projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
genetlink: pass family to functions using groups
[deliverable/linux.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 struct genl_multicast_group *grp)
67 {
68 genl_notify(family, skb, genl_info_net(info), info->snd_portid,
69 grp->id, info->nlhdr, GFP_KERNEL);
70 }
71
72 /**
73 * DOC: Locking:
74 *
75 * All writes e.g. Writes to device state (add/remove datapath, port, set
76 * operations on vports, etc.), Writes to other state (flow table
77 * modifications, set miscellaneous datapath parameters, etc.) are protected
78 * by ovs_lock.
79 *
80 * Reads are protected by RCU.
81 *
82 * There are a few special cases (mostly stats) that have their own
83 * synchronization but they nest under all of above and don't interact with
84 * each other.
85 *
86 * The RTNL lock nests inside ovs_mutex.
87 */
88
89 static DEFINE_MUTEX(ovs_mutex);
90
91 void ovs_lock(void)
92 {
93 mutex_lock(&ovs_mutex);
94 }
95
96 void ovs_unlock(void)
97 {
98 mutex_unlock(&ovs_mutex);
99 }
100
101 #ifdef CONFIG_LOCKDEP
102 int lockdep_ovsl_is_held(void)
103 {
104 if (debug_locks)
105 return lockdep_is_held(&ovs_mutex);
106 else
107 return 1;
108 }
109 #endif
110
111 static struct vport *new_vport(const struct vport_parms *);
112 static int queue_gso_packets(struct net *, int dp_ifindex, struct sk_buff *,
113 const struct dp_upcall_info *);
114 static int queue_userspace_packet(struct net *, int dp_ifindex,
115 struct sk_buff *,
116 const struct dp_upcall_info *);
117
118 /* Must be called with rcu_read_lock or ovs_mutex. */
119 static struct datapath *get_dp(struct net *net, int dp_ifindex)
120 {
121 struct datapath *dp = NULL;
122 struct net_device *dev;
123
124 rcu_read_lock();
125 dev = dev_get_by_index_rcu(net, dp_ifindex);
126 if (dev) {
127 struct vport *vport = ovs_internal_dev_get_vport(dev);
128 if (vport)
129 dp = vport->dp;
130 }
131 rcu_read_unlock();
132
133 return dp;
134 }
135
136 /* Must be called with rcu_read_lock or ovs_mutex. */
137 const char *ovs_dp_name(const struct datapath *dp)
138 {
139 struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
140 return vport->ops->get_name(vport);
141 }
142
143 static int get_dpifindex(struct datapath *dp)
144 {
145 struct vport *local;
146 int ifindex;
147
148 rcu_read_lock();
149
150 local = ovs_vport_rcu(dp, OVSP_LOCAL);
151 if (local)
152 ifindex = netdev_vport_priv(local)->dev->ifindex;
153 else
154 ifindex = 0;
155
156 rcu_read_unlock();
157
158 return ifindex;
159 }
160
161 static void destroy_dp_rcu(struct rcu_head *rcu)
162 {
163 struct datapath *dp = container_of(rcu, struct datapath, rcu);
164
165 ovs_flow_tbl_destroy(&dp->table);
166 free_percpu(dp->stats_percpu);
167 release_net(ovs_dp_get_net(dp));
168 kfree(dp->ports);
169 kfree(dp);
170 }
171
172 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
173 u16 port_no)
174 {
175 return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
176 }
177
178 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
179 {
180 struct vport *vport;
181 struct hlist_head *head;
182
183 head = vport_hash_bucket(dp, port_no);
184 hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
185 if (vport->port_no == port_no)
186 return vport;
187 }
188 return NULL;
189 }
190
191 /* Called with ovs_mutex. */
192 static struct vport *new_vport(const struct vport_parms *parms)
193 {
194 struct vport *vport;
195
196 vport = ovs_vport_add(parms);
197 if (!IS_ERR(vport)) {
198 struct datapath *dp = parms->dp;
199 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
200
201 hlist_add_head_rcu(&vport->dp_hash_node, head);
202 }
203 return vport;
204 }
205
206 void ovs_dp_detach_port(struct vport *p)
207 {
208 ASSERT_OVSL();
209
210 /* First drop references to device. */
211 hlist_del_rcu(&p->dp_hash_node);
212
213 /* Then destroy it. */
214 ovs_vport_del(p);
215 }
216
217 /* Must be called with rcu_read_lock. */
218 void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
219 {
220 struct datapath *dp = p->dp;
221 struct sw_flow *flow;
222 struct dp_stats_percpu *stats;
223 struct sw_flow_key key;
224 u64 *stats_counter;
225 u32 n_mask_hit;
226 int error;
227
228 stats = this_cpu_ptr(dp->stats_percpu);
229
230 /* Extract flow from 'skb' into 'key'. */
231 error = ovs_flow_extract(skb, p->port_no, &key);
232 if (unlikely(error)) {
233 kfree_skb(skb);
234 return;
235 }
236
237 /* Look up flow. */
238 flow = ovs_flow_tbl_lookup(&dp->table, &key, &n_mask_hit);
239 if (unlikely(!flow)) {
240 struct dp_upcall_info upcall;
241
242 upcall.cmd = OVS_PACKET_CMD_MISS;
243 upcall.key = &key;
244 upcall.userdata = NULL;
245 upcall.portid = p->upcall_portid;
246 ovs_dp_upcall(dp, skb, &upcall);
247 consume_skb(skb);
248 stats_counter = &stats->n_missed;
249 goto out;
250 }
251
252 OVS_CB(skb)->flow = flow;
253 OVS_CB(skb)->pkt_key = &key;
254
255 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 stats->n_mask_hit += n_mask_hit;
264 u64_stats_update_end(&stats->sync);
265 }
266
267 static struct genl_family dp_packet_genl_family = {
268 .id = GENL_ID_GENERATE,
269 .hdrsize = sizeof(struct ovs_header),
270 .name = OVS_PACKET_FAMILY,
271 .version = OVS_PACKET_VERSION,
272 .maxattr = OVS_PACKET_ATTR_MAX,
273 .netnsok = true,
274 .parallel_ops = true,
275 };
276
277 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
278 const struct dp_upcall_info *upcall_info)
279 {
280 struct dp_stats_percpu *stats;
281 int dp_ifindex;
282 int err;
283
284 if (upcall_info->portid == 0) {
285 err = -ENOTCONN;
286 goto err;
287 }
288
289 dp_ifindex = get_dpifindex(dp);
290 if (!dp_ifindex) {
291 err = -ENODEV;
292 goto err;
293 }
294
295 if (!skb_is_gso(skb))
296 err = queue_userspace_packet(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
297 else
298 err = queue_gso_packets(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
299 if (err)
300 goto err;
301
302 return 0;
303
304 err:
305 stats = this_cpu_ptr(dp->stats_percpu);
306
307 u64_stats_update_begin(&stats->sync);
308 stats->n_lost++;
309 u64_stats_update_end(&stats->sync);
310
311 return err;
312 }
313
314 static int queue_gso_packets(struct net *net, int dp_ifindex,
315 struct sk_buff *skb,
316 const struct dp_upcall_info *upcall_info)
317 {
318 unsigned short gso_type = skb_shinfo(skb)->gso_type;
319 struct dp_upcall_info later_info;
320 struct sw_flow_key later_key;
321 struct sk_buff *segs, *nskb;
322 int err;
323
324 segs = __skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM, false);
325 if (IS_ERR(segs))
326 return PTR_ERR(segs);
327
328 /* Queue all of the segments. */
329 skb = segs;
330 do {
331 err = queue_userspace_packet(net, dp_ifindex, skb, upcall_info);
332 if (err)
333 break;
334
335 if (skb == segs && gso_type & SKB_GSO_UDP) {
336 /* The initial flow key extracted by ovs_flow_extract()
337 * in this case is for a first fragment, so we need to
338 * properly mark later fragments.
339 */
340 later_key = *upcall_info->key;
341 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
342
343 later_info = *upcall_info;
344 later_info.key = &later_key;
345 upcall_info = &later_info;
346 }
347 } while ((skb = skb->next));
348
349 /* Free all of the segments. */
350 skb = segs;
351 do {
352 nskb = skb->next;
353 if (err)
354 kfree_skb(skb);
355 else
356 consume_skb(skb);
357 } while ((skb = nskb));
358 return err;
359 }
360
361 static size_t key_attr_size(void)
362 {
363 return nla_total_size(4) /* OVS_KEY_ATTR_PRIORITY */
364 + nla_total_size(0) /* OVS_KEY_ATTR_TUNNEL */
365 + nla_total_size(8) /* OVS_TUNNEL_KEY_ATTR_ID */
366 + nla_total_size(4) /* OVS_TUNNEL_KEY_ATTR_IPV4_SRC */
367 + nla_total_size(4) /* OVS_TUNNEL_KEY_ATTR_IPV4_DST */
368 + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TOS */
369 + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TTL */
370 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
371 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_CSUM */
372 + nla_total_size(4) /* OVS_KEY_ATTR_IN_PORT */
373 + nla_total_size(4) /* OVS_KEY_ATTR_SKB_MARK */
374 + nla_total_size(12) /* OVS_KEY_ATTR_ETHERNET */
375 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
376 + nla_total_size(4) /* OVS_KEY_ATTR_8021Q */
377 + nla_total_size(0) /* OVS_KEY_ATTR_ENCAP */
378 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
379 + nla_total_size(40) /* OVS_KEY_ATTR_IPV6 */
380 + nla_total_size(2) /* OVS_KEY_ATTR_ICMPV6 */
381 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
382 }
383
384 static size_t upcall_msg_size(const struct sk_buff *skb,
385 const struct nlattr *userdata)
386 {
387 size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
388 + nla_total_size(skb->len) /* OVS_PACKET_ATTR_PACKET */
389 + nla_total_size(key_attr_size()); /* OVS_PACKET_ATTR_KEY */
390
391 /* OVS_PACKET_ATTR_USERDATA */
392 if (userdata)
393 size += NLA_ALIGN(userdata->nla_len);
394
395 return size;
396 }
397
398 static int queue_userspace_packet(struct net *net, int dp_ifindex,
399 struct sk_buff *skb,
400 const struct dp_upcall_info *upcall_info)
401 {
402 struct ovs_header *upcall;
403 struct sk_buff *nskb = NULL;
404 struct sk_buff *user_skb; /* to be queued to userspace */
405 struct nlattr *nla;
406 int err;
407
408 if (vlan_tx_tag_present(skb)) {
409 nskb = skb_clone(skb, GFP_ATOMIC);
410 if (!nskb)
411 return -ENOMEM;
412
413 nskb = __vlan_put_tag(nskb, nskb->vlan_proto, vlan_tx_tag_get(nskb));
414 if (!nskb)
415 return -ENOMEM;
416
417 nskb->vlan_tci = 0;
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 const 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 struct ovs_dp_megaflow_stats *mega_stats)
571 {
572 int i;
573
574 memset(mega_stats, 0, sizeof(*mega_stats));
575
576 stats->n_flows = ovs_flow_tbl_count(&dp->table);
577 mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
578
579 stats->n_hit = stats->n_missed = stats->n_lost = 0;
580
581 for_each_possible_cpu(i) {
582 const struct dp_stats_percpu *percpu_stats;
583 struct dp_stats_percpu local_stats;
584 unsigned int start;
585
586 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
587
588 do {
589 start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
590 local_stats = *percpu_stats;
591 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
592
593 stats->n_hit += local_stats.n_hit;
594 stats->n_missed += local_stats.n_missed;
595 stats->n_lost += local_stats.n_lost;
596 mega_stats->n_mask_hit += local_stats.n_mask_hit;
597 }
598 }
599
600 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
601 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
602 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
603 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
604 };
605
606 static struct genl_family dp_flow_genl_family = {
607 .id = GENL_ID_GENERATE,
608 .hdrsize = sizeof(struct ovs_header),
609 .name = OVS_FLOW_FAMILY,
610 .version = OVS_FLOW_VERSION,
611 .maxattr = OVS_FLOW_ATTR_MAX,
612 .netnsok = true,
613 .parallel_ops = true,
614 };
615
616 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
617 .name = OVS_FLOW_MCGROUP
618 };
619
620 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
621 {
622 return NLMSG_ALIGN(sizeof(struct ovs_header))
623 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
624 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_MASK */
625 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
626 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
627 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
628 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
629 }
630
631 /* Called with ovs_mutex. */
632 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
633 struct sk_buff *skb, u32 portid,
634 u32 seq, u32 flags, u8 cmd)
635 {
636 const int skb_orig_len = skb->len;
637 struct nlattr *start;
638 struct ovs_flow_stats stats;
639 struct ovs_header *ovs_header;
640 struct nlattr *nla;
641 unsigned long used;
642 u8 tcp_flags;
643 int err;
644
645 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
646 if (!ovs_header)
647 return -EMSGSIZE;
648
649 ovs_header->dp_ifindex = get_dpifindex(dp);
650
651 /* Fill flow key. */
652 nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
653 if (!nla)
654 goto nla_put_failure;
655
656 err = ovs_nla_put_flow(&flow->unmasked_key, &flow->unmasked_key, skb);
657 if (err)
658 goto error;
659 nla_nest_end(skb, nla);
660
661 nla = nla_nest_start(skb, OVS_FLOW_ATTR_MASK);
662 if (!nla)
663 goto nla_put_failure;
664
665 err = ovs_nla_put_flow(&flow->key, &flow->mask->key, skb);
666 if (err)
667 goto error;
668
669 nla_nest_end(skb, nla);
670
671 spin_lock_bh(&flow->lock);
672 used = flow->used;
673 stats.n_packets = flow->packet_count;
674 stats.n_bytes = flow->byte_count;
675 tcp_flags = (u8)ntohs(flow->tcp_flags);
676 spin_unlock_bh(&flow->lock);
677
678 if (used &&
679 nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
680 goto nla_put_failure;
681
682 if (stats.n_packets &&
683 nla_put(skb, OVS_FLOW_ATTR_STATS,
684 sizeof(struct ovs_flow_stats), &stats))
685 goto nla_put_failure;
686
687 if (tcp_flags &&
688 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags))
689 goto nla_put_failure;
690
691 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
692 * this is the first flow to be dumped into 'skb'. This is unusual for
693 * Netlink but individual action lists can be longer than
694 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
695 * The userspace caller can always fetch the actions separately if it
696 * really wants them. (Most userspace callers in fact don't care.)
697 *
698 * This can only fail for dump operations because the skb is always
699 * properly sized for single flows.
700 */
701 start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
702 if (start) {
703 const struct sw_flow_actions *sf_acts;
704
705 sf_acts = rcu_dereference_check(flow->sf_acts,
706 lockdep_ovsl_is_held());
707
708 err = ovs_nla_put_actions(sf_acts->actions,
709 sf_acts->actions_len, skb);
710 if (!err)
711 nla_nest_end(skb, start);
712 else {
713 if (skb_orig_len)
714 goto error;
715
716 nla_nest_cancel(skb, start);
717 }
718 } else if (skb_orig_len)
719 goto nla_put_failure;
720
721 return genlmsg_end(skb, ovs_header);
722
723 nla_put_failure:
724 err = -EMSGSIZE;
725 error:
726 genlmsg_cancel(skb, ovs_header);
727 return err;
728 }
729
730 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
731 {
732 const struct sw_flow_actions *sf_acts;
733
734 sf_acts = ovsl_dereference(flow->sf_acts);
735
736 return genlmsg_new(ovs_flow_cmd_msg_size(sf_acts), GFP_KERNEL);
737 }
738
739 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
740 struct datapath *dp,
741 u32 portid, u32 seq, u8 cmd)
742 {
743 struct sk_buff *skb;
744 int retval;
745
746 skb = ovs_flow_cmd_alloc_info(flow);
747 if (!skb)
748 return ERR_PTR(-ENOMEM);
749
750 retval = ovs_flow_cmd_fill_info(flow, dp, skb, portid, seq, 0, cmd);
751 BUG_ON(retval < 0);
752 return skb;
753 }
754
755 static struct sw_flow *__ovs_flow_tbl_lookup(struct flow_table *tbl,
756 const struct sw_flow_key *key)
757 {
758 u32 __always_unused n_mask_hit;
759
760 return ovs_flow_tbl_lookup(tbl, key, &n_mask_hit);
761 }
762
763 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
764 {
765 struct nlattr **a = info->attrs;
766 struct ovs_header *ovs_header = info->userhdr;
767 struct sw_flow_key key, masked_key;
768 struct sw_flow *flow = NULL;
769 struct sw_flow_mask mask;
770 struct sk_buff *reply;
771 struct datapath *dp;
772 struct sw_flow_actions *acts = NULL;
773 struct sw_flow_match match;
774 int error;
775
776 /* Extract key. */
777 error = -EINVAL;
778 if (!a[OVS_FLOW_ATTR_KEY])
779 goto error;
780
781 ovs_match_init(&match, &key, &mask);
782 error = ovs_nla_get_match(&match,
783 a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
784 if (error)
785 goto error;
786
787 /* Validate actions. */
788 if (a[OVS_FLOW_ATTR_ACTIONS]) {
789 acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
790 error = PTR_ERR(acts);
791 if (IS_ERR(acts))
792 goto error;
793
794 ovs_flow_mask_key(&masked_key, &key, &mask);
795 error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS],
796 &masked_key, 0, &acts);
797 if (error) {
798 OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
799 goto err_kfree;
800 }
801 } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
802 error = -EINVAL;
803 goto error;
804 }
805
806 ovs_lock();
807 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
808 error = -ENODEV;
809 if (!dp)
810 goto err_unlock_ovs;
811
812 /* Check if this is a duplicate flow */
813 flow = __ovs_flow_tbl_lookup(&dp->table, &key);
814 if (!flow) {
815 /* Bail out if we're not allowed to create a new flow. */
816 error = -ENOENT;
817 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
818 goto err_unlock_ovs;
819
820 /* Allocate flow. */
821 flow = ovs_flow_alloc();
822 if (IS_ERR(flow)) {
823 error = PTR_ERR(flow);
824 goto err_unlock_ovs;
825 }
826 clear_stats(flow);
827
828 flow->key = masked_key;
829 flow->unmasked_key = key;
830 rcu_assign_pointer(flow->sf_acts, acts);
831
832 /* Put flow in bucket. */
833 error = ovs_flow_tbl_insert(&dp->table, flow, &mask);
834 if (error) {
835 acts = NULL;
836 goto err_flow_free;
837 }
838
839 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
840 info->snd_seq, OVS_FLOW_CMD_NEW);
841 } else {
842 /* We found a matching flow. */
843 struct sw_flow_actions *old_acts;
844
845 /* Bail out if we're not allowed to modify an existing flow.
846 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
847 * because Generic Netlink treats the latter as a dump
848 * request. We also accept NLM_F_EXCL in case that bug ever
849 * gets fixed.
850 */
851 error = -EEXIST;
852 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
853 info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
854 goto err_unlock_ovs;
855
856 /* The unmasked key has to be the same for flow updates. */
857 error = -EINVAL;
858 if (!ovs_flow_cmp_unmasked_key(flow, &match)) {
859 OVS_NLERR("Flow modification message rejected, unmasked key does not match.\n");
860 goto err_unlock_ovs;
861 }
862
863 /* Update actions. */
864 old_acts = ovsl_dereference(flow->sf_acts);
865 rcu_assign_pointer(flow->sf_acts, acts);
866 ovs_nla_free_flow_actions(old_acts);
867
868 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
869 info->snd_seq, OVS_FLOW_CMD_NEW);
870
871 /* Clear stats. */
872 if (a[OVS_FLOW_ATTR_CLEAR]) {
873 spin_lock_bh(&flow->lock);
874 clear_stats(flow);
875 spin_unlock_bh(&flow->lock);
876 }
877 }
878 ovs_unlock();
879
880 if (!IS_ERR(reply))
881 ovs_notify(&dp_flow_genl_family, reply, info,
882 &ovs_dp_flow_multicast_group);
883 else
884 genl_set_err(&dp_flow_genl_family, sock_net(skb->sk), 0,
885 ovs_dp_flow_multicast_group.id, PTR_ERR(reply));
886 return 0;
887
888 err_flow_free:
889 ovs_flow_free(flow, false);
890 err_unlock_ovs:
891 ovs_unlock();
892 err_kfree:
893 kfree(acts);
894 error:
895 return error;
896 }
897
898 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
899 {
900 struct nlattr **a = info->attrs;
901 struct ovs_header *ovs_header = info->userhdr;
902 struct sw_flow_key key;
903 struct sk_buff *reply;
904 struct sw_flow *flow;
905 struct datapath *dp;
906 struct sw_flow_match match;
907 int err;
908
909 if (!a[OVS_FLOW_ATTR_KEY]) {
910 OVS_NLERR("Flow get message rejected, Key attribute missing.\n");
911 return -EINVAL;
912 }
913
914 ovs_match_init(&match, &key, NULL);
915 err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
916 if (err)
917 return err;
918
919 ovs_lock();
920 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
921 if (!dp) {
922 err = -ENODEV;
923 goto unlock;
924 }
925
926 flow = __ovs_flow_tbl_lookup(&dp->table, &key);
927 if (!flow || !ovs_flow_cmp_unmasked_key(flow, &match)) {
928 err = -ENOENT;
929 goto unlock;
930 }
931
932 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
933 info->snd_seq, OVS_FLOW_CMD_NEW);
934 if (IS_ERR(reply)) {
935 err = PTR_ERR(reply);
936 goto unlock;
937 }
938
939 ovs_unlock();
940 return genlmsg_reply(reply, info);
941 unlock:
942 ovs_unlock();
943 return err;
944 }
945
946 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
947 {
948 struct nlattr **a = info->attrs;
949 struct ovs_header *ovs_header = info->userhdr;
950 struct sw_flow_key key;
951 struct sk_buff *reply;
952 struct sw_flow *flow;
953 struct datapath *dp;
954 struct sw_flow_match match;
955 int err;
956
957 ovs_lock();
958 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
959 if (!dp) {
960 err = -ENODEV;
961 goto unlock;
962 }
963
964 if (!a[OVS_FLOW_ATTR_KEY]) {
965 err = ovs_flow_tbl_flush(&dp->table);
966 goto unlock;
967 }
968
969 ovs_match_init(&match, &key, NULL);
970 err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
971 if (err)
972 goto unlock;
973
974 flow = __ovs_flow_tbl_lookup(&dp->table, &key);
975 if (!flow || !ovs_flow_cmp_unmasked_key(flow, &match)) {
976 err = -ENOENT;
977 goto unlock;
978 }
979
980 reply = ovs_flow_cmd_alloc_info(flow);
981 if (!reply) {
982 err = -ENOMEM;
983 goto unlock;
984 }
985
986 ovs_flow_tbl_remove(&dp->table, flow);
987
988 err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_portid,
989 info->snd_seq, 0, OVS_FLOW_CMD_DEL);
990 BUG_ON(err < 0);
991
992 ovs_flow_free(flow, true);
993 ovs_unlock();
994
995 ovs_notify(&dp_flow_genl_family, reply, info,
996 &ovs_dp_flow_multicast_group);
997 return 0;
998 unlock:
999 ovs_unlock();
1000 return err;
1001 }
1002
1003 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1004 {
1005 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1006 struct table_instance *ti;
1007 struct datapath *dp;
1008
1009 rcu_read_lock();
1010 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1011 if (!dp) {
1012 rcu_read_unlock();
1013 return -ENODEV;
1014 }
1015
1016 ti = rcu_dereference(dp->table.ti);
1017 for (;;) {
1018 struct sw_flow *flow;
1019 u32 bucket, obj;
1020
1021 bucket = cb->args[0];
1022 obj = cb->args[1];
1023 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1024 if (!flow)
1025 break;
1026
1027 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1028 NETLINK_CB(cb->skb).portid,
1029 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1030 OVS_FLOW_CMD_NEW) < 0)
1031 break;
1032
1033 cb->args[0] = bucket;
1034 cb->args[1] = obj;
1035 }
1036 rcu_read_unlock();
1037 return skb->len;
1038 }
1039
1040 static const struct genl_ops dp_flow_genl_ops[] = {
1041 { .cmd = OVS_FLOW_CMD_NEW,
1042 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1043 .policy = flow_policy,
1044 .doit = ovs_flow_cmd_new_or_set
1045 },
1046 { .cmd = OVS_FLOW_CMD_DEL,
1047 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1048 .policy = flow_policy,
1049 .doit = ovs_flow_cmd_del
1050 },
1051 { .cmd = OVS_FLOW_CMD_GET,
1052 .flags = 0, /* OK for unprivileged users. */
1053 .policy = flow_policy,
1054 .doit = ovs_flow_cmd_get,
1055 .dumpit = ovs_flow_cmd_dump
1056 },
1057 { .cmd = OVS_FLOW_CMD_SET,
1058 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1059 .policy = flow_policy,
1060 .doit = ovs_flow_cmd_new_or_set,
1061 },
1062 };
1063
1064 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1065 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1066 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1067 };
1068
1069 static struct genl_family dp_datapath_genl_family = {
1070 .id = GENL_ID_GENERATE,
1071 .hdrsize = sizeof(struct ovs_header),
1072 .name = OVS_DATAPATH_FAMILY,
1073 .version = OVS_DATAPATH_VERSION,
1074 .maxattr = OVS_DP_ATTR_MAX,
1075 .netnsok = true,
1076 .parallel_ops = true,
1077 };
1078
1079 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1080 .name = OVS_DATAPATH_MCGROUP
1081 };
1082
1083 static size_t ovs_dp_cmd_msg_size(void)
1084 {
1085 size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1086
1087 msgsize += nla_total_size(IFNAMSIZ);
1088 msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1089 msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1090
1091 return msgsize;
1092 }
1093
1094 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1095 u32 portid, u32 seq, u32 flags, u8 cmd)
1096 {
1097 struct ovs_header *ovs_header;
1098 struct ovs_dp_stats dp_stats;
1099 struct ovs_dp_megaflow_stats dp_megaflow_stats;
1100 int err;
1101
1102 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1103 flags, cmd);
1104 if (!ovs_header)
1105 goto error;
1106
1107 ovs_header->dp_ifindex = get_dpifindex(dp);
1108
1109 rcu_read_lock();
1110 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1111 rcu_read_unlock();
1112 if (err)
1113 goto nla_put_failure;
1114
1115 get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1116 if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1117 &dp_stats))
1118 goto nla_put_failure;
1119
1120 if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1121 sizeof(struct ovs_dp_megaflow_stats),
1122 &dp_megaflow_stats))
1123 goto nla_put_failure;
1124
1125 return genlmsg_end(skb, ovs_header);
1126
1127 nla_put_failure:
1128 genlmsg_cancel(skb, ovs_header);
1129 error:
1130 return -EMSGSIZE;
1131 }
1132
1133 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 portid,
1134 u32 seq, u8 cmd)
1135 {
1136 struct sk_buff *skb;
1137 int retval;
1138
1139 skb = genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1140 if (!skb)
1141 return ERR_PTR(-ENOMEM);
1142
1143 retval = ovs_dp_cmd_fill_info(dp, skb, portid, seq, 0, cmd);
1144 if (retval < 0) {
1145 kfree_skb(skb);
1146 return ERR_PTR(retval);
1147 }
1148 return skb;
1149 }
1150
1151 /* Called with ovs_mutex. */
1152 static struct datapath *lookup_datapath(struct net *net,
1153 struct ovs_header *ovs_header,
1154 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1155 {
1156 struct datapath *dp;
1157
1158 if (!a[OVS_DP_ATTR_NAME])
1159 dp = get_dp(net, ovs_header->dp_ifindex);
1160 else {
1161 struct vport *vport;
1162
1163 rcu_read_lock();
1164 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1165 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1166 rcu_read_unlock();
1167 }
1168 return dp ? dp : ERR_PTR(-ENODEV);
1169 }
1170
1171 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1172 {
1173 struct nlattr **a = info->attrs;
1174 struct vport_parms parms;
1175 struct sk_buff *reply;
1176 struct datapath *dp;
1177 struct vport *vport;
1178 struct ovs_net *ovs_net;
1179 int err, i;
1180
1181 err = -EINVAL;
1182 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1183 goto err;
1184
1185 ovs_lock();
1186
1187 err = -ENOMEM;
1188 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1189 if (dp == NULL)
1190 goto err_unlock_ovs;
1191
1192 ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1193
1194 /* Allocate table. */
1195 err = ovs_flow_tbl_init(&dp->table);
1196 if (err)
1197 goto err_free_dp;
1198
1199 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1200 if (!dp->stats_percpu) {
1201 err = -ENOMEM;
1202 goto err_destroy_table;
1203 }
1204
1205 dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1206 GFP_KERNEL);
1207 if (!dp->ports) {
1208 err = -ENOMEM;
1209 goto err_destroy_percpu;
1210 }
1211
1212 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1213 INIT_HLIST_HEAD(&dp->ports[i]);
1214
1215 /* Set up our datapath device. */
1216 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1217 parms.type = OVS_VPORT_TYPE_INTERNAL;
1218 parms.options = NULL;
1219 parms.dp = dp;
1220 parms.port_no = OVSP_LOCAL;
1221 parms.upcall_portid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1222
1223 vport = new_vport(&parms);
1224 if (IS_ERR(vport)) {
1225 err = PTR_ERR(vport);
1226 if (err == -EBUSY)
1227 err = -EEXIST;
1228
1229 goto err_destroy_ports_array;
1230 }
1231
1232 reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1233 info->snd_seq, OVS_DP_CMD_NEW);
1234 err = PTR_ERR(reply);
1235 if (IS_ERR(reply))
1236 goto err_destroy_local_port;
1237
1238 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1239 list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1240
1241 ovs_unlock();
1242
1243 ovs_notify(&dp_datapath_genl_family, reply, info,
1244 &ovs_dp_datapath_multicast_group);
1245 return 0;
1246
1247 err_destroy_local_port:
1248 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1249 err_destroy_ports_array:
1250 kfree(dp->ports);
1251 err_destroy_percpu:
1252 free_percpu(dp->stats_percpu);
1253 err_destroy_table:
1254 ovs_flow_tbl_destroy(&dp->table);
1255 err_free_dp:
1256 release_net(ovs_dp_get_net(dp));
1257 kfree(dp);
1258 err_unlock_ovs:
1259 ovs_unlock();
1260 err:
1261 return err;
1262 }
1263
1264 /* Called with ovs_mutex. */
1265 static void __dp_destroy(struct datapath *dp)
1266 {
1267 int i;
1268
1269 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1270 struct vport *vport;
1271 struct hlist_node *n;
1272
1273 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1274 if (vport->port_no != OVSP_LOCAL)
1275 ovs_dp_detach_port(vport);
1276 }
1277
1278 list_del_rcu(&dp->list_node);
1279
1280 /* OVSP_LOCAL is datapath internal port. We need to make sure that
1281 * all port in datapath are destroyed first before freeing datapath.
1282 */
1283 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1284
1285 call_rcu(&dp->rcu, destroy_dp_rcu);
1286 }
1287
1288 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1289 {
1290 struct sk_buff *reply;
1291 struct datapath *dp;
1292 int err;
1293
1294 ovs_lock();
1295 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1296 err = PTR_ERR(dp);
1297 if (IS_ERR(dp))
1298 goto unlock;
1299
1300 reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1301 info->snd_seq, OVS_DP_CMD_DEL);
1302 err = PTR_ERR(reply);
1303 if (IS_ERR(reply))
1304 goto unlock;
1305
1306 __dp_destroy(dp);
1307 ovs_unlock();
1308
1309 ovs_notify(&dp_datapath_genl_family, reply, info,
1310 &ovs_dp_datapath_multicast_group);
1311
1312 return 0;
1313 unlock:
1314 ovs_unlock();
1315 return err;
1316 }
1317
1318 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1319 {
1320 struct sk_buff *reply;
1321 struct datapath *dp;
1322 int err;
1323
1324 ovs_lock();
1325 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1326 err = PTR_ERR(dp);
1327 if (IS_ERR(dp))
1328 goto unlock;
1329
1330 reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1331 info->snd_seq, OVS_DP_CMD_NEW);
1332 if (IS_ERR(reply)) {
1333 err = PTR_ERR(reply);
1334 genl_set_err(&dp_datapath_genl_family, sock_net(skb->sk), 0,
1335 ovs_dp_datapath_multicast_group.id, err);
1336 err = 0;
1337 goto unlock;
1338 }
1339
1340 ovs_unlock();
1341 ovs_notify(&dp_datapath_genl_family, reply, info,
1342 &ovs_dp_datapath_multicast_group);
1343
1344 return 0;
1345 unlock:
1346 ovs_unlock();
1347 return err;
1348 }
1349
1350 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1351 {
1352 struct sk_buff *reply;
1353 struct datapath *dp;
1354 int err;
1355
1356 ovs_lock();
1357 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1358 if (IS_ERR(dp)) {
1359 err = PTR_ERR(dp);
1360 goto unlock;
1361 }
1362
1363 reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1364 info->snd_seq, OVS_DP_CMD_NEW);
1365 if (IS_ERR(reply)) {
1366 err = PTR_ERR(reply);
1367 goto unlock;
1368 }
1369
1370 ovs_unlock();
1371 return genlmsg_reply(reply, info);
1372
1373 unlock:
1374 ovs_unlock();
1375 return err;
1376 }
1377
1378 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1379 {
1380 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1381 struct datapath *dp;
1382 int skip = cb->args[0];
1383 int i = 0;
1384
1385 rcu_read_lock();
1386 list_for_each_entry_rcu(dp, &ovs_net->dps, list_node) {
1387 if (i >= skip &&
1388 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1389 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1390 OVS_DP_CMD_NEW) < 0)
1391 break;
1392 i++;
1393 }
1394 rcu_read_unlock();
1395
1396 cb->args[0] = i;
1397
1398 return skb->len;
1399 }
1400
1401 static const struct genl_ops dp_datapath_genl_ops[] = {
1402 { .cmd = OVS_DP_CMD_NEW,
1403 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1404 .policy = datapath_policy,
1405 .doit = ovs_dp_cmd_new
1406 },
1407 { .cmd = OVS_DP_CMD_DEL,
1408 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1409 .policy = datapath_policy,
1410 .doit = ovs_dp_cmd_del
1411 },
1412 { .cmd = OVS_DP_CMD_GET,
1413 .flags = 0, /* OK for unprivileged users. */
1414 .policy = datapath_policy,
1415 .doit = ovs_dp_cmd_get,
1416 .dumpit = ovs_dp_cmd_dump
1417 },
1418 { .cmd = OVS_DP_CMD_SET,
1419 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1420 .policy = datapath_policy,
1421 .doit = ovs_dp_cmd_set,
1422 },
1423 };
1424
1425 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1426 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1427 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1428 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1429 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1430 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1431 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1432 };
1433
1434 struct genl_family dp_vport_genl_family = {
1435 .id = GENL_ID_GENERATE,
1436 .hdrsize = sizeof(struct ovs_header),
1437 .name = OVS_VPORT_FAMILY,
1438 .version = OVS_VPORT_VERSION,
1439 .maxattr = OVS_VPORT_ATTR_MAX,
1440 .netnsok = true,
1441 .parallel_ops = true,
1442 };
1443
1444 struct genl_multicast_group ovs_dp_vport_multicast_group = {
1445 .name = OVS_VPORT_MCGROUP
1446 };
1447
1448 /* Called with ovs_mutex or RCU read lock. */
1449 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1450 u32 portid, u32 seq, u32 flags, u8 cmd)
1451 {
1452 struct ovs_header *ovs_header;
1453 struct ovs_vport_stats vport_stats;
1454 int err;
1455
1456 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1457 flags, cmd);
1458 if (!ovs_header)
1459 return -EMSGSIZE;
1460
1461 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1462
1463 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1464 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1465 nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1466 nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_portid))
1467 goto nla_put_failure;
1468
1469 ovs_vport_get_stats(vport, &vport_stats);
1470 if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1471 &vport_stats))
1472 goto nla_put_failure;
1473
1474 err = ovs_vport_get_options(vport, skb);
1475 if (err == -EMSGSIZE)
1476 goto error;
1477
1478 return genlmsg_end(skb, ovs_header);
1479
1480 nla_put_failure:
1481 err = -EMSGSIZE;
1482 error:
1483 genlmsg_cancel(skb, ovs_header);
1484 return err;
1485 }
1486
1487 /* Called with ovs_mutex or RCU read lock. */
1488 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1489 u32 seq, u8 cmd)
1490 {
1491 struct sk_buff *skb;
1492 int retval;
1493
1494 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1495 if (!skb)
1496 return ERR_PTR(-ENOMEM);
1497
1498 retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1499 BUG_ON(retval < 0);
1500
1501 return skb;
1502 }
1503
1504 /* Called with ovs_mutex or RCU read lock. */
1505 static struct vport *lookup_vport(struct net *net,
1506 struct ovs_header *ovs_header,
1507 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1508 {
1509 struct datapath *dp;
1510 struct vport *vport;
1511
1512 if (a[OVS_VPORT_ATTR_NAME]) {
1513 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1514 if (!vport)
1515 return ERR_PTR(-ENODEV);
1516 if (ovs_header->dp_ifindex &&
1517 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1518 return ERR_PTR(-ENODEV);
1519 return vport;
1520 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1521 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1522
1523 if (port_no >= DP_MAX_PORTS)
1524 return ERR_PTR(-EFBIG);
1525
1526 dp = get_dp(net, ovs_header->dp_ifindex);
1527 if (!dp)
1528 return ERR_PTR(-ENODEV);
1529
1530 vport = ovs_vport_ovsl_rcu(dp, port_no);
1531 if (!vport)
1532 return ERR_PTR(-ENODEV);
1533 return vport;
1534 } else
1535 return ERR_PTR(-EINVAL);
1536 }
1537
1538 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1539 {
1540 struct nlattr **a = info->attrs;
1541 struct ovs_header *ovs_header = info->userhdr;
1542 struct vport_parms parms;
1543 struct sk_buff *reply;
1544 struct vport *vport;
1545 struct datapath *dp;
1546 u32 port_no;
1547 int err;
1548
1549 err = -EINVAL;
1550 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1551 !a[OVS_VPORT_ATTR_UPCALL_PID])
1552 goto exit;
1553
1554 ovs_lock();
1555 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1556 err = -ENODEV;
1557 if (!dp)
1558 goto exit_unlock;
1559
1560 if (a[OVS_VPORT_ATTR_PORT_NO]) {
1561 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1562
1563 err = -EFBIG;
1564 if (port_no >= DP_MAX_PORTS)
1565 goto exit_unlock;
1566
1567 vport = ovs_vport_ovsl(dp, port_no);
1568 err = -EBUSY;
1569 if (vport)
1570 goto exit_unlock;
1571 } else {
1572 for (port_no = 1; ; port_no++) {
1573 if (port_no >= DP_MAX_PORTS) {
1574 err = -EFBIG;
1575 goto exit_unlock;
1576 }
1577 vport = ovs_vport_ovsl(dp, port_no);
1578 if (!vport)
1579 break;
1580 }
1581 }
1582
1583 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1584 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1585 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1586 parms.dp = dp;
1587 parms.port_no = port_no;
1588 parms.upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1589
1590 vport = new_vport(&parms);
1591 err = PTR_ERR(vport);
1592 if (IS_ERR(vport))
1593 goto exit_unlock;
1594
1595 err = 0;
1596 reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
1597 OVS_VPORT_CMD_NEW);
1598 if (IS_ERR(reply)) {
1599 err = PTR_ERR(reply);
1600 ovs_dp_detach_port(vport);
1601 goto exit_unlock;
1602 }
1603
1604 ovs_notify(&dp_vport_genl_family, reply, info,
1605 &ovs_dp_vport_multicast_group);
1606
1607 exit_unlock:
1608 ovs_unlock();
1609 exit:
1610 return err;
1611 }
1612
1613 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1614 {
1615 struct nlattr **a = info->attrs;
1616 struct sk_buff *reply;
1617 struct vport *vport;
1618 int err;
1619
1620 ovs_lock();
1621 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1622 err = PTR_ERR(vport);
1623 if (IS_ERR(vport))
1624 goto exit_unlock;
1625
1626 if (a[OVS_VPORT_ATTR_TYPE] &&
1627 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
1628 err = -EINVAL;
1629 goto exit_unlock;
1630 }
1631
1632 reply = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1633 if (!reply) {
1634 err = -ENOMEM;
1635 goto exit_unlock;
1636 }
1637
1638 if (a[OVS_VPORT_ATTR_OPTIONS]) {
1639 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1640 if (err)
1641 goto exit_free;
1642 }
1643
1644 if (a[OVS_VPORT_ATTR_UPCALL_PID])
1645 vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1646
1647 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1648 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1649 BUG_ON(err < 0);
1650
1651 ovs_unlock();
1652 ovs_notify(&dp_vport_genl_family, reply, info,
1653 &ovs_dp_vport_multicast_group);
1654 return 0;
1655
1656 exit_free:
1657 kfree_skb(reply);
1658 exit_unlock:
1659 ovs_unlock();
1660 return err;
1661 }
1662
1663 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1664 {
1665 struct nlattr **a = info->attrs;
1666 struct sk_buff *reply;
1667 struct vport *vport;
1668 int err;
1669
1670 ovs_lock();
1671 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1672 err = PTR_ERR(vport);
1673 if (IS_ERR(vport))
1674 goto exit_unlock;
1675
1676 if (vport->port_no == OVSP_LOCAL) {
1677 err = -EINVAL;
1678 goto exit_unlock;
1679 }
1680
1681 reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
1682 info->snd_seq, OVS_VPORT_CMD_DEL);
1683 err = PTR_ERR(reply);
1684 if (IS_ERR(reply))
1685 goto exit_unlock;
1686
1687 err = 0;
1688 ovs_dp_detach_port(vport);
1689
1690 ovs_notify(&dp_vport_genl_family, reply, info,
1691 &ovs_dp_vport_multicast_group);
1692
1693 exit_unlock:
1694 ovs_unlock();
1695 return err;
1696 }
1697
1698 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1699 {
1700 struct nlattr **a = info->attrs;
1701 struct ovs_header *ovs_header = info->userhdr;
1702 struct sk_buff *reply;
1703 struct vport *vport;
1704 int err;
1705
1706 rcu_read_lock();
1707 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1708 err = PTR_ERR(vport);
1709 if (IS_ERR(vport))
1710 goto exit_unlock;
1711
1712 reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
1713 info->snd_seq, OVS_VPORT_CMD_NEW);
1714 err = PTR_ERR(reply);
1715 if (IS_ERR(reply))
1716 goto exit_unlock;
1717
1718 rcu_read_unlock();
1719
1720 return genlmsg_reply(reply, info);
1721
1722 exit_unlock:
1723 rcu_read_unlock();
1724 return err;
1725 }
1726
1727 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1728 {
1729 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1730 struct datapath *dp;
1731 int bucket = cb->args[0], skip = cb->args[1];
1732 int i, j = 0;
1733
1734 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1735 if (!dp)
1736 return -ENODEV;
1737
1738 rcu_read_lock();
1739 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
1740 struct vport *vport;
1741
1742 j = 0;
1743 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1744 if (j >= skip &&
1745 ovs_vport_cmd_fill_info(vport, skb,
1746 NETLINK_CB(cb->skb).portid,
1747 cb->nlh->nlmsg_seq,
1748 NLM_F_MULTI,
1749 OVS_VPORT_CMD_NEW) < 0)
1750 goto out;
1751
1752 j++;
1753 }
1754 skip = 0;
1755 }
1756 out:
1757 rcu_read_unlock();
1758
1759 cb->args[0] = i;
1760 cb->args[1] = j;
1761
1762 return skb->len;
1763 }
1764
1765 static const struct genl_ops dp_vport_genl_ops[] = {
1766 { .cmd = OVS_VPORT_CMD_NEW,
1767 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1768 .policy = vport_policy,
1769 .doit = ovs_vport_cmd_new
1770 },
1771 { .cmd = OVS_VPORT_CMD_DEL,
1772 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1773 .policy = vport_policy,
1774 .doit = ovs_vport_cmd_del
1775 },
1776 { .cmd = OVS_VPORT_CMD_GET,
1777 .flags = 0, /* OK for unprivileged users. */
1778 .policy = vport_policy,
1779 .doit = ovs_vport_cmd_get,
1780 .dumpit = ovs_vport_cmd_dump
1781 },
1782 { .cmd = OVS_VPORT_CMD_SET,
1783 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1784 .policy = vport_policy,
1785 .doit = ovs_vport_cmd_set,
1786 },
1787 };
1788
1789 struct genl_family_and_ops {
1790 struct genl_family *family;
1791 const struct genl_ops *ops;
1792 int n_ops;
1793 struct genl_multicast_group *group;
1794 };
1795
1796 static const struct genl_family_and_ops dp_genl_families[] = {
1797 { &dp_datapath_genl_family,
1798 dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1799 &ovs_dp_datapath_multicast_group },
1800 { &dp_vport_genl_family,
1801 dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1802 &ovs_dp_vport_multicast_group },
1803 { &dp_flow_genl_family,
1804 dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1805 &ovs_dp_flow_multicast_group },
1806 { &dp_packet_genl_family,
1807 dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1808 NULL },
1809 };
1810
1811 static void dp_unregister_genl(int n_families)
1812 {
1813 int i;
1814
1815 for (i = 0; i < n_families; i++)
1816 genl_unregister_family(dp_genl_families[i].family);
1817 }
1818
1819 static int dp_register_genl(void)
1820 {
1821 int n_registered;
1822 int err;
1823 int i;
1824
1825 n_registered = 0;
1826 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
1827 const struct genl_family_and_ops *f = &dp_genl_families[i];
1828
1829 f->family->ops = f->ops;
1830 f->family->n_ops = f->n_ops;
1831 err = genl_register_family(f->family);
1832 if (err)
1833 goto error;
1834 n_registered++;
1835
1836 if (f->group) {
1837 err = genl_register_mc_group(f->family, f->group);
1838 if (err)
1839 goto error;
1840 }
1841 }
1842
1843 return 0;
1844
1845 error:
1846 dp_unregister_genl(n_registered);
1847 return err;
1848 }
1849
1850 static int __net_init ovs_init_net(struct net *net)
1851 {
1852 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1853
1854 INIT_LIST_HEAD(&ovs_net->dps);
1855 INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
1856 return 0;
1857 }
1858
1859 static void __net_exit ovs_exit_net(struct net *net)
1860 {
1861 struct datapath *dp, *dp_next;
1862 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1863
1864 ovs_lock();
1865 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
1866 __dp_destroy(dp);
1867 ovs_unlock();
1868
1869 cancel_work_sync(&ovs_net->dp_notify_work);
1870 }
1871
1872 static struct pernet_operations ovs_net_ops = {
1873 .init = ovs_init_net,
1874 .exit = ovs_exit_net,
1875 .id = &ovs_net_id,
1876 .size = sizeof(struct ovs_net),
1877 };
1878
1879 static int __init dp_init(void)
1880 {
1881 int err;
1882
1883 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
1884
1885 pr_info("Open vSwitch switching datapath\n");
1886
1887 err = ovs_flow_init();
1888 if (err)
1889 goto error;
1890
1891 err = ovs_vport_init();
1892 if (err)
1893 goto error_flow_exit;
1894
1895 err = register_pernet_device(&ovs_net_ops);
1896 if (err)
1897 goto error_vport_exit;
1898
1899 err = register_netdevice_notifier(&ovs_dp_device_notifier);
1900 if (err)
1901 goto error_netns_exit;
1902
1903 err = dp_register_genl();
1904 if (err < 0)
1905 goto error_unreg_notifier;
1906
1907 return 0;
1908
1909 error_unreg_notifier:
1910 unregister_netdevice_notifier(&ovs_dp_device_notifier);
1911 error_netns_exit:
1912 unregister_pernet_device(&ovs_net_ops);
1913 error_vport_exit:
1914 ovs_vport_exit();
1915 error_flow_exit:
1916 ovs_flow_exit();
1917 error:
1918 return err;
1919 }
1920
1921 static void dp_cleanup(void)
1922 {
1923 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
1924 unregister_netdevice_notifier(&ovs_dp_device_notifier);
1925 unregister_pernet_device(&ovs_net_ops);
1926 rcu_barrier();
1927 ovs_vport_exit();
1928 ovs_flow_exit();
1929 }
1930
1931 module_init(dp_init);
1932 module_exit(dp_cleanup);
1933
1934 MODULE_DESCRIPTION("Open vSwitch switching datapath");
1935 MODULE_LICENSE("GPL");
Linux kernel - Deliverables
This page took 0.070471 seconds and 6 git commands to generate.