?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jmorris...
[linux/fpc-iii.git] / net / openvswitch / datapath.c
blobdf4692826ead9ef8b5761f58ef7218bf4b3f5119
1 /*
2 * Copyright (c) 2007-2013 Nicira, Inc.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of version 2 of the GNU General Public
6 * License as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public License
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
16 * 02110-1301, USA
17 */
18
19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
20
21 #include <linux/init.h>
22 #include <linux/module.h>
23 #include <linux/if_arp.h>
24 #include <linux/if_vlan.h>
25 #include <linux/in.h>
26 #include <linux/ip.h>
27 #include <linux/jhash.h>
28 #include <linux/delay.h>
29 #include <linux/time.h>
30 #include <linux/etherdevice.h>
31 #include <linux/genetlink.h>
32 #include <linux/kernel.h>
33 #include <linux/kthread.h>
34 #include <linux/mutex.h>
35 #include <linux/percpu.h>
36 #include <linux/rcupdate.h>
37 #include <linux/tcp.h>
38 #include <linux/udp.h>
39 #include <linux/ethtool.h>
40 #include <linux/wait.h>
41 #include <asm/div64.h>
42 #include <linux/highmem.h>
43 #include <linux/netfilter_bridge.h>
44 #include <linux/netfilter_ipv4.h>
45 #include <linux/inetdevice.h>
46 #include <linux/list.h>
47 #include <linux/lockdep.h>
48 #include <linux/openvswitch.h>
49 #include <linux/rculist.h>
50 #include <linux/dmi.h>
51 #include <linux/workqueue.h>
52 #include <net/genetlink.h>
53 #include <net/net_namespace.h>
54 #include <net/netns/generic.h>
55
56 #include "datapath.h"
57 #include "flow.h"
58 #include "flow_netlink.h"
59 #include "vport-internal_dev.h"
60 #include "vport-netdev.h"
61
62 int ovs_net_id __read_mostly;
63
64 static void ovs_notify(struct genl_family *family,
65 struct sk_buff *skb, struct genl_info *info)
66 {
67 genl_notify(family, skb, genl_info_net(info), info->snd_portid,
68 0, info->nlhdr, GFP_KERNEL);
69 }
70
71 /**
72 * DOC: Locking:
73 *
74 * All writes e.g. Writes to device state (add/remove datapath, port, set
75 * operations on vports, etc.), Writes to other state (flow table
76 * modifications, set miscellaneous datapath parameters, etc.) are protected
77 * by ovs_lock.
78 *
79 * Reads are protected by RCU.
80 *
81 * There are a few special cases (mostly stats) that have their own
82 * synchronization but they nest under all of above and don't interact with
83 * each other.
84 *
85 * The RTNL lock nests inside ovs_mutex.
86 */
87
88 static DEFINE_MUTEX(ovs_mutex);
89
90 void ovs_lock(void)
91 {
92 mutex_lock(&ovs_mutex);
93 }
94
95 void ovs_unlock(void)
96 {
97 mutex_unlock(&ovs_mutex);
98 }
99
100 #ifdef CONFIG_LOCKDEP
101 int lockdep_ovsl_is_held(void)
102 {
103 if (debug_locks)
104 return lockdep_is_held(&ovs_mutex);
105 else
106 return 1;
107 }
108 #endif
109
110 static struct vport *new_vport(const struct vport_parms *);
111 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
112 const struct dp_upcall_info *);
113 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
114 const struct dp_upcall_info *);
115
116 /* Must be called with rcu_read_lock or ovs_mutex. */
117 static struct datapath *get_dp(struct net *net, int dp_ifindex)
118 {
119 struct datapath *dp = NULL;
120 struct net_device *dev;
121
122 rcu_read_lock();
123 dev = dev_get_by_index_rcu(net, dp_ifindex);
124 if (dev) {
125 struct vport *vport = ovs_internal_dev_get_vport(dev);
126 if (vport)
127 dp = vport->dp;
128 }
129 rcu_read_unlock();
130
131 return dp;
132 }
133
134 /* Must be called with rcu_read_lock or ovs_mutex. */
135 static const char *ovs_dp_name(const struct datapath *dp)
136 {
137 struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
138 return vport->ops->get_name(vport);
139 }
140
141 static int get_dpifindex(struct datapath *dp)
142 {
143 struct vport *local;
144 int ifindex;
145
146 rcu_read_lock();
147
148 local = ovs_vport_rcu(dp, OVSP_LOCAL);
149 if (local)
150 ifindex = netdev_vport_priv(local)->dev->ifindex;
151 else
152 ifindex = 0;
153
154 rcu_read_unlock();
155
156 return ifindex;
157 }
158
159 static void destroy_dp_rcu(struct rcu_head *rcu)
160 {
161 struct datapath *dp = container_of(rcu, struct datapath, rcu);
162
163 ovs_flow_tbl_destroy(&dp->table);
164 free_percpu(dp->stats_percpu);
165 release_net(ovs_dp_get_net(dp));
166 kfree(dp->ports);
167 kfree(dp);
168 }
169
170 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
171 u16 port_no)
172 {
173 return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
174 }
175
176 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
177 {
178 struct vport *vport;
179 struct hlist_head *head;
180
181 head = vport_hash_bucket(dp, port_no);
182 hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
183 if (vport->port_no == port_no)
184 return vport;
185 }
186 return NULL;
187 }
188
189 /* Called with ovs_mutex. */
190 static struct vport *new_vport(const struct vport_parms *parms)
191 {
192 struct vport *vport;
193
194 vport = ovs_vport_add(parms);
195 if (!IS_ERR(vport)) {
196 struct datapath *dp = parms->dp;
197 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
198
199 hlist_add_head_rcu(&vport->dp_hash_node, head);
200 }
201 return vport;
202 }
203
204 void ovs_dp_detach_port(struct vport *p)
205 {
206 ASSERT_OVSL();
207
208 /* First drop references to device. */
209 hlist_del_rcu(&p->dp_hash_node);
210
211 /* Then destroy it. */
212 ovs_vport_del(p);
213 }
214
215 /* Must be called with rcu_read_lock. */
216 void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
217 {
218 struct datapath *dp = p->dp;
219 struct sw_flow *flow;
220 struct dp_stats_percpu *stats;
221 struct sw_flow_key key;
222 u64 *stats_counter;
223 u32 n_mask_hit;
224 int error;
225
226 stats = this_cpu_ptr(dp->stats_percpu);
227
228 /* Extract flow from 'skb' into 'key'. */
229 error = ovs_flow_extract(skb, p->port_no, &key);
230 if (unlikely(error)) {
231 kfree_skb(skb);
232 return;
233 }
234
235 /* Look up flow. */
236 flow = ovs_flow_tbl_lookup_stats(&dp->table, &key, &n_mask_hit);
237 if (unlikely(!flow)) {
238 struct dp_upcall_info upcall;
239
240 upcall.cmd = OVS_PACKET_CMD_MISS;
241 upcall.key = &key;
242 upcall.userdata = NULL;
243 upcall.portid = p->upcall_portid;
244 ovs_dp_upcall(dp, skb, &upcall);
245 consume_skb(skb);
246 stats_counter = &stats->n_missed;
247 goto out;
248 }
249
250 OVS_CB(skb)->flow = flow;
251 OVS_CB(skb)->pkt_key = &key;
252
253 ovs_flow_stats_update(OVS_CB(skb)->flow, skb);
254 ovs_execute_actions(dp, skb);
255 stats_counter = &stats->n_hit;
256
257 out:
258 /* Update datapath statistics. */
259 u64_stats_update_begin(&stats->sync);
260 (*stats_counter)++;
261 stats->n_mask_hit += n_mask_hit;
262 u64_stats_update_end(&stats->sync);
263 }
264
265 static struct genl_family dp_packet_genl_family = {
266 .id = GENL_ID_GENERATE,
267 .hdrsize = sizeof(struct ovs_header),
268 .name = OVS_PACKET_FAMILY,
269 .version = OVS_PACKET_VERSION,
270 .maxattr = OVS_PACKET_ATTR_MAX,
271 .netnsok = true,
272 .parallel_ops = true,
273 };
274
275 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
276 const struct dp_upcall_info *upcall_info)
277 {
278 struct dp_stats_percpu *stats;
279 int err;
280
281 if (upcall_info->portid == 0) {
282 err = -ENOTCONN;
283 goto err;
284 }
285
286 if (!skb_is_gso(skb))
287 err = queue_userspace_packet(dp, skb, upcall_info);
288 else
289 err = queue_gso_packets(dp, skb, upcall_info);
290 if (err)
291 goto err;
292
293 return 0;
294
295 err:
296 stats = this_cpu_ptr(dp->stats_percpu);
297
298 u64_stats_update_begin(&stats->sync);
299 stats->n_lost++;
300 u64_stats_update_end(&stats->sync);
301
302 return err;
303 }
304
305 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
306 const struct dp_upcall_info *upcall_info)
307 {
308 unsigned short gso_type = skb_shinfo(skb)->gso_type;
309 struct dp_upcall_info later_info;
310 struct sw_flow_key later_key;
311 struct sk_buff *segs, *nskb;
312 int err;
313
314 segs = __skb_gso_segment(skb, NETIF_F_SG, false);
315 if (IS_ERR(segs))
316 return PTR_ERR(segs);
317
318 /* Queue all of the segments. */
319 skb = segs;
320 do {
321 err = queue_userspace_packet(dp, skb, upcall_info);
322 if (err)
323 break;
324
325 if (skb == segs && gso_type & SKB_GSO_UDP) {
326 /* The initial flow key extracted by ovs_flow_extract()
327 * in this case is for a first fragment, so we need to
328 * properly mark later fragments.
329 */
330 later_key = *upcall_info->key;
331 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
332
333 later_info = *upcall_info;
334 later_info.key = &later_key;
335 upcall_info = &later_info;
336 }
337 } while ((skb = skb->next));
338
339 /* Free all of the segments. */
340 skb = segs;
341 do {
342 nskb = skb->next;
343 if (err)
344 kfree_skb(skb);
345 else
346 consume_skb(skb);
347 } while ((skb = nskb));
348 return err;
349 }
350
351 static size_t key_attr_size(void)
352 {
353 return nla_total_size(4) /* OVS_KEY_ATTR_PRIORITY */
354 + nla_total_size(0) /* OVS_KEY_ATTR_TUNNEL */
355 + nla_total_size(8) /* OVS_TUNNEL_KEY_ATTR_ID */
356 + nla_total_size(4) /* OVS_TUNNEL_KEY_ATTR_IPV4_SRC */
357 + nla_total_size(4) /* OVS_TUNNEL_KEY_ATTR_IPV4_DST */
358 + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TOS */
359 + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TTL */
360 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
361 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_CSUM */
362 + nla_total_size(4) /* OVS_KEY_ATTR_IN_PORT */
363 + nla_total_size(4) /* OVS_KEY_ATTR_SKB_MARK */
364 + nla_total_size(12) /* OVS_KEY_ATTR_ETHERNET */
365 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
366 + nla_total_size(4) /* OVS_KEY_ATTR_8021Q */
367 + nla_total_size(0) /* OVS_KEY_ATTR_ENCAP */
368 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
369 + nla_total_size(40) /* OVS_KEY_ATTR_IPV6 */
370 + nla_total_size(2) /* OVS_KEY_ATTR_ICMPV6 */
371 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
372 }
373
374 static size_t upcall_msg_size(const struct nlattr *userdata,
375 unsigned int hdrlen)
376 {
377 size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
378 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
379 + nla_total_size(key_attr_size()); /* OVS_PACKET_ATTR_KEY */
380
381 /* OVS_PACKET_ATTR_USERDATA */
382 if (userdata)
383 size += NLA_ALIGN(userdata->nla_len);
384
385 return size;
386 }
387
388 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
389 const struct dp_upcall_info *upcall_info)
390 {
391 struct ovs_header *upcall;
392 struct sk_buff *nskb = NULL;
393 struct sk_buff *user_skb; /* to be queued to userspace */
394 struct nlattr *nla;
395 struct genl_info info = {
396 .dst_sk = ovs_dp_get_net(dp)->genl_sock,
397 .snd_portid = upcall_info->portid,
398 };
399 size_t len;
400 unsigned int hlen;
401 int err, dp_ifindex;
402
403 dp_ifindex = get_dpifindex(dp);
404 if (!dp_ifindex)
405 return -ENODEV;
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 nskb->vlan_tci = 0;
417 skb = nskb;
418 }
419
420 if (nla_attr_size(skb->len) > USHRT_MAX) {
421 err = -EFBIG;
422 goto out;
423 }
424
425 /* Complete checksum if needed */
426 if (skb->ip_summed == CHECKSUM_PARTIAL &&
427 (err = skb_checksum_help(skb)))
428 goto out;
429
430 /* Older versions of OVS user space enforce alignment of the last
431 * Netlink attribute to NLA_ALIGNTO which would require extensive
432 * padding logic. Only perform zerocopy if padding is not required.
433 */
434 if (dp->user_features & OVS_DP_F_UNALIGNED)
435 hlen = skb_zerocopy_headlen(skb);
436 else
437 hlen = skb->len;
438
439 len = upcall_msg_size(upcall_info->userdata, hlen);
440 user_skb = genlmsg_new_unicast(len, &info, GFP_ATOMIC);
441 if (!user_skb) {
442 err = -ENOMEM;
443 goto out;
444 }
445
446 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
447 0, upcall_info->cmd);
448 upcall->dp_ifindex = dp_ifindex;
449
450 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
451 ovs_nla_put_flow(upcall_info->key, upcall_info->key, user_skb);
452 nla_nest_end(user_skb, nla);
453
454 if (upcall_info->userdata)
455 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
456 nla_len(upcall_info->userdata),
457 nla_data(upcall_info->userdata));
458
459 /* Only reserve room for attribute header, packet data is added
460 * in skb_zerocopy() */
461 if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
462 err = -ENOBUFS;
463 goto out;
464 }
465 nla->nla_len = nla_attr_size(skb->len);
466
467 skb_zerocopy(user_skb, skb, skb->len, hlen);
468
469 ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
470
471 err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
472 out:
473 kfree_skb(nskb);
474 return err;
475 }
476
477 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
478 {
479 struct ovs_header *ovs_header = info->userhdr;
480 struct nlattr **a = info->attrs;
481 struct sw_flow_actions *acts;
482 struct sk_buff *packet;
483 struct sw_flow *flow;
484 struct datapath *dp;
485 struct ethhdr *eth;
486 int len;
487 int err;
488
489 err = -EINVAL;
490 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
491 !a[OVS_PACKET_ATTR_ACTIONS])
492 goto err;
493
494 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
495 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
496 err = -ENOMEM;
497 if (!packet)
498 goto err;
499 skb_reserve(packet, NET_IP_ALIGN);
500
501 nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
502
503 skb_reset_mac_header(packet);
504 eth = eth_hdr(packet);
505
506 /* Normally, setting the skb 'protocol' field would be handled by a
507 * call to eth_type_trans(), but it assumes there's a sending
508 * device, which we may not have. */
509 if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
510 packet->protocol = eth->h_proto;
511 else
512 packet->protocol = htons(ETH_P_802_2);
513
514 /* Build an sw_flow for sending this packet. */
515 flow = ovs_flow_alloc(false);
516 err = PTR_ERR(flow);
517 if (IS_ERR(flow))
518 goto err_kfree_skb;
519
520 err = ovs_flow_extract(packet, -1, &flow->key);
521 if (err)
522 goto err_flow_free;
523
524 err = ovs_nla_get_flow_metadata(flow, a[OVS_PACKET_ATTR_KEY]);
525 if (err)
526 goto err_flow_free;
527 acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_PACKET_ATTR_ACTIONS]));
528 err = PTR_ERR(acts);
529 if (IS_ERR(acts))
530 goto err_flow_free;
531
532 err = ovs_nla_copy_actions(a[OVS_PACKET_ATTR_ACTIONS],
533 &flow->key, 0, &acts);
534 rcu_assign_pointer(flow->sf_acts, acts);
535 if (err)
536 goto err_flow_free;
537
538 OVS_CB(packet)->flow = flow;
539 OVS_CB(packet)->pkt_key = &flow->key;
540 packet->priority = flow->key.phy.priority;
541 packet->mark = flow->key.phy.skb_mark;
542
543 rcu_read_lock();
544 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
545 err = -ENODEV;
546 if (!dp)
547 goto err_unlock;
548
549 local_bh_disable();
550 err = ovs_execute_actions(dp, packet);
551 local_bh_enable();
552 rcu_read_unlock();
553
554 ovs_flow_free(flow, false);
555 return err;
556
557 err_unlock:
558 rcu_read_unlock();
559 err_flow_free:
560 ovs_flow_free(flow, false);
561 err_kfree_skb:
562 kfree_skb(packet);
563 err:
564 return err;
565 }
566
567 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
568 [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
569 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
570 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
571 };
572
573 static const struct genl_ops dp_packet_genl_ops[] = {
574 { .cmd = OVS_PACKET_CMD_EXECUTE,
575 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
576 .policy = packet_policy,
577 .doit = ovs_packet_cmd_execute
578 }
579 };
580
581 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats,
582 struct ovs_dp_megaflow_stats *mega_stats)
583 {
584 int i;
585
586 memset(mega_stats, 0, sizeof(*mega_stats));
587
588 stats->n_flows = ovs_flow_tbl_count(&dp->table);
589 mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
590
591 stats->n_hit = stats->n_missed = stats->n_lost = 0;
592
593 for_each_possible_cpu(i) {
594 const struct dp_stats_percpu *percpu_stats;
595 struct dp_stats_percpu local_stats;
596 unsigned int start;
597
598 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
599
600 do {
601 start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
602 local_stats = *percpu_stats;
603 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
604
605 stats->n_hit += local_stats.n_hit;
606 stats->n_missed += local_stats.n_missed;
607 stats->n_lost += local_stats.n_lost;
608 mega_stats->n_mask_hit += local_stats.n_mask_hit;
609 }
610 }
611
612 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
613 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
614 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
615 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
616 };
617
618 static struct genl_family dp_flow_genl_family = {
619 .id = GENL_ID_GENERATE,
620 .hdrsize = sizeof(struct ovs_header),
621 .name = OVS_FLOW_FAMILY,
622 .version = OVS_FLOW_VERSION,
623 .maxattr = OVS_FLOW_ATTR_MAX,
624 .netnsok = true,
625 .parallel_ops = true,
626 };
627
628 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
629 .name = OVS_FLOW_MCGROUP
630 };
631
632 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
633 {
634 return NLMSG_ALIGN(sizeof(struct ovs_header))
635 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
636 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_MASK */
637 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
638 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
639 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
640 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
641 }
642
643 /* Called with ovs_mutex. */
644 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
645 struct sk_buff *skb, u32 portid,
646 u32 seq, u32 flags, u8 cmd)
647 {
648 const int skb_orig_len = skb->len;
649 struct nlattr *start;
650 struct ovs_flow_stats stats;
651 __be16 tcp_flags;
652 unsigned long used;
653 struct ovs_header *ovs_header;
654 struct nlattr *nla;
655 int err;
656
657 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
658 if (!ovs_header)
659 return -EMSGSIZE;
660
661 ovs_header->dp_ifindex = get_dpifindex(dp);
662
663 /* Fill flow key. */
664 nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
665 if (!nla)
666 goto nla_put_failure;
667
668 err = ovs_nla_put_flow(&flow->unmasked_key, &flow->unmasked_key, skb);
669 if (err)
670 goto error;
671 nla_nest_end(skb, nla);
672
673 nla = nla_nest_start(skb, OVS_FLOW_ATTR_MASK);
674 if (!nla)
675 goto nla_put_failure;
676
677 err = ovs_nla_put_flow(&flow->key, &flow->mask->key, skb);
678 if (err)
679 goto error;
680
681 nla_nest_end(skb, nla);
682
683 ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
684 if (used &&
685 nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
686 goto nla_put_failure;
687
688 if (stats.n_packets &&
689 nla_put(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats))
690 goto nla_put_failure;
691
692 if ((u8)ntohs(tcp_flags) &&
693 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
694 goto nla_put_failure;
695
696 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
697 * this is the first flow to be dumped into 'skb'. This is unusual for
698 * Netlink but individual action lists can be longer than
699 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
700 * The userspace caller can always fetch the actions separately if it
701 * really wants them. (Most userspace callers in fact don't care.)
702 *
703 * This can only fail for dump operations because the skb is always
704 * properly sized for single flows.
705 */
706 start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
707 if (start) {
708 const struct sw_flow_actions *sf_acts;
709
710 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
711
712 err = ovs_nla_put_actions(sf_acts->actions,
713 sf_acts->actions_len, skb);
714 if (!err)
715 nla_nest_end(skb, start);
716 else {
717 if (skb_orig_len)
718 goto error;
719
720 nla_nest_cancel(skb, start);
721 }
722 } else if (skb_orig_len)
723 goto nla_put_failure;
724
725 return genlmsg_end(skb, ovs_header);
726
727 nla_put_failure:
728 err = -EMSGSIZE;
729 error:
730 genlmsg_cancel(skb, ovs_header);
731 return err;
732 }
733
734 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow,
735 struct genl_info *info)
736 {
737 size_t len;
738
739 len = ovs_flow_cmd_msg_size(ovsl_dereference(flow->sf_acts));
740
741 return genlmsg_new_unicast(len, info, GFP_KERNEL);
742 }
743
744 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
745 struct datapath *dp,
746 struct genl_info *info,
747 u8 cmd)
748 {
749 struct sk_buff *skb;
750 int retval;
751
752 skb = ovs_flow_cmd_alloc_info(flow, info);
753 if (!skb)
754 return ERR_PTR(-ENOMEM);
755
756 retval = ovs_flow_cmd_fill_info(flow, dp, skb, info->snd_portid,
757 info->snd_seq, 0, cmd);
758 BUG_ON(retval < 0);
759 return skb;
760 }
761
762 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
763 {
764 struct nlattr **a = info->attrs;
765 struct ovs_header *ovs_header = info->userhdr;
766 struct sw_flow_key key, masked_key;
767 struct sw_flow *flow = NULL;
768 struct sw_flow_mask mask;
769 struct sk_buff *reply;
770 struct datapath *dp;
771 struct sw_flow_actions *acts = NULL;
772 struct sw_flow_match match;
773 bool exact_5tuple;
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, &exact_5tuple,
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(!exact_5tuple);
822 if (IS_ERR(flow)) {
823 error = PTR_ERR(flow);
824 goto err_unlock_ovs;
825 }
826
827 flow->key = masked_key;
828 flow->unmasked_key = key;
829 rcu_assign_pointer(flow->sf_acts, acts);
830
831 /* Put flow in bucket. */
832 error = ovs_flow_tbl_insert(&dp->table, flow, &mask);
833 if (error) {
834 acts = NULL;
835 goto err_flow_free;
836 }
837
838 reply = ovs_flow_cmd_build_info(flow, dp, info, OVS_FLOW_CMD_NEW);
839 } else {
840 /* We found a matching flow. */
841 struct sw_flow_actions *old_acts;
842
843 /* Bail out if we're not allowed to modify an existing flow.
844 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
845 * because Generic Netlink treats the latter as a dump
846 * request. We also accept NLM_F_EXCL in case that bug ever
847 * gets fixed.
848 */
849 error = -EEXIST;
850 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
851 info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
852 goto err_unlock_ovs;
853
854 /* The unmasked key has to be the same for flow updates. */
855 error = -EINVAL;
856 if (!ovs_flow_cmp_unmasked_key(flow, &match)) {
857 OVS_NLERR("Flow modification message rejected, unmasked key does not match.\n");
858 goto err_unlock_ovs;
859 }
860
861 /* Update actions. */
862 old_acts = ovsl_dereference(flow->sf_acts);
863 rcu_assign_pointer(flow->sf_acts, acts);
864 ovs_nla_free_flow_actions(old_acts);
865
866 reply = ovs_flow_cmd_build_info(flow, dp, info, OVS_FLOW_CMD_NEW);
867
868 /* Clear stats. */
869 if (a[OVS_FLOW_ATTR_CLEAR])
870 ovs_flow_stats_clear(flow);
871 }
872 ovs_unlock();
873
874 if (!IS_ERR(reply))
875 ovs_notify(&dp_flow_genl_family, reply, info);
876 else
877 genl_set_err(&dp_flow_genl_family, sock_net(skb->sk), 0,
878 0, PTR_ERR(reply));
879 return 0;
880
881 err_flow_free:
882 ovs_flow_free(flow, false);
883 err_unlock_ovs:
884 ovs_unlock();
885 err_kfree:
886 kfree(acts);
887 error:
888 return error;
889 }
890
891 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
892 {
893 struct nlattr **a = info->attrs;
894 struct ovs_header *ovs_header = info->userhdr;
895 struct sw_flow_key key;
896 struct sk_buff *reply;
897 struct sw_flow *flow;
898 struct datapath *dp;
899 struct sw_flow_match match;
900 int err;
901
902 if (!a[OVS_FLOW_ATTR_KEY]) {
903 OVS_NLERR("Flow get message rejected, Key attribute missing.\n");
904 return -EINVAL;
905 }
906
907 ovs_match_init(&match, &key, NULL);
908 err = ovs_nla_get_match(&match, NULL, a[OVS_FLOW_ATTR_KEY], NULL);
909 if (err)
910 return err;
911
912 ovs_lock();
913 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
914 if (!dp) {
915 err = -ENODEV;
916 goto unlock;
917 }
918
919 flow = ovs_flow_tbl_lookup(&dp->table, &key);
920 if (!flow || !ovs_flow_cmp_unmasked_key(flow, &match)) {
921 err = -ENOENT;
922 goto unlock;
923 }
924
925 reply = ovs_flow_cmd_build_info(flow, dp, info, OVS_FLOW_CMD_NEW);
926 if (IS_ERR(reply)) {
927 err = PTR_ERR(reply);
928 goto unlock;
929 }
930
931 ovs_unlock();
932 return genlmsg_reply(reply, info);
933 unlock:
934 ovs_unlock();
935 return err;
936 }
937
938 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
939 {
940 struct nlattr **a = info->attrs;
941 struct ovs_header *ovs_header = info->userhdr;
942 struct sw_flow_key key;
943 struct sk_buff *reply;
944 struct sw_flow *flow;
945 struct datapath *dp;
946 struct sw_flow_match match;
947 int err;
948
949 ovs_lock();
950 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
951 if (!dp) {
952 err = -ENODEV;
953 goto unlock;
954 }
955
956 if (!a[OVS_FLOW_ATTR_KEY]) {
957 err = ovs_flow_tbl_flush(&dp->table);
958 goto unlock;
959 }
960
961 ovs_match_init(&match, &key, NULL);
962 err = ovs_nla_get_match(&match, NULL, a[OVS_FLOW_ATTR_KEY], NULL);
963 if (err)
964 goto unlock;
965
966 flow = ovs_flow_tbl_lookup(&dp->table, &key);
967 if (!flow || !ovs_flow_cmp_unmasked_key(flow, &match)) {
968 err = -ENOENT;
969 goto unlock;
970 }
971
972 reply = ovs_flow_cmd_alloc_info(flow, info);
973 if (!reply) {
974 err = -ENOMEM;
975 goto unlock;
976 }
977
978 ovs_flow_tbl_remove(&dp->table, flow);
979
980 err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_portid,
981 info->snd_seq, 0, OVS_FLOW_CMD_DEL);
982 BUG_ON(err < 0);
983
984 ovs_flow_free(flow, true);
985 ovs_unlock();
986
987 ovs_notify(&dp_flow_genl_family, reply, info);
988 return 0;
989 unlock:
990 ovs_unlock();
991 return err;
992 }
993
994 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
995 {
996 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
997 struct table_instance *ti;
998 struct datapath *dp;
999
1000 rcu_read_lock();
1001 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1002 if (!dp) {
1003 rcu_read_unlock();
1004 return -ENODEV;
1005 }
1006
1007 ti = rcu_dereference(dp->table.ti);
1008 for (;;) {
1009 struct sw_flow *flow;
1010 u32 bucket, obj;
1011
1012 bucket = cb->args[0];
1013 obj = cb->args[1];
1014 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1015 if (!flow)
1016 break;
1017
1018 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1019 NETLINK_CB(cb->skb).portid,
1020 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1021 OVS_FLOW_CMD_NEW) < 0)
1022 break;
1023
1024 cb->args[0] = bucket;
1025 cb->args[1] = obj;
1026 }
1027 rcu_read_unlock();
1028 return skb->len;
1029 }
1030
1031 static const struct genl_ops dp_flow_genl_ops[] = {
1032 { .cmd = OVS_FLOW_CMD_NEW,
1033 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1034 .policy = flow_policy,
1035 .doit = ovs_flow_cmd_new_or_set
1036 },
1037 { .cmd = OVS_FLOW_CMD_DEL,
1038 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1039 .policy = flow_policy,
1040 .doit = ovs_flow_cmd_del
1041 },
1042 { .cmd = OVS_FLOW_CMD_GET,
1043 .flags = 0, /* OK for unprivileged users. */
1044 .policy = flow_policy,
1045 .doit = ovs_flow_cmd_get,
1046 .dumpit = ovs_flow_cmd_dump
1047 },
1048 { .cmd = OVS_FLOW_CMD_SET,
1049 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1050 .policy = flow_policy,
1051 .doit = ovs_flow_cmd_new_or_set,
1052 },
1053 };
1054
1055 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1056 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1057 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1058 [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1059 };
1060
1061 static struct genl_family dp_datapath_genl_family = {
1062 .id = GENL_ID_GENERATE,
1063 .hdrsize = sizeof(struct ovs_header),
1064 .name = OVS_DATAPATH_FAMILY,
1065 .version = OVS_DATAPATH_VERSION,
1066 .maxattr = OVS_DP_ATTR_MAX,
1067 .netnsok = true,
1068 .parallel_ops = true,
1069 };
1070
1071 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1072 .name = OVS_DATAPATH_MCGROUP
1073 };
1074
1075 static size_t ovs_dp_cmd_msg_size(void)
1076 {
1077 size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1078
1079 msgsize += nla_total_size(IFNAMSIZ);
1080 msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1081 msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1082
1083 return msgsize;
1084 }
1085
1086 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1087 u32 portid, u32 seq, u32 flags, u8 cmd)
1088 {
1089 struct ovs_header *ovs_header;
1090 struct ovs_dp_stats dp_stats;
1091 struct ovs_dp_megaflow_stats dp_megaflow_stats;
1092 int err;
1093
1094 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1095 flags, cmd);
1096 if (!ovs_header)
1097 goto error;
1098
1099 ovs_header->dp_ifindex = get_dpifindex(dp);
1100
1101 rcu_read_lock();
1102 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1103 rcu_read_unlock();
1104 if (err)
1105 goto nla_put_failure;
1106
1107 get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1108 if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1109 &dp_stats))
1110 goto nla_put_failure;
1111
1112 if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1113 sizeof(struct ovs_dp_megaflow_stats),
1114 &dp_megaflow_stats))
1115 goto nla_put_failure;
1116
1117 if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1118 goto nla_put_failure;
1119
1120 return genlmsg_end(skb, ovs_header);
1121
1122 nla_put_failure:
1123 genlmsg_cancel(skb, ovs_header);
1124 error:
1125 return -EMSGSIZE;
1126 }
1127
1128 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp,
1129 struct genl_info *info, u8 cmd)
1130 {
1131 struct sk_buff *skb;
1132 int retval;
1133
1134 skb = genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info, GFP_KERNEL);
1135 if (!skb)
1136 return ERR_PTR(-ENOMEM);
1137
1138 retval = ovs_dp_cmd_fill_info(dp, skb, info->snd_portid, info->snd_seq, 0, cmd);
1139 if (retval < 0) {
1140 kfree_skb(skb);
1141 return ERR_PTR(retval);
1142 }
1143 return skb;
1144 }
1145
1146 /* Called with ovs_mutex. */
1147 static struct datapath *lookup_datapath(struct net *net,
1148 struct ovs_header *ovs_header,
1149 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1150 {
1151 struct datapath *dp;
1152
1153 if (!a[OVS_DP_ATTR_NAME])
1154 dp = get_dp(net, ovs_header->dp_ifindex);
1155 else {
1156 struct vport *vport;
1157
1158 rcu_read_lock();
1159 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1160 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1161 rcu_read_unlock();
1162 }
1163 return dp ? dp : ERR_PTR(-ENODEV);
1164 }
1165
1166 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1167 {
1168 struct datapath *dp;
1169
1170 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1171 if (!dp)
1172 return;
1173
1174 WARN(dp->user_features, "Dropping previously announced user features\n");
1175 dp->user_features = 0;
1176 }
1177
1178 static void ovs_dp_change(struct datapath *dp, struct nlattr **a)
1179 {
1180 if (a[OVS_DP_ATTR_USER_FEATURES])
1181 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1182 }
1183
1184 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1185 {
1186 struct nlattr **a = info->attrs;
1187 struct vport_parms parms;
1188 struct sk_buff *reply;
1189 struct datapath *dp;
1190 struct vport *vport;
1191 struct ovs_net *ovs_net;
1192 int err, i;
1193
1194 err = -EINVAL;
1195 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1196 goto err;
1197
1198 ovs_lock();
1199
1200 err = -ENOMEM;
1201 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1202 if (dp == NULL)
1203 goto err_unlock_ovs;
1204
1205 ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1206
1207 /* Allocate table. */
1208 err = ovs_flow_tbl_init(&dp->table);
1209 if (err)
1210 goto err_free_dp;
1211
1212 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1213 if (!dp->stats_percpu) {
1214 err = -ENOMEM;
1215 goto err_destroy_table;
1216 }
1217
1218 for_each_possible_cpu(i) {
1219 struct dp_stats_percpu *dpath_stats;
1220 dpath_stats = per_cpu_ptr(dp->stats_percpu, i);
1221 u64_stats_init(&dpath_stats->sync);
1222 }
1223
1224 dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1225 GFP_KERNEL);
1226 if (!dp->ports) {
1227 err = -ENOMEM;
1228 goto err_destroy_percpu;
1229 }
1230
1231 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1232 INIT_HLIST_HEAD(&dp->ports[i]);
1233
1234 /* Set up our datapath device. */
1235 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1236 parms.type = OVS_VPORT_TYPE_INTERNAL;
1237 parms.options = NULL;
1238 parms.dp = dp;
1239 parms.port_no = OVSP_LOCAL;
1240 parms.upcall_portid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1241
1242 ovs_dp_change(dp, a);
1243
1244 vport = new_vport(&parms);
1245 if (IS_ERR(vport)) {
1246 err = PTR_ERR(vport);
1247 if (err == -EBUSY)
1248 err = -EEXIST;
1249
1250 if (err == -EEXIST) {
1251 /* An outdated user space instance that does not understand
1252 * the concept of user_features has attempted to create a new
1253 * datapath and is likely to reuse it. Drop all user features.
1254 */
1255 if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1256 ovs_dp_reset_user_features(skb, info);
1257 }
1258
1259 goto err_destroy_ports_array;
1260 }
1261
1262 reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_NEW);
1263 err = PTR_ERR(reply);
1264 if (IS_ERR(reply))
1265 goto err_destroy_local_port;
1266
1267 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1268 list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1269
1270 ovs_unlock();
1271
1272 ovs_notify(&dp_datapath_genl_family, reply, info);
1273 return 0;
1274
1275 err_destroy_local_port:
1276 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1277 err_destroy_ports_array:
1278 kfree(dp->ports);
1279 err_destroy_percpu:
1280 free_percpu(dp->stats_percpu);
1281 err_destroy_table:
1282 ovs_flow_tbl_destroy(&dp->table);
1283 err_free_dp:
1284 release_net(ovs_dp_get_net(dp));
1285 kfree(dp);
1286 err_unlock_ovs:
1287 ovs_unlock();
1288 err:
1289 return err;
1290 }
1291
1292 /* Called with ovs_mutex. */
1293 static void __dp_destroy(struct datapath *dp)
1294 {
1295 int i;
1296
1297 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1298 struct vport *vport;
1299 struct hlist_node *n;
1300
1301 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1302 if (vport->port_no != OVSP_LOCAL)
1303 ovs_dp_detach_port(vport);
1304 }
1305
1306 list_del_rcu(&dp->list_node);
1307
1308 /* OVSP_LOCAL is datapath internal port. We need to make sure that
1309 * all port in datapath are destroyed first before freeing datapath.
1310 */
1311 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1312
1313 call_rcu(&dp->rcu, destroy_dp_rcu);
1314 }
1315
1316 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1317 {
1318 struct sk_buff *reply;
1319 struct datapath *dp;
1320 int err;
1321
1322 ovs_lock();
1323 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1324 err = PTR_ERR(dp);
1325 if (IS_ERR(dp))
1326 goto unlock;
1327
1328 reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_DEL);
1329 err = PTR_ERR(reply);
1330 if (IS_ERR(reply))
1331 goto unlock;
1332
1333 __dp_destroy(dp);
1334 ovs_unlock();
1335
1336 ovs_notify(&dp_datapath_genl_family, reply, info);
1337
1338 return 0;
1339 unlock:
1340 ovs_unlock();
1341 return err;
1342 }
1343
1344 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1345 {
1346 struct sk_buff *reply;
1347 struct datapath *dp;
1348 int err;
1349
1350 ovs_lock();
1351 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1352 err = PTR_ERR(dp);
1353 if (IS_ERR(dp))
1354 goto unlock;
1355
1356 ovs_dp_change(dp, info->attrs);
1357
1358 reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_NEW);
1359 if (IS_ERR(reply)) {
1360 err = PTR_ERR(reply);
1361 genl_set_err(&dp_datapath_genl_family, sock_net(skb->sk), 0,
1362 0, err);
1363 err = 0;
1364 goto unlock;
1365 }
1366
1367 ovs_unlock();
1368 ovs_notify(&dp_datapath_genl_family, reply, info);
1369
1370 return 0;
1371 unlock:
1372 ovs_unlock();
1373 return err;
1374 }
1375
1376 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1377 {
1378 struct sk_buff *reply;
1379 struct datapath *dp;
1380 int err;
1381
1382 ovs_lock();
1383 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1384 if (IS_ERR(dp)) {
1385 err = PTR_ERR(dp);
1386 goto unlock;
1387 }
1388
1389 reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_NEW);
1390 if (IS_ERR(reply)) {
1391 err = PTR_ERR(reply);
1392 goto unlock;
1393 }
1394
1395 ovs_unlock();
1396 return genlmsg_reply(reply, info);
1397
1398 unlock:
1399 ovs_unlock();
1400 return err;
1401 }
1402
1403 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1404 {
1405 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1406 struct datapath *dp;
1407 int skip = cb->args[0];
1408 int i = 0;
1409
1410 rcu_read_lock();
1411 list_for_each_entry_rcu(dp, &ovs_net->dps, list_node) {
1412 if (i >= skip &&
1413 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1414 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1415 OVS_DP_CMD_NEW) < 0)
1416 break;
1417 i++;
1418 }
1419 rcu_read_unlock();
1420
1421 cb->args[0] = i;
1422
1423 return skb->len;
1424 }
1425
1426 static const struct genl_ops dp_datapath_genl_ops[] = {
1427 { .cmd = OVS_DP_CMD_NEW,
1428 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1429 .policy = datapath_policy,
1430 .doit = ovs_dp_cmd_new
1431 },
1432 { .cmd = OVS_DP_CMD_DEL,
1433 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1434 .policy = datapath_policy,
1435 .doit = ovs_dp_cmd_del
1436 },
1437 { .cmd = OVS_DP_CMD_GET,
1438 .flags = 0, /* OK for unprivileged users. */
1439 .policy = datapath_policy,
1440 .doit = ovs_dp_cmd_get,
1441 .dumpit = ovs_dp_cmd_dump
1442 },
1443 { .cmd = OVS_DP_CMD_SET,
1444 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1445 .policy = datapath_policy,
1446 .doit = ovs_dp_cmd_set,
1447 },
1448 };
1449
1450 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1451 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1452 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1453 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1454 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1455 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1456 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1457 };
1458
1459 struct genl_family dp_vport_genl_family = {
1460 .id = GENL_ID_GENERATE,
1461 .hdrsize = sizeof(struct ovs_header),
1462 .name = OVS_VPORT_FAMILY,
1463 .version = OVS_VPORT_VERSION,
1464 .maxattr = OVS_VPORT_ATTR_MAX,
1465 .netnsok = true,
1466 .parallel_ops = true,
1467 };
1468
1469 static struct genl_multicast_group ovs_dp_vport_multicast_group = {
1470 .name = OVS_VPORT_MCGROUP
1471 };
1472
1473 /* Called with ovs_mutex or RCU read lock. */
1474 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1475 u32 portid, u32 seq, u32 flags, u8 cmd)
1476 {
1477 struct ovs_header *ovs_header;
1478 struct ovs_vport_stats vport_stats;
1479 int err;
1480
1481 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1482 flags, cmd);
1483 if (!ovs_header)
1484 return -EMSGSIZE;
1485
1486 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1487
1488 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1489 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1490 nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1491 nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_portid))
1492 goto nla_put_failure;
1493
1494 ovs_vport_get_stats(vport, &vport_stats);
1495 if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1496 &vport_stats))
1497 goto nla_put_failure;
1498
1499 err = ovs_vport_get_options(vport, skb);
1500 if (err == -EMSGSIZE)
1501 goto error;
1502
1503 return genlmsg_end(skb, ovs_header);
1504
1505 nla_put_failure:
1506 err = -EMSGSIZE;
1507 error:
1508 genlmsg_cancel(skb, ovs_header);
1509 return err;
1510 }
1511
1512 /* Called with ovs_mutex or RCU read lock. */
1513 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1514 u32 seq, u8 cmd)
1515 {
1516 struct sk_buff *skb;
1517 int retval;
1518
1519 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1520 if (!skb)
1521 return ERR_PTR(-ENOMEM);
1522
1523 retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1524 BUG_ON(retval < 0);
1525
1526 return skb;
1527 }
1528
1529 /* Called with ovs_mutex or RCU read lock. */
1530 static struct vport *lookup_vport(struct net *net,
1531 struct ovs_header *ovs_header,
1532 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1533 {
1534 struct datapath *dp;
1535 struct vport *vport;
1536
1537 if (a[OVS_VPORT_ATTR_NAME]) {
1538 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1539 if (!vport)
1540 return ERR_PTR(-ENODEV);
1541 if (ovs_header->dp_ifindex &&
1542 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1543 return ERR_PTR(-ENODEV);
1544 return vport;
1545 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1546 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1547
1548 if (port_no >= DP_MAX_PORTS)
1549 return ERR_PTR(-EFBIG);
1550
1551 dp = get_dp(net, ovs_header->dp_ifindex);
1552 if (!dp)
1553 return ERR_PTR(-ENODEV);
1554
1555 vport = ovs_vport_ovsl_rcu(dp, port_no);
1556 if (!vport)
1557 return ERR_PTR(-ENODEV);
1558 return vport;
1559 } else
1560 return ERR_PTR(-EINVAL);
1561 }
1562
1563 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1564 {
1565 struct nlattr **a = info->attrs;
1566 struct ovs_header *ovs_header = info->userhdr;
1567 struct vport_parms parms;
1568 struct sk_buff *reply;
1569 struct vport *vport;
1570 struct datapath *dp;
1571 u32 port_no;
1572 int err;
1573
1574 err = -EINVAL;
1575 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1576 !a[OVS_VPORT_ATTR_UPCALL_PID])
1577 goto exit;
1578
1579 ovs_lock();
1580 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1581 err = -ENODEV;
1582 if (!dp)
1583 goto exit_unlock;
1584
1585 if (a[OVS_VPORT_ATTR_PORT_NO]) {
1586 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1587
1588 err = -EFBIG;
1589 if (port_no >= DP_MAX_PORTS)
1590 goto exit_unlock;
1591
1592 vport = ovs_vport_ovsl(dp, port_no);
1593 err = -EBUSY;
1594 if (vport)
1595 goto exit_unlock;
1596 } else {
1597 for (port_no = 1; ; port_no++) {
1598 if (port_no >= DP_MAX_PORTS) {
1599 err = -EFBIG;
1600 goto exit_unlock;
1601 }
1602 vport = ovs_vport_ovsl(dp, port_no);
1603 if (!vport)
1604 break;
1605 }
1606 }
1607
1608 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1609 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1610 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1611 parms.dp = dp;
1612 parms.port_no = port_no;
1613 parms.upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1614
1615 vport = new_vport(&parms);
1616 err = PTR_ERR(vport);
1617 if (IS_ERR(vport))
1618 goto exit_unlock;
1619
1620 err = 0;
1621 reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
1622 OVS_VPORT_CMD_NEW);
1623 if (IS_ERR(reply)) {
1624 err = PTR_ERR(reply);
1625 ovs_dp_detach_port(vport);
1626 goto exit_unlock;
1627 }
1628
1629 ovs_notify(&dp_vport_genl_family, reply, info);
1630
1631 exit_unlock:
1632 ovs_unlock();
1633 exit:
1634 return err;
1635 }
1636
1637 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1638 {
1639 struct nlattr **a = info->attrs;
1640 struct sk_buff *reply;
1641 struct vport *vport;
1642 int err;
1643
1644 ovs_lock();
1645 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1646 err = PTR_ERR(vport);
1647 if (IS_ERR(vport))
1648 goto exit_unlock;
1649
1650 if (a[OVS_VPORT_ATTR_TYPE] &&
1651 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
1652 err = -EINVAL;
1653 goto exit_unlock;
1654 }
1655
1656 reply = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1657 if (!reply) {
1658 err = -ENOMEM;
1659 goto exit_unlock;
1660 }
1661
1662 if (a[OVS_VPORT_ATTR_OPTIONS]) {
1663 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1664 if (err)
1665 goto exit_free;
1666 }
1667
1668 if (a[OVS_VPORT_ATTR_UPCALL_PID])
1669 vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1670
1671 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1672 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1673 BUG_ON(err < 0);
1674
1675 ovs_unlock();
1676 ovs_notify(&dp_vport_genl_family, reply, info);
1677 return 0;
1678
1679 exit_free:
1680 kfree_skb(reply);
1681 exit_unlock:
1682 ovs_unlock();
1683 return err;
1684 }
1685
1686 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1687 {
1688 struct nlattr **a = info->attrs;
1689 struct sk_buff *reply;
1690 struct vport *vport;
1691 int err;
1692
1693 ovs_lock();
1694 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1695 err = PTR_ERR(vport);
1696 if (IS_ERR(vport))
1697 goto exit_unlock;
1698
1699 if (vport->port_no == OVSP_LOCAL) {
1700 err = -EINVAL;
1701 goto exit_unlock;
1702 }
1703
1704 reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
1705 info->snd_seq, OVS_VPORT_CMD_DEL);
1706 err = PTR_ERR(reply);
1707 if (IS_ERR(reply))
1708 goto exit_unlock;
1709
1710 err = 0;
1711 ovs_dp_detach_port(vport);
1712
1713 ovs_notify(&dp_vport_genl_family, reply, info);
1714
1715 exit_unlock:
1716 ovs_unlock();
1717 return err;
1718 }
1719
1720 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1721 {
1722 struct nlattr **a = info->attrs;
1723 struct ovs_header *ovs_header = info->userhdr;
1724 struct sk_buff *reply;
1725 struct vport *vport;
1726 int err;
1727
1728 rcu_read_lock();
1729 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1730 err = PTR_ERR(vport);
1731 if (IS_ERR(vport))
1732 goto exit_unlock;
1733
1734 reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
1735 info->snd_seq, OVS_VPORT_CMD_NEW);
1736 err = PTR_ERR(reply);
1737 if (IS_ERR(reply))
1738 goto exit_unlock;
1739
1740 rcu_read_unlock();
1741
1742 return genlmsg_reply(reply, info);
1743
1744 exit_unlock:
1745 rcu_read_unlock();
1746 return err;
1747 }
1748
1749 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1750 {
1751 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1752 struct datapath *dp;
1753 int bucket = cb->args[0], skip = cb->args[1];
1754 int i, j = 0;
1755
1756 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1757 if (!dp)
1758 return -ENODEV;
1759
1760 rcu_read_lock();
1761 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
1762 struct vport *vport;
1763
1764 j = 0;
1765 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1766 if (j >= skip &&
1767 ovs_vport_cmd_fill_info(vport, skb,
1768 NETLINK_CB(cb->skb).portid,
1769 cb->nlh->nlmsg_seq,
1770 NLM_F_MULTI,
1771 OVS_VPORT_CMD_NEW) < 0)
1772 goto out;
1773
1774 j++;
1775 }
1776 skip = 0;
1777 }
1778 out:
1779 rcu_read_unlock();
1780
1781 cb->args[0] = i;
1782 cb->args[1] = j;
1783
1784 return skb->len;
1785 }
1786
1787 static const struct genl_ops dp_vport_genl_ops[] = {
1788 { .cmd = OVS_VPORT_CMD_NEW,
1789 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1790 .policy = vport_policy,
1791 .doit = ovs_vport_cmd_new
1792 },
1793 { .cmd = OVS_VPORT_CMD_DEL,
1794 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1795 .policy = vport_policy,
1796 .doit = ovs_vport_cmd_del
1797 },
1798 { .cmd = OVS_VPORT_CMD_GET,
1799 .flags = 0, /* OK for unprivileged users. */
1800 .policy = vport_policy,
1801 .doit = ovs_vport_cmd_get,
1802 .dumpit = ovs_vport_cmd_dump
1803 },
1804 { .cmd = OVS_VPORT_CMD_SET,
1805 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1806 .policy = vport_policy,
1807 .doit = ovs_vport_cmd_set,
1808 },
1809 };
1810
1811 struct genl_family_and_ops {
1812 struct genl_family *family;
1813 const struct genl_ops *ops;
1814 int n_ops;
1815 const struct genl_multicast_group *group;
1816 };
1817
1818 static const struct genl_family_and_ops dp_genl_families[] = {
1819 { &dp_datapath_genl_family,
1820 dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1821 &ovs_dp_datapath_multicast_group },
1822 { &dp_vport_genl_family,
1823 dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1824 &ovs_dp_vport_multicast_group },
1825 { &dp_flow_genl_family,
1826 dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1827 &ovs_dp_flow_multicast_group },
1828 { &dp_packet_genl_family,
1829 dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1830 NULL },
1831 };
1832
1833 static void dp_unregister_genl(int n_families)
1834 {
1835 int i;
1836
1837 for (i = 0; i < n_families; i++)
1838 genl_unregister_family(dp_genl_families[i].family);
1839 }
1840
1841 static int dp_register_genl(void)
1842 {
1843 int n_registered;
1844 int err;
1845 int i;
1846
1847 n_registered = 0;
1848 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
1849 const struct genl_family_and_ops *f = &dp_genl_families[i];
1850
1851 f->family->ops = f->ops;
1852 f->family->n_ops = f->n_ops;
1853 f->family->mcgrps = f->group;
1854 f->family->n_mcgrps = f->group ? 1 : 0;
1855 err = genl_register_family(f->family);
1856 if (err)
1857 goto error;
1858 n_registered++;
1859 }
1860
1861 return 0;
1862
1863 error:
1864 dp_unregister_genl(n_registered);
1865 return err;
1866 }
1867
1868 static int __net_init ovs_init_net(struct net *net)
1869 {
1870 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1871
1872 INIT_LIST_HEAD(&ovs_net->dps);
1873 INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
1874 return 0;
1875 }
1876
1877 static void __net_exit ovs_exit_net(struct net *net)
1878 {
1879 struct datapath *dp, *dp_next;
1880 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1881
1882 ovs_lock();
1883 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
1884 __dp_destroy(dp);
1885 ovs_unlock();
1886
1887 cancel_work_sync(&ovs_net->dp_notify_work);
1888 }
1889
1890 static struct pernet_operations ovs_net_ops = {
1891 .init = ovs_init_net,
1892 .exit = ovs_exit_net,
1893 .id = &ovs_net_id,
1894 .size = sizeof(struct ovs_net),
1895 };
1896
1897 static int __init dp_init(void)
1898 {
1899 int err;
1900
1901 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
1902
1903 pr_info("Open vSwitch switching datapath\n");
1904
1905 err = ovs_flow_init();
1906 if (err)
1907 goto error;
1908
1909 err = ovs_vport_init();
1910 if (err)
1911 goto error_flow_exit;
1912
1913 err = register_pernet_device(&ovs_net_ops);
1914 if (err)
1915 goto error_vport_exit;
1916
1917 err = register_netdevice_notifier(&ovs_dp_device_notifier);
1918 if (err)
1919 goto error_netns_exit;
1920
1921 err = dp_register_genl();
1922 if (err < 0)
1923 goto error_unreg_notifier;
1924
1925 return 0;
1926
1927 error_unreg_notifier:
1928 unregister_netdevice_notifier(&ovs_dp_device_notifier);
1929 error_netns_exit:
1930 unregister_pernet_device(&ovs_net_ops);
1931 error_vport_exit:
1932 ovs_vport_exit();
1933 error_flow_exit:
1934 ovs_flow_exit();
1935 error:
1936 return err;
1937 }
1938
1939 static void dp_cleanup(void)
1940 {
1941 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
1942 unregister_netdevice_notifier(&ovs_dp_device_notifier);
1943 unregister_pernet_device(&ovs_net_ops);
1944 rcu_barrier();
1945 ovs_vport_exit();
1946 ovs_flow_exit();
1947 }
1948
1949 module_init(dp_init);
1950 module_exit(dp_cleanup);
1951
1952 MODULE_DESCRIPTION("Open vSwitch switching datapath");
1953 MODULE_LICENSE("GPL");
Linux with added FPC-III support