                         OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
811                         goto err_kfree;
812                 }
813         } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
814                 error = -EINVAL;
815                 goto error;
816         }
817
818         ovs_lock();
819         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
820         error = -ENODEV;
821         if (!dp)
822                 goto err_unlock_ovs;
823
824         /* Check if this is a duplicate flow */
825         flow = ovs_flow_tbl_lookup(&dp->table, &key);
826         if (!flow) {
827                 /* Bail out if we're not allowed to create a new flow. */
828                 error = -ENOENT;
829                 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
830                         goto err_unlock_ovs;
831
832                 /* Allocate flow. */
833                 flow = ovs_flow_alloc();
834                 if (IS_ERR(flow)) {
835                         error = PTR_ERR(flow);
836                         goto err_unlock_ovs;
837                 }
838
839                 flow->key = masked_key;
840                 flow->unmasked_key = key;
841                 rcu_assign_pointer(flow->sf_acts, acts);
842
843                 /* Put flow in bucket. */
844                 error = ovs_flow_tbl_insert(&dp->table, flow, &mask);
845                 if (error) {
846                         acts = NULL;
847                         goto err_flow_free;
848                 }
849
850                 reply = ovs_flow_cmd_build_info(flow, dp, info, OVS_FLOW_CMD_NEW);
851         } else {
852                 /* We found a matching flow. */
853                 struct sw_flow_actions *old_acts;
854
855                 /* Bail out if we're not allowed to modify an existing flow.
856                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
857                  * because Generic Netlink treats the latter as a dump
858                  * request.  We also accept NLM_F_EXCL in case that bug ever
859                  * gets fixed.
860                  */
861                 error = -EEXIST;
862                 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
863                     info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
864                         goto err_unlock_ovs;
865
866                 /* The unmasked key has to be the same for flow updates. */
867                 if (!ovs_flow_cmp_unmasked_key(flow, &match))
868                         goto err_unlock_ovs;
869
870                 /* Update actions. */
871                 old_acts = ovsl_dereference(flow->sf_acts);
872                 rcu_assign_pointer(flow->sf_acts, acts);
873                 ovs_nla_free_flow_actions(old_acts);
874
875                 reply = ovs_flow_cmd_build_info(flow, dp, info, OVS_FLOW_CMD_NEW);
876
877                 /* Clear stats. */
878                 if (a[OVS_FLOW_ATTR_CLEAR])
879                         ovs_flow_stats_clear(flow);
880         }
881         ovs_unlock();
882
883         if (!IS_ERR(reply))
884                 ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
885         else
886                 netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
887                                 ovs_dp_flow_multicast_group.id, PTR_ERR(reply));
888         return 0;
889
890 err_flow_free:
891         ovs_flow_free(flow, false);
892 err_unlock_ovs:
893         ovs_unlock();
894 err_kfree:
895         kfree(acts);
896 error:
897         return error;
898 }
899
900 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
901 {
902         struct nlattr **a = info->attrs;
903         struct ovs_header *ovs_header = info->userhdr;
904         struct sw_flow_key key;
905         struct sk_buff *reply;
906         struct sw_flow *flow;
907         struct datapath *dp;
908         struct sw_flow_match match;
909         int err;
910
911         if (!a[OVS_FLOW_ATTR_KEY]) {
912                 OVS_NLERR("Flow get message rejected, Key attribute missing.\n");
913                 return -EINVAL;
914         }
915
916         ovs_match_init(&match, &key, NULL);
917         err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
918         if (err)
919                 return err;
920
921         ovs_lock();
922         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
923         if (!dp) {
924                 err = -ENODEV;
925                 goto unlock;
926         }
927
928         flow = ovs_flow_tbl_lookup(&dp->table, &key);
929         if (!flow || !ovs_flow_cmp_unmasked_key(flow, &match)) {
930                 err = -ENOENT;
931                 goto unlock;
932         }
933
934         reply = ovs_flow_cmd_build_info(flow, dp, info, OVS_FLOW_CMD_NEW);
935         if (IS_ERR(reply)) {
936                 err = PTR_ERR(reply);
937                 goto unlock;
938         }
939
940         ovs_unlock();
941         return genlmsg_reply(reply, info);
942 unlock:
943         ovs_unlock();
944         return err;
945 }
946
947 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
948 {
949         struct nlattr **a = info->attrs;
950         struct ovs_header *ovs_header = info->userhdr;
951         struct sw_flow_key key;
952         struct sk_buff *reply;
953         struct sw_flow *flow;
954         struct datapath *dp;
955         struct sw_flow_match match;
956         int err;
957
958         ovs_lock();
959         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
960         if (!dp) {
961                 err = -ENODEV;
962                 goto unlock;
963         }
964
965         if (!a[OVS_FLOW_ATTR_KEY]) {
966                 err = ovs_flow_tbl_flush(&dp->table);
967                 goto unlock;
968         }
969
970         ovs_match_init(&match, &key, NULL);
971         err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
972         if (err)
973                 goto unlock;
974
975         flow = ovs_flow_tbl_lookup(&dp->table, &key);
976         if (!flow || !ovs_flow_cmp_unmasked_key(flow, &match)) {
977                 err = -ENOENT;
978                 goto unlock;
979         }
980
981         reply = ovs_flow_cmd_alloc_info(flow, info);
982         if (!reply) {
983                 err = -ENOMEM;
984                 goto unlock;
985         }
986
987         ovs_flow_tbl_remove(&dp->table, flow);
988
989         err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_portid,
990                                      info->snd_seq, 0, OVS_FLOW_CMD_DEL);
991         BUG_ON(err < 0);
992
993         ovs_flow_free(flow, true);
994         ovs_unlock();
995
996         ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
997         return 0;
998 unlock:
999         ovs_unlock();
1000         return err;
1001 }
1002
1003 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1004 {
1005         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1006         struct table_instance *ti;
1007         struct datapath *dp;
1008
1009         rcu_read_lock();
1010         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1011         if (!dp) {
1012                 rcu_read_unlock();
1013                 return -ENODEV;
1014         }
1015
1016         ti = rcu_dereference(dp->table.ti);
1017         for (;;) {
1018                 struct sw_flow *flow;
1019                 u32 bucket, obj;
1020
1021                 bucket = cb->args[0];
1022                 obj = cb->args[1];
1023                 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1024                 if (!flow)
1025                         break;
1026
1027                 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1028                                            NETLINK_CB(cb->skb).portid,
1029                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1030                                            OVS_FLOW_CMD_NEW) < 0)
1031                         break;
1032
1033                 cb->args[0] = bucket;
1034                 cb->args[1] = obj;
1035         }
1036         rcu_read_unlock();
1037         return skb->len;
1038 }
1039
1040 static struct genl_ops dp_flow_genl_ops[] = {
1041         { .cmd = OVS_FLOW_CMD_NEW,
1042           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1043           .policy = flow_policy,
1044           .doit = ovs_flow_cmd_new_or_set
1045         },
1046         { .cmd = OVS_FLOW_CMD_DEL,
1047           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1048           .policy = flow_policy,
1049           .doit = ovs_flow_cmd_del
1050         },
1051         { .cmd = OVS_FLOW_CMD_GET,
1052           .flags = 0,               /* OK for unprivileged users. */
1053           .policy = flow_policy,
1054           .doit = ovs_flow_cmd_get,
1055           .dumpit = ovs_flow_cmd_dump
1056         },
1057         { .cmd = OVS_FLOW_CMD_SET,
1058           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1059           .policy = flow_policy,
1060           .doit = ovs_flow_cmd_new_or_set,
1061         },
1062 };
1063
1064 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1065         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1066         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1067         [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1068 };
1069
1070 static struct genl_family dp_datapath_genl_family = {
1071         .id = GENL_ID_GENERATE,
1072         .hdrsize = sizeof(struct ovs_header),
1073         .name = OVS_DATAPATH_FAMILY,
1074         .version = OVS_DATAPATH_VERSION,
1075         .maxattr = OVS_DP_ATTR_MAX,
1076         .netnsok = true,
1077          SET_PARALLEL_OPS
1078 };
1079
1080 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1081         .name = OVS_DATAPATH_MCGROUP
1082 };
1083
1084 static size_t ovs_dp_cmd_msg_size(void)
1085 {
1086         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1087
1088         msgsize += nla_total_size(IFNAMSIZ);
1089         msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1090         msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1091         msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1092
1093         return msgsize;
1094 }
1095
1096 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1097                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1098 {
1099         struct ovs_header *ovs_header;
1100         struct ovs_dp_stats dp_stats;
1101         struct ovs_dp_megaflow_stats dp_megaflow_stats;
1102         int err;
1103
1104         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1105                                    flags, cmd);
1106         if (!ovs_header)
1107                 goto error;
1108
1109         ovs_header->dp_ifindex = get_dpifindex(dp);
1110
1111         rcu_read_lock();
1112         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1113         rcu_read_unlock();
1114         if (err)
1115                 goto nla_put_failure;
1116
1117         get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1118         if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1119                         &dp_stats))
1120                 goto nla_put_failure;
1121
1122         if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1123                         sizeof(struct ovs_dp_megaflow_stats),
1124                         &dp_megaflow_stats))
1125                 goto nla_put_failure;
1126
1127         if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1128                 goto nla_put_failure;
1129
1130         return genlmsg_end(skb, ovs_header);
1131
1132 nla_put_failure:
1133         genlmsg_cancel(skb, ovs_header);
1134 error:
1135         return -EMSGSIZE;
1136 }
1137
1138 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp,
1139                                              struct genl_info *info, u8 cmd)
1140 {
1141         struct sk_buff *skb;
1142         int retval;
1143
1144         skb = genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info, GFP_KERNEL);
1145         if (!skb)
1146                 return ERR_PTR(-ENOMEM);
1147
1148         retval = ovs_dp_cmd_fill_info(dp, skb, info->snd_portid, info->snd_seq, 0, cmd);
1149         if (retval < 0) {
1150                 kfree_skb(skb);
1151                 return ERR_PTR(retval);
1152         }
1153         return skb;
1154 }
1155
1156 /* Called with ovs_mutex. */
1157 static struct datapath *lookup_datapath(struct net *net,
1158                                         struct ovs_header *ovs_header,
1159                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1160 {
1161         struct datapath *dp;
1162
1163         if (!a[OVS_DP_ATTR_NAME])
1164                 dp = get_dp(net, ovs_header->dp_ifindex);
1165         else {
1166                 struct vport *vport;
1167
1168                 rcu_read_lock();
1169                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1170                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1171                 rcu_read_unlock();
1172         }
1173         return dp ? dp : ERR_PTR(-ENODEV);
1174 }
1175
1176 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1177 {
1178         struct datapath *dp;
1179
1180         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1181         if (IS_ERR(dp))
1182                 return;
1183
1184         WARN(dp->user_features, "Dropping previously announced user features\n");
1185         dp->user_features = 0;
1186 }
1187
1188 static void ovs_dp_change(struct datapath *dp, struct nlattr **a)
1189 {
1190         if (a[OVS_DP_ATTR_USER_FEATURES])
1191                 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1192 }
1193
1194 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1195 {
1196         struct nlattr **a = info->attrs;
1197         struct vport_parms parms;
1198         struct sk_buff *reply;
1199         struct datapath *dp;
1200         struct vport *vport;
1201         struct ovs_net *ovs_net;
1202         int err, i;
1203
1204         err = -EINVAL;
1205         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1206                 goto err;
1207
1208         ovs_lock();
1209
1210         err = -ENOMEM;
1211         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1212         if (dp == NULL)
1213                 goto err_unlock_ovs;
1214
1215         ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1216
1217         /* Allocate table. */
1218         err = ovs_flow_tbl_init(&dp->table);
1219         if (err)
1220                 goto err_free_dp;
1221
1222         dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1223         if (!dp->stats_percpu) {
1224                 err = -ENOMEM;
1225                 goto err_destroy_table;
1226         }
1227
1228         dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1229                             GFP_KERNEL);
1230         if (!dp->ports) {
1231                 err = -ENOMEM;
1232                 goto err_destroy_percpu;
1233         }
1234
1235         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1236                 INIT_HLIST_HEAD(&dp->ports[i]);
1237
1238         /* Set up our datapath device. */
1239         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1240         parms.type = OVS_VPORT_TYPE_INTERNAL;
1241         parms.options = NULL;
1242         parms.dp = dp;
1243         parms.port_no = OVSP_LOCAL;
1244         parms.upcall_portid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1245
1246         ovs_dp_change(dp, a);
1247
1248         vport = new_vport(&parms);
1249         if (IS_ERR(vport)) {
1250                 err = PTR_ERR(vport);
1251                 if (err == -EBUSY)
1252                         err = -EEXIST;
1253
1254                 if (err == -EEXIST) {
1255                         /* An outdated user space instance that does not understand
1256                          * the concept of user_features has attempted to create a new
1257                          * datapath and is likely to reuse it. Drop all user features.
1258                          */
1259                         if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1260                                 ovs_dp_reset_user_features(skb, info);
1261                 }
1262
1263                 goto err_destroy_ports_array;
1264         }
1265
1266         reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_NEW);
1267         err = PTR_ERR(reply);
1268         if (IS_ERR(reply))
1269                 goto err_destroy_local_port;
1270
1271         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1272         list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1273
1274         ovs_unlock();
1275
1276         ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1277         return 0;
1278
1279 err_destroy_local_port:
1280         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1281 err_destroy_ports_array:
1282         kfree(dp->ports);
1283 err_destroy_percpu:
1284         free_percpu(dp->stats_percpu);
1285 err_destroy_table:
1286         ovs_flow_tbl_destroy(&dp->table, false);
1287 err_free_dp:
1288         release_net(ovs_dp_get_net(dp));
1289         kfree(dp);
1290 err_unlock_ovs:
1291         ovs_unlock();
1292 err:
1293         return err;
1294 }
1295
1296 /* Called with ovs_mutex. */
1297 static void __dp_destroy(struct datapath *dp)
1298 {
1299         int i;
1300
1301         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1302                 struct vport *vport;
1303                 struct hlist_node *n;
1304
1305                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1306                         if (vport->port_no != OVSP_LOCAL)
1307                                 ovs_dp_detach_port(vport);
1308         }
1309
1310         list_del_rcu(&dp->list_node);
1311
1312         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1313          * all ports in datapath are destroyed first before freeing datapath.
1314          */
1315         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1316
1317         /* RCU destroy the flow table */
1318         ovs_flow_tbl_destroy(&dp->table, true);
1319
1320         call_rcu(&dp->rcu, destroy_dp_rcu);
1321 }
1322
1323 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1324 {
1325         struct sk_buff *reply;
1326         struct datapath *dp;
1327         int err;
1328
1329         ovs_lock();
1330         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1331         err = PTR_ERR(dp);
1332         if (IS_ERR(dp))
1333                 goto unlock;
1334
1335         reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_DEL);
1336         err = PTR_ERR(reply);
1337         if (IS_ERR(reply))
1338                 goto unlock;
1339
1340         __dp_destroy(dp);
1341         ovs_unlock();
1342
1343         ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1344
1345         return 0;
1346 unlock:
1347         ovs_unlock();
1348         return err;
1349 }
1350
1351 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1352 {
1353         struct sk_buff *reply;
1354         struct datapath *dp;
1355         int err;
1356
1357         ovs_lock();
1358         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1359         err = PTR_ERR(dp);
1360         if (IS_ERR(dp))
1361                 goto unlock;
1362
1363         ovs_dp_change(dp, info->attrs);
1364
1365         reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_NEW);
1366         if (IS_ERR(reply)) {
1367                 err = PTR_ERR(reply);
1368                 netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
1369                                 ovs_dp_datapath_multicast_group.id, err);
1370                 err = 0;
1371                 goto unlock;
1372         }
1373
1374         ovs_unlock();
1375         ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1376
1377         return 0;
1378 unlock:
1379         ovs_unlock();
1380         return err;
1381 }
1382
1383 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1384 {
1385         struct sk_buff *reply;
1386         struct datapath *dp;
1387         int err;
1388
1389         ovs_lock();
1390         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1391         if (IS_ERR(dp)) {
1392                 err = PTR_ERR(dp);
1393                 goto unlock;
1394         }
1395
1396         reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_NEW);
1397         if (IS_ERR(reply)) {
1398                 err = PTR_ERR(reply);
1399                 goto unlock;
1400         }
1401
1402         ovs_unlock();
1403         return genlmsg_reply(reply, info);
1404
1405 unlock:
1406         ovs_unlock();
1407         return err;
1408 }
1409
1410 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1411 {
1412         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1413         struct datapath *dp;
1414         int skip = cb->args[0];
1415         int i = 0;
1416
1417         rcu_read_lock();
1418         list_for_each_entry_rcu(dp, &ovs_net->dps, list_node) {
1419                 if (i >= skip &&
1420                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1421                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1422                                          OVS_DP_CMD_NEW) < 0)
1423                         break;
1424                 i++;
1425         }
1426         rcu_read_unlock();
1427
1428         cb->args[0] = i;
1429
1430         return skb->len;
1431 }
1432
1433 static struct genl_ops dp_datapath_genl_ops[] = {
1434         { .cmd = OVS_DP_CMD_NEW,
1435           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1436           .policy = datapath_policy,
1437           .doit = ovs_dp_cmd_new
1438         },
1439         { .cmd = OVS_DP_CMD_DEL,
1440           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1441           .policy = datapath_policy,
1442           .doit = ovs_dp_cmd_del
1443         },
1444         { .cmd = OVS_DP_CMD_GET,
1445           .flags = 0,               /* OK for unprivileged users. */
1446           .policy = datapath_policy,
1447           .doit = ovs_dp_cmd_get,
1448           .dumpit = ovs_dp_cmd_dump
1449         },
1450         { .cmd = OVS_DP_CMD_SET,
1451           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1452           .policy = datapath_policy,
1453           .doit = ovs_dp_cmd_set,
1454         },
1455 };
1456
1457 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1458         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1459         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1460         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1461         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1462         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1463         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1464 };
1465
1466 static struct genl_family dp_vport_genl_family = {
1467         .id = GENL_ID_GENERATE,
1468         .hdrsize = sizeof(struct ovs_header),
1469         .name = OVS_VPORT_FAMILY,
1470         .version = OVS_VPORT_VERSION,
1471         .maxattr = OVS_VPORT_ATTR_MAX,
1472         .netnsok = true,
1473          SET_PARALLEL_OPS
1474 };
1475
1476 struct genl_multicast_group ovs_dp_vport_multicast_group = {
1477         .name = OVS_VPORT_MCGROUP
1478 };
1479
1480 /* Called with ovs_mutex or RCU read lock. */
1481 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1482                                    u32 portid, u32 seq, u32 flags, u8 cmd)
1483 {
1484         struct ovs_header *ovs_header;
1485         struct ovs_vport_stats vport_stats;
1486         int err;
1487
1488         ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1489                                  flags, cmd);
1490         if (!ovs_header)
1491                 return -EMSGSIZE;
1492
1493         ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1494
1495         if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1496             nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1497             nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1498             nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_portid))
1499                 goto nla_put_failure;
1500
1501         ovs_vport_get_stats(vport, &vport_stats);
1502         if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1503                     &vport_stats))
1504                 goto nla_put_failure;
1505
1506         err = ovs_vport_get_options(vport, skb);
1507         if (err == -EMSGSIZE)
1508                 goto error;
1509
1510         return genlmsg_end(skb, ovs_header);
1511
1512 nla_put_failure:
1513         err = -EMSGSIZE;
1514 error:
1515         genlmsg_cancel(skb, ovs_header);
1516         return err;
1517 }
1518
1519 /* Called with ovs_mutex or RCU read lock. */
1520 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1521                                          u32 seq, u8 cmd)
1522 {
1523         struct sk_buff *skb;
1524         int retval;
1525
1526         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1527         if (!skb)
1528                 return ERR_PTR(-ENOMEM);
1529
1530         retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1531         BUG_ON(retval < 0);
1532
1533         return skb;
1534 }
1535
1536 /* Called with ovs_mutex or RCU read lock. */
1537 static struct vport *lookup_vport(struct net *net,
1538                                   struct ovs_header *ovs_header,
1539                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1540 {
1541         struct datapath *dp;
1542         struct vport *vport;
1543
1544         if (a[OVS_VPORT_ATTR_NAME]) {
1545                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1546                 if (!vport)
1547                         return ERR_PTR(-ENODEV);
1548                 if (ovs_header->dp_ifindex &&
1549                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1550                         return ERR_PTR(-ENODEV);
1551                 return vport;
1552         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1553                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1554
1555                 if (port_no >= DP_MAX_PORTS)
1556                         return ERR_PTR(-EFBIG);
1557
1558                 dp = get_dp(net, ovs_header->dp_ifindex);
1559                 if (!dp)
1560                         return ERR_PTR(-ENODEV);
1561
1562                 vport = ovs_vport_ovsl_rcu(dp, port_no);
1563                 if (!vport)
1564                         return ERR_PTR(-ENODEV);
1565                 return vport;
1566         } else
1567                 return ERR_PTR(-EINVAL);
1568 }
1569
1570 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1571 {
1572         struct nlattr **a = info->attrs;
1573         struct ovs_header *ovs_header = info->userhdr;
1574         struct vport_parms parms;
1575         struct sk_buff *reply;
1576         struct vport *vport;
1577         struct datapath *dp;
1578         u32 port_no;
1579         int err;
1580
1581         err = -EINVAL;
1582         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1583             !a[OVS_VPORT_ATTR_UPCALL_PID])
1584                 goto exit;
1585
1586         ovs_lock();
1587         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1588         err = -ENODEV;
1589         if (!dp)
1590                 goto exit_unlock;
1591
1592         if (a[OVS_VPORT_ATTR_PORT_NO]) {
1593                 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1594
1595                 err = -EFBIG;
1596                 if (port_no >= DP_MAX_PORTS)
1597                         goto exit_unlock;
1598
1599                 vport = ovs_vport_ovsl(dp, port_no);
1600                 err = -EBUSY;
1601                 if (vport)
1602                         goto exit_unlock;
1603         } else {
1604                 for (port_no = 1; ; port_no++) {
1605                         if (port_no >= DP_MAX_PORTS) {
1606                                 err = -EFBIG;
1607                                 goto exit_unlock;
1608                         }
1609                         vport = ovs_vport_ovsl(dp, port_no);
1610                         if (!vport)
1611                                 break;
1612                 }
1613         }
1614
1615         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1616         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1617         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1618         parms.dp = dp;
1619         parms.port_no = port_no;
1620         parms.upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1621
1622         vport = new_vport(&parms);
1623         err = PTR_ERR(vport);
1624         if (IS_ERR(vport))
1625                 goto exit_unlock;
1626
1627         err = 0;
1628         if (a[OVS_VPORT_ATTR_STATS])
1629                 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1630
1631         reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
1632                                          OVS_VPORT_CMD_NEW);
1633         if (IS_ERR(reply)) {
1634                 err = PTR_ERR(reply);
1635                 ovs_dp_detach_port(vport);
1636                 goto exit_unlock;
1637         }
1638
1639         ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
1640
1641 exit_unlock:
1642         ovs_unlock();
1643 exit:
1644         return err;
1645 }
1646
1647 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1648 {
1649         struct nlattr **a = info->attrs;
1650         struct sk_buff *reply;
1651         struct vport *vport;
1652         int err;
1653
1654         ovs_lock();
1655         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1656         err = PTR_ERR(vport);
1657         if (IS_ERR(vport))
1658                 goto exit_unlock;
1659
1660         if (a[OVS_VPORT_ATTR_TYPE] &&
1661             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
1662                 err = -EINVAL;
1663                 goto exit_unlock;
1664         }
1665
1666         reply = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1667         if (!reply) {
1668                 err = -ENOMEM;
1669                 goto exit_unlock;
1670         }
1671
1672         if (a[OVS_VPORT_ATTR_OPTIONS]) {
1673                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1674                 if (err)
1675                         goto exit_free;
1676         }
1677
1678         if (a[OVS_VPORT_ATTR_STATS])
1679                 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1680
1681         if (a[OVS_VPORT_ATTR_UPCALL_PID])
1682                 vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1683
1684         err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1685                                       info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1686         BUG_ON(err < 0);
1687
1688         ovs_unlock();
1689         ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
1690         return 0;
1691
1692 exit_free:
1693         kfree_skb(reply);
1694 exit_unlock:
1695         ovs_unlock();
1696         return err;
1697 }
1698
1699 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1700 {
1701         struct nlattr **a = info->attrs;
1702         struct sk_buff *reply;
1703         struct vport *vport;
1704         int err;
1705
1706         ovs_lock();
1707         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1708         err = PTR_ERR(vport);
1709         if (IS_ERR(vport))
1710                 goto exit_unlock;
1711
1712         if (vport->port_no == OVSP_LOCAL) {
1713                 err = -EINVAL;
1714                 goto exit_unlock;
1715         }
1716
1717         reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
1718                                          info->snd_seq, OVS_VPORT_CMD_DEL);
1719         err = PTR_ERR(reply);
1720         if (IS_ERR(reply))
1721                 goto exit_unlock;
1722
1723         err = 0;
1724         ovs_dp_detach_port(vport);
1725
1726         ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
1727
1728 exit_unlock:
1729         ovs_unlock();
1730         return err;
1731 }
1732
1733 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1734 {
1735         struct nlattr **a = info->attrs;
1736         struct ovs_header *ovs_header = info->userhdr;
1737         struct sk_buff *reply;
1738         struct vport *vport;
1739         int err;
1740
1741         rcu_read_lock();
1742         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1743         err = PTR_ERR(vport);
1744         if (IS_ERR(vport))
1745                 goto exit_unlock;
1746
1747         reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
1748                                          info->snd_seq, OVS_VPORT_CMD_NEW);
1749         err = PTR_ERR(reply);
1750         if (IS_ERR(reply))
1751                 goto exit_unlock;
1752
1753         rcu_read_unlock();
1754
1755         return genlmsg_reply(reply, info);
1756
1757 exit_unlock:
1758         rcu_read_unlock();
1759         return err;
1760 }
1761
1762 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1763 {
1764         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1765         struct datapath *dp;
1766         int bucket = cb->args[0], skip = cb->args[1];
1767         int i, j = 0;
1768
1769         rcu_read_lock();
1770         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1771         if (!dp) {
1772                 rcu_read_unlock();
1773                 return -ENODEV;
1774         }
1775         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
1776                 struct vport *vport;
1777
1778                 j = 0;
1779                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1780                         if (j >= skip &&
1781                             ovs_vport_cmd_fill_info(vport, skb,
1782                                                     NETLINK_CB(cb->skb).portid,
1783                                                     cb->nlh->nlmsg_seq,
1784                                                     NLM_F_MULTI,
1785                                                     OVS_VPORT_CMD_NEW) < 0)
1786                                 goto out;
1787
1788                         j++;
1789                 }
1790                 skip = 0;
1791         }
1792 out:
1793         rcu_read_unlock();
1794
1795         cb->args[0] = i;
1796         cb->args[1] = j;
1797
1798         return skb->len;
1799 }
1800
1801 static struct genl_ops dp_vport_genl_ops[] = {
1802         { .cmd = OVS_VPORT_CMD_NEW,
1803           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1804           .policy = vport_policy,
1805           .doit = ovs_vport_cmd_new
1806         },
1807         { .cmd = OVS_VPORT_CMD_DEL,
1808           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1809           .policy = vport_policy,
1810           .doit = ovs_vport_cmd_del
1811         },
1812         { .cmd = OVS_VPORT_CMD_GET,
1813           .flags = 0,               /* OK for unprivileged users. */
1814           .policy = vport_policy,
1815           .doit = ovs_vport_cmd_get,
1816           .dumpit = ovs_vport_cmd_dump
1817         },
1818         { .cmd = OVS_VPORT_CMD_SET,
1819           .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1820           .policy = vport_policy,
1821           .doit = ovs_vport_cmd_set,
1822         },
1823 };
1824
1825 struct genl_family_and_ops {
1826         struct genl_family *family;
1827         struct genl_ops *ops;
1828         int n_ops;
1829         struct genl_multicast_group *group;
1830 };
1831
1832 static const struct genl_family_and_ops dp_genl_families[] = {
1833         { &dp_datapath_genl_family,
1834           dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1835           &ovs_dp_datapath_multicast_group },
1836         { &dp_vport_genl_family,
1837           dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1838           &ovs_dp_vport_multicast_group },
1839         { &dp_flow_genl_family,
1840           dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1841           &ovs_dp_flow_multicast_group },
1842         { &dp_packet_genl_family,
1843           dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1844           NULL },
1845 };
1846
1847 static void dp_unregister_genl(int n_families)
1848 {
1849         int i;
1850
1851         for (i = 0; i < n_families; i++)
1852                 genl_unregister_family(dp_genl_families[i].family);
1853 }
1854
1855 static int dp_register_genl(void)
1856 {
1857         int n_registered;
1858         int err;
1859         int i;
1860
1861         n_registered = 0;
1862         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
1863                 const struct genl_family_and_ops *f = &dp_genl_families[i];
1864
1865                 err = genl_register_family_with_ops(f->family, f->ops,
1866                                                     f->n_ops);
1867                 if (err)
1868                         goto error;
1869                 n_registered++;
1870
1871                 if (f->group) {
1872                         err = genl_register_mc_group(f->family, f->group);
1873                         if (err)
1874                                 goto error;
1875                 }
1876         }
1877
1878         return 0;
1879
1880 error:
1881         dp_unregister_genl(n_registered);
1882         return err;
1883 }
1884
1885 static int __net_init ovs_init_net(struct net *net)
1886 {
1887         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1888
1889         INIT_LIST_HEAD(&ovs_net->dps);
1890         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
1891         return 0;
1892 }
1893
1894 static void __net_exit ovs_exit_net(struct net *net)
1895 {
1896         struct datapath *dp, *dp_next;
1897         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1898
1899         ovs_lock();
1900         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
1901                 __dp_destroy(dp);
1902         ovs_unlock();
1903
1904         cancel_work_sync(&ovs_net->dp_notify_work);
1905 }
1906
1907 static struct pernet_operations ovs_net_ops = {
1908         .init = ovs_init_net,
1909         .exit = ovs_exit_net,
1910         .id   = &ovs_net_id,
1911         .size = sizeof(struct ovs_net),
1912 };
1913
1914 DEFINE_COMPAT_PNET_REG_FUNC(device);
1915
1916 static int __init dp_init(void)
1917 {
1918         int err;
1919
1920         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
1921
1922         pr_info("Open vSwitch switching datapath %s, built "__DATE__" "__TIME__"\n",
1923                 VERSION);
1924
1925         err = ovs_flow_init();
1926         if (err)
1927                 goto error;
1928
1929         err = ovs_vport_init();
1930         if (err)
1931                 goto error_flow_exit;
1932
1933         err = register_pernet_device(&ovs_net_ops);
1934         if (err)
1935                 goto error_vport_exit;
1936
1937         err = register_netdevice_notifier(&ovs_dp_device_notifier);
1938         if (err)
1939                 goto error_netns_exit;
1940
1941         err = dp_register_genl();
1942         if (err < 0)
1943                 goto error_unreg_notifier;
1944
1945         return 0;
1946
1947 error_unreg_notifier:
1948         unregister_netdevice_notifier(&ovs_dp_device_notifier);
1949 error_netns_exit:
1950         unregister_pernet_device(&ovs_net_ops);
1951 error_vport_exit:
1952         ovs_vport_exit();
1953 error_flow_exit:
1954         ovs_flow_exit();
1955 error:
1956         return err;
1957 }
1958
1959 static void dp_cleanup(void)
1960 {
1961         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
1962         unregister_netdevice_notifier(&ovs_dp_device_notifier);
1963         unregister_pernet_device(&ovs_net_ops);
1964         rcu_barrier();
1965         ovs_vport_exit();
1966         ovs_flow_exit();
1967 }
1968
1969 module_init(dp_init);
1970 module_exit(dp_cleanup);
1971
1972 MODULE_DESCRIPTION("Open vSwitch switching datapath");
1973 MODULE_LICENSE("GPL");
1974 MODULE_VERSION(VERSION);