]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blob - net/openvswitch/datapath.c
projects / mirror_ubuntu-zesty-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
openvswitch: Allow user space to announce ability to accept unaligned Netlink messages
[mirror_ubuntu-zesty-kernel.git] / net / openvswitch / 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/ethtool.h>
40 #include <linux/wait.h>
41 #include <asm/div64.h>
42 #include <linux/highmem.h>
43 #include <linux/netfilter_bridge.h>
44 #include <linux/netfilter_ipv4.h>
45 #include <linux/inetdevice.h>
46 #include <linux/list.h>
47 #include <linux/lockdep.h>
48 #include <linux/openvswitch.h>
49 #include <linux/rculist.h>
50 #include <linux/dmi.h>
51 #include <linux/workqueue.h>
52 #include <net/genetlink.h>
53 #include <net/net_namespace.h>
54 #include <net/netns/generic.h>
55
56 #include "datapath.h"
57 #include "flow.h"
58 #include "flow_netlink.h"
59 #include "vport-internal_dev.h"
60 #include "vport-netdev.h"
61
62 int ovs_net_id __read_mostly;
63
64 static void ovs_notify(struct genl_family *family,
65 struct sk_buff *skb, struct genl_info *info)
66 {
67 genl_notify(family, skb, genl_info_net(info), info->snd_portid,
68 0, info->nlhdr, GFP_KERNEL);
69 }
70
71 /**
72 * DOC: Locking:
73 *
74 * All writes e.g. Writes to device state (add/remove datapath, port, set
75 * operations on vports, etc.), Writes to other state (flow table
76 * modifications, set miscellaneous datapath parameters, etc.) are protected
77 * by ovs_lock.
78 *
79 * Reads are protected by RCU.
80 *
81 * There are a few special cases (mostly stats) that have their own
82 * synchronization but they nest under all of above and don't interact with
83 * each other.
84 *
85 * The RTNL lock nests inside ovs_mutex.
86 */
87
88 static DEFINE_MUTEX(ovs_mutex);
89
90 void ovs_lock(void)
91 {
92 mutex_lock(&ovs_mutex);
93 }
94
95 void ovs_unlock(void)
96 {
97 mutex_unlock(&ovs_mutex);
98 }
99
100 #ifdef CONFIG_LOCKDEP
101 int lockdep_ovsl_is_held(void)
102 {
103 if (debug_locks)
104 return lockdep_is_held(&ovs_mutex);
105 else
106 return 1;
107 }
108 #endif
109
110 static struct vport *new_vport(const struct vport_parms *);
111 static int queue_gso_packets(struct net *, int dp_ifindex, struct sk_buff *,
112 const struct dp_upcall_info *);
113 static int queue_userspace_packet(struct net *, int dp_ifindex,
114 struct sk_buff *,
115 const struct dp_upcall_info *);
116
117 /* Must be called with rcu_read_lock or ovs_mutex. */
118 static struct datapath *get_dp(struct net *net, int dp_ifindex)
119 {
120 struct datapath *dp = NULL;
121 struct net_device *dev;
122
123 rcu_read_lock();
124 dev = dev_get_by_index_rcu(net, dp_ifindex);
125 if (dev) {
126 struct vport *vport = ovs_internal_dev_get_vport(dev);
127 if (vport)
128 dp = vport->dp;
129 }
130 rcu_read_unlock();
131
132 return dp;
133 }
134
135 /* Must be called with rcu_read_lock or ovs_mutex. */
136 const char *ovs_dp_name(const struct datapath *dp)
137 {
138 struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
139 return vport->ops->get_name(vport);
140 }
141
142 static int get_dpifindex(struct datapath *dp)
143 {
144 struct vport *local;
145 int ifindex;
146
147 rcu_read_lock();
148
149 local = ovs_vport_rcu(dp, OVSP_LOCAL);
150 if (local)
151 ifindex = netdev_vport_priv(local)->dev->ifindex;
152 else
153 ifindex = 0;
154
155 rcu_read_unlock();
156
157 return ifindex;
158 }
159
160 static void destroy_dp_rcu(struct rcu_head *rcu)
161 {
162 struct datapath *dp = container_of(rcu, struct datapath, rcu);
163
164 ovs_flow_tbl_destroy(&dp->table);
165 free_percpu(dp->stats_percpu);
166 release_net(ovs_dp_get_net(dp));
167 kfree(dp->ports);
168 kfree(dp);
169 }
170
171 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
172 u16 port_no)
173 {
174 return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
175 }
176
177 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
178 {
179 struct vport *vport;
180 struct hlist_head *head;
181
182 head = vport_hash_bucket(dp, port_no);
183 hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
184 if (vport->port_no == port_no)
185 return vport;
186 }
187 return NULL;
188 }
189
190 /* Called with ovs_mutex. */
191 static struct vport *new_vport(const struct vport_parms *parms)
192 {
193 struct vport *vport;
194
195 vport = ovs_vport_add(parms);
196 if (!IS_ERR(vport)) {
197 struct datapath *dp = parms->dp;
198 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
199
200 hlist_add_head_rcu(&vport->dp_hash_node, head);
201 }
202 return vport;
203 }
204
205 void ovs_dp_detach_port(struct vport *p)
206 {
207 ASSERT_OVSL();
208
209 /* First drop references to device. */
210 hlist_del_rcu(&p->dp_hash_node);
211
212 /* Then destroy it. */
213 ovs_vport_del(p);
214 }
215
216 /* Must be called with rcu_read_lock. */
217 void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
218 {
219 struct datapath *dp = p->dp;
220 struct sw_flow *flow;
221 struct dp_stats_percpu *stats;
222 struct sw_flow_key key;
223 u64 *stats_counter;
224 u32 n_mask_hit;
225 int error;
226
227 stats = this_cpu_ptr(dp->stats_percpu);
228
229 /* Extract flow from 'skb' into 'key'. */
230 error = ovs_flow_extract(skb, p->port_no, &key);
231 if (unlikely(error)) {
232 kfree_skb(skb);
233 return;
234 }
235
236 /* Look up flow. */
237 flow = ovs_flow_tbl_lookup_stats(&dp->table, &key, &n_mask_hit);
238 if (unlikely(!flow)) {
239 struct dp_upcall_info upcall;
240
241 upcall.cmd = OVS_PACKET_CMD_MISS;
242 upcall.key = &key;
243 upcall.userdata = NULL;
244 upcall.portid = p->upcall_portid;
245 ovs_dp_upcall(dp, skb, &upcall);
246 consume_skb(skb);
247 stats_counter = &stats->n_missed;
248 goto out;
249 }
250
251 OVS_CB(skb)->flow = flow;
252 OVS_CB(skb)->pkt_key = &key;
253
254 ovs_flow_stats_update(OVS_CB(skb)->flow, skb);
255 ovs_execute_actions(dp, skb);
256 stats_counter = &stats->n_hit;
257
258 out:
259 /* Update datapath statistics. */
260 u64_stats_update_begin(&stats->sync);
261 (*stats_counter)++;
262 stats->n_mask_hit += n_mask_hit;
263 u64_stats_update_end(&stats->sync);
264 }
265
266 static struct genl_family dp_packet_genl_family = {
267 .id = GENL_ID_GENERATE,
268 .hdrsize = sizeof(struct ovs_header),
269 .name = OVS_PACKET_FAMILY,
270 .version = OVS_PACKET_VERSION,
271 .maxattr = OVS_PACKET_ATTR_MAX,
272 .netnsok = true,
273 .parallel_ops = true,
274 };
275
276 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
277 const struct dp_upcall_info *upcall_info)
278 {
279 struct dp_stats_percpu *stats;
280 int dp_ifindex;
281 int err;
282
283 if (upcall_info->portid == 0) {
284 err = -ENOTCONN;
285 goto err;
286 }
287
288 dp_ifindex = get_dpifindex(dp);
289 if (!dp_ifindex) {
290 err = -ENODEV;
291 goto err;
292 }
293
294 if (!skb_is_gso(skb))
295 err = queue_userspace_packet(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
296 else
297 err = queue_gso_packets(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
298 if (err)
299 goto err;
300
301 return 0;
302
303 err:
304 stats = this_cpu_ptr(dp->stats_percpu);
305
306 u64_stats_update_begin(&stats->sync);
307 stats->n_lost++;
308 u64_stats_update_end(&stats->sync);
309
310 return err;
311 }
312
313 static int queue_gso_packets(struct net *net, int dp_ifindex,
314 struct sk_buff *skb,
315 const struct dp_upcall_info *upcall_info)
316 {
317 unsigned short gso_type = skb_shinfo(skb)->gso_type;
318 struct dp_upcall_info later_info;
319 struct sw_flow_key later_key;
320 struct sk_buff *segs, *nskb;
321 int err;
322
323 segs = __skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM, false);
324 if (IS_ERR(segs))
325 return PTR_ERR(segs);
326
327 /* Queue all of the segments. */
328 skb = segs;
329 do {
330 err = queue_userspace_packet(net, dp_ifindex, skb, upcall_info);
331 if (err)
332 break;
333
334 if (skb == segs && gso_type & SKB_GSO_UDP) {
335 /* The initial flow key extracted by ovs_flow_extract()
336 * in this case is for a first fragment, so we need to
337 * properly mark later fragments.
338 */
339 later_key = *upcall_info->key;
340 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
341
342 later_info = *upcall_info;
343 later_info.key = &later_key;
344 upcall_info = &later_info;
345 }
346 } while ((skb = skb->next));
347
348 /* Free all of the segments. */
349 skb = segs;
350 do {
351 nskb = skb->next;
352 if (err)
353 kfree_skb(skb);
354 else
355 consume_skb(skb);
356 } while ((skb = nskb));
357 return err;
358 }
359
360 static size_t key_attr_size(void)
361 {
362 return nla_total_size(4) /* OVS_KEY_ATTR_PRIORITY */
363 + nla_total_size(0) /* OVS_KEY_ATTR_TUNNEL */
364 + nla_total_size(8) /* OVS_TUNNEL_KEY_ATTR_ID */
365 + nla_total_size(4) /* OVS_TUNNEL_KEY_ATTR_IPV4_SRC */
366 + nla_total_size(4) /* OVS_TUNNEL_KEY_ATTR_IPV4_DST */
367 + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TOS */
368 + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TTL */
369 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
370 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_CSUM */
371 + nla_total_size(4) /* OVS_KEY_ATTR_IN_PORT */
372 + nla_total_size(4) /* OVS_KEY_ATTR_SKB_MARK */
373 + nla_total_size(12) /* OVS_KEY_ATTR_ETHERNET */
374 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
375 + nla_total_size(4) /* OVS_KEY_ATTR_8021Q */
376 + nla_total_size(0) /* OVS_KEY_ATTR_ENCAP */
377 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
378 + nla_total_size(40) /* OVS_KEY_ATTR_IPV6 */
379 + nla_total_size(2) /* OVS_KEY_ATTR_ICMPV6 */
380 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
381 }
382
383 static size_t upcall_msg_size(const struct sk_buff *skb,
384 const struct nlattr *userdata)
385 {
386 size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
387 + nla_total_size(skb->len) /* OVS_PACKET_ATTR_PACKET */
388 + nla_total_size(key_attr_size()); /* OVS_PACKET_ATTR_KEY */
389
390 /* OVS_PACKET_ATTR_USERDATA */
391 if (userdata)
392 size += NLA_ALIGN(userdata->nla_len);
393
394 return size;
395 }
396
397 static int queue_userspace_packet(struct net *net, int dp_ifindex,
398 struct sk_buff *skb,
399 const struct dp_upcall_info *upcall_info)
400 {
401 struct ovs_header *upcall;
402 struct sk_buff *nskb = NULL;
403 struct sk_buff *user_skb; /* to be queued to userspace */
404 struct nlattr *nla;
405 struct genl_info info = {
406 .dst_sk = net->genl_sock,
407 .snd_portid = upcall_info->portid,
408 };
409 size_t len;
410 int err;
411
412 if (vlan_tx_tag_present(skb)) {
413 nskb = skb_clone(skb, GFP_ATOMIC);
414 if (!nskb)
415 return -ENOMEM;
416
417 nskb = __vlan_put_tag(nskb, nskb->vlan_proto, vlan_tx_tag_get(nskb));
418 if (!nskb)
419 return -ENOMEM;
420
421 nskb->vlan_tci = 0;
422 skb = nskb;
423 }
424
425 if (nla_attr_size(skb->len) > USHRT_MAX) {
426 err = -EFBIG;
427 goto out;
428 }
429
430 len = upcall_msg_size(skb, upcall_info->userdata);
431 user_skb = genlmsg_new_unicast(len, &info, GFP_ATOMIC);
432 if (!user_skb) {
433 err = -ENOMEM;
434 goto out;
435 }
436
437 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
438 0, upcall_info->cmd);
439 upcall->dp_ifindex = dp_ifindex;
440
441 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
442 ovs_nla_put_flow(upcall_info->key, upcall_info->key, user_skb);
443 nla_nest_end(user_skb, nla);
444
445 if (upcall_info->userdata)
446 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
447 nla_len(upcall_info->userdata),
448 nla_data(upcall_info->userdata));
449
450 nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
451
452 skb_copy_and_csum_dev(skb, nla_data(nla));
453
454 genlmsg_end(user_skb, upcall);
455 err = genlmsg_unicast(net, user_skb, upcall_info->portid);
456
457 out:
458 kfree_skb(nskb);
459 return err;
460 }
461
462 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
463 {
464 struct ovs_header *ovs_header = info->userhdr;
465 struct nlattr **a = info->attrs;
466 struct sw_flow_actions *acts;
467 struct sk_buff *packet;
468 struct sw_flow *flow;
469 struct datapath *dp;
470 struct ethhdr *eth;
471 int len;
472 int err;
473
474 err = -EINVAL;
475 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
476 !a[OVS_PACKET_ATTR_ACTIONS])
477 goto err;
478
479 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
480 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
481 err = -ENOMEM;
482 if (!packet)
483 goto err;
484 skb_reserve(packet, NET_IP_ALIGN);
485
486 nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
487
488 skb_reset_mac_header(packet);
489 eth = eth_hdr(packet);
490
491 /* Normally, setting the skb 'protocol' field would be handled by a
492 * call to eth_type_trans(), but it assumes there's a sending
493 * device, which we may not have. */
494 if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
495 packet->protocol = eth->h_proto;
496 else
497 packet->protocol = htons(ETH_P_802_2);
498
499 /* Build an sw_flow for sending this packet. */
500 flow = ovs_flow_alloc(false);
501 err = PTR_ERR(flow);
502 if (IS_ERR(flow))
503 goto err_kfree_skb;
504
505 err = ovs_flow_extract(packet, -1, &flow->key);
506 if (err)
507 goto err_flow_free;
508
509 err = ovs_nla_get_flow_metadata(flow, a[OVS_PACKET_ATTR_KEY]);
510 if (err)
511 goto err_flow_free;
512 acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_PACKET_ATTR_ACTIONS]));
513 err = PTR_ERR(acts);
514 if (IS_ERR(acts))
515 goto err_flow_free;
516
517 err = ovs_nla_copy_actions(a[OVS_PACKET_ATTR_ACTIONS],
518 &flow->key, 0, &acts);
519 rcu_assign_pointer(flow->sf_acts, acts);
520 if (err)
521 goto err_flow_free;
522
523 OVS_CB(packet)->flow = flow;
524 OVS_CB(packet)->pkt_key = &flow->key;
525 packet->priority = flow->key.phy.priority;
526 packet->mark = flow->key.phy.skb_mark;
527
528 rcu_read_lock();
529 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
530 err = -ENODEV;
531 if (!dp)
532 goto err_unlock;
533
534 local_bh_disable();
535 err = ovs_execute_actions(dp, packet);
536 local_bh_enable();
537 rcu_read_unlock();
538
539 ovs_flow_free(flow, false);
540 return err;
541
542 err_unlock:
543 rcu_read_unlock();
544 err_flow_free:
545 ovs_flow_free(flow, false);
546 err_kfree_skb:
547 kfree_skb(packet);
548 err:
549 return err;
550 }
551
552 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
553 [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
554 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
555 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
556 };
557
558 static const struct genl_ops dp_packet_genl_ops[] = {
559 { .cmd = OVS_PACKET_CMD_EXECUTE,
560 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
561 .policy = packet_policy,
562 .doit = ovs_packet_cmd_execute
563 }
564 };
565
566 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats,
567 struct ovs_dp_megaflow_stats *mega_stats)
568 {
569 int i;
570
571 memset(mega_stats, 0, sizeof(*mega_stats));
572
573 stats->n_flows = ovs_flow_tbl_count(&dp->table);
574 mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
575
576 stats->n_hit = stats->n_missed = stats->n_lost = 0;
577
578 for_each_possible_cpu(i) {
579 const struct dp_stats_percpu *percpu_stats;
580 struct dp_stats_percpu local_stats;
581 unsigned int start;
582
583 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
584
585 do {
586 start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
587 local_stats = *percpu_stats;
588 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
589
590 stats->n_hit += local_stats.n_hit;
591 stats->n_missed += local_stats.n_missed;
592 stats->n_lost += local_stats.n_lost;
593 mega_stats->n_mask_hit += local_stats.n_mask_hit;
594 }
595 }
596
597 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
598 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
599 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
600 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
601 };
602
603 static struct genl_family dp_flow_genl_family = {
604 .id = GENL_ID_GENERATE,
605 .hdrsize = sizeof(struct ovs_header),
606 .name = OVS_FLOW_FAMILY,
607 .version = OVS_FLOW_VERSION,
608 .maxattr = OVS_FLOW_ATTR_MAX,
609 .netnsok = true,
610 .parallel_ops = true,
611 };
612
613 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
614 .name = OVS_FLOW_MCGROUP
615 };
616
617 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
618 {
619 return NLMSG_ALIGN(sizeof(struct ovs_header))
620 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
621 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_MASK */
622 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
623 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
624 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
625 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
626 }
627
628 /* Called with ovs_mutex. */
629 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
630 struct sk_buff *skb, u32 portid,
631 u32 seq, u32 flags, u8 cmd)
632 {
633 const int skb_orig_len = skb->len;
634 struct nlattr *start;
635 struct ovs_flow_stats stats;
636 __be16 tcp_flags;
637 unsigned long used;
638 struct ovs_header *ovs_header;
639 struct nlattr *nla;
640 int err;
641
642 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
643 if (!ovs_header)
644 return -EMSGSIZE;
645
646 ovs_header->dp_ifindex = get_dpifindex(dp);
647
648 /* Fill flow key. */
649 nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
650 if (!nla)
651 goto nla_put_failure;
652
653 err = ovs_nla_put_flow(&flow->unmasked_key, &flow->unmasked_key, skb);
654 if (err)
655 goto error;
656 nla_nest_end(skb, nla);
657
658 nla = nla_nest_start(skb, OVS_FLOW_ATTR_MASK);
659 if (!nla)
660 goto nla_put_failure;
661
662 err = ovs_nla_put_flow(&flow->key, &flow->mask->key, skb);
663 if (err)
664 goto error;
665
666 nla_nest_end(skb, nla);
667
668 ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
669 if (used &&
670 nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
671 goto nla_put_failure;
672
673 if (stats.n_packets &&
674 nla_put(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats))
675 goto nla_put_failure;
676
677 if ((u8)ntohs(tcp_flags) &&
678 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
679 goto nla_put_failure;
680
681 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
682 * this is the first flow to be dumped into 'skb'. This is unusual for
683 * Netlink but individual action lists can be longer than
684 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
685 * The userspace caller can always fetch the actions separately if it
686 * really wants them. (Most userspace callers in fact don't care.)
687 *
688 * This can only fail for dump operations because the skb is always
689 * properly sized for single flows.
690 */
691 start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
692 if (start) {
693 const struct sw_flow_actions *sf_acts;
694
695 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
696
697 err = ovs_nla_put_actions(sf_acts->actions,
698 sf_acts->actions_len, skb);
699 if (!err)
700 nla_nest_end(skb, start);
701 else {
702 if (skb_orig_len)
703 goto error;
704
705 nla_nest_cancel(skb, start);
706 }
707 } else if (skb_orig_len)
708 goto nla_put_failure;
709
710 return genlmsg_end(skb, ovs_header);
711
712 nla_put_failure:
713 err = -EMSGSIZE;
714 error:
715 genlmsg_cancel(skb, ovs_header);
716 return err;
717 }
718
719 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow,
720 struct genl_info *info)
721 {
722 size_t len;
723
724 len = ovs_flow_cmd_msg_size(ovsl_dereference(flow->sf_acts));
725
726 return genlmsg_new_unicast(len, info, GFP_KERNEL);
727 }
728
729 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
730 struct datapath *dp,
731 struct genl_info *info,
732 u8 cmd)
733 {
734 struct sk_buff *skb;
735 int retval;
736
737 skb = ovs_flow_cmd_alloc_info(flow, info);
738 if (!skb)
739 return ERR_PTR(-ENOMEM);
740
741 retval = ovs_flow_cmd_fill_info(flow, dp, skb, info->snd_portid,
742 info->snd_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 bool exact_5tuple;
759 int error;
760
761 /* Extract key. */
762 error = -EINVAL;
763 if (!a[OVS_FLOW_ATTR_KEY])
764 goto error;
765
766 ovs_match_init(&match, &key, &mask);
767 error = ovs_nla_get_match(&match, &exact_5tuple,
768 a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
769 if (error)
770 goto error;
771
772 /* Validate actions. */
773 if (a[OVS_FLOW_ATTR_ACTIONS]) {
774 acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
775 error = PTR_ERR(acts);
776 if (IS_ERR(acts))
777 goto error;
778
779 ovs_flow_mask_key(&masked_key, &key, &mask);
780 error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS],
781 &masked_key, 0, &acts);
782 if (error) {
783 OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
784 goto err_kfree;
785 }
786 } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
787 error = -EINVAL;
788 goto error;
789 }
790
791 ovs_lock();
792 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
793 error = -ENODEV;
794 if (!dp)
795 goto err_unlock_ovs;
796
797 /* Check if this is a duplicate flow */
798 flow = ovs_flow_tbl_lookup(&dp->table, &key);
799 if (!flow) {
800 /* Bail out if we're not allowed to create a new flow. */
801 error = -ENOENT;
802 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
803 goto err_unlock_ovs;
804
805 /* Allocate flow. */
806 flow = ovs_flow_alloc(!exact_5tuple);
807 if (IS_ERR(flow)) {
808 error = PTR_ERR(flow);
809 goto err_unlock_ovs;
810 }
811
812 flow->key = masked_key;
813 flow->unmasked_key = key;
814 rcu_assign_pointer(flow->sf_acts, acts);
815
816 /* Put flow in bucket. */
817 error = ovs_flow_tbl_insert(&dp->table, flow, &mask);
818 if (error) {
819 acts = NULL;
820 goto err_flow_free;
821 }
822
823 reply = ovs_flow_cmd_build_info(flow, dp, info, 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, OVS_FLOW_CMD_NEW);
852
853 /* Clear stats. */
854 if (a[OVS_FLOW_ATTR_CLEAR])
855 ovs_flow_stats_clear(flow);
856 }
857 ovs_unlock();
858
859 if (!IS_ERR(reply))
860 ovs_notify(&dp_flow_genl_family, reply, info);
861 else
862 genl_set_err(&dp_flow_genl_family, sock_net(skb->sk), 0,
863 0, PTR_ERR(reply));
864 return 0;
865
866 err_flow_free:
867 ovs_flow_free(flow, false);
868 err_unlock_ovs:
869 ovs_unlock();
870 err_kfree:
871 kfree(acts);
872 error:
873 return error;
874 }
875
876 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
877 {
878 struct nlattr **a = info->attrs;
879 struct ovs_header *ovs_header = info->userhdr;
880 struct sw_flow_key key;
881 struct sk_buff *reply;
882 struct sw_flow *flow;
883 struct datapath *dp;
884 struct sw_flow_match match;
885 int err;
886
887 if (!a[OVS_FLOW_ATTR_KEY]) {
888 OVS_NLERR("Flow get message rejected, Key attribute missing.\n");
889 return -EINVAL;
890 }
891
892 ovs_match_init(&match, &key, NULL);
893 err = ovs_nla_get_match(&match, NULL, a[OVS_FLOW_ATTR_KEY], NULL);
894 if (err)
895 return err;
896
897 ovs_lock();
898 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
899 if (!dp) {
900 err = -ENODEV;
901 goto unlock;
902 }
903
904 flow = ovs_flow_tbl_lookup(&dp->table, &key);
905 if (!flow || !ovs_flow_cmp_unmasked_key(flow, &match)) {
906 err = -ENOENT;
907 goto unlock;
908 }
909
910 reply = ovs_flow_cmd_build_info(flow, dp, info, OVS_FLOW_CMD_NEW);
911 if (IS_ERR(reply)) {
912 err = PTR_ERR(reply);
913 goto unlock;
914 }
915
916 ovs_unlock();
917 return genlmsg_reply(reply, info);
918 unlock:
919 ovs_unlock();
920 return err;
921 }
922
923 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
924 {
925 struct nlattr **a = info->attrs;
926 struct ovs_header *ovs_header = info->userhdr;
927 struct sw_flow_key key;
928 struct sk_buff *reply;
929 struct sw_flow *flow;
930 struct datapath *dp;
931 struct sw_flow_match match;
932 int err;
933
934 ovs_lock();
935 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
936 if (!dp) {
937 err = -ENODEV;
938 goto unlock;
939 }
940
941 if (!a[OVS_FLOW_ATTR_KEY]) {
942 err = ovs_flow_tbl_flush(&dp->table);
943 goto unlock;
944 }
945
946 ovs_match_init(&match, &key, NULL);
947 err = ovs_nla_get_match(&match, NULL, a[OVS_FLOW_ATTR_KEY], NULL);
948 if (err)
949 goto unlock;
950
951 flow = ovs_flow_tbl_lookup(&dp->table, &key);
952 if (!flow || !ovs_flow_cmp_unmasked_key(flow, &match)) {
953 err = -ENOENT;
954 goto unlock;
955 }
956
957 reply = ovs_flow_cmd_alloc_info(flow, info);
958 if (!reply) {
959 err = -ENOMEM;
960 goto unlock;
961 }
962
963 ovs_flow_tbl_remove(&dp->table, flow);
964
965 err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_portid,
966 info->snd_seq, 0, OVS_FLOW_CMD_DEL);
967 BUG_ON(err < 0);
968
969 ovs_flow_free(flow, true);
970 ovs_unlock();
971
972 ovs_notify(&dp_flow_genl_family, reply, info);
973 return 0;
974 unlock:
975 ovs_unlock();
976 return err;
977 }
978
979 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
980 {
981 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
982 struct table_instance *ti;
983 struct datapath *dp;
984
985 rcu_read_lock();
986 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
987 if (!dp) {
988 rcu_read_unlock();
989 return -ENODEV;
990 }
991
992 ti = rcu_dereference(dp->table.ti);
993 for (;;) {
994 struct sw_flow *flow;
995 u32 bucket, obj;
996
997 bucket = cb->args[0];
998 obj = cb->args[1];
999 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1000 if (!flow)
1001 break;
1002
1003 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1004 NETLINK_CB(cb->skb).portid,
1005 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1006 OVS_FLOW_CMD_NEW) < 0)
1007 break;
1008
1009 cb->args[0] = bucket;
1010 cb->args[1] = obj;
1011 }
1012 rcu_read_unlock();
1013 return skb->len;
1014 }
1015
1016 static const struct genl_ops dp_flow_genl_ops[] = {
1017 { .cmd = OVS_FLOW_CMD_NEW,
1018 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1019 .policy = flow_policy,
1020 .doit = ovs_flow_cmd_new_or_set
1021 },
1022 { .cmd = OVS_FLOW_CMD_DEL,
1023 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1024 .policy = flow_policy,
1025 .doit = ovs_flow_cmd_del
1026 },
1027 { .cmd = OVS_FLOW_CMD_GET,
1028 .flags = 0, /* OK for unprivileged users. */
1029 .policy = flow_policy,
1030 .doit = ovs_flow_cmd_get,
1031 .dumpit = ovs_flow_cmd_dump
1032 },
1033 { .cmd = OVS_FLOW_CMD_SET,
1034 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1035 .policy = flow_policy,
1036 .doit = ovs_flow_cmd_new_or_set,
1037 },
1038 };
1039
1040 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1041 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1042 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1043 [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1044 };
1045
1046 static struct genl_family dp_datapath_genl_family = {
1047 .id = GENL_ID_GENERATE,
1048 .hdrsize = sizeof(struct ovs_header),
1049 .name = OVS_DATAPATH_FAMILY,
1050 .version = OVS_DATAPATH_VERSION,
1051 .maxattr = OVS_DP_ATTR_MAX,
1052 .netnsok = true,
1053 .parallel_ops = true,
1054 };
1055
1056 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1057 .name = OVS_DATAPATH_MCGROUP
1058 };
1059
1060 static size_t ovs_dp_cmd_msg_size(void)
1061 {
1062 size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1063
1064 msgsize += nla_total_size(IFNAMSIZ);
1065 msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1066 msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1067
1068 return msgsize;
1069 }
1070
1071 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1072 u32 portid, u32 seq, u32 flags, u8 cmd)
1073 {
1074 struct ovs_header *ovs_header;
1075 struct ovs_dp_stats dp_stats;
1076 struct ovs_dp_megaflow_stats dp_megaflow_stats;
1077 int err;
1078
1079 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1080 flags, cmd);
1081 if (!ovs_header)
1082 goto error;
1083
1084 ovs_header->dp_ifindex = get_dpifindex(dp);
1085
1086 rcu_read_lock();
1087 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1088 rcu_read_unlock();
1089 if (err)
1090 goto nla_put_failure;
1091
1092 get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1093 if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1094 &dp_stats))
1095 goto nla_put_failure;
1096
1097 if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1098 sizeof(struct ovs_dp_megaflow_stats),
1099 &dp_megaflow_stats))
1100 goto nla_put_failure;
1101
1102 if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1103 goto nla_put_failure;
1104
1105 return genlmsg_end(skb, ovs_header);
1106
1107 nla_put_failure:
1108 genlmsg_cancel(skb, ovs_header);
1109 error:
1110 return -EMSGSIZE;
1111 }
1112
1113 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp,
1114 struct genl_info *info, u8 cmd)
1115 {
1116 struct sk_buff *skb;
1117 int retval;
1118
1119 skb = genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info, GFP_KERNEL);
1120 if (!skb)
1121 return ERR_PTR(-ENOMEM);
1122
1123 retval = ovs_dp_cmd_fill_info(dp, skb, info->snd_portid, info->snd_seq, 0, cmd);
1124 if (retval < 0) {
1125 kfree_skb(skb);
1126 return ERR_PTR(retval);
1127 }
1128 return skb;
1129 }
1130
1131 /* Called with ovs_mutex. */
1132 static struct datapath *lookup_datapath(struct net *net,
1133 struct ovs_header *ovs_header,
1134 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1135 {
1136 struct datapath *dp;
1137
1138 if (!a[OVS_DP_ATTR_NAME])
1139 dp = get_dp(net, ovs_header->dp_ifindex);
1140 else {
1141 struct vport *vport;
1142
1143 rcu_read_lock();
1144 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1145 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1146 rcu_read_unlock();
1147 }
1148 return dp ? dp : ERR_PTR(-ENODEV);
1149 }
1150
1151 static void ovs_dp_change(struct datapath *dp, struct nlattr **a)
1152 {
1153 if (a[OVS_DP_ATTR_USER_FEATURES])
1154 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1155 }
1156
1157 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1158 {
1159 struct nlattr **a = info->attrs;
1160 struct vport_parms parms;
1161 struct sk_buff *reply;
1162 struct datapath *dp;
1163 struct vport *vport;
1164 struct ovs_net *ovs_net;
1165 int err, i;
1166
1167 err = -EINVAL;
1168 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1169 goto err;
1170
1171 ovs_lock();
1172
1173 err = -ENOMEM;
1174 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1175 if (dp == NULL)
1176 goto err_unlock_ovs;
1177
1178 ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1179
1180 /* Allocate table. */
1181 err = ovs_flow_tbl_init(&dp->table);
1182 if (err)
1183 goto err_free_dp;
1184
1185 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1186 if (!dp->stats_percpu) {
1187 err = -ENOMEM;
1188 goto err_destroy_table;
1189 }
1190
1191 for_each_possible_cpu(i) {
1192 struct dp_stats_percpu *dpath_stats;
1193 dpath_stats = per_cpu_ptr(dp->stats_percpu, i);
1194 u64_stats_init(&dpath_stats->sync);
1195 }
1196
1197 dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1198 GFP_KERNEL);
1199 if (!dp->ports) {
1200 err = -ENOMEM;
1201 goto err_destroy_percpu;
1202 }
1203
1204 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1205 INIT_HLIST_HEAD(&dp->ports[i]);
1206
1207 /* Set up our datapath device. */
1208 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1209 parms.type = OVS_VPORT_TYPE_INTERNAL;
1210 parms.options = NULL;
1211 parms.dp = dp;
1212 parms.port_no = OVSP_LOCAL;
1213 parms.upcall_portid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1214
1215 ovs_dp_change(dp, a);
1216
1217 vport = new_vport(&parms);
1218 if (IS_ERR(vport)) {
1219 err = PTR_ERR(vport);
1220 if (err == -EBUSY)
1221 err = -EEXIST;
1222
1223 goto err_destroy_ports_array;
1224 }
1225
1226 reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_NEW);
1227 err = PTR_ERR(reply);
1228 if (IS_ERR(reply))
1229 goto err_destroy_local_port;
1230
1231 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1232 list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1233
1234 ovs_unlock();
1235
1236 ovs_notify(&dp_datapath_genl_family, reply, info);
1237 return 0;
1238
1239 err_destroy_local_port:
1240 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1241 err_destroy_ports_array:
1242 kfree(dp->ports);
1243 err_destroy_percpu:
1244 free_percpu(dp->stats_percpu);
1245 err_destroy_table:
1246 ovs_flow_tbl_destroy(&dp->table);
1247 err_free_dp:
1248 release_net(ovs_dp_get_net(dp));
1249 kfree(dp);
1250 err_unlock_ovs:
1251 ovs_unlock();
1252 err:
1253 return err;
1254 }
1255
1256 /* Called with ovs_mutex. */
1257 static void __dp_destroy(struct datapath *dp)
1258 {
1259 int i;
1260
1261 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1262 struct vport *vport;
1263 struct hlist_node *n;
1264
1265 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1266 if (vport->port_no != OVSP_LOCAL)
1267 ovs_dp_detach_port(vport);
1268 }
1269
1270 list_del_rcu(&dp->list_node);
1271
1272 /* OVSP_LOCAL is datapath internal port. We need to make sure that
1273 * all port in datapath are destroyed first before freeing datapath.
1274 */
1275 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1276
1277 call_rcu(&dp->rcu, destroy_dp_rcu);
1278 }
1279
1280 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1281 {
1282 struct sk_buff *reply;
1283 struct datapath *dp;
1284 int err;
1285
1286 ovs_lock();
1287 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1288 err = PTR_ERR(dp);
1289 if (IS_ERR(dp))
1290 goto unlock;
1291
1292 reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_DEL);
1293 err = PTR_ERR(reply);
1294 if (IS_ERR(reply))
1295 goto unlock;
1296
1297 __dp_destroy(dp);
1298 ovs_unlock();
1299
1300 ovs_notify(&dp_datapath_genl_family, reply, info);
1301
1302 return 0;
1303 unlock:
1304 ovs_unlock();
1305 return err;
1306 }
1307
1308 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1309 {
1310 struct sk_buff *reply;
1311 struct datapath *dp;
1312 int err;
1313
1314 ovs_lock();
1315 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1316 err = PTR_ERR(dp);
1317 if (IS_ERR(dp))
1318 goto unlock;
1319
1320 ovs_dp_change(dp, info->attrs);
1321
1322 reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_NEW);
1323 if (IS_ERR(reply)) {
1324 err = PTR_ERR(reply);
1325 genl_set_err(&dp_datapath_genl_family, sock_net(skb->sk), 0,
1326 0, err);
1327 err = 0;
1328 goto unlock;
1329 }
1330
1331 ovs_unlock();
1332 ovs_notify(&dp_datapath_genl_family, reply, info);
1333
1334 return 0;
1335 unlock:
1336 ovs_unlock();
1337 return err;
1338 }
1339
1340 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1341 {
1342 struct sk_buff *reply;
1343 struct datapath *dp;
1344 int err;
1345
1346 ovs_lock();
1347 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1348 if (IS_ERR(dp)) {
1349 err = PTR_ERR(dp);
1350 goto unlock;
1351 }
1352
1353 reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_NEW);
1354 if (IS_ERR(reply)) {
1355 err = PTR_ERR(reply);
1356 goto unlock;
1357 }
1358
1359 ovs_unlock();
1360 return genlmsg_reply(reply, info);
1361
1362 unlock:
1363 ovs_unlock();
1364 return err;
1365 }
1366
1367 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1368 {
1369 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1370 struct datapath *dp;
1371 int skip = cb->args[0];
1372 int i = 0;
1373
1374 rcu_read_lock();
1375 list_for_each_entry_rcu(dp, &ovs_net->dps, list_node) {
1376 if (i >= skip &&
1377 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1378 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1379 OVS_DP_CMD_NEW) < 0)
1380 break;
1381 i++;
1382 }
1383 rcu_read_unlock();
1384
1385 cb->args[0] = i;
1386
1387 return skb->len;
1388 }
1389
1390 static const struct genl_ops dp_datapath_genl_ops[] = {
1391 { .cmd = OVS_DP_CMD_NEW,
1392 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1393 .policy = datapath_policy,
1394 .doit = ovs_dp_cmd_new
1395 },
1396 { .cmd = OVS_DP_CMD_DEL,
1397 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1398 .policy = datapath_policy,
1399 .doit = ovs_dp_cmd_del
1400 },
1401 { .cmd = OVS_DP_CMD_GET,
1402 .flags = 0, /* OK for unprivileged users. */
1403 .policy = datapath_policy,
1404 .doit = ovs_dp_cmd_get,
1405 .dumpit = ovs_dp_cmd_dump
1406 },
1407 { .cmd = OVS_DP_CMD_SET,
1408 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1409 .policy = datapath_policy,
1410 .doit = ovs_dp_cmd_set,
1411 },
1412 };
1413
1414 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1415 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1416 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1417 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1418 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1419 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1420 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1421 };
1422
1423 struct genl_family dp_vport_genl_family = {
1424 .id = GENL_ID_GENERATE,
1425 .hdrsize = sizeof(struct ovs_header),
1426 .name = OVS_VPORT_FAMILY,
1427 .version = OVS_VPORT_VERSION,
1428 .maxattr = OVS_VPORT_ATTR_MAX,
1429 .netnsok = true,
1430 .parallel_ops = true,
1431 };
1432
1433 struct genl_multicast_group ovs_dp_vport_multicast_group = {
1434 .name = OVS_VPORT_MCGROUP
1435 };
1436
1437 /* Called with ovs_mutex or RCU read lock. */
1438 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1439 u32 portid, u32 seq, u32 flags, u8 cmd)
1440 {
1441 struct ovs_header *ovs_header;
1442 struct ovs_vport_stats vport_stats;
1443 int err;
1444
1445 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1446 flags, cmd);
1447 if (!ovs_header)
1448 return -EMSGSIZE;
1449
1450 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1451
1452 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1453 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1454 nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1455 nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_portid))
1456 goto nla_put_failure;
1457
1458 ovs_vport_get_stats(vport, &vport_stats);
1459 if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1460 &vport_stats))
1461 goto nla_put_failure;
1462
1463 err = ovs_vport_get_options(vport, skb);
1464 if (err == -EMSGSIZE)
1465 goto error;
1466
1467 return genlmsg_end(skb, ovs_header);
1468
1469 nla_put_failure:
1470 err = -EMSGSIZE;
1471 error:
1472 genlmsg_cancel(skb, ovs_header);
1473 return err;
1474 }
1475
1476 /* Called with ovs_mutex or RCU read lock. */
1477 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1478 u32 seq, u8 cmd)
1479 {
1480 struct sk_buff *skb;
1481 int retval;
1482
1483 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1484 if (!skb)
1485 return ERR_PTR(-ENOMEM);
1486
1487 retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1488 BUG_ON(retval < 0);
1489
1490 return skb;
1491 }
1492
1493 /* Called with ovs_mutex or RCU read lock. */
1494 static struct vport *lookup_vport(struct net *net,
1495 struct ovs_header *ovs_header,
1496 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1497 {
1498 struct datapath *dp;
1499 struct vport *vport;
1500
1501 if (a[OVS_VPORT_ATTR_NAME]) {
1502 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1503 if (!vport)
1504 return ERR_PTR(-ENODEV);
1505 if (ovs_header->dp_ifindex &&
1506 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1507 return ERR_PTR(-ENODEV);
1508 return vport;
1509 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1510 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1511
1512 if (port_no >= DP_MAX_PORTS)
1513 return ERR_PTR(-EFBIG);
1514
1515 dp = get_dp(net, ovs_header->dp_ifindex);
1516 if (!dp)
1517 return ERR_PTR(-ENODEV);
1518
1519 vport = ovs_vport_ovsl_rcu(dp, port_no);
1520 if (!vport)
1521 return ERR_PTR(-ENODEV);
1522 return vport;
1523 } else
1524 return ERR_PTR(-EINVAL);
1525 }
1526
1527 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1528 {
1529 struct nlattr **a = info->attrs;
1530 struct ovs_header *ovs_header = info->userhdr;
1531 struct vport_parms parms;
1532 struct sk_buff *reply;
1533 struct vport *vport;
1534 struct datapath *dp;
1535 u32 port_no;
1536 int err;
1537
1538 err = -EINVAL;
1539 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1540 !a[OVS_VPORT_ATTR_UPCALL_PID])
1541 goto exit;
1542
1543 ovs_lock();
1544 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1545 err = -ENODEV;
1546 if (!dp)
1547 goto exit_unlock;
1548
1549 if (a[OVS_VPORT_ATTR_PORT_NO]) {
1550 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1551
1552 err = -EFBIG;
1553 if (port_no >= DP_MAX_PORTS)
1554 goto exit_unlock;
1555
1556 vport = ovs_vport_ovsl(dp, port_no);
1557 err = -EBUSY;
1558 if (vport)
1559 goto exit_unlock;
1560 } else {
1561 for (port_no = 1; ; port_no++) {
1562 if (port_no >= DP_MAX_PORTS) {
1563 err = -EFBIG;
1564 goto exit_unlock;
1565 }
1566 vport = ovs_vport_ovsl(dp, port_no);
1567 if (!vport)
1568 break;
1569 }
1570 }
1571
1572 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1573 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1574 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1575 parms.dp = dp;
1576 parms.port_no = port_no;
1577 parms.upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1578
1579 vport = new_vport(&parms);
1580 err = PTR_ERR(vport);
1581 if (IS_ERR(vport))
1582 goto exit_unlock;
1583
1584 err = 0;
1585 reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
1586 OVS_VPORT_CMD_NEW);
1587 if (IS_ERR(reply)) {
1588 err = PTR_ERR(reply);
1589 ovs_dp_detach_port(vport);
1590 goto exit_unlock;
1591 }
1592
1593 ovs_notify(&dp_vport_genl_family, reply, info);
1594
1595 exit_unlock:
1596 ovs_unlock();
1597 exit:
1598 return err;
1599 }
1600
1601 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1602 {
1603 struct nlattr **a = info->attrs;
1604 struct sk_buff *reply;
1605 struct vport *vport;
1606 int err;
1607
1608 ovs_lock();
1609 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1610 err = PTR_ERR(vport);
1611 if (IS_ERR(vport))
1612 goto exit_unlock;
1613
1614 if (a[OVS_VPORT_ATTR_TYPE] &&
1615 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
1616 err = -EINVAL;
1617 goto exit_unlock;
1618 }
1619
1620 reply = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1621 if (!reply) {
1622 err = -ENOMEM;
1623 goto exit_unlock;
1624 }
1625
1626 if (a[OVS_VPORT_ATTR_OPTIONS]) {
1627 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1628 if (err)
1629 goto exit_free;
1630 }
1631
1632 if (a[OVS_VPORT_ATTR_UPCALL_PID])
1633 vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1634
1635 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1636 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1637 BUG_ON(err < 0);
1638
1639 ovs_unlock();
1640 ovs_notify(&dp_vport_genl_family, reply, info);
1641 return 0;
1642
1643 exit_free:
1644 kfree_skb(reply);
1645 exit_unlock:
1646 ovs_unlock();
1647 return err;
1648 }
1649
1650 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1651 {
1652 struct nlattr **a = info->attrs;
1653 struct sk_buff *reply;
1654 struct vport *vport;
1655 int err;
1656
1657 ovs_lock();
1658 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1659 err = PTR_ERR(vport);
1660 if (IS_ERR(vport))
1661 goto exit_unlock;
1662
1663 if (vport->port_no == OVSP_LOCAL) {
1664 err = -EINVAL;
1665 goto exit_unlock;
1666 }
1667
1668 reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
1669 info->snd_seq, OVS_VPORT_CMD_DEL);
1670 err = PTR_ERR(reply);
1671 if (IS_ERR(reply))
1672 goto exit_unlock;
1673
1674 err = 0;
1675 ovs_dp_detach_port(vport);
1676
1677 ovs_notify(&dp_vport_genl_family, reply, info);
1678
1679 exit_unlock:
1680 ovs_unlock();
1681 return err;
1682 }
1683
1684 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1685 {
1686 struct nlattr **a = info->attrs;
1687 struct ovs_header *ovs_header = info->userhdr;
1688 struct sk_buff *reply;
1689 struct vport *vport;
1690 int err;
1691
1692 rcu_read_lock();
1693 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1694 err = PTR_ERR(vport);
1695 if (IS_ERR(vport))
1696 goto exit_unlock;
1697
1698 reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
1699 info->snd_seq, OVS_VPORT_CMD_NEW);
1700 err = PTR_ERR(reply);
1701 if (IS_ERR(reply))
1702 goto exit_unlock;
1703
1704 rcu_read_unlock();
1705
1706 return genlmsg_reply(reply, info);
1707
1708 exit_unlock:
1709 rcu_read_unlock();
1710 return err;
1711 }
1712
1713 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1714 {
1715 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1716 struct datapath *dp;
1717 int bucket = cb->args[0], skip = cb->args[1];
1718 int i, j = 0;
1719
1720 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1721 if (!dp)
1722 return -ENODEV;
1723
1724 rcu_read_lock();
1725 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
1726 struct vport *vport;
1727
1728 j = 0;
1729 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1730 if (j >= skip &&
1731 ovs_vport_cmd_fill_info(vport, skb,
1732 NETLINK_CB(cb->skb).portid,
1733 cb->nlh->nlmsg_seq,
1734 NLM_F_MULTI,
1735 OVS_VPORT_CMD_NEW) < 0)
1736 goto out;
1737
1738 j++;
1739 }
1740 skip = 0;
1741 }
1742 out:
1743 rcu_read_unlock();
1744
1745 cb->args[0] = i;
1746 cb->args[1] = j;
1747
1748 return skb->len;
1749 }
1750
1751 static const struct genl_ops dp_vport_genl_ops[] = {
1752 { .cmd = OVS_VPORT_CMD_NEW,
1753 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1754 .policy = vport_policy,
1755 .doit = ovs_vport_cmd_new
1756 },
1757 { .cmd = OVS_VPORT_CMD_DEL,
1758 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1759 .policy = vport_policy,
1760 .doit = ovs_vport_cmd_del
1761 },
1762 { .cmd = OVS_VPORT_CMD_GET,
1763 .flags = 0, /* OK for unprivileged users. */
1764 .policy = vport_policy,
1765 .doit = ovs_vport_cmd_get,
1766 .dumpit = ovs_vport_cmd_dump
1767 },
1768 { .cmd = OVS_VPORT_CMD_SET,
1769 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1770 .policy = vport_policy,
1771 .doit = ovs_vport_cmd_set,
1772 },
1773 };
1774
1775 struct genl_family_and_ops {
1776 struct genl_family *family;
1777 const struct genl_ops *ops;
1778 int n_ops;
1779 const struct genl_multicast_group *group;
1780 };
1781
1782 static const struct genl_family_and_ops dp_genl_families[] = {
1783 { &dp_datapath_genl_family,
1784 dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1785 &ovs_dp_datapath_multicast_group },
1786 { &dp_vport_genl_family,
1787 dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1788 &ovs_dp_vport_multicast_group },
1789 { &dp_flow_genl_family,
1790 dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1791 &ovs_dp_flow_multicast_group },
1792 { &dp_packet_genl_family,
1793 dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1794 NULL },
1795 };
1796
1797 static void dp_unregister_genl(int n_families)
1798 {
1799 int i;
1800
1801 for (i = 0; i < n_families; i++)
1802 genl_unregister_family(dp_genl_families[i].family);
1803 }
1804
1805 static int dp_register_genl(void)
1806 {
1807 int n_registered;
1808 int err;
1809 int i;
1810
1811 n_registered = 0;
1812 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
1813 const struct genl_family_and_ops *f = &dp_genl_families[i];
1814
1815 f->family->ops = f->ops;
1816 f->family->n_ops = f->n_ops;
1817 f->family->mcgrps = f->group;
1818 f->family->n_mcgrps = f->group ? 1 : 0;
1819 err = genl_register_family(f->family);
1820 if (err)
1821 goto error;
1822 n_registered++;
1823 }
1824
1825 return 0;
1826
1827 error:
1828 dp_unregister_genl(n_registered);
1829 return err;
1830 }
1831
1832 static int __net_init ovs_init_net(struct net *net)
1833 {
1834 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1835
1836 INIT_LIST_HEAD(&ovs_net->dps);
1837 INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
1838 return 0;
1839 }
1840
1841 static void __net_exit ovs_exit_net(struct net *net)
1842 {
1843 struct datapath *dp, *dp_next;
1844 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1845
1846 ovs_lock();
1847 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
1848 __dp_destroy(dp);
1849 ovs_unlock();
1850
1851 cancel_work_sync(&ovs_net->dp_notify_work);
1852 }
1853
1854 static struct pernet_operations ovs_net_ops = {
1855 .init = ovs_init_net,
1856 .exit = ovs_exit_net,
1857 .id = &ovs_net_id,
1858 .size = sizeof(struct ovs_net),
1859 };
1860
1861 static int __init dp_init(void)
1862 {
1863 int err;
1864
1865 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
1866
1867 pr_info("Open vSwitch switching datapath\n");
1868
1869 err = ovs_flow_init();
1870 if (err)
1871 goto error;
1872
1873 err = ovs_vport_init();
1874 if (err)
1875 goto error_flow_exit;
1876
1877 err = register_pernet_device(&ovs_net_ops);
1878 if (err)
1879 goto error_vport_exit;
1880
1881 err = register_netdevice_notifier(&ovs_dp_device_notifier);
1882 if (err)
1883 goto error_netns_exit;
1884
1885 err = dp_register_genl();
1886 if (err < 0)
1887 goto error_unreg_notifier;
1888
1889 return 0;
1890
1891 error_unreg_notifier:
1892 unregister_netdevice_notifier(&ovs_dp_device_notifier);
1893 error_netns_exit:
1894 unregister_pernet_device(&ovs_net_ops);
1895 error_vport_exit:
1896 ovs_vport_exit();
1897 error_flow_exit:
1898 ovs_flow_exit();
1899 error:
1900 return err;
1901 }
1902
1903 static void dp_cleanup(void)
1904 {
1905 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
1906 unregister_netdevice_notifier(&ovs_dp_device_notifier);
1907 unregister_pernet_device(&ovs_net_ops);
1908 rcu_barrier();
1909 ovs_vport_exit();
1910 ovs_flow_exit();
1911 }
1912
1913 module_init(dp_init);
1914 module_exit(dp_cleanup);
1915
1916 MODULE_DESCRIPTION("Open vSwitch switching datapath");
1917 MODULE_LICENSE("GPL");
ubuntu-zesty-kernel mirror