?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge branch 'clk-next-unregister' into clk-next
[linux-2.6/btrfs-unstable.git] / net / openvswitch / datapath.c
blob6f5e1dd3be2dbdcfdb591d64323c3558d0636e6d
1 /*
2 * Copyright (c) 2007-2013 Nicira, Inc.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of version 2 of the GNU General Public
6 * License as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public License
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
16 * 02110-1301, USA
17 */
18
19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
20
21 #include <linux/init.h>
22 #include <linux/module.h>
23 #include <linux/if_arp.h>
24 #include <linux/if_vlan.h>
25 #include <linux/in.h>
26 #include <linux/ip.h>
27 #include <linux/jhash.h>
28 #include <linux/delay.h>
29 #include <linux/time.h>
30 #include <linux/etherdevice.h>
31 #include <linux/genetlink.h>
32 #include <linux/kernel.h>
33 #include <linux/kthread.h>
34 #include <linux/mutex.h>
35 #include <linux/percpu.h>
36 #include <linux/rcupdate.h>
37 #include <linux/tcp.h>
38 #include <linux/udp.h>
39 #include <linux/ethtool.h>
40 #include <linux/wait.h>
41 #include <asm/div64.h>
42 #include <linux/highmem.h>
43 #include <linux/netfilter_bridge.h>
44 #include <linux/netfilter_ipv4.h>
45 #include <linux/inetdevice.h>
46 #include <linux/list.h>
47 #include <linux/lockdep.h>
48 #include <linux/openvswitch.h>
49 #include <linux/rculist.h>
50 #include <linux/dmi.h>
51 #include <linux/workqueue.h>
52 #include <net/genetlink.h>
53 #include <net/net_namespace.h>
54 #include <net/netns/generic.h>
55
56 #include "datapath.h"
57 #include "flow.h"
58 #include "flow_netlink.h"
59 #include "vport-internal_dev.h"
60 #include "vport-netdev.h"
61
62 int ovs_net_id __read_mostly;
63
64 static void ovs_notify(struct genl_family *family,
65 struct sk_buff *skb, struct genl_info *info)
66 {
67 genl_notify(family, skb, genl_info_net(info), info->snd_portid,
68 0, info->nlhdr, GFP_KERNEL);
69 }
70
71 /**
72 * DOC: Locking:
73 *
74 * All writes e.g. Writes to device state (add/remove datapath, port, set
75 * operations on vports, etc.), Writes to other state (flow table
76 * modifications, set miscellaneous datapath parameters, etc.) are protected
77 * by ovs_lock.
78 *
79 * Reads are protected by RCU.
80 *
81 * There are a few special cases (mostly stats) that have their own
82 * synchronization but they nest under all of above and don't interact with
83 * each other.
84 *
85 * The RTNL lock nests inside ovs_mutex.
86 */
87
88 static DEFINE_MUTEX(ovs_mutex);
89
90 void ovs_lock(void)
91 {
92 mutex_lock(&ovs_mutex);
93 }
94
95 void ovs_unlock(void)
96 {
97 mutex_unlock(&ovs_mutex);
98 }
99
100 #ifdef CONFIG_LOCKDEP
101 int lockdep_ovsl_is_held(void)
102 {
103 if (debug_locks)
104 return lockdep_is_held(&ovs_mutex);
105 else
106 return 1;
107 }
108 #endif
109
110 static struct vport *new_vport(const struct vport_parms *);
111 static int queue_gso_packets(struct net *, int dp_ifindex, struct sk_buff *,
112 const struct dp_upcall_info *);
113 static int queue_userspace_packet(struct net *, int dp_ifindex,
114 struct sk_buff *,
115 const struct dp_upcall_info *);
116
117 /* Must be called with rcu_read_lock or ovs_mutex. */
118 static struct datapath *get_dp(struct net *net, int dp_ifindex)
119 {
120 struct datapath *dp = NULL;
121 struct net_device *dev;
122
123 rcu_read_lock();
124 dev = dev_get_by_index_rcu(net, dp_ifindex);
125 if (dev) {
126 struct vport *vport = ovs_internal_dev_get_vport(dev);
127 if (vport)
128 dp = vport->dp;
129 }
130 rcu_read_unlock();
131
132 return dp;
133 }
134
135 /* Must be called with rcu_read_lock or ovs_mutex. */
136 const char *ovs_dp_name(const struct datapath *dp)
137 {
138 struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
139 return vport->ops->get_name(vport);
140 }
141
142 static int get_dpifindex(struct datapath *dp)
143 {
144 struct vport *local;
145 int ifindex;
146
147 rcu_read_lock();
148
149 local = ovs_vport_rcu(dp, OVSP_LOCAL);
150 if (local)
151 ifindex = netdev_vport_priv(local)->dev->ifindex;
152 else
153 ifindex = 0;
154
155 rcu_read_unlock();
156
157 return ifindex;
158 }
159
160 static void destroy_dp_rcu(struct rcu_head *rcu)
161 {
162 struct datapath *dp = container_of(rcu, struct datapath, rcu);
163
164 ovs_flow_tbl_destroy(&dp->table);
165 free_percpu(dp->stats_percpu);
166 release_net(ovs_dp_get_net(dp));
167 kfree(dp->ports);
168 kfree(dp);
169 }
170
171 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
172 u16 port_no)
173 {
174 return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
175 }
176
177 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
178 {
179 struct vport *vport;
180 struct hlist_head *head;
181
182 head = vport_hash_bucket(dp, port_no);
183 hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
184 if (vport->port_no == port_no)
185 return vport;
186 }
187 return NULL;
188 }
189
190 /* Called with ovs_mutex. */
191 static struct vport *new_vport(const struct vport_parms *parms)
192 {
193 struct vport *vport;
194
195 vport = ovs_vport_add(parms);
196 if (!IS_ERR(vport)) {
197 struct datapath *dp = parms->dp;
198 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
199
200 hlist_add_head_rcu(&vport->dp_hash_node, head);
201 }
202 return vport;
203 }
204
205 void ovs_dp_detach_port(struct vport *p)
206 {
207 ASSERT_OVSL();
208
209 /* First drop references to device. */
210 hlist_del_rcu(&p->dp_hash_node);
211
212 /* Then destroy it. */
213 ovs_vport_del(p);
214 }
215
216 /* Must be called with rcu_read_lock. */
217 void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
218 {
219 struct datapath *dp = p->dp;
220 struct sw_flow *flow;
221 struct dp_stats_percpu *stats;
222 struct sw_flow_key key;
223 u64 *stats_counter;
224 u32 n_mask_hit;
225 int error;
226
227 stats = this_cpu_ptr(dp->stats_percpu);
228
229 /* Extract flow from 'skb' into 'key'. */
230 error = ovs_flow_extract(skb, p->port_no, &key);
231 if (unlikely(error)) {
232 kfree_skb(skb);
233 return;
234 }
235
236 /* Look up flow. */
237 flow = ovs_flow_tbl_lookup(&dp->table, &key, &n_mask_hit);
238 if (unlikely(!flow)) {
239 struct dp_upcall_info upcall;
240
241 upcall.cmd = OVS_PACKET_CMD_MISS;
242 upcall.key = &key;
243 upcall.userdata = NULL;
244 upcall.portid = p->upcall_portid;
245 ovs_dp_upcall(dp, skb, &upcall);
246 consume_skb(skb);
247 stats_counter = &stats->n_missed;
248 goto out;
249 }
250
251 OVS_CB(skb)->flow = flow;
252 OVS_CB(skb)->pkt_key = &key;
253
254 stats_counter = &stats->n_hit;
255 ovs_flow_used(OVS_CB(skb)->flow, skb);
256 ovs_execute_actions(dp, skb);
257
258 out:
259 /* Update datapath statistics. */
260 u64_stats_update_begin(&stats->sync);
261 (*stats_counter)++;
262 stats->n_mask_hit += n_mask_hit;
263 u64_stats_update_end(&stats->sync);
264 }
265
266 static struct genl_family dp_packet_genl_family = {
267 .id = GENL_ID_GENERATE,
268 .hdrsize = sizeof(struct ovs_header),
269 .name = OVS_PACKET_FAMILY,
270 .version = OVS_PACKET_VERSION,
271 .maxattr = OVS_PACKET_ATTR_MAX,
272 .netnsok = true,
273 .parallel_ops = true,
274 };
275
276 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
277 const struct dp_upcall_info *upcall_info)
278 {
279 struct dp_stats_percpu *stats;
280 int dp_ifindex;
281 int err;
282
283 if (upcall_info->portid == 0) {
284 err = -ENOTCONN;
285 goto err;
286 }
287
288 dp_ifindex = get_dpifindex(dp);
289 if (!dp_ifindex) {
290 err = -ENODEV;
291 goto err;
292 }
293
294 if (!skb_is_gso(skb))
295 err = queue_userspace_packet(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
296 else
297 err = queue_gso_packets(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
298 if (err)
299 goto err;
300
301 return 0;
302
303 err:
304 stats = this_cpu_ptr(dp->stats_percpu);
305
306 u64_stats_update_begin(&stats->sync);
307 stats->n_lost++;
308 u64_stats_update_end(&stats->sync);
309
310 return err;
311 }
312
313 static int queue_gso_packets(struct net *net, int dp_ifindex,
314 struct sk_buff *skb,
315 const struct dp_upcall_info *upcall_info)
316 {
317 unsigned short gso_type = skb_shinfo(skb)->gso_type;
318 struct dp_upcall_info later_info;
319 struct sw_flow_key later_key;
320 struct sk_buff *segs, *nskb;
321 int err;
322
323 segs = __skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM, false);
324 if (IS_ERR(segs))
325 return PTR_ERR(segs);
326
327 /* Queue all of the segments. */
328 skb = segs;
329 do {
330 err = queue_userspace_packet(net, dp_ifindex, skb, upcall_info);
331 if (err)
332 break;
333
334 if (skb == segs && gso_type & SKB_GSO_UDP) {
335 /* The initial flow key extracted by ovs_flow_extract()
336 * in this case is for a first fragment, so we need to
337 * properly mark later fragments.
338 */
339 later_key = *upcall_info->key;
340 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
341
342 later_info = *upcall_info;
343 later_info.key = &later_key;
344 upcall_info = &later_info;
345 }
346 } while ((skb = skb->next));
347
348 /* Free all of the segments. */
349 skb = segs;
350 do {
351 nskb = skb->next;
352 if (err)
353 kfree_skb(skb);
354 else
355 consume_skb(skb);
356 } while ((skb = nskb));
357 return err;
358 }
359
360 static size_t key_attr_size(void)
361 {
362 return nla_total_size(4) /* OVS_KEY_ATTR_PRIORITY */
363 + nla_total_size(0) /* OVS_KEY_ATTR_TUNNEL */
364 + nla_total_size(8) /* OVS_TUNNEL_KEY_ATTR_ID */
365 + nla_total_size(4) /* OVS_TUNNEL_KEY_ATTR_IPV4_SRC */
366 + nla_total_size(4) /* OVS_TUNNEL_KEY_ATTR_IPV4_DST */
367 + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TOS */
368 + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TTL */
369 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
370 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_CSUM */
371 + nla_total_size(4) /* OVS_KEY_ATTR_IN_PORT */
372 + nla_total_size(4) /* OVS_KEY_ATTR_SKB_MARK */
373 + nla_total_size(12) /* OVS_KEY_ATTR_ETHERNET */
374 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
375 + nla_total_size(4) /* OVS_KEY_ATTR_8021Q */
376 + nla_total_size(0) /* OVS_KEY_ATTR_ENCAP */
377 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
378 + nla_total_size(40) /* OVS_KEY_ATTR_IPV6 */
379 + nla_total_size(2) /* OVS_KEY_ATTR_ICMPV6 */
380 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
381 }
382
383 static size_t upcall_msg_size(const struct sk_buff *skb,
384 const struct nlattr *userdata)
385 {
386 size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
387 + nla_total_size(skb->len) /* OVS_PACKET_ATTR_PACKET */
388 + nla_total_size(key_attr_size()); /* OVS_PACKET_ATTR_KEY */
389
390 /* OVS_PACKET_ATTR_USERDATA */
391 if (userdata)
392 size += NLA_ALIGN(userdata->nla_len);
393
394 return size;
395 }
396
397 static int queue_userspace_packet(struct net *net, int dp_ifindex,
398 struct sk_buff *skb,
399 const struct dp_upcall_info *upcall_info)
400 {
401 struct ovs_header *upcall;
402 struct sk_buff *nskb = NULL;
403 struct sk_buff *user_skb; /* to be queued to userspace */
404 struct nlattr *nla;
405 int err;
406
407 if (vlan_tx_tag_present(skb)) {
408 nskb = skb_clone(skb, GFP_ATOMIC);
409 if (!nskb)
410 return -ENOMEM;
411
412 nskb = __vlan_put_tag(nskb, nskb->vlan_proto, vlan_tx_tag_get(nskb));
413 if (!nskb)
414 return -ENOMEM;
415
416 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 user_skb = genlmsg_new(upcall_msg_size(skb, upcall_info->userdata), GFP_ATOMIC);
426 if (!user_skb) {
427 err = -ENOMEM;
428 goto out;
429 }
430
431 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
432 0, upcall_info->cmd);
433 upcall->dp_ifindex = dp_ifindex;
434
435 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
436 ovs_nla_put_flow(upcall_info->key, upcall_info->key, user_skb);
437 nla_nest_end(user_skb, nla);
438
439 if (upcall_info->userdata)
440 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
441 nla_len(upcall_info->userdata),
442 nla_data(upcall_info->userdata));
443
444 nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
445
446 skb_copy_and_csum_dev(skb, nla_data(nla));
447
448 genlmsg_end(user_skb, upcall);
449 err = genlmsg_unicast(net, user_skb, upcall_info->portid);
450
451 out:
452 kfree_skb(nskb);
453 return err;
454 }
455
456 static void clear_stats(struct sw_flow *flow)
457 {
458 flow->used = 0;
459 flow->tcp_flags = 0;
460 flow->packet_count = 0;
461 flow->byte_count = 0;
462 }
463
464 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
465 {
466 struct ovs_header *ovs_header = info->userhdr;
467 struct nlattr **a = info->attrs;
468 struct sw_flow_actions *acts;
469 struct sk_buff *packet;
470 struct sw_flow *flow;
471 struct datapath *dp;
472 struct ethhdr *eth;
473 int len;
474 int err;
475
476 err = -EINVAL;
477 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
478 !a[OVS_PACKET_ATTR_ACTIONS])
479 goto err;
480
481 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
482 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
483 err = -ENOMEM;
484 if (!packet)
485 goto err;
486 skb_reserve(packet, NET_IP_ALIGN);
487
488 nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
489
490 skb_reset_mac_header(packet);
491 eth = eth_hdr(packet);
492
493 /* Normally, setting the skb 'protocol' field would be handled by a
494 * call to eth_type_trans(), but it assumes there's a sending
495 * device, which we may not have. */
496 if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
497 packet->protocol = eth->h_proto;
498 else
499 packet->protocol = htons(ETH_P_802_2);
500
501 /* Build an sw_flow for sending this packet. */
502 flow = ovs_flow_alloc();
503 err = PTR_ERR(flow);
504 if (IS_ERR(flow))
505 goto err_kfree_skb;
506
507 err = ovs_flow_extract(packet, -1, &flow->key);
508 if (err)
509 goto err_flow_free;
510
511 err = ovs_nla_get_flow_metadata(flow, a[OVS_PACKET_ATTR_KEY]);
512 if (err)
513 goto err_flow_free;
514 acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_PACKET_ATTR_ACTIONS]));
515 err = PTR_ERR(acts);
516 if (IS_ERR(acts))
517 goto err_flow_free;
518
519 err = ovs_nla_copy_actions(a[OVS_PACKET_ATTR_ACTIONS],
520 &flow->key, 0, &acts);
521 rcu_assign_pointer(flow->sf_acts, acts);
522 if (err)
523 goto err_flow_free;
524
525 OVS_CB(packet)->flow = flow;
526 OVS_CB(packet)->pkt_key = &flow->key;
527 packet->priority = flow->key.phy.priority;
528 packet->mark = flow->key.phy.skb_mark;
529
530 rcu_read_lock();
531 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
532 err = -ENODEV;
533 if (!dp)
534 goto err_unlock;
535
536 local_bh_disable();
537 err = ovs_execute_actions(dp, packet);
538 local_bh_enable();
539 rcu_read_unlock();
540
541 ovs_flow_free(flow, false);
542 return err;
543
544 err_unlock:
545 rcu_read_unlock();
546 err_flow_free:
547 ovs_flow_free(flow, false);
548 err_kfree_skb:
549 kfree_skb(packet);
550 err:
551 return err;
552 }
553
554 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
555 [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
556 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
557 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
558 };
559
560 static const struct genl_ops dp_packet_genl_ops[] = {
561 { .cmd = OVS_PACKET_CMD_EXECUTE,
562 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
563 .policy = packet_policy,
564 .doit = ovs_packet_cmd_execute
565 }
566 };
567
568 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats,
569 struct ovs_dp_megaflow_stats *mega_stats)
570 {
571 int i;
572
573 memset(mega_stats, 0, sizeof(*mega_stats));
574
575 stats->n_flows = ovs_flow_tbl_count(&dp->table);
576 mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
577
578 stats->n_hit = stats->n_missed = stats->n_lost = 0;
579
580 for_each_possible_cpu(i) {
581 const struct dp_stats_percpu *percpu_stats;
582 struct dp_stats_percpu local_stats;
583 unsigned int start;
584
585 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
586
587 do {
588 start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
589 local_stats = *percpu_stats;
590 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
591
592 stats->n_hit += local_stats.n_hit;
593 stats->n_missed += local_stats.n_missed;
594 stats->n_lost += local_stats.n_lost;
595 mega_stats->n_mask_hit += local_stats.n_mask_hit;
596 }
597 }
598
599 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
600 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
601 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
602 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
603 };
604
605 static struct genl_family dp_flow_genl_family = {
606 .id = GENL_ID_GENERATE,
607 .hdrsize = sizeof(struct ovs_header),
608 .name = OVS_FLOW_FAMILY,
609 .version = OVS_FLOW_VERSION,
610 .maxattr = OVS_FLOW_ATTR_MAX,
611 .netnsok = true,
612 .parallel_ops = true,
613 };
614
615 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
616 .name = OVS_FLOW_MCGROUP
617 };
618
619 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
620 {
621 return NLMSG_ALIGN(sizeof(struct ovs_header))
622 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
623 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_MASK */
624 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
625 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
626 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
627 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
628 }
629
630 /* Called with ovs_mutex. */
631 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
632 struct sk_buff *skb, u32 portid,
633 u32 seq, u32 flags, u8 cmd)
634 {
635 const int skb_orig_len = skb->len;
636 struct nlattr *start;
637 struct ovs_flow_stats stats;
638 struct ovs_header *ovs_header;
639 struct nlattr *nla;
640 unsigned long used;
641 u8 tcp_flags;
642 int err;
643
644 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
645 if (!ovs_header)
646 return -EMSGSIZE;
647
648 ovs_header->dp_ifindex = get_dpifindex(dp);
649
650 /* Fill flow key. */
651 nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
652 if (!nla)
653 goto nla_put_failure;
654
655 err = ovs_nla_put_flow(&flow->unmasked_key, &flow->unmasked_key, skb);
656 if (err)
657 goto error;
658 nla_nest_end(skb, nla);
659
660 nla = nla_nest_start(skb, OVS_FLOW_ATTR_MASK);
661 if (!nla)
662 goto nla_put_failure;
663
664 err = ovs_nla_put_flow(&flow->key, &flow->mask->key, skb);
665 if (err)
666 goto error;
667
668 nla_nest_end(skb, nla);
669
670 spin_lock_bh(&flow->lock);
671 used = flow->used;
672 stats.n_packets = flow->packet_count;
673 stats.n_bytes = flow->byte_count;
674 tcp_flags = (u8)ntohs(flow->tcp_flags);
675 spin_unlock_bh(&flow->lock);
676
677 if (used &&
678 nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
679 goto nla_put_failure;
680
681 if (stats.n_packets &&
682 nla_put(skb, OVS_FLOW_ATTR_STATS,
683 sizeof(struct ovs_flow_stats), &stats))
684 goto nla_put_failure;
685
686 if (tcp_flags &&
687 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags))
688 goto nla_put_failure;
689
690 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
691 * this is the first flow to be dumped into 'skb'. This is unusual for
692 * Netlink but individual action lists can be longer than
693 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
694 * The userspace caller can always fetch the actions separately if it
695 * really wants them. (Most userspace callers in fact don't care.)
696 *
697 * This can only fail for dump operations because the skb is always
698 * properly sized for single flows.
699 */
700 start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
701 if (start) {
702 const struct sw_flow_actions *sf_acts;
703
704 sf_acts = rcu_dereference_check(flow->sf_acts,
705 lockdep_ovsl_is_held());
706
707 err = ovs_nla_put_actions(sf_acts->actions,
708 sf_acts->actions_len, skb);
709 if (!err)
710 nla_nest_end(skb, start);
711 else {
712 if (skb_orig_len)
713 goto error;
714
715 nla_nest_cancel(skb, start);
716 }
717 } else if (skb_orig_len)
718 goto nla_put_failure;
719
720 return genlmsg_end(skb, ovs_header);
721
722 nla_put_failure:
723 err = -EMSGSIZE;
724 error:
725 genlmsg_cancel(skb, ovs_header);
726 return err;
727 }
728
729 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
730 {
731 const struct sw_flow_actions *sf_acts;
732
733 sf_acts = ovsl_dereference(flow->sf_acts);
734
735 return genlmsg_new(ovs_flow_cmd_msg_size(sf_acts), GFP_KERNEL);
736 }
737
738 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
739 struct datapath *dp,
740 u32 portid, u32 seq, u8 cmd)
741 {
742 struct sk_buff *skb;
743 int retval;
744
745 skb = ovs_flow_cmd_alloc_info(flow);
746 if (!skb)
747 return ERR_PTR(-ENOMEM);
748
749 retval = ovs_flow_cmd_fill_info(flow, dp, skb, portid, seq, 0, cmd);
750 BUG_ON(retval < 0);
751 return skb;
752 }
753
754 static struct sw_flow *__ovs_flow_tbl_lookup(struct flow_table *tbl,
755 const struct sw_flow_key *key)
756 {
757 u32 __always_unused n_mask_hit;
758
759 return ovs_flow_tbl_lookup(tbl, key, &n_mask_hit);
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 int error;
774
775 /* Extract key. */
776 error = -EINVAL;
777 if (!a[OVS_FLOW_ATTR_KEY])
778 goto error;
779
780 ovs_match_init(&match, &key, &mask);
781 error = ovs_nla_get_match(&match,
782 a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
783 if (error)
784 goto error;
785
786 /* Validate actions. */
787 if (a[OVS_FLOW_ATTR_ACTIONS]) {
788 acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
789 error = PTR_ERR(acts);
790 if (IS_ERR(acts))
791 goto error;
792
793 ovs_flow_mask_key(&masked_key, &key, &mask);
794 error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS],
795 &masked_key, 0, &acts);
796 if (error) {
797 OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
798 goto err_kfree;
799 }
800 } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
801 error = -EINVAL;
802 goto error;
803 }
804
805 ovs_lock();
806 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
807 error = -ENODEV;
808 if (!dp)
809 goto err_unlock_ovs;
810
811 /* Check if this is a duplicate flow */
812 flow = __ovs_flow_tbl_lookup(&dp->table, &key);
813 if (!flow) {
814 /* Bail out if we're not allowed to create a new flow. */
815 error = -ENOENT;
816 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
817 goto err_unlock_ovs;
818
819 /* Allocate flow. */
820 flow = ovs_flow_alloc();
821 if (IS_ERR(flow)) {
822 error = PTR_ERR(flow);
823 goto err_unlock_ovs;
824 }
825 clear_stats(flow);
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->snd_portid,
839 info->snd_seq, OVS_FLOW_CMD_NEW);
840 } else {
841 /* We found a matching flow. */
842 struct sw_flow_actions *old_acts;
843
844 /* Bail out if we're not allowed to modify an existing flow.
845 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
846 * because Generic Netlink treats the latter as a dump
847 * request. We also accept NLM_F_EXCL in case that bug ever
848 * gets fixed.
849 */
850 error = -EEXIST;
851 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
852 info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
853 goto err_unlock_ovs;
854
855 /* The unmasked key has to be the same for flow updates. */
856 error = -EINVAL;
857 if (!ovs_flow_cmp_unmasked_key(flow, &match)) {
858 OVS_NLERR("Flow modification message rejected, unmasked key does not match.\n");
859 goto err_unlock_ovs;
860 }
861
862 /* Update actions. */
863 old_acts = ovsl_dereference(flow->sf_acts);
864 rcu_assign_pointer(flow->sf_acts, acts);
865 ovs_nla_free_flow_actions(old_acts);
866
867 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
868 info->snd_seq, OVS_FLOW_CMD_NEW);
869
870 /* Clear stats. */
871 if (a[OVS_FLOW_ATTR_CLEAR]) {
872 spin_lock_bh(&flow->lock);
873 clear_stats(flow);
874 spin_unlock_bh(&flow->lock);
875 }
876 }
877 ovs_unlock();
878
879 if (!IS_ERR(reply))
880 ovs_notify(&dp_flow_genl_family, reply, info);
881 else
882 genl_set_err(&dp_flow_genl_family, sock_net(skb->sk), 0,
883 0, PTR_ERR(reply));
884 return 0;
885
886 err_flow_free:
887 ovs_flow_free(flow, false);
888 err_unlock_ovs:
889 ovs_unlock();
890 err_kfree:
891 kfree(acts);
892 error:
893 return error;
894 }
895
896 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
897 {
898 struct nlattr **a = info->attrs;
899 struct ovs_header *ovs_header = info->userhdr;
900 struct sw_flow_key key;
901 struct sk_buff *reply;
902 struct sw_flow *flow;
903 struct datapath *dp;
904 struct sw_flow_match match;
905 int err;
906
907 if (!a[OVS_FLOW_ATTR_KEY]) {
908 OVS_NLERR("Flow get message rejected, Key attribute missing.\n");
909 return -EINVAL;
910 }
911
912 ovs_match_init(&match, &key, NULL);
913 err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
914 if (err)
915 return err;
916
917 ovs_lock();
918 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
919 if (!dp) {
920 err = -ENODEV;
921 goto unlock;
922 }
923
924 flow = __ovs_flow_tbl_lookup(&dp->table, &key);
925 if (!flow || !ovs_flow_cmp_unmasked_key(flow, &match)) {
926 err = -ENOENT;
927 goto unlock;
928 }
929
930 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
931 info->snd_seq, OVS_FLOW_CMD_NEW);
932 if (IS_ERR(reply)) {
933 err = PTR_ERR(reply);
934 goto unlock;
935 }
936
937 ovs_unlock();
938 return genlmsg_reply(reply, info);
939 unlock:
940 ovs_unlock();
941 return err;
942 }
943
944 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
945 {
946 struct nlattr **a = info->attrs;
947 struct ovs_header *ovs_header = info->userhdr;
948 struct sw_flow_key key;
949 struct sk_buff *reply;
950 struct sw_flow *flow;
951 struct datapath *dp;
952 struct sw_flow_match match;
953 int err;
954
955 ovs_lock();
956 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
957 if (!dp) {
958 err = -ENODEV;
959 goto unlock;
960 }
961
962 if (!a[OVS_FLOW_ATTR_KEY]) {
963 err = ovs_flow_tbl_flush(&dp->table);
964 goto unlock;
965 }
966
967 ovs_match_init(&match, &key, NULL);
968 err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
969 if (err)
970 goto unlock;
971
972 flow = __ovs_flow_tbl_lookup(&dp->table, &key);
973 if (!flow || !ovs_flow_cmp_unmasked_key(flow, &match)) {
974 err = -ENOENT;
975 goto unlock;
976 }
977
978 reply = ovs_flow_cmd_alloc_info(flow);
979 if (!reply) {
980 err = -ENOMEM;
981 goto unlock;
982 }
983
984 ovs_flow_tbl_remove(&dp->table, flow);
985
986 err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_portid,
987 info->snd_seq, 0, OVS_FLOW_CMD_DEL);
988 BUG_ON(err < 0);
989
990 ovs_flow_free(flow, true);
991 ovs_unlock();
992
993 ovs_notify(&dp_flow_genl_family, reply, info);
994 return 0;
995 unlock:
996 ovs_unlock();
997 return err;
998 }
999
1000 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1001 {
1002 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1003 struct table_instance *ti;
1004 struct datapath *dp;
1005
1006 rcu_read_lock();
1007 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1008 if (!dp) {
1009 rcu_read_unlock();
1010 return -ENODEV;
1011 }
1012
1013 ti = rcu_dereference(dp->table.ti);
1014 for (;;) {
1015 struct sw_flow *flow;
1016 u32 bucket, obj;
1017
1018 bucket = cb->args[0];
1019 obj = cb->args[1];
1020 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1021 if (!flow)
1022 break;
1023
1024 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1025 NETLINK_CB(cb->skb).portid,
1026 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1027 OVS_FLOW_CMD_NEW) < 0)
1028 break;
1029
1030 cb->args[0] = bucket;
1031 cb->args[1] = obj;
1032 }
1033 rcu_read_unlock();
1034 return skb->len;
1035 }
1036
1037 static const struct genl_ops dp_flow_genl_ops[] = {
1038 { .cmd = OVS_FLOW_CMD_NEW,
1039 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1040 .policy = flow_policy,
1041 .doit = ovs_flow_cmd_new_or_set
1042 },
1043 { .cmd = OVS_FLOW_CMD_DEL,
1044 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1045 .policy = flow_policy,
1046 .doit = ovs_flow_cmd_del
1047 },
1048 { .cmd = OVS_FLOW_CMD_GET,
1049 .flags = 0, /* OK for unprivileged users. */
1050 .policy = flow_policy,
1051 .doit = ovs_flow_cmd_get,
1052 .dumpit = ovs_flow_cmd_dump
1053 },
1054 { .cmd = OVS_FLOW_CMD_SET,
1055 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1056 .policy = flow_policy,
1057 .doit = ovs_flow_cmd_new_or_set,
1058 },
1059 };
1060
1061 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1062 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1063 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1064 };
1065
1066 static struct genl_family dp_datapath_genl_family = {
1067 .id = GENL_ID_GENERATE,
1068 .hdrsize = sizeof(struct ovs_header),
1069 .name = OVS_DATAPATH_FAMILY,
1070 .version = OVS_DATAPATH_VERSION,
1071 .maxattr = OVS_DP_ATTR_MAX,
1072 .netnsok = true,
1073 .parallel_ops = true,
1074 };
1075
1076 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1077 .name = OVS_DATAPATH_MCGROUP
1078 };
1079
1080 static size_t ovs_dp_cmd_msg_size(void)
1081 {
1082 size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1083
1084 msgsize += nla_total_size(IFNAMSIZ);
1085 msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1086 msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1087
1088 return msgsize;
1089 }
1090
1091 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1092 u32 portid, u32 seq, u32 flags, u8 cmd)
1093 {
1094 struct ovs_header *ovs_header;
1095 struct ovs_dp_stats dp_stats;
1096 struct ovs_dp_megaflow_stats dp_megaflow_stats;
1097 int err;
1098
1099 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1100 flags, cmd);
1101 if (!ovs_header)
1102 goto error;
1103
1104 ovs_header->dp_ifindex = get_dpifindex(dp);
1105
1106 rcu_read_lock();
1107 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1108 rcu_read_unlock();
1109 if (err)
1110 goto nla_put_failure;
1111
1112 get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1113 if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1114 &dp_stats))
1115 goto nla_put_failure;
1116
1117 if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1118 sizeof(struct ovs_dp_megaflow_stats),
1119 &dp_megaflow_stats))
1120 goto nla_put_failure;
1121
1122 return genlmsg_end(skb, ovs_header);
1123
1124 nla_put_failure:
1125 genlmsg_cancel(skb, ovs_header);
1126 error:
1127 return -EMSGSIZE;
1128 }
1129
1130 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 portid,
1131 u32 seq, u8 cmd)
1132 {
1133 struct sk_buff *skb;
1134 int retval;
1135
1136 skb = genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1137 if (!skb)
1138 return ERR_PTR(-ENOMEM);
1139
1140 retval = ovs_dp_cmd_fill_info(dp, skb, portid, seq, 0, cmd);
1141 if (retval < 0) {
1142 kfree_skb(skb);
1143 return ERR_PTR(retval);
1144 }
1145 return skb;
1146 }
1147
1148 /* Called with ovs_mutex. */
1149 static struct datapath *lookup_datapath(struct net *net,
1150 struct ovs_header *ovs_header,
1151 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1152 {
1153 struct datapath *dp;
1154
1155 if (!a[OVS_DP_ATTR_NAME])
1156 dp = get_dp(net, ovs_header->dp_ifindex);
1157 else {
1158 struct vport *vport;
1159
1160 rcu_read_lock();
1161 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1162 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1163 rcu_read_unlock();
1164 }
1165 return dp ? dp : ERR_PTR(-ENODEV);
1166 }
1167
1168 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1169 {
1170 struct nlattr **a = info->attrs;
1171 struct vport_parms parms;
1172 struct sk_buff *reply;
1173 struct datapath *dp;
1174 struct vport *vport;
1175 struct ovs_net *ovs_net;
1176 int err, i;
1177
1178 err = -EINVAL;
1179 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1180 goto err;
1181
1182 ovs_lock();
1183
1184 err = -ENOMEM;
1185 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1186 if (dp == NULL)
1187 goto err_unlock_ovs;
1188
1189 ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1190
1191 /* Allocate table. */
1192 err = ovs_flow_tbl_init(&dp->table);
1193 if (err)
1194 goto err_free_dp;
1195
1196 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1197 if (!dp->stats_percpu) {
1198 err = -ENOMEM;
1199 goto err_destroy_table;
1200 }
1201
1202 for_each_possible_cpu(i) {
1203 struct dp_stats_percpu *dpath_stats;
1204 dpath_stats = per_cpu_ptr(dp->stats_percpu, i);
1205 u64_stats_init(&dpath_stats->sync);
1206 }
1207
1208 dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1209 GFP_KERNEL);
1210 if (!dp->ports) {
1211 err = -ENOMEM;
1212 goto err_destroy_percpu;
1213 }
1214
1215 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1216 INIT_HLIST_HEAD(&dp->ports[i]);
1217
1218 /* Set up our datapath device. */
1219 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1220 parms.type = OVS_VPORT_TYPE_INTERNAL;
1221 parms.options = NULL;
1222 parms.dp = dp;
1223 parms.port_no = OVSP_LOCAL;
1224 parms.upcall_portid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1225
1226 vport = new_vport(&parms);
1227 if (IS_ERR(vport)) {
1228 err = PTR_ERR(vport);
1229 if (err == -EBUSY)
1230 err = -EEXIST;
1231
1232 goto err_destroy_ports_array;
1233 }
1234
1235 reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1236 info->snd_seq, OVS_DP_CMD_NEW);
1237 err = PTR_ERR(reply);
1238 if (IS_ERR(reply))
1239 goto err_destroy_local_port;
1240
1241 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1242 list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1243
1244 ovs_unlock();
1245
1246 ovs_notify(&dp_datapath_genl_family, reply, info);
1247 return 0;
1248
1249 err_destroy_local_port:
1250 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1251 err_destroy_ports_array:
1252 kfree(dp->ports);
1253 err_destroy_percpu:
1254 free_percpu(dp->stats_percpu);
1255 err_destroy_table:
1256 ovs_flow_tbl_destroy(&dp->table);
1257 err_free_dp:
1258 release_net(ovs_dp_get_net(dp));
1259 kfree(dp);
1260 err_unlock_ovs:
1261 ovs_unlock();
1262 err:
1263 return err;
1264 }
1265
1266 /* Called with ovs_mutex. */
1267 static void __dp_destroy(struct datapath *dp)
1268 {
1269 int i;
1270
1271 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1272 struct vport *vport;
1273 struct hlist_node *n;
1274
1275 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1276 if (vport->port_no != OVSP_LOCAL)
1277 ovs_dp_detach_port(vport);
1278 }
1279
1280 list_del_rcu(&dp->list_node);
1281
1282 /* OVSP_LOCAL is datapath internal port. We need to make sure that
1283 * all port in datapath are destroyed first before freeing datapath.
1284 */
1285 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1286
1287 call_rcu(&dp->rcu, destroy_dp_rcu);
1288 }
1289
1290 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1291 {
1292 struct sk_buff *reply;
1293 struct datapath *dp;
1294 int err;
1295
1296 ovs_lock();
1297 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1298 err = PTR_ERR(dp);
1299 if (IS_ERR(dp))
1300 goto unlock;
1301
1302 reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1303 info->snd_seq, OVS_DP_CMD_DEL);
1304 err = PTR_ERR(reply);
1305 if (IS_ERR(reply))
1306 goto unlock;
1307
1308 __dp_destroy(dp);
1309 ovs_unlock();
1310
1311 ovs_notify(&dp_datapath_genl_family, reply, info);
1312
1313 return 0;
1314 unlock:
1315 ovs_unlock();
1316 return err;
1317 }
1318
1319 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1320 {
1321 struct sk_buff *reply;
1322 struct datapath *dp;
1323 int err;
1324
1325 ovs_lock();
1326 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1327 err = PTR_ERR(dp);
1328 if (IS_ERR(dp))
1329 goto unlock;
1330
1331 reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1332 info->snd_seq, OVS_DP_CMD_NEW);
1333 if (IS_ERR(reply)) {
1334 err = PTR_ERR(reply);
1335 genl_set_err(&dp_datapath_genl_family, sock_net(skb->sk), 0,
1336 0, err);
1337 err = 0;
1338 goto unlock;
1339 }
1340
1341 ovs_unlock();
1342 ovs_notify(&dp_datapath_genl_family, reply, info);
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
1606 exit_unlock:
1607 ovs_unlock();
1608 exit:
1609 return err;
1610 }
1611
1612 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1613 {
1614 struct nlattr **a = info->attrs;
1615 struct sk_buff *reply;
1616 struct vport *vport;
1617 int err;
1618
1619 ovs_lock();
1620 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1621 err = PTR_ERR(vport);
1622 if (IS_ERR(vport))
1623 goto exit_unlock;
1624
1625 if (a[OVS_VPORT_ATTR_TYPE] &&
1626 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
1627 err = -EINVAL;
1628 goto exit_unlock;
1629 }
1630
1631 reply = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1632 if (!reply) {
1633 err = -ENOMEM;
1634 goto exit_unlock;
1635 }
1636
1637 if (a[OVS_VPORT_ATTR_OPTIONS]) {
1638 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1639 if (err)
1640 goto exit_free;
1641 }
1642
1643 if (a[OVS_VPORT_ATTR_UPCALL_PID])
1644 vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1645
1646 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1647 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1648 BUG_ON(err < 0);
1649
1650 ovs_unlock();
1651 ovs_notify(&dp_vport_genl_family, reply, info);
1652 return 0;
1653
1654 exit_free:
1655 kfree_skb(reply);
1656 exit_unlock:
1657 ovs_unlock();
1658 return err;
1659 }
1660
1661 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1662 {
1663 struct nlattr **a = info->attrs;
1664 struct sk_buff *reply;
1665 struct vport *vport;
1666 int err;
1667
1668 ovs_lock();
1669 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1670 err = PTR_ERR(vport);
1671 if (IS_ERR(vport))
1672 goto exit_unlock;
1673
1674 if (vport->port_no == OVSP_LOCAL) {
1675 err = -EINVAL;
1676 goto exit_unlock;
1677 }
1678
1679 reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
1680 info->snd_seq, OVS_VPORT_CMD_DEL);
1681 err = PTR_ERR(reply);
1682 if (IS_ERR(reply))
1683 goto exit_unlock;
1684
1685 err = 0;
1686 ovs_dp_detach_port(vport);
1687
1688 ovs_notify(&dp_vport_genl_family, reply, info);
1689
1690 exit_unlock:
1691 ovs_unlock();
1692 return err;
1693 }
1694
1695 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1696 {
1697 struct nlattr **a = info->attrs;
1698 struct ovs_header *ovs_header = info->userhdr;
1699 struct sk_buff *reply;
1700 struct vport *vport;
1701 int err;
1702
1703 rcu_read_lock();
1704 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1705 err = PTR_ERR(vport);
1706 if (IS_ERR(vport))
1707 goto exit_unlock;
1708
1709 reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
1710 info->snd_seq, OVS_VPORT_CMD_NEW);
1711 err = PTR_ERR(reply);
1712 if (IS_ERR(reply))
1713 goto exit_unlock;
1714
1715 rcu_read_unlock();
1716
1717 return genlmsg_reply(reply, info);
1718
1719 exit_unlock:
1720 rcu_read_unlock();
1721 return err;
1722 }
1723
1724 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1725 {
1726 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1727 struct datapath *dp;
1728 int bucket = cb->args[0], skip = cb->args[1];
1729 int i, j = 0;
1730
1731 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1732 if (!dp)
1733 return -ENODEV;
1734
1735 rcu_read_lock();
1736 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
1737 struct vport *vport;
1738
1739 j = 0;
1740 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1741 if (j >= skip &&
1742 ovs_vport_cmd_fill_info(vport, skb,
1743 NETLINK_CB(cb->skb).portid,
1744 cb->nlh->nlmsg_seq,
1745 NLM_F_MULTI,
1746 OVS_VPORT_CMD_NEW) < 0)
1747 goto out;
1748
1749 j++;
1750 }
1751 skip = 0;
1752 }
1753 out:
1754 rcu_read_unlock();
1755
1756 cb->args[0] = i;
1757 cb->args[1] = j;
1758
1759 return skb->len;
1760 }
1761
1762 static const struct genl_ops dp_vport_genl_ops[] = {
1763 { .cmd = OVS_VPORT_CMD_NEW,
1764 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1765 .policy = vport_policy,
1766 .doit = ovs_vport_cmd_new
1767 },
1768 { .cmd = OVS_VPORT_CMD_DEL,
1769 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1770 .policy = vport_policy,
1771 .doit = ovs_vport_cmd_del
1772 },
1773 { .cmd = OVS_VPORT_CMD_GET,
1774 .flags = 0, /* OK for unprivileged users. */
1775 .policy = vport_policy,
1776 .doit = ovs_vport_cmd_get,
1777 .dumpit = ovs_vport_cmd_dump
1778 },
1779 { .cmd = OVS_VPORT_CMD_SET,
1780 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1781 .policy = vport_policy,
1782 .doit = ovs_vport_cmd_set,
1783 },
1784 };
1785
1786 struct genl_family_and_ops {
1787 struct genl_family *family;
1788 const struct genl_ops *ops;
1789 int n_ops;
1790 const struct genl_multicast_group *group;
1791 };
1792
1793 static const struct genl_family_and_ops dp_genl_families[] = {
1794 { &dp_datapath_genl_family,
1795 dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1796 &ovs_dp_datapath_multicast_group },
1797 { &dp_vport_genl_family,
1798 dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1799 &ovs_dp_vport_multicast_group },
1800 { &dp_flow_genl_family,
1801 dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1802 &ovs_dp_flow_multicast_group },
1803 { &dp_packet_genl_family,
1804 dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1805 NULL },
1806 };
1807
1808 static void dp_unregister_genl(int n_families)
1809 {
1810 int i;
1811
1812 for (i = 0; i < n_families; i++)
1813 genl_unregister_family(dp_genl_families[i].family);
1814 }
1815
1816 static int dp_register_genl(void)
1817 {
1818 int n_registered;
1819 int err;
1820 int i;
1821
1822 n_registered = 0;
1823 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
1824 const struct genl_family_and_ops *f = &dp_genl_families[i];
1825
1826 f->family->ops = f->ops;
1827 f->family->n_ops = f->n_ops;
1828 f->family->mcgrps = f->group;
1829 f->family->n_mcgrps = f->group ? 1 : 0;
1830 err = genl_register_family(f->family);
1831 if (err)
1832 goto error;
1833 n_registered++;
1834 }
1835
1836 return 0;
1837
1838 error:
1839 dp_unregister_genl(n_registered);
1840 return err;
1841 }
1842
1843 static int __net_init ovs_init_net(struct net *net)
1844 {
1845 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1846
1847 INIT_LIST_HEAD(&ovs_net->dps);
1848 INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
1849 return 0;
1850 }
1851
1852 static void __net_exit ovs_exit_net(struct net *net)
1853 {
1854 struct datapath *dp, *dp_next;
1855 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1856
1857 ovs_lock();
1858 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
1859 __dp_destroy(dp);
1860 ovs_unlock();
1861
1862 cancel_work_sync(&ovs_net->dp_notify_work);
1863 }
1864
1865 static struct pernet_operations ovs_net_ops = {
1866 .init = ovs_init_net,
1867 .exit = ovs_exit_net,
1868 .id = &ovs_net_id,
1869 .size = sizeof(struct ovs_net),
1870 };
1871
1872 static int __init dp_init(void)
1873 {
1874 int err;
1875
1876 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
1877
1878 pr_info("Open vSwitch switching datapath\n");
1879
1880 err = ovs_flow_init();
1881 if (err)
1882 goto error;
1883
1884 err = ovs_vport_init();
1885 if (err)
1886 goto error_flow_exit;
1887
1888 err = register_pernet_device(&ovs_net_ops);
1889 if (err)
1890 goto error_vport_exit;
1891
1892 err = register_netdevice_notifier(&ovs_dp_device_notifier);
1893 if (err)
1894 goto error_netns_exit;
1895
1896 err = dp_register_genl();
1897 if (err < 0)
1898 goto error_unreg_notifier;
1899
1900 return 0;
1901
1902 error_unreg_notifier:
1903 unregister_netdevice_notifier(&ovs_dp_device_notifier);
1904 error_netns_exit:
1905 unregister_pernet_device(&ovs_net_ops);
1906 error_vport_exit:
1907 ovs_vport_exit();
1908 error_flow_exit:
1909 ovs_flow_exit();
1910 error:
1911 return err;
1912 }
1913
1914 static void dp_cleanup(void)
1915 {
1916 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
1917 unregister_netdevice_notifier(&ovs_dp_device_notifier);
1918 unregister_pernet_device(&ovs_net_ops);
1919 rcu_barrier();
1920 ovs_vport_exit();
1921 ovs_flow_exit();
1922 }
1923
1924 module_init(dp_init);
1925 module_exit(dp_cleanup);
1926
1927 MODULE_DESCRIPTION("Open vSwitch switching datapath");
1928 MODULE_LICENSE("GPL");
(deprecated) Btrfs devel tree