Linux Kernel  3.7.1
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
datapath.c
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2007-2012 Nicira, Inc.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of version 2 of the GNU General Public
6  * License as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful, but
9  * WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public License
14  * along with this program; if not, write to the Free Software
15  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
16  * 02110-1301, USA
17  */
18 
19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
20 
21 #include <linux/init.h>
22 #include <linux/module.h>
23 #include <linux/if_arp.h>
24 #include <linux/if_vlan.h>
25 #include <linux/in.h>
26 #include <linux/ip.h>
27 #include <linux/jhash.h>
28 #include <linux/delay.h>
29 #include <linux/time.h>
30 #include <linux/etherdevice.h>
31 #include <linux/genetlink.h>
32 #include <linux/kernel.h>
33 #include <linux/kthread.h>
34 #include <linux/mutex.h>
35 #include <linux/percpu.h>
36 #include <linux/rcupdate.h>
37 #include <linux/tcp.h>
38 #include <linux/udp.h>
39 #include <linux/ethtool.h>
40 #include <linux/wait.h>
41 #include <asm/div64.h>
42 #include <linux/highmem.h>
43 #include <linux/netfilter_bridge.h>
44 #include <linux/netfilter_ipv4.h>
45 #include <linux/inetdevice.h>
46 #include <linux/list.h>
47 #include <linux/openvswitch.h>
48 #include <linux/rculist.h>
49 #include <linux/dmi.h>
50 #include <linux/workqueue.h>
51 #include <net/genetlink.h>
52 #include <net/net_namespace.h>
53 #include <net/netns/generic.h>
54 
55 #include "datapath.h"
56 #include "flow.h"
57 #include "vport-internal_dev.h"
58 
64 struct ovs_net {
65  struct list_head dps;
66 };
67 
68 static int ovs_net_id __read_mostly;
69 
70 #define REHASH_FLOW_INTERVAL (10 * 60 * HZ)
71 static void rehash_flow_table(struct work_struct *work);
72 static DECLARE_DELAYED_WORK(rehash_flow_wq, rehash_flow_table);
73 
91 static struct vport *new_vport(const struct vport_parms *);
92 static int queue_gso_packets(struct net *, int dp_ifindex, struct sk_buff *,
93  const struct dp_upcall_info *);
94 static int queue_userspace_packet(struct net *, int dp_ifindex,
95  struct sk_buff *,
96  const struct dp_upcall_info *);
97 
98 /* Must be called with rcu_read_lock, genl_mutex, or RTNL lock. */
99 static struct datapath *get_dp(struct net *net, int dp_ifindex)
100 {
101  struct datapath *dp = NULL;
102  struct net_device *dev;
103 
104  rcu_read_lock();
105  dev = dev_get_by_index_rcu(net, dp_ifindex);
106  if (dev) {
107  struct vport *vport = ovs_internal_dev_get_vport(dev);
108  if (vport)
109  dp = vport->dp;
110  }
111  rcu_read_unlock();
112 
113  return dp;
114 }
115 
116 /* Must be called with rcu_read_lock or RTNL lock. */
117 const char *ovs_dp_name(const struct datapath *dp)
118 {
119  struct vport *vport = ovs_vport_rtnl_rcu(dp, OVSP_LOCAL);
120  return vport->ops->get_name(vport);
121 }
122 
123 static int get_dpifindex(struct datapath *dp)
124 {
125  struct vport *local;
126  int ifindex;
127 
128  rcu_read_lock();
129 
130  local = ovs_vport_rcu(dp, OVSP_LOCAL);
131  if (local)
132  ifindex = local->ops->get_ifindex(local);
133  else
134  ifindex = 0;
135 
136  rcu_read_unlock();
137 
138  return ifindex;
139 }
140 
141 static void destroy_dp_rcu(struct rcu_head *rcu)
142 {
143  struct datapath *dp = container_of(rcu, struct datapath, rcu);
144 
147  release_net(ovs_dp_get_net(dp));
148  kfree(dp->ports);
149  kfree(dp);
150 }
151 
152 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
153  u16 port_no)
154 {
155  return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
156 }
157 
158 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
159 {
160  struct vport *vport;
161  struct hlist_node *n;
162  struct hlist_head *head;
163 
164  head = vport_hash_bucket(dp, port_no);
165  hlist_for_each_entry_rcu(vport, n, head, dp_hash_node) {
166  if (vport->port_no == port_no)
167  return vport;
168  }
169  return NULL;
170 }
171 
172 /* Called with RTNL lock and genl_lock. */
173 static struct vport *new_vport(const struct vport_parms *parms)
174 {
175  struct vport *vport;
176 
177  vport = ovs_vport_add(parms);
178  if (!IS_ERR(vport)) {
179  struct datapath *dp = parms->dp;
180  struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
181 
182  hlist_add_head_rcu(&vport->dp_hash_node, head);
183  }
184 
185  return vport;
186 }
187 
188 /* Called with RTNL lock. */
189 void ovs_dp_detach_port(struct vport *p)
190 {
191  ASSERT_RTNL();
192 
193  /* First drop references to device. */
194  hlist_del_rcu(&p->dp_hash_node);
195 
196  /* Then destroy it. */
197  ovs_vport_del(p);
198 }
199 
200 /* Must be called with rcu_read_lock. */
201 void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
202 {
203  struct datapath *dp = p->dp;
204  struct sw_flow *flow;
205  struct dp_stats_percpu *stats;
206  struct sw_flow_key key;
208  int error;
209  int key_len;
210 
211  stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
212 
213  /* Extract flow from 'skb' into 'key'. */
214  error = ovs_flow_extract(skb, p->port_no, &key, &key_len);
215  if (unlikely(error)) {
216  kfree_skb(skb);
217  return;
218  }
219 
220  /* Look up flow. */
221  flow = ovs_flow_tbl_lookup(rcu_dereference(dp->table), &key, key_len);
222  if (unlikely(!flow)) {
223  struct dp_upcall_info upcall;
224 
225  upcall.cmd = OVS_PACKET_CMD_MISS;
226  upcall.key = &key;
227  upcall.userdata = NULL;
228  upcall.portid = p->upcall_portid;
229  ovs_dp_upcall(dp, skb, &upcall);
230  consume_skb(skb);
231  stats_counter = &stats->n_missed;
232  goto out;
233  }
234 
235  OVS_CB(skb)->flow = flow;
236 
237  stats_counter = &stats->n_hit;
238  ovs_flow_used(OVS_CB(skb)->flow, skb);
239  ovs_execute_actions(dp, skb);
240 
241 out:
242  /* Update datapath statistics. */
243  u64_stats_update_begin(&stats->sync);
244  (*stats_counter)++;
245  u64_stats_update_end(&stats->sync);
246 }
247 
248 static struct genl_family dp_packet_genl_family = {
249  .id = GENL_ID_GENERATE,
250  .hdrsize = sizeof(struct ovs_header),
252  .version = OVS_PACKET_VERSION,
253  .maxattr = OVS_PACKET_ATTR_MAX,
254  .netnsok = true
255 };
256 
257 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
258  const struct dp_upcall_info *upcall_info)
259 {
260  struct dp_stats_percpu *stats;
261  int dp_ifindex;
262  int err;
263 
264  if (upcall_info->portid == 0) {
265  err = -ENOTCONN;
266  goto err;
267  }
268 
269  dp_ifindex = get_dpifindex(dp);
270  if (!dp_ifindex) {
271  err = -ENODEV;
272  goto err;
273  }
274 
275  if (!skb_is_gso(skb))
276  err = queue_userspace_packet(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
277  else
278  err = queue_gso_packets(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
279  if (err)
280  goto err;
281 
282  return 0;
283 
284 err:
285  stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
286 
287  u64_stats_update_begin(&stats->sync);
288  stats->n_lost++;
289  u64_stats_update_end(&stats->sync);
290 
291  return err;
292 }
293 
294 static int queue_gso_packets(struct net *net, int dp_ifindex,
295  struct sk_buff *skb,
296  const struct dp_upcall_info *upcall_info)
297 {
298  unsigned short gso_type = skb_shinfo(skb)->gso_type;
299  struct dp_upcall_info later_info;
300  struct sw_flow_key later_key;
301  struct sk_buff *segs, *nskb;
302  int err;
303 
305  if (IS_ERR(segs))
306  return PTR_ERR(segs);
307 
308  /* Queue all of the segments. */
309  skb = segs;
310  do {
311  err = queue_userspace_packet(net, dp_ifindex, skb, upcall_info);
312  if (err)
313  break;
314 
315  if (skb == segs && gso_type & SKB_GSO_UDP) {
316  /* The initial flow key extracted by ovs_flow_extract()
317  * in this case is for a first fragment, so we need to
318  * properly mark later fragments.
319  */
320  later_key = *upcall_info->key;
321  later_key.ip.frag = OVS_FRAG_TYPE_LATER;
322 
323  later_info = *upcall_info;
324  later_info.key = &later_key;
325  upcall_info = &later_info;
326  }
327  } while ((skb = skb->next));
328 
329  /* Free all of the segments. */
330  skb = segs;
331  do {
332  nskb = skb->next;
333  if (err)
334  kfree_skb(skb);
335  else
336  consume_skb(skb);
337  } while ((skb = nskb));
338  return err;
339 }
340 
341 static int queue_userspace_packet(struct net *net, int dp_ifindex,
342  struct sk_buff *skb,
343  const struct dp_upcall_info *upcall_info)
344 {
345  struct ovs_header *upcall;
346  struct sk_buff *nskb = NULL;
347  struct sk_buff *user_skb; /* to be queued to userspace */
348  struct nlattr *nla;
349  unsigned int len;
350  int err;
351 
352  if (vlan_tx_tag_present(skb)) {
353  nskb = skb_clone(skb, GFP_ATOMIC);
354  if (!nskb)
355  return -ENOMEM;
356 
357  nskb = __vlan_put_tag(nskb, vlan_tx_tag_get(nskb));
358  if (!nskb)
359  return -ENOMEM;
360 
361  nskb->vlan_tci = 0;
362  skb = nskb;
363  }
364 
365  if (nla_attr_size(skb->len) > USHRT_MAX) {
366  err = -EFBIG;
367  goto out;
368  }
369 
370  len = sizeof(struct ovs_header);
371  len += nla_total_size(skb->len);
372  len += nla_total_size(FLOW_BUFSIZE);
373  if (upcall_info->cmd == OVS_PACKET_CMD_ACTION)
374  len += nla_total_size(8);
375 
376  user_skb = genlmsg_new(len, GFP_ATOMIC);
377  if (!user_skb) {
378  err = -ENOMEM;
379  goto out;
380  }
381 
382  upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
383  0, upcall_info->cmd);
384  upcall->dp_ifindex = dp_ifindex;
385 
386  nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
387  ovs_flow_to_nlattrs(upcall_info->key, user_skb);
388  nla_nest_end(user_skb, nla);
389 
390  if (upcall_info->userdata)
391  nla_put_u64(user_skb, OVS_PACKET_ATTR_USERDATA,
392  nla_get_u64(upcall_info->userdata));
393 
394  nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
395 
396  skb_copy_and_csum_dev(skb, nla_data(nla));
397 
398  err = genlmsg_unicast(net, user_skb, upcall_info->portid);
399 
400 out:
401  kfree_skb(nskb);
402  return err;
403 }
404 
405 /* Called with genl_mutex. */
406 static int flush_flows(struct datapath *dp)
407 {
408  struct flow_table *old_table;
409  struct flow_table *new_table;
410 
411  old_table = genl_dereference(dp->table);
412  new_table = ovs_flow_tbl_alloc(TBL_MIN_BUCKETS);
413  if (!new_table)
414  return -ENOMEM;
415 
416  rcu_assign_pointer(dp->table, new_table);
417 
419  return 0;
420 }
421 
422 static int validate_actions(const struct nlattr *attr,
423  const struct sw_flow_key *key, int depth);
424 
425 static int validate_sample(const struct nlattr *attr,
426  const struct sw_flow_key *key, int depth)
427 {
428  const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
429  const struct nlattr *probability, *actions;
430  const struct nlattr *a;
431  int rem;
432 
433  memset(attrs, 0, sizeof(attrs));
434  nla_for_each_nested(a, attr, rem) {
435  int type = nla_type(a);
436  if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
437  return -EINVAL;
438  attrs[type] = a;
439  }
440  if (rem)
441  return -EINVAL;
442 
443  probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
444  if (!probability || nla_len(probability) != sizeof(u32))
445  return -EINVAL;
446 
447  actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
448  if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
449  return -EINVAL;
450  return validate_actions(actions, key, depth + 1);
451 }
452 
453 static int validate_tp_port(const struct sw_flow_key *flow_key)
454 {
455  if (flow_key->eth.type == htons(ETH_P_IP)) {
456  if (flow_key->ipv4.tp.src || flow_key->ipv4.tp.dst)
457  return 0;
458  } else if (flow_key->eth.type == htons(ETH_P_IPV6)) {
459  if (flow_key->ipv6.tp.src || flow_key->ipv6.tp.dst)
460  return 0;
461  }
462 
463  return -EINVAL;
464 }
465 
466 static int validate_set(const struct nlattr *a,
467  const struct sw_flow_key *flow_key)
468 {
469  const struct nlattr *ovs_key = nla_data(a);
470  int key_type = nla_type(ovs_key);
471 
472  /* There can be only one key in a action */
473  if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
474  return -EINVAL;
475 
476  if (key_type > OVS_KEY_ATTR_MAX ||
477  nla_len(ovs_key) != ovs_key_lens[key_type])
478  return -EINVAL;
479 
480  switch (key_type) {
481  const struct ovs_key_ipv4 *ipv4_key;
482 
485  break;
486 
487  case OVS_KEY_ATTR_IPV4:
488  if (flow_key->eth.type != htons(ETH_P_IP))
489  return -EINVAL;
490 
491  if (!flow_key->ip.proto)
492  return -EINVAL;
493 
494  ipv4_key = nla_data(ovs_key);
495  if (ipv4_key->ipv4_proto != flow_key->ip.proto)
496  return -EINVAL;
497 
498  if (ipv4_key->ipv4_frag != flow_key->ip.frag)
499  return -EINVAL;
500 
501  break;
502 
503  case OVS_KEY_ATTR_TCP:
504  if (flow_key->ip.proto != IPPROTO_TCP)
505  return -EINVAL;
506 
507  return validate_tp_port(flow_key);
508 
509  case OVS_KEY_ATTR_UDP:
510  if (flow_key->ip.proto != IPPROTO_UDP)
511  return -EINVAL;
512 
513  return validate_tp_port(flow_key);
514 
515  default:
516  return -EINVAL;
517  }
518 
519  return 0;
520 }
521 
522 static int validate_userspace(const struct nlattr *attr)
523 {
524  static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] = {
526  [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_U64 },
527  };
528  struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
529  int error;
530 
531  error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
532  attr, userspace_policy);
533  if (error)
534  return error;
535 
536  if (!a[OVS_USERSPACE_ATTR_PID] ||
537  !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
538  return -EINVAL;
539 
540  return 0;
541 }
542 
543 static int validate_actions(const struct nlattr *attr,
544  const struct sw_flow_key *key, int depth)
545 {
546  const struct nlattr *a;
547  int rem, err;
548 
549  if (depth >= SAMPLE_ACTION_DEPTH)
550  return -EOVERFLOW;
551 
552  nla_for_each_nested(a, attr, rem) {
553  /* Expected argument lengths, (u32)-1 for variable length. */
554  static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
555  [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
559  [OVS_ACTION_ATTR_SET] = (u32)-1,
561  };
562  const struct ovs_action_push_vlan *vlan;
563  int type = nla_type(a);
564 
565  if (type > OVS_ACTION_ATTR_MAX ||
566  (action_lens[type] != nla_len(a) &&
567  action_lens[type] != (u32)-1))
568  return -EINVAL;
569 
570  switch (type) {
572  return -EINVAL;
573 
575  err = validate_userspace(a);
576  if (err)
577  return err;
578  break;
579 
581  if (nla_get_u32(a) >= DP_MAX_PORTS)
582  return -EINVAL;
583  break;
584 
585 
587  break;
588 
590  vlan = nla_data(a);
591  if (vlan->vlan_tpid != htons(ETH_P_8021Q))
592  return -EINVAL;
593  if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
594  return -EINVAL;
595  break;
596 
597  case OVS_ACTION_ATTR_SET:
598  err = validate_set(a, key);
599  if (err)
600  return err;
601  break;
602 
604  err = validate_sample(a, key, depth);
605  if (err)
606  return err;
607  break;
608 
609  default:
610  return -EINVAL;
611  }
612  }
613 
614  if (rem > 0)
615  return -EINVAL;
616 
617  return 0;
618 }
619 
620 static void clear_stats(struct sw_flow *flow)
621 {
622  flow->used = 0;
623  flow->tcp_flags = 0;
624  flow->packet_count = 0;
625  flow->byte_count = 0;
626 }
627 
628 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
629 {
630  struct ovs_header *ovs_header = info->userhdr;
631  struct nlattr **a = info->attrs;
632  struct sw_flow_actions *acts;
633  struct sk_buff *packet;
634  struct sw_flow *flow;
635  struct datapath *dp;
636  struct ethhdr *eth;
637  int len;
638  int err;
639  int key_len;
640 
641  err = -EINVAL;
644  nla_len(a[OVS_PACKET_ATTR_PACKET]) < ETH_HLEN)
645  goto err;
646 
647  len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
648  packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
649  err = -ENOMEM;
650  if (!packet)
651  goto err;
652  skb_reserve(packet, NET_IP_ALIGN);
653 
654  memcpy(__skb_put(packet, len), nla_data(a[OVS_PACKET_ATTR_PACKET]), len);
655 
656  skb_reset_mac_header(packet);
657  eth = eth_hdr(packet);
658 
659  /* Normally, setting the skb 'protocol' field would be handled by a
660  * call to eth_type_trans(), but it assumes there's a sending
661  * device, which we may not have. */
662  if (ntohs(eth->h_proto) >= 1536)
663  packet->protocol = eth->h_proto;
664  else
665  packet->protocol = htons(ETH_P_802_2);
666 
667  /* Build an sw_flow for sending this packet. */
668  flow = ovs_flow_alloc();
669  err = PTR_ERR(flow);
670  if (IS_ERR(flow))
671  goto err_kfree_skb;
672 
673  err = ovs_flow_extract(packet, -1, &flow->key, &key_len);
674  if (err)
675  goto err_flow_free;
676 
677  err = ovs_flow_metadata_from_nlattrs(&flow->key.phy.priority,
678  &flow->key.phy.in_port,
680  if (err)
681  goto err_flow_free;
682 
683  err = validate_actions(a[OVS_PACKET_ATTR_ACTIONS], &flow->key, 0);
684  if (err)
685  goto err_flow_free;
686 
687  flow->hash = ovs_flow_hash(&flow->key, key_len);
688 
690  err = PTR_ERR(acts);
691  if (IS_ERR(acts))
692  goto err_flow_free;
693  rcu_assign_pointer(flow->sf_acts, acts);
694 
695  OVS_CB(packet)->flow = flow;
696  packet->priority = flow->key.phy.priority;
697 
698  rcu_read_lock();
699  dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
700  err = -ENODEV;
701  if (!dp)
702  goto err_unlock;
703 
705  err = ovs_execute_actions(dp, packet);
706  local_bh_enable();
707  rcu_read_unlock();
708 
709  ovs_flow_free(flow);
710  return err;
711 
712 err_unlock:
713  rcu_read_unlock();
714 err_flow_free:
715  ovs_flow_free(flow);
716 err_kfree_skb:
717  kfree_skb(packet);
718 err:
719  return err;
720 }
721 
722 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
723  [OVS_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
724  [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
725  [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
726 };
727 
728 static struct genl_ops dp_packet_genl_ops[] = {
729  { .cmd = OVS_PACKET_CMD_EXECUTE,
730  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
731  .policy = packet_policy,
732  .doit = ovs_packet_cmd_execute
733  }
734 };
735 
736 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
737 {
738  int i;
739  struct flow_table *table = genl_dereference(dp->table);
740 
741  stats->n_flows = ovs_flow_tbl_count(table);
742 
743  stats->n_hit = stats->n_missed = stats->n_lost = 0;
745  const struct dp_stats_percpu *percpu_stats;
746  struct dp_stats_percpu local_stats;
747  unsigned int start;
748 
749  percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
750 
751  do {
752  start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
753  local_stats = *percpu_stats;
754  } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
755 
756  stats->n_hit += local_stats.n_hit;
757  stats->n_missed += local_stats.n_missed;
758  stats->n_lost += local_stats.n_lost;
759  }
760 }
761 
762 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
763  [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
764  [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
765  [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
766 };
767 
768 static struct genl_family dp_flow_genl_family = {
769  .id = GENL_ID_GENERATE,
770  .hdrsize = sizeof(struct ovs_header),
772  .version = OVS_FLOW_VERSION,
773  .maxattr = OVS_FLOW_ATTR_MAX,
774  .netnsok = true
775 };
776 
777 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
778  .name = OVS_FLOW_MCGROUP
779 };
780 
781 /* Called with genl_lock. */
782 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
783  struct sk_buff *skb, u32 portid,
784  u32 seq, u32 flags, u8 cmd)
785 {
786  const int skb_orig_len = skb->len;
787  const struct sw_flow_actions *sf_acts;
788  struct ovs_flow_stats stats;
789  struct ovs_header *ovs_header;
790  struct nlattr *nla;
791  unsigned long used;
792  u8 tcp_flags;
793  int err;
794 
795  sf_acts = rcu_dereference_protected(flow->sf_acts,
796  lockdep_genl_is_held());
797 
798  ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
799  if (!ovs_header)
800  return -EMSGSIZE;
801 
802  ovs_header->dp_ifindex = get_dpifindex(dp);
803 
804  nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
805  if (!nla)
806  goto nla_put_failure;
807  err = ovs_flow_to_nlattrs(&flow->key, skb);
808  if (err)
809  goto error;
810  nla_nest_end(skb, nla);
811 
812  spin_lock_bh(&flow->lock);
813  used = flow->used;
814  stats.n_packets = flow->packet_count;
815  stats.n_bytes = flow->byte_count;
816  tcp_flags = flow->tcp_flags;
817  spin_unlock_bh(&flow->lock);
818 
819  if (used &&
820  nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
821  goto nla_put_failure;
822 
823  if (stats.n_packets &&
825  sizeof(struct ovs_flow_stats), &stats))
826  goto nla_put_failure;
827 
828  if (tcp_flags &&
829  nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags))
830  goto nla_put_failure;
831 
832  /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
833  * this is the first flow to be dumped into 'skb'. This is unusual for
834  * Netlink but individual action lists can be longer than
835  * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
836  * The userspace caller can always fetch the actions separately if it
837  * really wants them. (Most userspace callers in fact don't care.)
838  *
839  * This can only fail for dump operations because the skb is always
840  * properly sized for single flows.
841  */
842  err = nla_put(skb, OVS_FLOW_ATTR_ACTIONS, sf_acts->actions_len,
843  sf_acts->actions);
844  if (err < 0 && skb_orig_len)
845  goto error;
846 
847  return genlmsg_end(skb, ovs_header);
848 
849 nla_put_failure:
850  err = -EMSGSIZE;
851 error:
852  genlmsg_cancel(skb, ovs_header);
853  return err;
854 }
855 
856 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
857 {
858  const struct sw_flow_actions *sf_acts;
859  int len;
860 
861  sf_acts = rcu_dereference_protected(flow->sf_acts,
862  lockdep_genl_is_held());
863 
864  /* OVS_FLOW_ATTR_KEY */
865  len = nla_total_size(FLOW_BUFSIZE);
866  /* OVS_FLOW_ATTR_ACTIONS */
867  len += nla_total_size(sf_acts->actions_len);
868  /* OVS_FLOW_ATTR_STATS */
869  len += nla_total_size(sizeof(struct ovs_flow_stats));
870  /* OVS_FLOW_ATTR_TCP_FLAGS */
871  len += nla_total_size(1);
872  /* OVS_FLOW_ATTR_USED */
873  len += nla_total_size(8);
874 
875  len += NLMSG_ALIGN(sizeof(struct ovs_header));
876 
877  return genlmsg_new(len, GFP_KERNEL);
878 }
879 
880 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
881  struct datapath *dp,
882  u32 portid, u32 seq, u8 cmd)
883 {
884  struct sk_buff *skb;
885  int retval;
886 
887  skb = ovs_flow_cmd_alloc_info(flow);
888  if (!skb)
889  return ERR_PTR(-ENOMEM);
890 
891  retval = ovs_flow_cmd_fill_info(flow, dp, skb, portid, seq, 0, cmd);
892  BUG_ON(retval < 0);
893  return skb;
894 }
895 
896 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
897 {
898  struct nlattr **a = info->attrs;
899  struct ovs_header *ovs_header = info->userhdr;
900  struct sw_flow_key key;
901  struct sw_flow *flow;
902  struct sk_buff *reply;
903  struct datapath *dp;
904  struct flow_table *table;
905  int error;
906  int key_len;
907 
908  /* Extract key. */
909  error = -EINVAL;
910  if (!a[OVS_FLOW_ATTR_KEY])
911  goto error;
912  error = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
913  if (error)
914  goto error;
915 
916  /* Validate actions. */
917  if (a[OVS_FLOW_ATTR_ACTIONS]) {
918  error = validate_actions(a[OVS_FLOW_ATTR_ACTIONS], &key, 0);
919  if (error)
920  goto error;
921  } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
922  error = -EINVAL;
923  goto error;
924  }
925 
926  dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
927  error = -ENODEV;
928  if (!dp)
929  goto error;
930 
931  table = genl_dereference(dp->table);
932  flow = ovs_flow_tbl_lookup(table, &key, key_len);
933  if (!flow) {
934  struct sw_flow_actions *acts;
935 
936  /* Bail out if we're not allowed to create a new flow. */
937  error = -ENOENT;
938  if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
939  goto error;
940 
941  /* Expand table, if necessary, to make room. */
942  if (ovs_flow_tbl_need_to_expand(table)) {
943  struct flow_table *new_table;
944 
945  new_table = ovs_flow_tbl_expand(table);
946  if (!IS_ERR(new_table)) {
947  rcu_assign_pointer(dp->table, new_table);
949  table = genl_dereference(dp->table);
950  }
951  }
952 
953  /* Allocate flow. */
954  flow = ovs_flow_alloc();
955  if (IS_ERR(flow)) {
956  error = PTR_ERR(flow);
957  goto error;
958  }
959  flow->key = key;
960  clear_stats(flow);
961 
962  /* Obtain actions. */
963  acts = ovs_flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]);
964  error = PTR_ERR(acts);
965  if (IS_ERR(acts))
966  goto error_free_flow;
967  rcu_assign_pointer(flow->sf_acts, acts);
968 
969  /* Put flow in bucket. */
970  flow->hash = ovs_flow_hash(&key, key_len);
971  ovs_flow_tbl_insert(table, flow);
972 
973  reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
974  info->snd_seq,
976  } else {
977  /* We found a matching flow. */
978  struct sw_flow_actions *old_acts;
979  struct nlattr *acts_attrs;
980 
981  /* Bail out if we're not allowed to modify an existing flow.
982  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
983  * because Generic Netlink treats the latter as a dump
984  * request. We also accept NLM_F_EXCL in case that bug ever
985  * gets fixed.
986  */
987  error = -EEXIST;
988  if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
989  info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
990  goto error;
991 
992  /* Update actions. */
993  old_acts = rcu_dereference_protected(flow->sf_acts,
994  lockdep_genl_is_held());
995  acts_attrs = a[OVS_FLOW_ATTR_ACTIONS];
996  if (acts_attrs &&
997  (old_acts->actions_len != nla_len(acts_attrs) ||
998  memcmp(old_acts->actions, nla_data(acts_attrs),
999  old_acts->actions_len))) {
1000  struct sw_flow_actions *new_acts;
1001 
1002  new_acts = ovs_flow_actions_alloc(acts_attrs);
1003  error = PTR_ERR(new_acts);
1004  if (IS_ERR(new_acts))
1005  goto error;
1006 
1007  rcu_assign_pointer(flow->sf_acts, new_acts);
1008  ovs_flow_deferred_free_acts(old_acts);
1009  }
1010 
1011  reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1012  info->snd_seq, OVS_FLOW_CMD_NEW);
1013 
1014  /* Clear stats. */
1015  if (a[OVS_FLOW_ATTR_CLEAR]) {
1016  spin_lock_bh(&flow->lock);
1017  clear_stats(flow);
1018  spin_unlock_bh(&flow->lock);
1019  }
1020  }
1021 
1022  if (!IS_ERR(reply))
1023  genl_notify(reply, genl_info_net(info), info->snd_portid,
1024  ovs_dp_flow_multicast_group.id, info->nlhdr,
1025  GFP_KERNEL);
1026  else
1027  netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
1028  ovs_dp_flow_multicast_group.id, PTR_ERR(reply));
1029  return 0;
1030 
1031 error_free_flow:
1032  ovs_flow_free(flow);
1033 error:
1034  return error;
1035 }
1036 
1037 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1038 {
1039  struct nlattr **a = info->attrs;
1040  struct ovs_header *ovs_header = info->userhdr;
1041  struct sw_flow_key key;
1042  struct sk_buff *reply;
1043  struct sw_flow *flow;
1044  struct datapath *dp;
1045  struct flow_table *table;
1046  int err;
1047  int key_len;
1048 
1049  if (!a[OVS_FLOW_ATTR_KEY])
1050  return -EINVAL;
1051  err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1052  if (err)
1053  return err;
1054 
1055  dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1056  if (!dp)
1057  return -ENODEV;
1058 
1059  table = genl_dereference(dp->table);
1060  flow = ovs_flow_tbl_lookup(table, &key, key_len);
1061  if (!flow)
1062  return -ENOENT;
1063 
1064  reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1065  info->snd_seq, OVS_FLOW_CMD_NEW);
1066  if (IS_ERR(reply))
1067  return PTR_ERR(reply);
1068 
1069  return genlmsg_reply(reply, info);
1070 }
1071 
1072 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1073 {
1074  struct nlattr **a = info->attrs;
1075  struct ovs_header *ovs_header = info->userhdr;
1076  struct sw_flow_key key;
1077  struct sk_buff *reply;
1078  struct sw_flow *flow;
1079  struct datapath *dp;
1080  struct flow_table *table;
1081  int err;
1082  int key_len;
1083 
1084  dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1085  if (!dp)
1086  return -ENODEV;
1087 
1088  if (!a[OVS_FLOW_ATTR_KEY])
1089  return flush_flows(dp);
1090 
1091  err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1092  if (err)
1093  return err;
1094 
1095  table = genl_dereference(dp->table);
1096  flow = ovs_flow_tbl_lookup(table, &key, key_len);
1097  if (!flow)
1098  return -ENOENT;
1099 
1100  reply = ovs_flow_cmd_alloc_info(flow);
1101  if (!reply)
1102  return -ENOMEM;
1103 
1104  ovs_flow_tbl_remove(table, flow);
1105 
1106  err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_portid,
1107  info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1108  BUG_ON(err < 0);
1109 
1110  ovs_flow_deferred_free(flow);
1111 
1112  genl_notify(reply, genl_info_net(info), info->snd_portid,
1113  ovs_dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1114  return 0;
1115 }
1116 
1117 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1118 {
1119  struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1120  struct datapath *dp;
1121  struct flow_table *table;
1122 
1123  dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1124  if (!dp)
1125  return -ENODEV;
1126 
1127  table = genl_dereference(dp->table);
1128 
1129  for (;;) {
1130  struct sw_flow *flow;
1131  u32 bucket, obj;
1132 
1133  bucket = cb->args[0];
1134  obj = cb->args[1];
1135  flow = ovs_flow_tbl_next(table, &bucket, &obj);
1136  if (!flow)
1137  break;
1138 
1139  if (ovs_flow_cmd_fill_info(flow, dp, skb,
1140  NETLINK_CB(cb->skb).portid,
1141  cb->nlh->nlmsg_seq, NLM_F_MULTI,
1142  OVS_FLOW_CMD_NEW) < 0)
1143  break;
1144 
1145  cb->args[0] = bucket;
1146  cb->args[1] = obj;
1147  }
1148  return skb->len;
1149 }
1150 
1151 static struct genl_ops dp_flow_genl_ops[] = {
1152  { .cmd = OVS_FLOW_CMD_NEW,
1153  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1154  .policy = flow_policy,
1155  .doit = ovs_flow_cmd_new_or_set
1156  },
1157  { .cmd = OVS_FLOW_CMD_DEL,
1158  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1159  .policy = flow_policy,
1160  .doit = ovs_flow_cmd_del
1161  },
1162  { .cmd = OVS_FLOW_CMD_GET,
1163  .flags = 0, /* OK for unprivileged users. */
1164  .policy = flow_policy,
1165  .doit = ovs_flow_cmd_get,
1166  .dumpit = ovs_flow_cmd_dump
1167  },
1168  { .cmd = OVS_FLOW_CMD_SET,
1169  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1170  .policy = flow_policy,
1171  .doit = ovs_flow_cmd_new_or_set,
1172  },
1173 };
1174 
1175 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1176  [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1177  [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1178 };
1179 
1180 static struct genl_family dp_datapath_genl_family = {
1181  .id = GENL_ID_GENERATE,
1182  .hdrsize = sizeof(struct ovs_header),
1184  .version = OVS_DATAPATH_VERSION,
1185  .maxattr = OVS_DP_ATTR_MAX,
1186  .netnsok = true
1187 };
1188 
1189 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1190  .name = OVS_DATAPATH_MCGROUP
1191 };
1192 
1193 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1194  u32 portid, u32 seq, u32 flags, u8 cmd)
1195 {
1196  struct ovs_header *ovs_header;
1197  struct ovs_dp_stats dp_stats;
1198  int err;
1199 
1200  ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1201  flags, cmd);
1202  if (!ovs_header)
1203  goto error;
1204 
1205  ovs_header->dp_ifindex = get_dpifindex(dp);
1206 
1207  rcu_read_lock();
1208  err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1209  rcu_read_unlock();
1210  if (err)
1211  goto nla_put_failure;
1212 
1213  get_dp_stats(dp, &dp_stats);
1214  if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats), &dp_stats))
1215  goto nla_put_failure;
1216 
1217  return genlmsg_end(skb, ovs_header);
1218 
1219 nla_put_failure:
1220  genlmsg_cancel(skb, ovs_header);
1221 error:
1222  return -EMSGSIZE;
1223 }
1224 
1225 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 portid,
1226  u32 seq, u8 cmd)
1227 {
1228  struct sk_buff *skb;
1229  int retval;
1230 
1231  skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1232  if (!skb)
1233  return ERR_PTR(-ENOMEM);
1234 
1235  retval = ovs_dp_cmd_fill_info(dp, skb, portid, seq, 0, cmd);
1236  if (retval < 0) {
1237  kfree_skb(skb);
1238  return ERR_PTR(retval);
1239  }
1240  return skb;
1241 }
1242 
1243 /* Called with genl_mutex and optionally with RTNL lock also. */
1244 static struct datapath *lookup_datapath(struct net *net,
1245  struct ovs_header *ovs_header,
1246  struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1247 {
1248  struct datapath *dp;
1249 
1250  if (!a[OVS_DP_ATTR_NAME])
1251  dp = get_dp(net, ovs_header->dp_ifindex);
1252  else {
1253  struct vport *vport;
1254 
1255  rcu_read_lock();
1256  vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1257  dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1258  rcu_read_unlock();
1259  }
1260  return dp ? dp : ERR_PTR(-ENODEV);
1261 }
1262 
1263 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1264 {
1265  struct nlattr **a = info->attrs;
1266  struct vport_parms parms;
1267  struct sk_buff *reply;
1268  struct datapath *dp;
1269  struct vport *vport;
1270  struct ovs_net *ovs_net;
1271  int err, i;
1272 
1273  err = -EINVAL;
1274  if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1275  goto err;
1276 
1277  rtnl_lock();
1278 
1279  err = -ENOMEM;
1280  dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1281  if (dp == NULL)
1282  goto err_unlock_rtnl;
1283 
1284  ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1285 
1286  /* Allocate table. */
1287  err = -ENOMEM;
1289  if (!dp->table)
1290  goto err_free_dp;
1291 
1293  if (!dp->stats_percpu) {
1294  err = -ENOMEM;
1295  goto err_destroy_table;
1296  }
1297 
1298  dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1299  GFP_KERNEL);
1300  if (!dp->ports) {
1301  err = -ENOMEM;
1302  goto err_destroy_percpu;
1303  }
1304 
1305  for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1306  INIT_HLIST_HEAD(&dp->ports[i]);
1307 
1308  /* Set up our datapath device. */
1309  parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1310  parms.type = OVS_VPORT_TYPE_INTERNAL;
1311  parms.options = NULL;
1312  parms.dp = dp;
1313  parms.port_no = OVSP_LOCAL;
1314  parms.upcall_portid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1315 
1316  vport = new_vport(&parms);
1317  if (IS_ERR(vport)) {
1318  err = PTR_ERR(vport);
1319  if (err == -EBUSY)
1320  err = -EEXIST;
1321 
1322  goto err_destroy_ports_array;
1323  }
1324 
1325  reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1326  info->snd_seq, OVS_DP_CMD_NEW);
1327  err = PTR_ERR(reply);
1328  if (IS_ERR(reply))
1329  goto err_destroy_local_port;
1330 
1331  ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1332  list_add_tail(&dp->list_node, &ovs_net->dps);
1333  rtnl_unlock();
1334 
1335  genl_notify(reply, genl_info_net(info), info->snd_portid,
1336  ovs_dp_datapath_multicast_group.id, info->nlhdr,
1337  GFP_KERNEL);
1338  return 0;
1339 
1340 err_destroy_local_port:
1341  ovs_dp_detach_port(ovs_vport_rtnl(dp, OVSP_LOCAL));
1342 err_destroy_ports_array:
1343  kfree(dp->ports);
1344 err_destroy_percpu:
1346 err_destroy_table:
1348 err_free_dp:
1349  release_net(ovs_dp_get_net(dp));
1350  kfree(dp);
1351 err_unlock_rtnl:
1352  rtnl_unlock();
1353 err:
1354  return err;
1355 }
1356 
1357 /* Called with genl_mutex. */
1358 static void __dp_destroy(struct datapath *dp)
1359 {
1360  int i;
1361 
1362  rtnl_lock();
1363 
1364  for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1365  struct vport *vport;
1366  struct hlist_node *node, *n;
1367 
1368  hlist_for_each_entry_safe(vport, node, n, &dp->ports[i], dp_hash_node)
1369  if (vport->port_no != OVSP_LOCAL)
1370  ovs_dp_detach_port(vport);
1371  }
1372 
1373  list_del(&dp->list_node);
1374  ovs_dp_detach_port(ovs_vport_rtnl(dp, OVSP_LOCAL));
1375 
1376  /* rtnl_unlock() will wait until all the references to devices that
1377  * are pending unregistration have been dropped. We do it here to
1378  * ensure that any internal devices (which contain DP pointers) are
1379  * fully destroyed before freeing the datapath.
1380  */
1381  rtnl_unlock();
1382 
1383  call_rcu(&dp->rcu, destroy_dp_rcu);
1384 }
1385 
1386 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1387 {
1388  struct sk_buff *reply;
1389  struct datapath *dp;
1390  int err;
1391 
1392  dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1393  err = PTR_ERR(dp);
1394  if (IS_ERR(dp))
1395  return err;
1396 
1397  reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1398  info->snd_seq, OVS_DP_CMD_DEL);
1399  err = PTR_ERR(reply);
1400  if (IS_ERR(reply))
1401  return err;
1402 
1403  __dp_destroy(dp);
1404 
1405  genl_notify(reply, genl_info_net(info), info->snd_portid,
1406  ovs_dp_datapath_multicast_group.id, info->nlhdr,
1407  GFP_KERNEL);
1408 
1409  return 0;
1410 }
1411 
1412 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1413 {
1414  struct sk_buff *reply;
1415  struct datapath *dp;
1416  int err;
1417 
1418  dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1419  if (IS_ERR(dp))
1420  return PTR_ERR(dp);
1421 
1422  reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1423  info->snd_seq, OVS_DP_CMD_NEW);
1424  if (IS_ERR(reply)) {
1425  err = PTR_ERR(reply);
1426  netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
1427  ovs_dp_datapath_multicast_group.id, err);
1428  return 0;
1429  }
1430 
1431  genl_notify(reply, genl_info_net(info), info->snd_portid,
1432  ovs_dp_datapath_multicast_group.id, info->nlhdr,
1433  GFP_KERNEL);
1434 
1435  return 0;
1436 }
1437 
1438 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1439 {
1440  struct sk_buff *reply;
1441  struct datapath *dp;
1442 
1443  dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1444  if (IS_ERR(dp))
1445  return PTR_ERR(dp);
1446 
1447  reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1448  info->snd_seq, OVS_DP_CMD_NEW);
1449  if (IS_ERR(reply))
1450  return PTR_ERR(reply);
1451 
1452  return genlmsg_reply(reply, info);
1453 }
1454 
1455 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1456 {
1457  struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1458  struct datapath *dp;
1459  int skip = cb->args[0];
1460  int i = 0;
1461 
1462  list_for_each_entry(dp, &ovs_net->dps, list_node) {
1463  if (i >= skip &&
1464  ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1465  cb->nlh->nlmsg_seq, NLM_F_MULTI,
1466  OVS_DP_CMD_NEW) < 0)
1467  break;
1468  i++;
1469  }
1470 
1471  cb->args[0] = i;
1472 
1473  return skb->len;
1474 }
1475 
1476 static struct genl_ops dp_datapath_genl_ops[] = {
1477  { .cmd = OVS_DP_CMD_NEW,
1478  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1479  .policy = datapath_policy,
1480  .doit = ovs_dp_cmd_new
1481  },
1482  { .cmd = OVS_DP_CMD_DEL,
1483  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1484  .policy = datapath_policy,
1485  .doit = ovs_dp_cmd_del
1486  },
1487  { .cmd = OVS_DP_CMD_GET,
1488  .flags = 0, /* OK for unprivileged users. */
1489  .policy = datapath_policy,
1490  .doit = ovs_dp_cmd_get,
1491  .dumpit = ovs_dp_cmd_dump
1492  },
1493  { .cmd = OVS_DP_CMD_SET,
1494  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1495  .policy = datapath_policy,
1496  .doit = ovs_dp_cmd_set,
1497  },
1498 };
1499 
1500 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1501  [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1502  [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1503  [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1504  [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1505  [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1506  [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1507 };
1508 
1509 static struct genl_family dp_vport_genl_family = {
1510  .id = GENL_ID_GENERATE,
1511  .hdrsize = sizeof(struct ovs_header),
1513  .version = OVS_VPORT_VERSION,
1514  .maxattr = OVS_VPORT_ATTR_MAX,
1515  .netnsok = true
1516 };
1517 
1519  .name = OVS_VPORT_MCGROUP
1520 };
1521 
1522 /* Called with RTNL lock or RCU read lock. */
1523 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1524  u32 portid, u32 seq, u32 flags, u8 cmd)
1525 {
1526  struct ovs_header *ovs_header;
1527  struct ovs_vport_stats vport_stats;
1528  int err;
1529 
1530  ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1531  flags, cmd);
1532  if (!ovs_header)
1533  return -EMSGSIZE;
1534 
1535  ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1536 
1537  if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1538  nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1539  nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1540  nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_portid))
1541  goto nla_put_failure;
1542 
1543  ovs_vport_get_stats(vport, &vport_stats);
1544  if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1545  &vport_stats))
1546  goto nla_put_failure;
1547 
1548  err = ovs_vport_get_options(vport, skb);
1549  if (err == -EMSGSIZE)
1550  goto error;
1551 
1552  return genlmsg_end(skb, ovs_header);
1553 
1554 nla_put_failure:
1555  err = -EMSGSIZE;
1556 error:
1557  genlmsg_cancel(skb, ovs_header);
1558  return err;
1559 }
1560 
1561 /* Called with RTNL lock or RCU read lock. */
1562 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1563  u32 seq, u8 cmd)
1564 {
1565  struct sk_buff *skb;
1566  int retval;
1567 
1568  skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1569  if (!skb)
1570  return ERR_PTR(-ENOMEM);
1571 
1572  retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1573  if (retval < 0) {
1574  kfree_skb(skb);
1575  return ERR_PTR(retval);
1576  }
1577  return skb;
1578 }
1579 
1580 /* Called with RTNL lock or RCU read lock. */
1581 static struct vport *lookup_vport(struct net *net,
1582  struct ovs_header *ovs_header,
1583  struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1584 {
1585  struct datapath *dp;
1586  struct vport *vport;
1587 
1588  if (a[OVS_VPORT_ATTR_NAME]) {
1589  vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1590  if (!vport)
1591  return ERR_PTR(-ENODEV);
1592  if (ovs_header->dp_ifindex &&
1593  ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1594  return ERR_PTR(-ENODEV);
1595  return vport;
1596  } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1597  u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1598 
1599  if (port_no >= DP_MAX_PORTS)
1600  return ERR_PTR(-EFBIG);
1601 
1602  dp = get_dp(net, ovs_header->dp_ifindex);
1603  if (!dp)
1604  return ERR_PTR(-ENODEV);
1605 
1606  vport = ovs_vport_rtnl_rcu(dp, port_no);
1607  if (!vport)
1608  return ERR_PTR(-ENOENT);
1609  return vport;
1610  } else
1611  return ERR_PTR(-EINVAL);
1612 }
1613 
1614 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1615 {
1616  struct nlattr **a = info->attrs;
1617  struct ovs_header *ovs_header = info->userhdr;
1618  struct vport_parms parms;
1619  struct sk_buff *reply;
1620  struct vport *vport;
1621  struct datapath *dp;
1622  u32 port_no;
1623  int err;
1624 
1625  err = -EINVAL;
1626  if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1628  goto exit;
1629 
1630  rtnl_lock();
1631  dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1632  err = -ENODEV;
1633  if (!dp)
1634  goto exit_unlock;
1635 
1636  if (a[OVS_VPORT_ATTR_PORT_NO]) {
1637  port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1638 
1639  err = -EFBIG;
1640  if (port_no >= DP_MAX_PORTS)
1641  goto exit_unlock;
1642 
1643  vport = ovs_vport_rtnl_rcu(dp, port_no);
1644  err = -EBUSY;
1645  if (vport)
1646  goto exit_unlock;
1647  } else {
1648  for (port_no = 1; ; port_no++) {
1649  if (port_no >= DP_MAX_PORTS) {
1650  err = -EFBIG;
1651  goto exit_unlock;
1652  }
1653  vport = ovs_vport_rtnl(dp, port_no);
1654  if (!vport)
1655  break;
1656  }
1657  }
1658 
1659  parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1660  parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1661  parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1662  parms.dp = dp;
1663  parms.port_no = port_no;
1664  parms.upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1665 
1666  vport = new_vport(&parms);
1667  err = PTR_ERR(vport);
1668  if (IS_ERR(vport))
1669  goto exit_unlock;
1670 
1671  reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
1673  if (IS_ERR(reply)) {
1674  err = PTR_ERR(reply);
1675  ovs_dp_detach_port(vport);
1676  goto exit_unlock;
1677  }
1678  genl_notify(reply, genl_info_net(info), info->snd_portid,
1679  ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1680 
1681 exit_unlock:
1682  rtnl_unlock();
1683 exit:
1684  return err;
1685 }
1686 
1687 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1688 {
1689  struct nlattr **a = info->attrs;
1690  struct sk_buff *reply;
1691  struct vport *vport;
1692  int err;
1693 
1694  rtnl_lock();
1695  vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1696  err = PTR_ERR(vport);
1697  if (IS_ERR(vport))
1698  goto exit_unlock;
1699 
1700  err = 0;
1701  if (a[OVS_VPORT_ATTR_TYPE] &&
1702  nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type)
1703  err = -EINVAL;
1704 
1705  if (!err && a[OVS_VPORT_ATTR_OPTIONS])
1706  err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1707  if (err)
1708  goto exit_unlock;
1709  if (a[OVS_VPORT_ATTR_UPCALL_PID])
1710  vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1711 
1712  reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
1714  if (IS_ERR(reply)) {
1715  netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
1716  ovs_dp_vport_multicast_group.id, PTR_ERR(reply));
1717  goto exit_unlock;
1718  }
1719 
1720  genl_notify(reply, genl_info_net(info), info->snd_portid,
1721  ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1722 
1723 exit_unlock:
1724  rtnl_unlock();
1725  return err;
1726 }
1727 
1728 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1729 {
1730  struct nlattr **a = info->attrs;
1731  struct sk_buff *reply;
1732  struct vport *vport;
1733  int err;
1734 
1735  rtnl_lock();
1736  vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1737  err = PTR_ERR(vport);
1738  if (IS_ERR(vport))
1739  goto exit_unlock;
1740 
1741  if (vport->port_no == OVSP_LOCAL) {
1742  err = -EINVAL;
1743  goto exit_unlock;
1744  }
1745 
1746  reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
1748  err = PTR_ERR(reply);
1749  if (IS_ERR(reply))
1750  goto exit_unlock;
1751 
1752  ovs_dp_detach_port(vport);
1753 
1754  genl_notify(reply, genl_info_net(info), info->snd_portid,
1755  ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1756 
1757 exit_unlock:
1758  rtnl_unlock();
1759  return err;
1760 }
1761 
1762 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1763 {
1764  struct nlattr **a = info->attrs;
1765  struct ovs_header *ovs_header = info->userhdr;
1766  struct sk_buff *reply;
1767  struct vport *vport;
1768  int err;
1769 
1770  rcu_read_lock();
1771  vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1772  err = PTR_ERR(vport);
1773  if (IS_ERR(vport))
1774  goto exit_unlock;
1775 
1776  reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
1778  err = PTR_ERR(reply);
1779  if (IS_ERR(reply))
1780  goto exit_unlock;
1781 
1782  rcu_read_unlock();
1783 
1784  return genlmsg_reply(reply, info);
1785 
1786 exit_unlock:
1787  rcu_read_unlock();
1788  return err;
1789 }
1790 
1791 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1792 {
1793  struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1794  struct datapath *dp;
1795  int bucket = cb->args[0], skip = cb->args[1];
1796  int i, j = 0;
1797 
1798  dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1799  if (!dp)
1800  return -ENODEV;
1801 
1802  rcu_read_lock();
1803  for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
1804  struct vport *vport;
1805  struct hlist_node *n;
1806 
1807  j = 0;
1808  hlist_for_each_entry_rcu(vport, n, &dp->ports[i], dp_hash_node) {
1809  if (j >= skip &&
1810  ovs_vport_cmd_fill_info(vport, skb,
1811  NETLINK_CB(cb->skb).portid,
1812  cb->nlh->nlmsg_seq,
1813  NLM_F_MULTI,
1814  OVS_VPORT_CMD_NEW) < 0)
1815  goto out;
1816 
1817  j++;
1818  }
1819  skip = 0;
1820  }
1821 out:
1822  rcu_read_unlock();
1823 
1824  cb->args[0] = i;
1825  cb->args[1] = j;
1826 
1827  return skb->len;
1828 }
1829 
1830 static struct genl_ops dp_vport_genl_ops[] = {
1831  { .cmd = OVS_VPORT_CMD_NEW,
1832  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1833  .policy = vport_policy,
1834  .doit = ovs_vport_cmd_new
1835  },
1836  { .cmd = OVS_VPORT_CMD_DEL,
1837  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1838  .policy = vport_policy,
1839  .doit = ovs_vport_cmd_del
1840  },
1841  { .cmd = OVS_VPORT_CMD_GET,
1842  .flags = 0, /* OK for unprivileged users. */
1843  .policy = vport_policy,
1844  .doit = ovs_vport_cmd_get,
1845  .dumpit = ovs_vport_cmd_dump
1846  },
1847  { .cmd = OVS_VPORT_CMD_SET,
1848  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1849  .policy = vport_policy,
1850  .doit = ovs_vport_cmd_set,
1851  },
1852 };
1853 
1856  struct genl_ops *ops;
1857  int n_ops;
1859 };
1860 
1861 static const struct genl_family_and_ops dp_genl_families[] = {
1862  { &dp_datapath_genl_family,
1863  dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1864  &ovs_dp_datapath_multicast_group },
1865  { &dp_vport_genl_family,
1866  dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1867  &ovs_dp_vport_multicast_group },
1868  { &dp_flow_genl_family,
1869  dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1870  &ovs_dp_flow_multicast_group },
1871  { &dp_packet_genl_family,
1872  dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1873  NULL },
1874 };
1875 
1876 static void dp_unregister_genl(int n_families)
1877 {
1878  int i;
1879 
1880  for (i = 0; i < n_families; i++)
1881  genl_unregister_family(dp_genl_families[i].family);
1882 }
1883 
1884 static int dp_register_genl(void)
1885 {
1886  int n_registered;
1887  int err;
1888  int i;
1889 
1890  n_registered = 0;
1891  for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
1892  const struct genl_family_and_ops *f = &dp_genl_families[i];
1893 
1895  f->n_ops);
1896  if (err)
1897  goto error;
1898  n_registered++;
1899 
1900  if (f->group) {
1901  err = genl_register_mc_group(f->family, f->group);
1902  if (err)
1903  goto error;
1904  }
1905  }
1906 
1907  return 0;
1908 
1909 error:
1910  dp_unregister_genl(n_registered);
1911  return err;
1912 }
1913 
1914 static void rehash_flow_table(struct work_struct *work)
1915 {
1916  struct datapath *dp;
1917  struct net *net;
1918 
1919  genl_lock();
1920  rtnl_lock();
1921  for_each_net(net) {
1922  struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1923 
1924  list_for_each_entry(dp, &ovs_net->dps, list_node) {
1925  struct flow_table *old_table = genl_dereference(dp->table);
1926  struct flow_table *new_table;
1927 
1928  new_table = ovs_flow_tbl_rehash(old_table);
1929  if (!IS_ERR(new_table)) {
1930  rcu_assign_pointer(dp->table, new_table);
1931  ovs_flow_tbl_deferred_destroy(old_table);
1932  }
1933  }
1934  }
1935  rtnl_unlock();
1936  genl_unlock();
1937 
1938  schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
1939 }
1940 
1941 static int __net_init ovs_init_net(struct net *net)
1942 {
1943  struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1944 
1945  INIT_LIST_HEAD(&ovs_net->dps);
1946  return 0;
1947 }
1948 
1949 static void __net_exit ovs_exit_net(struct net *net)
1950 {
1951  struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1952  struct datapath *dp, *dp_next;
1953 
1954  genl_lock();
1955  list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
1956  __dp_destroy(dp);
1957  genl_unlock();
1958 }
1959 
1960 static struct pernet_operations ovs_net_ops = {
1961  .init = ovs_init_net,
1962  .exit = ovs_exit_net,
1963  .id = &ovs_net_id,
1964  .size = sizeof(struct ovs_net),
1965 };
1966 
1967 static int __init dp_init(void)
1968 {
1969  struct sk_buff *dummy_skb;
1970  int err;
1971 
1972  BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
1973 
1974  pr_info("Open vSwitch switching datapath\n");
1975 
1976  err = ovs_flow_init();
1977  if (err)
1978  goto error;
1979 
1980  err = ovs_vport_init();
1981  if (err)
1982  goto error_flow_exit;
1983 
1984  err = register_pernet_device(&ovs_net_ops);
1985  if (err)
1986  goto error_vport_exit;
1987 
1989  if (err)
1990  goto error_netns_exit;
1991 
1992  err = dp_register_genl();
1993  if (err < 0)
1994  goto error_unreg_notifier;
1995 
1996  schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
1997 
1998  return 0;
1999 
2000 error_unreg_notifier:
2002 error_netns_exit:
2003  unregister_pernet_device(&ovs_net_ops);
2004 error_vport_exit:
2005  ovs_vport_exit();
2006 error_flow_exit:
2007  ovs_flow_exit();
2008 error:
2009  return err;
2010 }
2011 
2012 static void dp_cleanup(void)
2013 {
2014  cancel_delayed_work_sync(&rehash_flow_wq);
2015  dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2017  unregister_pernet_device(&ovs_net_ops);
2018  rcu_barrier();
2019  ovs_vport_exit();
2020  ovs_flow_exit();
2021 }
2022 
2023 module_init(dp_init);
2024 module_exit(dp_cleanup);
2025 
2026 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2027 MODULE_LICENSE("GPL");