]> git.proxmox.com Git - mirror_ovs.git/blob - datapath/datapath.c
projects / mirror_ovs.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
datapath: Use percpu allocator for flow-stats.
[mirror_ovs.git] / datapath / datapath.c
1 /*
2 * Copyright (c) 2007-2013 Nicira, Inc.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of version 2 of the GNU General Public
6 * License as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public License
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
16 * 02110-1301, USA
17 */
18
19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
20
21 #include <linux/init.h>
22 #include <linux/module.h>
23 #include <linux/if_arp.h>
24 #include <linux/if_vlan.h>
25 #include <linux/in.h>
26 #include <linux/ip.h>
27 #include <linux/jhash.h>
28 #include <linux/delay.h>
29 #include <linux/time.h>
30 #include <linux/etherdevice.h>
31 #include <linux/genetlink.h>
32 #include <linux/kernel.h>
33 #include <linux/kthread.h>
34 #include <linux/mutex.h>
35 #include <linux/percpu.h>
36 #include <linux/rcupdate.h>
37 #include <linux/tcp.h>
38 #include <linux/udp.h>
39 #include <linux/version.h>
40 #include <linux/ethtool.h>
41 #include <linux/wait.h>
42 #include <asm/div64.h>
43 #include <linux/highmem.h>
44 #include <linux/netfilter_bridge.h>
45 #include <linux/netfilter_ipv4.h>
46 #include <linux/inetdevice.h>
47 #include <linux/list.h>
48 #include <linux/openvswitch.h>
49 #include <linux/rculist.h>
50 #include <linux/dmi.h>
51 #include <linux/genetlink.h>
52 #include <net/genetlink.h>
53 #include <net/genetlink.h>
54 #include <net/net_namespace.h>
55 #include <net/netns/generic.h>
56
57 #include "datapath.h"
58 #include "flow.h"
59 #include "flow_netlink.h"
60 #include "vlan.h"
61 #include "vport-internal_dev.h"
62 #include "vport-netdev.h"
63
64 int ovs_net_id __read_mostly;
65
66 static void ovs_notify(struct sk_buff *skb, struct genl_info *info,
67 struct genl_multicast_group *grp)
68 {
69 genl_notify(skb, genl_info_net(info), info->snd_portid,
70 grp->id, info->nlhdr, GFP_KERNEL);
71 }
72
73 /**
74 * DOC: Locking:
75 *
76 * All writes e.g. Writes to device state (add/remove datapath, port, set
77 * operations on vports, etc.), Writes to other state (flow table
78 * modifications, set miscellaneous datapath parameters, etc.) are protected
79 * by ovs_lock.
80 *
81 * Reads are protected by RCU.
82 *
83 * There are a few special cases (mostly stats) that have their own
84 * synchronization but they nest under all of above and don't interact with
85 * each other.
86 *
87 * The RTNL lock nests inside ovs_mutex.
88 */
89
90 static DEFINE_MUTEX(ovs_mutex);
91
92 void ovs_lock(void)
93 {
94 mutex_lock(&ovs_mutex);
95 }
96
97 void ovs_unlock(void)
98 {
99 mutex_unlock(&ovs_mutex);
100 }
101
102 #ifdef CONFIG_LOCKDEP
103 int lockdep_ovsl_is_held(void)
104 {
105 if (debug_locks)
106 return lockdep_is_held(&ovs_mutex);
107 else
108 return 1;
109 }
110 #endif
111
112 static struct vport *new_vport(const struct vport_parms *);
113 static int queue_gso_packets(struct net *, int dp_ifindex, struct sk_buff *,
114 const struct dp_upcall_info *);
115 static int queue_userspace_packet(struct net *, int dp_ifindex,
116 struct sk_buff *,
117 const struct dp_upcall_info *);
118
119 /* Must be called with rcu_read_lock or ovs_mutex. */
120 static struct datapath *get_dp(struct net *net, int dp_ifindex)
121 {
122 struct datapath *dp = NULL;
123 struct net_device *dev;
124
125 rcu_read_lock();
126 dev = dev_get_by_index_rcu(net, dp_ifindex);
127 if (dev) {
128 struct vport *vport = ovs_internal_dev_get_vport(dev);
129 if (vport)
130 dp = vport->dp;
131 }
132 rcu_read_unlock();
133
134 return dp;
135 }
136
137 /* Must be called with rcu_read_lock or ovs_mutex. */
138 const char *ovs_dp_name(const struct datapath *dp)
139 {
140 struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
141 return vport->ops->get_name(vport);
142 }
143
144 static int get_dpifindex(struct datapath *dp)
145 {
146 struct vport *local;
147 int ifindex;
148
149 rcu_read_lock();
150
151 local = ovs_vport_rcu(dp, OVSP_LOCAL);
152 if (local)
153 ifindex = netdev_vport_priv(local)->dev->ifindex;
154 else
155 ifindex = 0;
156
157 rcu_read_unlock();
158
159 return ifindex;
160 }
161
162 static void destroy_dp_rcu(struct rcu_head *rcu)
163 {
164 struct datapath *dp = container_of(rcu, struct datapath, rcu);
165
166 ovs_flow_tbl_destroy(&dp->table);
167 free_percpu(dp->stats_percpu);
168 release_net(ovs_dp_get_net(dp));
169 kfree(dp->ports);
170 kfree(dp);
171 }
172
173 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
174 u16 port_no)
175 {
176 return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
177 }
178
179 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
180 {
181 struct vport *vport;
182 struct hlist_head *head;
183
184 head = vport_hash_bucket(dp, port_no);
185 hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
186 if (vport->port_no == port_no)
187 return vport;
188 }
189 return NULL;
190 }
191
192 /* Called with ovs_mutex. */
193 static struct vport *new_vport(const struct vport_parms *parms)
194 {
195 struct vport *vport;
196
197 vport = ovs_vport_add(parms);
198 if (!IS_ERR(vport)) {
199 struct datapath *dp = parms->dp;
200 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
201
202 hlist_add_head_rcu(&vport->dp_hash_node, head);
203 }
204 return vport;
205 }
206
207 void ovs_dp_detach_port(struct vport *p)
208 {
209 ASSERT_OVSL();
210
211 /* First drop references to device. */
212 hlist_del_rcu(&p->dp_hash_node);
213
214 /* Then destroy it. */
215 ovs_vport_del(p);
216 }
217
218 /* Must be called with rcu_read_lock. */
219 void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
220 {
221 struct datapath *dp = p->dp;
222 struct sw_flow *flow;
223 struct dp_stats_percpu *stats;
224 struct sw_flow_key key;
225 u64 *stats_counter;
226 u32 n_mask_hit;
227 int error;
228
229 stats = this_cpu_ptr(dp->stats_percpu);
230
231 /* Extract flow from 'skb' into 'key'. */
232 error = ovs_flow_extract(skb, p->port_no, &key);
233 if (unlikely(error)) {
234 kfree_skb(skb);
235 return;
236 }
237
238 /* Look up flow. */
239 flow = ovs_flow_tbl_lookup_stats(&dp->table, &key, &n_mask_hit);
240 if (unlikely(!flow)) {
241 struct dp_upcall_info upcall;
242
243 upcall.cmd = OVS_PACKET_CMD_MISS;
244 upcall.key = &key;
245 upcall.userdata = NULL;
246 upcall.portid = p->upcall_portid;
247 ovs_dp_upcall(dp, skb, &upcall);
248 consume_skb(skb);
249 stats_counter = &stats->n_missed;
250 goto out;
251 }
252
253 OVS_CB(skb)->flow = flow;
254 OVS_CB(skb)->pkt_key = &key;
255
256 ovs_flow_stats_update(OVS_CB(skb)->flow, skb);
257 ovs_execute_actions(dp, skb);
258 stats_counter = &stats->n_hit;
259
260 out:
261 /* Update datapath statistics. */
262 u64_stats_update_begin(&stats->sync);
263 (*stats_counter)++;
264 stats->n_mask_hit += n_mask_hit;
265 u64_stats_update_end(&stats->sync);
266 }
267
268 static struct genl_family dp_packet_genl_family = {
269 .id = GENL_ID_GENERATE,
270 .hdrsize = sizeof(struct ovs_header),
271 .name = OVS_PACKET_FAMILY,
272 .version = OVS_PACKET_VERSION,
273 .maxattr = OVS_PACKET_ATTR_MAX,
274 .netnsok = true,
275 SET_PARALLEL_OPS
276 };
277
278 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
279 const struct dp_upcall_info *upcall_info)
280 {
281 struct dp_stats_percpu *stats;
282 int dp_ifindex;
283 int err;
284
285 if (upcall_info->portid == 0) {
286 err = -ENOTCONN;
287 goto err;
288 }
289
290 dp_ifindex = get_dpifindex(dp);
291 if (!dp_ifindex) {
292 err = -ENODEV;
293 goto err;
294 }
295
296 if (!skb_is_gso(skb))
297 err = queue_userspace_packet(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
298 else
299 err = queue_gso_packets(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
300 if (err)
301 goto err;
302
303 return 0;
304
305 err:
306 stats = this_cpu_ptr(dp->stats_percpu);
307
308 u64_stats_update_begin(&stats->sync);
309 stats->n_lost++;
310 u64_stats_update_end(&stats->sync);
311
312 return err;
313 }
314
315 static int queue_gso_packets(struct net *net, int dp_ifindex,
316 struct sk_buff *skb,
317 const struct dp_upcall_info *upcall_info)
318 {
319 unsigned short gso_type = skb_shinfo(skb)->gso_type;
320 struct dp_upcall_info later_info;
321 struct sw_flow_key later_key;
322 struct sk_buff *segs, *nskb;
323 int err;
324
325 segs = __skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM, false);
326 if (IS_ERR(segs))
327 return PTR_ERR(segs);
328
329 /* Queue all of the segments. */
330 skb = segs;
331 do {
332 err = queue_userspace_packet(net, dp_ifindex, skb, upcall_info);
333 if (err)
334 break;
335
336 if (skb == segs && gso_type & SKB_GSO_UDP) {
337 /* The initial flow key extracted by ovs_flow_extract()
338 * in this case is for a first fragment, so we need to
339 * properly mark later fragments.
340 */
341 later_key = *upcall_info->key;
342 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
343
344 later_info = *upcall_info;
345 later_info.key = &later_key;
346 upcall_info = &later_info;
347 }
348 } while ((skb = skb->next));
349
350 /* Free all of the segments. */
351 skb = segs;
352 do {
353 nskb = skb->next;
354 if (err)
355 kfree_skb(skb);
356 else
357 consume_skb(skb);
358 } while ((skb = nskb));
359 return err;
360 }
361
362 static size_t key_attr_size(void)
363 {
364 return nla_total_size(4) /* OVS_KEY_ATTR_PRIORITY */
365 + nla_total_size(0) /* OVS_KEY_ATTR_TUNNEL */
366 + nla_total_size(8) /* OVS_TUNNEL_KEY_ATTR_ID */
367 + nla_total_size(4) /* OVS_TUNNEL_KEY_ATTR_IPV4_SRC */
368 + nla_total_size(4) /* OVS_TUNNEL_KEY_ATTR_IPV4_DST */
369 + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TOS */
370 + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TTL */
371 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
372 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_CSUM */
373 + nla_total_size(4) /* OVS_KEY_ATTR_IN_PORT */
374 + nla_total_size(4) /* OVS_KEY_ATTR_SKB_MARK */
375 + nla_total_size(12) /* OVS_KEY_ATTR_ETHERNET */
376 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
377 + nla_total_size(4) /* OVS_KEY_ATTR_8021Q */
378 + nla_total_size(0) /* OVS_KEY_ATTR_ENCAP */
379 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
380 + nla_total_size(40) /* OVS_KEY_ATTR_IPV6 */
381 + nla_total_size(2) /* OVS_KEY_ATTR_ICMPV6 */
382 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
383 }
384
385 static size_t upcall_msg_size(const struct sk_buff *skb,
386 const struct nlattr *userdata)
387 {
388 size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
389 + nla_total_size(skb->len) /* OVS_PACKET_ATTR_PACKET */
390 + nla_total_size(key_attr_size()); /* OVS_PACKET_ATTR_KEY */
391
392 /* OVS_PACKET_ATTR_USERDATA */
393 if (userdata)
394 size += NLA_ALIGN(userdata->nla_len);
395
396 return size;
397 }
398
399 static int queue_userspace_packet(struct net *net, int dp_ifindex,
400 struct sk_buff *skb,
401 const struct dp_upcall_info *upcall_info)
402 {
403 struct ovs_header *upcall;
404 struct sk_buff *nskb = NULL;
405 struct sk_buff *user_skb; /* to be queued to userspace */
406 struct nlattr *nla;
407 struct genl_info info = {
408 #if LINUX_VERSION_CODE >= KERNEL_VERSION(3,14,0)
409 .dst_sk = net->genl_sock,
410 #endif
411 .snd_portid = upcall_info->portid,
412 };
413 size_t len;
414 int err;
415
416 if (vlan_tx_tag_present(skb)) {
417 nskb = skb_clone(skb, GFP_ATOMIC);
418 if (!nskb)
419 return -ENOMEM;
420
421 nskb = __vlan_put_tag(nskb, nskb->vlan_proto, vlan_tx_tag_get(nskb));
422 if (!nskb)
423 return -ENOMEM;
424
425 vlan_set_tci(nskb, 0);
426
427 skb = nskb;
428 }
429
430 if (nla_attr_size(skb->len) > USHRT_MAX) {
431 err = -EFBIG;
432 goto out;
433 }
434
435 len = upcall_msg_size(skb, upcall_info->userdata);
436 user_skb = genlmsg_new_unicast(len, &info, GFP_ATOMIC);
437 if (!user_skb) {
438 err = -ENOMEM;
439 goto out;
440 }
441
442 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
443 0, upcall_info->cmd);
444 upcall->dp_ifindex = dp_ifindex;
445
446 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
447 ovs_nla_put_flow(upcall_info->key, upcall_info->key, user_skb);
448 nla_nest_end(user_skb, nla);
449
450 if (upcall_info->userdata)
451 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
452 nla_len(upcall_info->userdata),
453 nla_data(upcall_info->userdata));
454
455 nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
456
457 skb_copy_and_csum_dev(skb, nla_data(nla));
458
459 genlmsg_end(user_skb, upcall);
460 err = genlmsg_unicast(net, user_skb, upcall_info->portid);
461
462 out:
463 kfree_skb(nskb);
464 return err;
465 }
466
467 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
468 {
469 struct ovs_header *ovs_header = info->userhdr;
470 struct nlattr **a = info->attrs;
471 struct sw_flow_actions *acts;
472 struct sk_buff *packet;
473 struct sw_flow *flow;
474 struct datapath *dp;
475 struct ethhdr *eth;
476 int len;
477 int err;
478
479 err = -EINVAL;
480 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
481 !a[OVS_PACKET_ATTR_ACTIONS])
482 goto err;
483
484 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
485 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
486 err = -ENOMEM;
487 if (!packet)
488 goto err;
489 skb_reserve(packet, NET_IP_ALIGN);
490
491 nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
492
493 skb_reset_mac_header(packet);
494 eth = eth_hdr(packet);
495
496 /* Normally, setting the skb 'protocol' field would be handled by a
497 * call to eth_type_trans(), but it assumes there's a sending
498 * device, which we may not have. */
499 if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
500 packet->protocol = eth->h_proto;
501 else
502 packet->protocol = htons(ETH_P_802_2);
503
504 /* Build an sw_flow for sending this packet. */
505 flow = ovs_flow_alloc(false);
506 err = PTR_ERR(flow);
507 if (IS_ERR(flow))
508 goto err_kfree_skb;
509
510 err = ovs_flow_extract(packet, -1, &flow->key);
511 if (err)
512 goto err_flow_free;
513
514 err = ovs_nla_get_flow_metadata(flow, a[OVS_PACKET_ATTR_KEY]);
515 if (err)
516 goto err_flow_free;
517 acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_PACKET_ATTR_ACTIONS]));
518 err = PTR_ERR(acts);
519 if (IS_ERR(acts))
520 goto err_flow_free;
521
522 err = ovs_nla_copy_actions(a[OVS_PACKET_ATTR_ACTIONS],
523 &flow->key, 0, &acts);
524 rcu_assign_pointer(flow->sf_acts, acts);
525 if (err)
526 goto err_flow_free;
527
528 OVS_CB(packet)->flow = flow;
529 OVS_CB(packet)->pkt_key = &flow->key;
530 packet->priority = flow->key.phy.priority;
531 packet->mark = flow->key.phy.skb_mark;
532
533 rcu_read_lock();
534 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
535 err = -ENODEV;
536 if (!dp)
537 goto err_unlock;
538
539 local_bh_disable();
540 err = ovs_execute_actions(dp, packet);
541 local_bh_enable();
542 rcu_read_unlock();
543
544 ovs_flow_free(flow, false);
545 return err;
546
547 err_unlock:
548 rcu_read_unlock();
549 err_flow_free:
550 ovs_flow_free(flow, false);
551 err_kfree_skb:
552 kfree_skb(packet);
553 err:
554 return err;
555 }
556
557 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
558 [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
559 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
560 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
561 };
562
563 static struct genl_ops dp_packet_genl_ops[] = {
564 { .cmd = OVS_PACKET_CMD_EXECUTE,
565 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
566 .policy = packet_policy,
567 .doit = ovs_packet_cmd_execute
568 }
569 };
570
571 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats,
572 struct ovs_dp_megaflow_stats *mega_stats)
573 {
574 int i;
575
576 memset(mega_stats, 0, sizeof(*mega_stats));
577
578 stats->n_flows = ovs_flow_tbl_count(&dp->table);
579 mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
580
581 stats->n_hit = stats->n_missed = stats->n_lost = 0;
582
583 for_each_possible_cpu(i) {
584 const struct dp_stats_percpu *percpu_stats;
585 struct dp_stats_percpu local_stats;
586 unsigned int start;
587
588 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
589
590 do {
591 start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
592 local_stats = *percpu_stats;
593 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
594
595 stats->n_hit += local_stats.n_hit;
596 stats->n_missed += local_stats.n_missed;
597 stats->n_lost += local_stats.n_lost;
598 mega_stats->n_mask_hit += local_stats.n_mask_hit;
599 }
600 }
601
602 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
603 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
604 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
605 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
606 };
607
608 static struct genl_family dp_flow_genl_family = {
609 .id = GENL_ID_GENERATE,
610 .hdrsize = sizeof(struct ovs_header),
611 .name = OVS_FLOW_FAMILY,
612 .version = OVS_FLOW_VERSION,
613 .maxattr = OVS_FLOW_ATTR_MAX,
614 .netnsok = true,
615 SET_PARALLEL_OPS
616 };
617
618 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
619 .name = OVS_FLOW_MCGROUP
620 };
621
622 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
623 {
624 return NLMSG_ALIGN(sizeof(struct ovs_header))
625 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
626 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_MASK */
627 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
628 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
629 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
630 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
631 }
632
633 /* Called with ovs_mutex. */
634 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
635 struct sk_buff *skb, u32 portid,
636 u32 seq, u32 flags, u8 cmd)
637 {
638 const int skb_orig_len = skb->len;
639 struct nlattr *start;
640 struct ovs_flow_stats stats;
641 __be16 tcp_flags;
642 unsigned long used;
643 struct ovs_header *ovs_header;
644 struct nlattr *nla;
645 int err;
646
647 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
648 if (!ovs_header)
649 return -EMSGSIZE;
650
651 ovs_header->dp_ifindex = get_dpifindex(dp);
652
653 /* Fill flow key. */
654 nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
655 if (!nla)
656 goto nla_put_failure;
657
658 err = ovs_nla_put_flow(&flow->unmasked_key, &flow->unmasked_key, skb);
659 if (err)
660 goto error;
661 nla_nest_end(skb, nla);
662
663 nla = nla_nest_start(skb, OVS_FLOW_ATTR_MASK);
664 if (!nla)
665 goto nla_put_failure;
666
667 err = ovs_nla_put_flow(&flow->key, &flow->mask->key, skb);
668 if (err)
669 goto error;
670
671 nla_nest_end(skb, nla);
672
673 ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
674 if (used &&
675 nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
676 goto nla_put_failure;
677
678 if (stats.n_packets &&
679 nla_put(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats))
680 goto nla_put_failure;
681
682 if ((u8)ntohs(tcp_flags) &&
683 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
684 goto nla_put_failure;
685
686 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
687 * this is the first flow to be dumped into 'skb'. This is unusual for
688 * Netlink but individual action lists can be longer than
689 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
690 * The userspace caller can always fetch the actions separately if it
691 * really wants them. (Most userspace callers in fact don't care.)
692 *
693 * This can only fail for dump operations because the skb is always
694 * properly sized for single flows.
695 */
696 start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
697 if (start) {
698 const struct sw_flow_actions *sf_acts;
699
700 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
701
702 err = ovs_nla_put_actions(sf_acts->actions,
703 sf_acts->actions_len, skb);
704 if (!err)
705 nla_nest_end(skb, start);
706 else {
707 if (skb_orig_len)
708 goto error;
709
710 nla_nest_cancel(skb, start);
711 }
712 } else if (skb_orig_len)
713 goto nla_put_failure;
714
715 return genlmsg_end(skb, ovs_header);
716
717 nla_put_failure:
718 err = -EMSGSIZE;
719 error:
720 genlmsg_cancel(skb, ovs_header);
721 return err;
722 }
723
724 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow,
725 struct genl_info *info)
726 {
727 size_t len;
728
729 len = ovs_flow_cmd_msg_size(ovsl_dereference(flow->sf_acts));
730
731 return genlmsg_new_unicast(len, info, GFP_KERNEL);
732 }
733
734 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
735 struct datapath *dp,
736 struct genl_info *info,
737 u8 cmd)
738 {
739 struct sk_buff *skb;
740 int retval;
741
742 skb = ovs_flow_cmd_alloc_info(flow, info);
743 if (!skb)
744 return ERR_PTR(-ENOMEM);
745
746 retval = ovs_flow_cmd_fill_info(flow, dp, skb, info->snd_portid,
747 info->snd_seq, 0, cmd);
748 BUG_ON(retval < 0);
749 return skb;
750 }
751
752 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
753 {
754 struct nlattr **a = info->attrs;
755 struct ovs_header *ovs_header = info->userhdr;
756 struct sw_flow_key key, masked_key;
757 struct sw_flow *flow = NULL;
758 struct sw_flow_mask mask;
759 struct sk_buff *reply;
760 struct datapath *dp;
761 struct sw_flow_actions *acts = NULL;
762 struct sw_flow_match match;
763 bool exact_5tuple;
764 int error;
765
766 /* Extract key. */
767 error = -EINVAL;
768 if (!a[OVS_FLOW_ATTR_KEY])
769 goto error;
770
771 ovs_match_init(&match, &key, &mask);
772 error = ovs_nla_get_match(&match, &exact_5tuple,
773 a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
774 if (error)
775 goto error;
776
777 /* Validate actions. */
778 if (a[OVS_FLOW_ATTR_ACTIONS]) {
779 acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
780 error = PTR_ERR(acts);
781 if (IS_ERR(acts))
782 goto error;
783
784 ovs_flow_mask_key(&masked_key, &key, &mask);
785 error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS],
786 &masked_key, 0, &acts);
787 if (error) {
788 OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
789 goto err_kfree;
790 }
791 } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
792 error = -EINVAL;
793 goto error;
794 }
795
796 ovs_lock();
797 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
798 error = -ENODEV;
799 if (!dp)
800 goto err_unlock_ovs;
801
802 /* Check if this is a duplicate flow */
803 flow = ovs_flow_tbl_lookup(&dp->table, &key);
804 if (!flow) {
805 /* Bail out if we're not allowed to create a new flow. */
806 error = -ENOENT;
807 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
808 goto err_unlock_ovs;
809
810 /* Allocate flow. */
811 flow = ovs_flow_alloc(!exact_5tuple);
812 if (IS_ERR(flow)) {
813 error = PTR_ERR(flow);
814 goto err_unlock_ovs;
815 }
816
817 flow->key = masked_key;
818 flow->unmasked_key = key;
819 rcu_assign_pointer(flow->sf_acts, acts);
820
821 /* Put flow in bucket. */
822 error = ovs_flow_tbl_insert(&dp->table, flow, &mask);
823 if (error) {
824 acts = NULL;
825 goto err_flow_free;
826 }
827
828 reply = ovs_flow_cmd_build_info(flow, dp, info, OVS_FLOW_CMD_NEW);
829 } else {
830 /* We found a matching flow. */
831 struct sw_flow_actions *old_acts;
832
833 /* Bail out if we're not allowed to modify an existing flow.
834 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
835 * because Generic Netlink treats the latter as a dump
836 * request. We also accept NLM_F_EXCL in case that bug ever
837 * gets fixed.
838 */
839 error = -EEXIST;
840 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
841 info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
842 goto err_unlock_ovs;
843
844 /* The unmasked key has to be the same for flow updates. */
845 error = -EINVAL;
846 if (!ovs_flow_cmp_unmasked_key(flow, &match)) {
847 OVS_NLERR("Flow modification message rejected, unmasked key does not match.\n");
848 goto err_unlock_ovs;
849 }
850
851 /* Update actions. */
852 old_acts = ovsl_dereference(flow->sf_acts);
853 rcu_assign_pointer(flow->sf_acts, acts);
854 ovs_nla_free_flow_actions(old_acts);
855
856 reply = ovs_flow_cmd_build_info(flow, dp, info, OVS_FLOW_CMD_NEW);
857
858 /* Clear stats. */
859 if (a[OVS_FLOW_ATTR_CLEAR])
860 ovs_flow_stats_clear(flow);
861 }
862 ovs_unlock();
863
864 if (!IS_ERR(reply))
865 ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
866 else
867 netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
868 ovs_dp_flow_multicast_group.id, PTR_ERR(reply));
869 return 0;
870
871 err_flow_free:
872 ovs_flow_free(flow, false);
873 err_unlock_ovs:
874 ovs_unlock();
875 err_kfree:
876 kfree(acts);
877 error:
878 return error;
879 }
880
881 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
882 {
883 struct nlattr **a = info->attrs;
884 struct ovs_header *ovs_header = info->userhdr;
885 struct sw_flow_key key;
886 struct sk_buff *reply;
887 struct sw_flow *flow;
888 struct datapath *dp;
889 struct sw_flow_match match;
890 int err;
891
892 if (!a[OVS_FLOW_ATTR_KEY]) {
893 OVS_NLERR("Flow get message rejected, Key attribute missing.\n");
894 return -EINVAL;
895 }
896
897 ovs_match_init(&match, &key, NULL);
898 err = ovs_nla_get_match(&match, NULL, a[OVS_FLOW_ATTR_KEY], NULL);
899 if (err)
900 return err;
901
902 ovs_lock();
903 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
904 if (!dp) {
905 err = -ENODEV;
906 goto unlock;
907 }
908
909 flow = ovs_flow_tbl_lookup(&dp->table, &key);
910 if (!flow || !ovs_flow_cmp_unmasked_key(flow, &match)) {
911 err = -ENOENT;
912 goto unlock;
913 }
914
915 reply = ovs_flow_cmd_build_info(flow, dp, info, OVS_FLOW_CMD_NEW);
916 if (IS_ERR(reply)) {
917 err = PTR_ERR(reply);
918 goto unlock;
919 }
920
921 ovs_unlock();
922 return genlmsg_reply(reply, info);
923 unlock:
924 ovs_unlock();
925 return err;
926 }
927
928 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
929 {
930 struct nlattr **a = info->attrs;
931 struct ovs_header *ovs_header = info->userhdr;
932 struct sw_flow_key key;
933 struct sk_buff *reply;
934 struct sw_flow *flow;
935 struct datapath *dp;
936 struct sw_flow_match match;
937 int err;
938
939 ovs_lock();
940 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
941 if (!dp) {
942 err = -ENODEV;
943 goto unlock;
944 }
945
946 if (!a[OVS_FLOW_ATTR_KEY]) {
947 err = ovs_flow_tbl_flush(&dp->table);
948 goto unlock;
949 }
950
951 ovs_match_init(&match, &key, NULL);
952 err = ovs_nla_get_match(&match, NULL, a[OVS_FLOW_ATTR_KEY], NULL);
953 if (err)
954 goto unlock;
955
956 flow = ovs_flow_tbl_lookup(&dp->table, &key);
957 if (!flow || !ovs_flow_cmp_unmasked_key(flow, &match)) {
958 err = -ENOENT;
959 goto unlock;
960 }
961
962 reply = ovs_flow_cmd_alloc_info(flow, info);
963 if (!reply) {
964 err = -ENOMEM;
965 goto unlock;
966 }
967
968 ovs_flow_tbl_remove(&dp->table, flow);
969
970 err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_portid,
971 info->snd_seq, 0, OVS_FLOW_CMD_DEL);
972 BUG_ON(err < 0);
973
974 ovs_flow_free(flow, true);
975 ovs_unlock();
976
977 ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
978 return 0;
979 unlock:
980 ovs_unlock();
981 return err;
982 }
983
984 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
985 {
986 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
987 struct table_instance *ti;
988 struct datapath *dp;
989
990 rcu_read_lock();
991 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
992 if (!dp) {
993 rcu_read_unlock();
994 return -ENODEV;
995 }
996
997 ti = rcu_dereference(dp->table.ti);
998 for (;;) {
999 struct sw_flow *flow;
1000 u32 bucket, obj;
1001
1002 bucket = cb->args[0];
1003 obj = cb->args[1];
1004 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1005 if (!flow)
1006 break;
1007
1008 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1009 NETLINK_CB(cb->skb).portid,
1010 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1011 OVS_FLOW_CMD_NEW) < 0)
1012 break;
1013
1014 cb->args[0] = bucket;
1015 cb->args[1] = obj;
1016 }
1017 rcu_read_unlock();
1018 return skb->len;
1019 }
1020
1021 static struct genl_ops dp_flow_genl_ops[] = {
1022 { .cmd = OVS_FLOW_CMD_NEW,
1023 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1024 .policy = flow_policy,
1025 .doit = ovs_flow_cmd_new_or_set
1026 },
1027 { .cmd = OVS_FLOW_CMD_DEL,
1028 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1029 .policy = flow_policy,
1030 .doit = ovs_flow_cmd_del
1031 },
1032 { .cmd = OVS_FLOW_CMD_GET,
1033 .flags = 0, /* OK for unprivileged users. */
1034 .policy = flow_policy,
1035 .doit = ovs_flow_cmd_get,
1036 .dumpit = ovs_flow_cmd_dump
1037 },
1038 { .cmd = OVS_FLOW_CMD_SET,
1039 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1040 .policy = flow_policy,
1041 .doit = ovs_flow_cmd_new_or_set,
1042 },
1043 };
1044
1045 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1046 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1047 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1048 };
1049
1050 static struct genl_family dp_datapath_genl_family = {
1051 .id = GENL_ID_GENERATE,
1052 .hdrsize = sizeof(struct ovs_header),
1053 .name = OVS_DATAPATH_FAMILY,
1054 .version = OVS_DATAPATH_VERSION,
1055 .maxattr = OVS_DP_ATTR_MAX,
1056 .netnsok = true,
1057 SET_PARALLEL_OPS
1058 };
1059
1060 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1061 .name = OVS_DATAPATH_MCGROUP
1062 };
1063
1064 static size_t ovs_dp_cmd_msg_size(void)
1065 {
1066 size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1067
1068 msgsize += nla_total_size(IFNAMSIZ);
1069 msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1070 msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1071
1072 return msgsize;
1073 }
1074
1075 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1076 u32 portid, u32 seq, u32 flags, u8 cmd)
1077 {
1078 struct ovs_header *ovs_header;
1079 struct ovs_dp_stats dp_stats;
1080 struct ovs_dp_megaflow_stats dp_megaflow_stats;
1081 int err;
1082
1083 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1084 flags, cmd);
1085 if (!ovs_header)
1086 goto error;
1087
1088 ovs_header->dp_ifindex = get_dpifindex(dp);
1089
1090 rcu_read_lock();
1091 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1092 rcu_read_unlock();
1093 if (err)
1094 goto nla_put_failure;
1095
1096 get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1097 if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1098 &dp_stats))
1099 goto nla_put_failure;
1100
1101 if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1102 sizeof(struct ovs_dp_megaflow_stats),
1103 &dp_megaflow_stats))
1104 goto nla_put_failure;
1105
1106 return genlmsg_end(skb, ovs_header);
1107
1108 nla_put_failure:
1109 genlmsg_cancel(skb, ovs_header);
1110 error:
1111 return -EMSGSIZE;
1112 }
1113
1114 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp,
1115 struct genl_info *info, u8 cmd)
1116 {
1117 struct sk_buff *skb;
1118 int retval;
1119
1120 skb = genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info, GFP_KERNEL);
1121 if (!skb)
1122 return ERR_PTR(-ENOMEM);
1123
1124 retval = ovs_dp_cmd_fill_info(dp, skb, info->snd_portid, info->snd_seq, 0, cmd);
1125 if (retval < 0) {
1126 kfree_skb(skb);
1127 return ERR_PTR(retval);
1128 }
1129 return skb;
1130 }
1131
1132 /* Called with ovs_mutex. */
1133 static struct datapath *lookup_datapath(struct net *net,
1134 struct ovs_header *ovs_header,
1135 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1136 {
1137 struct datapath *dp;
1138
1139 if (!a[OVS_DP_ATTR_NAME])
1140 dp = get_dp(net, ovs_header->dp_ifindex);
1141 else {
1142 struct vport *vport;
1143
1144 rcu_read_lock();
1145 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1146 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1147 rcu_read_unlock();
1148 }
1149 return dp ? dp : ERR_PTR(-ENODEV);
1150 }
1151
1152 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1153 {
1154 struct nlattr **a = info->attrs;
1155 struct vport_parms parms;
1156 struct sk_buff *reply;
1157 struct datapath *dp;
1158 struct vport *vport;
1159 struct ovs_net *ovs_net;
1160 int err, i;
1161
1162 err = -EINVAL;
1163 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1164 goto err;
1165
1166 ovs_lock();
1167
1168 err = -ENOMEM;
1169 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1170 if (dp == NULL)
1171 goto err_unlock_ovs;
1172
1173 ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1174
1175 /* Allocate table. */
1176 err = ovs_flow_tbl_init(&dp->table);
1177 if (err)
1178 goto err_free_dp;
1179
1180 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1181 if (!dp->stats_percpu) {
1182 err = -ENOMEM;
1183 goto err_destroy_table;
1184 }
1185
1186 dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1187 GFP_KERNEL);
1188 if (!dp->ports) {
1189 err = -ENOMEM;
1190 goto err_destroy_percpu;
1191 }
1192
1193 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1194 INIT_HLIST_HEAD(&dp->ports[i]);
1195
1196 /* Set up our datapath device. */
1197 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1198 parms.type = OVS_VPORT_TYPE_INTERNAL;
1199 parms.options = NULL;
1200 parms.dp = dp;
1201 parms.port_no = OVSP_LOCAL;
1202 parms.upcall_portid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1203
1204 vport = new_vport(&parms);
1205 if (IS_ERR(vport)) {
1206 err = PTR_ERR(vport);
1207 if (err == -EBUSY)
1208 err = -EEXIST;
1209
1210 goto err_destroy_ports_array;
1211 }
1212
1213 reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_NEW);
1214 err = PTR_ERR(reply);
1215 if (IS_ERR(reply))
1216 goto err_destroy_local_port;
1217
1218 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1219 list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1220
1221 ovs_unlock();
1222
1223 ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1224 return 0;
1225
1226 err_destroy_local_port:
1227 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1228 err_destroy_ports_array:
1229 kfree(dp->ports);
1230 err_destroy_percpu:
1231 free_percpu(dp->stats_percpu);
1232 err_destroy_table:
1233 ovs_flow_tbl_destroy(&dp->table);
1234 err_free_dp:
1235 release_net(ovs_dp_get_net(dp));
1236 kfree(dp);
1237 err_unlock_ovs:
1238 ovs_unlock();
1239 err:
1240 return err;
1241 }
1242
1243 /* Called with ovs_mutex. */
1244 static void __dp_destroy(struct datapath *dp)
1245 {
1246 int i;
1247
1248 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1249 struct vport *vport;
1250 struct hlist_node *n;
1251
1252 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1253 if (vport->port_no != OVSP_LOCAL)
1254 ovs_dp_detach_port(vport);
1255 }
1256
1257 list_del_rcu(&dp->list_node);
1258
1259 /* OVSP_LOCAL is datapath internal port. We need to make sure that
1260 * all port in datapath are destroyed first before freeing datapath.
1261 */
1262 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1263
1264 call_rcu(&dp->rcu, destroy_dp_rcu);
1265 }
1266
1267 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1268 {
1269 struct sk_buff *reply;
1270 struct datapath *dp;
1271 int err;
1272
1273 ovs_lock();
1274 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1275 err = PTR_ERR(dp);
1276 if (IS_ERR(dp))
1277 goto unlock;
1278
1279 reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_DEL);
1280 err = PTR_ERR(reply);
1281 if (IS_ERR(reply))
1282 goto unlock;
1283
1284 __dp_destroy(dp);
1285 ovs_unlock();
1286
1287 ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1288
1289 return 0;
1290 unlock:
1291 ovs_unlock();
1292 return err;
1293 }
1294
1295 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1296 {
1297 struct sk_buff *reply;
1298 struct datapath *dp;
1299 int err;
1300
1301 ovs_lock();
1302 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1303 err = PTR_ERR(dp);
1304 if (IS_ERR(dp))
1305 goto unlock;
1306
1307 reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_NEW);
1308 if (IS_ERR(reply)) {
1309 err = PTR_ERR(reply);
1310 netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
1311 ovs_dp_datapath_multicast_group.id, err);
1312 err = 0;
1313 goto unlock;
1314 }
1315
1316 ovs_unlock();
1317 ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1318
1319 return 0;
1320 unlock:
1321 ovs_unlock();
1322 return err;
1323 }
1324
1325 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1326 {
1327 struct sk_buff *reply;
1328 struct datapath *dp;
1329 int err;
1330
1331 ovs_lock();
1332 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1333 if (IS_ERR(dp)) {
1334 err = PTR_ERR(dp);
1335 goto unlock;
1336 }
1337
1338 reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_NEW);
1339 if (IS_ERR(reply)) {
1340 err = PTR_ERR(reply);
1341 goto unlock;
1342 }
1343
1344 ovs_unlock();
1345 return genlmsg_reply(reply, info);
1346
1347 unlock:
1348 ovs_unlock();
1349 return err;
1350 }
1351
1352 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1353 {
1354 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1355 struct datapath *dp;
1356 int skip = cb->args[0];
1357 int i = 0;
1358
1359 rcu_read_lock();
1360 list_for_each_entry_rcu(dp, &ovs_net->dps, list_node) {
1361 if (i >= skip &&
1362 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1363 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1364 OVS_DP_CMD_NEW) < 0)
1365 break;
1366 i++;
1367 }
1368 rcu_read_unlock();
1369
1370 cb->args[0] = i;
1371
1372 return skb->len;
1373 }
1374
1375 static struct genl_ops dp_datapath_genl_ops[] = {
1376 { .cmd = OVS_DP_CMD_NEW,
1377 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1378 .policy = datapath_policy,
1379 .doit = ovs_dp_cmd_new
1380 },
1381 { .cmd = OVS_DP_CMD_DEL,
1382 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1383 .policy = datapath_policy,
1384 .doit = ovs_dp_cmd_del
1385 },
1386 { .cmd = OVS_DP_CMD_GET,
1387 .flags = 0, /* OK for unprivileged users. */
1388 .policy = datapath_policy,
1389 .doit = ovs_dp_cmd_get,
1390 .dumpit = ovs_dp_cmd_dump
1391 },
1392 { .cmd = OVS_DP_CMD_SET,
1393 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1394 .policy = datapath_policy,
1395 .doit = ovs_dp_cmd_set,
1396 },
1397 };
1398
1399 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1400 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1401 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1402 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1403 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1404 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1405 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1406 };
1407
1408 static struct genl_family dp_vport_genl_family = {
1409 .id = GENL_ID_GENERATE,
1410 .hdrsize = sizeof(struct ovs_header),
1411 .name = OVS_VPORT_FAMILY,
1412 .version = OVS_VPORT_VERSION,
1413 .maxattr = OVS_VPORT_ATTR_MAX,
1414 .netnsok = true,
1415 SET_PARALLEL_OPS
1416 };
1417
1418 struct genl_multicast_group ovs_dp_vport_multicast_group = {
1419 .name = OVS_VPORT_MCGROUP
1420 };
1421
1422 /* Called with ovs_mutex or RCU read lock. */
1423 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1424 u32 portid, u32 seq, u32 flags, u8 cmd)
1425 {
1426 struct ovs_header *ovs_header;
1427 struct ovs_vport_stats vport_stats;
1428 int err;
1429
1430 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1431 flags, cmd);
1432 if (!ovs_header)
1433 return -EMSGSIZE;
1434
1435 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1436
1437 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1438 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1439 nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1440 nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_portid))
1441 goto nla_put_failure;
1442
1443 ovs_vport_get_stats(vport, &vport_stats);
1444 if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1445 &vport_stats))
1446 goto nla_put_failure;
1447
1448 err = ovs_vport_get_options(vport, skb);
1449 if (err == -EMSGSIZE)
1450 goto error;
1451
1452 return genlmsg_end(skb, ovs_header);
1453
1454 nla_put_failure:
1455 err = -EMSGSIZE;
1456 error:
1457 genlmsg_cancel(skb, ovs_header);
1458 return err;
1459 }
1460
1461 /* Called with ovs_mutex or RCU read lock. */
1462 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1463 u32 seq, u8 cmd)
1464 {
1465 struct sk_buff *skb;
1466 int retval;
1467
1468 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1469 if (!skb)
1470 return ERR_PTR(-ENOMEM);
1471
1472 retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1473 BUG_ON(retval < 0);
1474
1475 return skb;
1476 }
1477
1478 /* Called with ovs_mutex or RCU read lock. */
1479 static struct vport *lookup_vport(struct net *net,
1480 struct ovs_header *ovs_header,
1481 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1482 {
1483 struct datapath *dp;
1484 struct vport *vport;
1485
1486 if (a[OVS_VPORT_ATTR_NAME]) {
1487 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1488 if (!vport)
1489 return ERR_PTR(-ENODEV);
1490 if (ovs_header->dp_ifindex &&
1491 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1492 return ERR_PTR(-ENODEV);
1493 return vport;
1494 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1495 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1496
1497 if (port_no >= DP_MAX_PORTS)
1498 return ERR_PTR(-EFBIG);
1499
1500 dp = get_dp(net, ovs_header->dp_ifindex);
1501 if (!dp)
1502 return ERR_PTR(-ENODEV);
1503
1504 vport = ovs_vport_ovsl_rcu(dp, port_no);
1505 if (!vport)
1506 return ERR_PTR(-ENODEV);
1507 return vport;
1508 } else
1509 return ERR_PTR(-EINVAL);
1510 }
1511
1512 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1513 {
1514 struct nlattr **a = info->attrs;
1515 struct ovs_header *ovs_header = info->userhdr;
1516 struct vport_parms parms;
1517 struct sk_buff *reply;
1518 struct vport *vport;
1519 struct datapath *dp;
1520 u32 port_no;
1521 int err;
1522
1523 err = -EINVAL;
1524 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1525 !a[OVS_VPORT_ATTR_UPCALL_PID])
1526 goto exit;
1527
1528 ovs_lock();
1529 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1530 err = -ENODEV;
1531 if (!dp)
1532 goto exit_unlock;
1533
1534 if (a[OVS_VPORT_ATTR_PORT_NO]) {
1535 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1536
1537 err = -EFBIG;
1538 if (port_no >= DP_MAX_PORTS)
1539 goto exit_unlock;
1540
1541 vport = ovs_vport_ovsl(dp, port_no);
1542 err = -EBUSY;
1543 if (vport)
1544 goto exit_unlock;
1545 } else {
1546 for (port_no = 1; ; port_no++) {
1547 if (port_no >= DP_MAX_PORTS) {
1548 err = -EFBIG;
1549 goto exit_unlock;
1550 }
1551 vport = ovs_vport_ovsl(dp, port_no);
1552 if (!vport)
1553 break;
1554 }
1555 }
1556
1557 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1558 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1559 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1560 parms.dp = dp;
1561 parms.port_no = port_no;
1562 parms.upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1563
1564 vport = new_vport(&parms);
1565 err = PTR_ERR(vport);
1566 if (IS_ERR(vport))
1567 goto exit_unlock;
1568
1569 err = 0;
1570 if (a[OVS_VPORT_ATTR_STATS])
1571 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1572
1573 reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
1574 OVS_VPORT_CMD_NEW);
1575 if (IS_ERR(reply)) {
1576 err = PTR_ERR(reply);
1577 ovs_dp_detach_port(vport);
1578 goto exit_unlock;
1579 }
1580
1581 ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
1582
1583 exit_unlock:
1584 ovs_unlock();
1585 exit:
1586 return err;
1587 }
1588
1589 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1590 {
1591 struct nlattr **a = info->attrs;
1592 struct sk_buff *reply;
1593 struct vport *vport;
1594 int err;
1595
1596 ovs_lock();
1597 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1598 err = PTR_ERR(vport);
1599 if (IS_ERR(vport))
1600 goto exit_unlock;
1601
1602 if (a[OVS_VPORT_ATTR_TYPE] &&
1603 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
1604 err = -EINVAL;
1605 goto exit_unlock;
1606 }
1607
1608 reply = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1609 if (!reply) {
1610 err = -ENOMEM;
1611 goto exit_unlock;
1612 }
1613
1614 if (a[OVS_VPORT_ATTR_OPTIONS]) {
1615 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1616 if (err)
1617 goto exit_free;
1618 }
1619
1620 if (a[OVS_VPORT_ATTR_STATS])
1621 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1622
1623 if (a[OVS_VPORT_ATTR_UPCALL_PID])
1624 vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1625
1626 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1627 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1628 BUG_ON(err < 0);
1629
1630 ovs_unlock();
1631 ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
1632 return 0;
1633
1634 exit_free:
1635 kfree_skb(reply);
1636 exit_unlock:
1637 ovs_unlock();
1638 return err;
1639 }
1640
1641 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1642 {
1643 struct nlattr **a = info->attrs;
1644 struct sk_buff *reply;
1645 struct vport *vport;
1646 int err;
1647
1648 ovs_lock();
1649 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1650 err = PTR_ERR(vport);
1651 if (IS_ERR(vport))
1652 goto exit_unlock;
1653
1654 if (vport->port_no == OVSP_LOCAL) {
1655 err = -EINVAL;
1656 goto exit_unlock;
1657 }
1658
1659 reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
1660 info->snd_seq, OVS_VPORT_CMD_DEL);
1661 err = PTR_ERR(reply);
1662 if (IS_ERR(reply))
1663 goto exit_unlock;
1664
1665 err = 0;
1666 ovs_dp_detach_port(vport);
1667
1668 ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
1669
1670 exit_unlock:
1671 ovs_unlock();
1672 return err;
1673 }
1674
1675 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1676 {
1677 struct nlattr **a = info->attrs;
1678 struct ovs_header *ovs_header = info->userhdr;
1679 struct sk_buff *reply;
1680 struct vport *vport;
1681 int err;
1682
1683 rcu_read_lock();
1684 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1685 err = PTR_ERR(vport);
1686 if (IS_ERR(vport))
1687 goto exit_unlock;
1688
1689 reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
1690 info->snd_seq, OVS_VPORT_CMD_NEW);
1691 err = PTR_ERR(reply);
1692 if (IS_ERR(reply))
1693 goto exit_unlock;
1694
1695 rcu_read_unlock();
1696
1697 return genlmsg_reply(reply, info);
1698
1699 exit_unlock:
1700 rcu_read_unlock();
1701 return err;
1702 }
1703
1704 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1705 {
1706 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1707 struct datapath *dp;
1708 int bucket = cb->args[0], skip = cb->args[1];
1709 int i, j = 0;
1710
1711 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1712 if (!dp)
1713 return -ENODEV;
1714
1715 rcu_read_lock();
1716 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
1717 struct vport *vport;
1718
1719 j = 0;
1720 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1721 if (j >= skip &&
1722 ovs_vport_cmd_fill_info(vport, skb,
1723 NETLINK_CB(cb->skb).portid,
1724 cb->nlh->nlmsg_seq,
1725 NLM_F_MULTI,
1726 OVS_VPORT_CMD_NEW) < 0)
1727 goto out;
1728
1729 j++;
1730 }
1731 skip = 0;
1732 }
1733 out:
1734 rcu_read_unlock();
1735
1736 cb->args[0] = i;
1737 cb->args[1] = j;
1738
1739 return skb->len;
1740 }
1741
1742 static struct genl_ops dp_vport_genl_ops[] = {
1743 { .cmd = OVS_VPORT_CMD_NEW,
1744 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1745 .policy = vport_policy,
1746 .doit = ovs_vport_cmd_new
1747 },
1748 { .cmd = OVS_VPORT_CMD_DEL,
1749 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1750 .policy = vport_policy,
1751 .doit = ovs_vport_cmd_del
1752 },
1753 { .cmd = OVS_VPORT_CMD_GET,
1754 .flags = 0, /* OK for unprivileged users. */
1755 .policy = vport_policy,
1756 .doit = ovs_vport_cmd_get,
1757 .dumpit = ovs_vport_cmd_dump
1758 },
1759 { .cmd = OVS_VPORT_CMD_SET,
1760 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1761 .policy = vport_policy,
1762 .doit = ovs_vport_cmd_set,
1763 },
1764 };
1765
1766 struct genl_family_and_ops {
1767 struct genl_family *family;
1768 struct genl_ops *ops;
1769 int n_ops;
1770 struct genl_multicast_group *group;
1771 };
1772
1773 static const struct genl_family_and_ops dp_genl_families[] = {
1774 { &dp_datapath_genl_family,
1775 dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1776 &ovs_dp_datapath_multicast_group },
1777 { &dp_vport_genl_family,
1778 dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1779 &ovs_dp_vport_multicast_group },
1780 { &dp_flow_genl_family,
1781 dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1782 &ovs_dp_flow_multicast_group },
1783 { &dp_packet_genl_family,
1784 dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1785 NULL },
1786 };
1787
1788 static void dp_unregister_genl(int n_families)
1789 {
1790 int i;
1791
1792 for (i = 0; i < n_families; i++)
1793 genl_unregister_family(dp_genl_families[i].family);
1794 }
1795
1796 static int dp_register_genl(void)
1797 {
1798 int n_registered;
1799 int err;
1800 int i;
1801
1802 n_registered = 0;
1803 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
1804 const struct genl_family_and_ops *f = &dp_genl_families[i];
1805
1806 err = genl_register_family_with_ops(f->family, f->ops,
1807 f->n_ops);
1808 if (err)
1809 goto error;
1810 n_registered++;
1811
1812 if (f->group) {
1813 err = genl_register_mc_group(f->family, f->group);
1814 if (err)
1815 goto error;
1816 }
1817 }
1818
1819 return 0;
1820
1821 error:
1822 dp_unregister_genl(n_registered);
1823 return err;
1824 }
1825
1826 static int __net_init ovs_init_net(struct net *net)
1827 {
1828 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1829
1830 INIT_LIST_HEAD(&ovs_net->dps);
1831 INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
1832 return 0;
1833 }
1834
1835 static void __net_exit ovs_exit_net(struct net *net)
1836 {
1837 struct datapath *dp, *dp_next;
1838 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1839
1840 ovs_lock();
1841 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
1842 __dp_destroy(dp);
1843 ovs_unlock();
1844
1845 cancel_work_sync(&ovs_net->dp_notify_work);
1846 }
1847
1848 static struct pernet_operations ovs_net_ops = {
1849 .init = ovs_init_net,
1850 .exit = ovs_exit_net,
1851 .id = &ovs_net_id,
1852 .size = sizeof(struct ovs_net),
1853 };
1854
1855 DEFINE_COMPAT_PNET_REG_FUNC(device);
1856
1857 static int __init dp_init(void)
1858 {
1859 int err;
1860
1861 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
1862
1863 pr_info("Open vSwitch switching datapath %s, built "__DATE__" "__TIME__"\n",
1864 VERSION);
1865
1866 err = ovs_flow_init();
1867 if (err)
1868 goto error;
1869
1870 err = ovs_vport_init();
1871 if (err)
1872 goto error_flow_exit;
1873
1874 err = register_pernet_device(&ovs_net_ops);
1875 if (err)
1876 goto error_vport_exit;
1877
1878 err = register_netdevice_notifier(&ovs_dp_device_notifier);
1879 if (err)
1880 goto error_netns_exit;
1881
1882 err = dp_register_genl();
1883 if (err < 0)
1884 goto error_unreg_notifier;
1885
1886 return 0;
1887
1888 error_unreg_notifier:
1889 unregister_netdevice_notifier(&ovs_dp_device_notifier);
1890 error_netns_exit:
1891 unregister_pernet_device(&ovs_net_ops);
1892 error_vport_exit:
1893 ovs_vport_exit();
1894 error_flow_exit:
1895 ovs_flow_exit();
1896 error:
1897 return err;
1898 }
1899
1900 static void dp_cleanup(void)
1901 {
1902 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
1903 unregister_netdevice_notifier(&ovs_dp_device_notifier);
1904 unregister_pernet_device(&ovs_net_ops);
1905 rcu_barrier();
1906 ovs_vport_exit();
1907 ovs_flow_exit();
1908 }
1909
1910 module_init(dp_init);
1911 module_exit(dp_cleanup);
1912
1913 MODULE_DESCRIPTION("Open vSwitch switching datapath");
1914 MODULE_LICENSE("GPL");
1915 MODULE_VERSION(VERSION);
openvswitch mirror