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