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