projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
net: introduce netdev_alloc_pcpu_stats() for drivers
[deliverable/linux.git] / net / openvswitch / datapath.c
1 /*
2 * Copyright (c) 2007-2013 Nicira, Inc.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of version 2 of the GNU General Public
6 * License as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public License
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
16 * 02110-1301, USA
17 */
18
19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
20
21 #include <linux/init.h>
22 #include <linux/module.h>
23 #include <linux/if_arp.h>
24 #include <linux/if_vlan.h>
25 #include <linux/in.h>
26 #include <linux/ip.h>
27 #include <linux/jhash.h>
28 #include <linux/delay.h>
29 #include <linux/time.h>
30 #include <linux/etherdevice.h>
31 #include <linux/genetlink.h>
32 #include <linux/kernel.h>
33 #include <linux/kthread.h>
34 #include <linux/mutex.h>
35 #include <linux/percpu.h>
36 #include <linux/rcupdate.h>
37 #include <linux/tcp.h>
38 #include <linux/udp.h>
39 #include <linux/ethtool.h>
40 #include <linux/wait.h>
41 #include <asm/div64.h>
42 #include <linux/highmem.h>
43 #include <linux/netfilter_bridge.h>
44 #include <linux/netfilter_ipv4.h>
45 #include <linux/inetdevice.h>
46 #include <linux/list.h>
47 #include <linux/lockdep.h>
48 #include <linux/openvswitch.h>
49 #include <linux/rculist.h>
50 #include <linux/dmi.h>
51 #include <linux/workqueue.h>
52 #include <net/genetlink.h>
53 #include <net/net_namespace.h>
54 #include <net/netns/generic.h>
55
56 #include "datapath.h"
57 #include "flow.h"
58 #include "flow_table.h"
59 #include "flow_netlink.h"
60 #include "vport-internal_dev.h"
61 #include "vport-netdev.h"
62
63 int ovs_net_id __read_mostly;
64
65 static void ovs_notify(struct genl_family *family,
66 struct sk_buff *skb, struct genl_info *info)
67 {
68 genl_notify(family, skb, genl_info_net(info), info->snd_portid,
69 0, info->nlhdr, GFP_KERNEL);
70 }
71
72 /**
73 * DOC: Locking:
74 *
75 * All writes e.g. Writes to device state (add/remove datapath, port, set
76 * operations on vports, etc.), Writes to other state (flow table
77 * modifications, set miscellaneous datapath parameters, etc.) are protected
78 * by ovs_lock.
79 *
80 * Reads are protected by RCU.
81 *
82 * There are a few special cases (mostly stats) that have their own
83 * synchronization but they nest under all of above and don't interact with
84 * each other.
85 *
86 * The RTNL lock nests inside ovs_mutex.
87 */
88
89 static DEFINE_MUTEX(ovs_mutex);
90
91 void ovs_lock(void)
92 {
93 mutex_lock(&ovs_mutex);
94 }
95
96 void ovs_unlock(void)
97 {
98 mutex_unlock(&ovs_mutex);
99 }
100
101 #ifdef CONFIG_LOCKDEP
102 int lockdep_ovsl_is_held(void)
103 {
104 if (debug_locks)
105 return lockdep_is_held(&ovs_mutex);
106 else
107 return 1;
108 }
109 #endif
110
111 static struct vport *new_vport(const struct vport_parms *);
112 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
113 const struct dp_upcall_info *);
114 static int queue_userspace_packet(struct datapath *dp, 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 static 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 free_percpu(dp->stats_percpu);
165 release_net(ovs_dp_get_net(dp));
166 kfree(dp->ports);
167 kfree(dp);
168 }
169
170 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
171 u16 port_no)
172 {
173 return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
174 }
175
176 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
177 {
178 struct vport *vport;
179 struct hlist_head *head;
180
181 head = vport_hash_bucket(dp, port_no);
182 hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
183 if (vport->port_no == port_no)
184 return vport;
185 }
186 return NULL;
187 }
188
189 /* Called with ovs_mutex. */
190 static struct vport *new_vport(const struct vport_parms *parms)
191 {
192 struct vport *vport;
193
194 vport = ovs_vport_add(parms);
195 if (!IS_ERR(vport)) {
196 struct datapath *dp = parms->dp;
197 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
198
199 hlist_add_head_rcu(&vport->dp_hash_node, head);
200 }
201 return vport;
202 }
203
204 void ovs_dp_detach_port(struct vport *p)
205 {
206 ASSERT_OVSL();
207
208 /* First drop references to device. */
209 hlist_del_rcu(&p->dp_hash_node);
210
211 /* Then destroy it. */
212 ovs_vport_del(p);
213 }
214
215 /* Must be called with rcu_read_lock. */
216 void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
217 {
218 struct datapath *dp = p->dp;
219 struct sw_flow *flow;
220 struct dp_stats_percpu *stats;
221 struct sw_flow_key key;
222 u64 *stats_counter;
223 u32 n_mask_hit;
224 int error;
225
226 stats = this_cpu_ptr(dp->stats_percpu);
227
228 /* Extract flow from 'skb' into 'key'. */
229 error = ovs_flow_extract(skb, p->port_no, &key);
230 if (unlikely(error)) {
231 kfree_skb(skb);
232 return;
233 }
234
235 /* Look up flow. */
236 flow = ovs_flow_tbl_lookup_stats(&dp->table, &key, &n_mask_hit);
237 if (unlikely(!flow)) {
238 struct dp_upcall_info upcall;
239
240 upcall.cmd = OVS_PACKET_CMD_MISS;
241 upcall.key = &key;
242 upcall.userdata = NULL;
243 upcall.portid = p->upcall_portid;
244 ovs_dp_upcall(dp, skb, &upcall);
245 consume_skb(skb);
246 stats_counter = &stats->n_missed;
247 goto out;
248 }
249
250 OVS_CB(skb)->flow = flow;
251 OVS_CB(skb)->pkt_key = &key;
252
253 ovs_flow_stats_update(OVS_CB(skb)->flow, skb);
254 ovs_execute_actions(dp, skb);
255 stats_counter = &stats->n_hit;
256
257 out:
258 /* Update datapath statistics. */
259 u64_stats_update_begin(&stats->sync);
260 (*stats_counter)++;
261 stats->n_mask_hit += n_mask_hit;
262 u64_stats_update_end(&stats->sync);
263 }
264
265 static struct genl_family dp_packet_genl_family = {
266 .id = GENL_ID_GENERATE,
267 .hdrsize = sizeof(struct ovs_header),
268 .name = OVS_PACKET_FAMILY,
269 .version = OVS_PACKET_VERSION,
270 .maxattr = OVS_PACKET_ATTR_MAX,
271 .netnsok = true,
272 .parallel_ops = true,
273 };
274
275 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
276 const struct dp_upcall_info *upcall_info)
277 {
278 struct dp_stats_percpu *stats;
279 int err;
280
281 if (upcall_info->portid == 0) {
282 err = -ENOTCONN;
283 goto err;
284 }
285
286 if (!skb_is_gso(skb))
287 err = queue_userspace_packet(dp, skb, upcall_info);
288 else
289 err = queue_gso_packets(dp, skb, upcall_info);
290 if (err)
291 goto err;
292
293 return 0;
294
295 err:
296 stats = this_cpu_ptr(dp->stats_percpu);
297
298 u64_stats_update_begin(&stats->sync);
299 stats->n_lost++;
300 u64_stats_update_end(&stats->sync);
301
302 return err;
303 }
304
305 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
306 const struct dp_upcall_info *upcall_info)
307 {
308 unsigned short gso_type = skb_shinfo(skb)->gso_type;
309 struct dp_upcall_info later_info;
310 struct sw_flow_key later_key;
311 struct sk_buff *segs, *nskb;
312 int err;
313
314 segs = __skb_gso_segment(skb, NETIF_F_SG, false);
315 if (IS_ERR(segs))
316 return PTR_ERR(segs);
317
318 /* Queue all of the segments. */
319 skb = segs;
320 do {
321 err = queue_userspace_packet(dp, skb, upcall_info);
322 if (err)
323 break;
324
325 if (skb == segs && gso_type & SKB_GSO_UDP) {
326 /* The initial flow key extracted by ovs_flow_extract()
327 * in this case is for a first fragment, so we need to
328 * properly mark later fragments.
329 */
330 later_key = *upcall_info->key;
331 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
332
333 later_info = *upcall_info;
334 later_info.key = &later_key;
335 upcall_info = &later_info;
336 }
337 } while ((skb = skb->next));
338
339 /* Free all of the segments. */
340 skb = segs;
341 do {
342 nskb = skb->next;
343 if (err)
344 kfree_skb(skb);
345 else
346 consume_skb(skb);
347 } while ((skb = nskb));
348 return err;
349 }
350
351 static size_t key_attr_size(void)
352 {
353 return nla_total_size(4) /* OVS_KEY_ATTR_PRIORITY */
354 + nla_total_size(0) /* OVS_KEY_ATTR_TUNNEL */
355 + nla_total_size(8) /* OVS_TUNNEL_KEY_ATTR_ID */
356 + nla_total_size(4) /* OVS_TUNNEL_KEY_ATTR_IPV4_SRC */
357 + nla_total_size(4) /* OVS_TUNNEL_KEY_ATTR_IPV4_DST */
358 + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TOS */
359 + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TTL */
360 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
361 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_CSUM */
362 + nla_total_size(4) /* OVS_KEY_ATTR_IN_PORT */
363 + nla_total_size(4) /* OVS_KEY_ATTR_SKB_MARK */
364 + nla_total_size(12) /* OVS_KEY_ATTR_ETHERNET */
365 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
366 + nla_total_size(4) /* OVS_KEY_ATTR_8021Q */
367 + nla_total_size(0) /* OVS_KEY_ATTR_ENCAP */
368 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
369 + nla_total_size(40) /* OVS_KEY_ATTR_IPV6 */
370 + nla_total_size(2) /* OVS_KEY_ATTR_ICMPV6 */
371 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
372 }
373
374 static size_t upcall_msg_size(const struct nlattr *userdata,
375 unsigned int hdrlen)
376 {
377 size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
378 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
379 + nla_total_size(key_attr_size()); /* OVS_PACKET_ATTR_KEY */
380
381 /* OVS_PACKET_ATTR_USERDATA */
382 if (userdata)
383 size += NLA_ALIGN(userdata->nla_len);
384
385 return size;
386 }
387
388 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
389 const struct dp_upcall_info *upcall_info)
390 {
391 struct ovs_header *upcall;
392 struct sk_buff *nskb = NULL;
393 struct sk_buff *user_skb; /* to be queued to userspace */
394 struct nlattr *nla;
395 struct genl_info info = {
396 .dst_sk = ovs_dp_get_net(dp)->genl_sock,
397 .snd_portid = upcall_info->portid,
398 };
399 size_t len;
400 unsigned int hlen;
401 int err, dp_ifindex;
402
403 dp_ifindex = get_dpifindex(dp);
404 if (!dp_ifindex)
405 return -ENODEV;
406
407 if (vlan_tx_tag_present(skb)) {
408 nskb = skb_clone(skb, GFP_ATOMIC);
409 if (!nskb)
410 return -ENOMEM;
411
412 nskb = __vlan_put_tag(nskb, nskb->vlan_proto, vlan_tx_tag_get(nskb));
413 if (!nskb)
414 return -ENOMEM;
415
416 nskb->vlan_tci = 0;
417 skb = nskb;
418 }
419
420 if (nla_attr_size(skb->len) > USHRT_MAX) {
421 err = -EFBIG;
422 goto out;
423 }
424
425 /* Complete checksum if needed */
426 if (skb->ip_summed == CHECKSUM_PARTIAL &&
427 (err = skb_checksum_help(skb)))
428 goto out;
429
430 /* Older versions of OVS user space enforce alignment of the last
431 * Netlink attribute to NLA_ALIGNTO which would require extensive
432 * padding logic. Only perform zerocopy if padding is not required.
433 */
434 if (dp->user_features & OVS_DP_F_UNALIGNED)
435 hlen = skb_zerocopy_headlen(skb);
436 else
437 hlen = skb->len;
438
439 len = upcall_msg_size(upcall_info->userdata, hlen);
440 user_skb = genlmsg_new_unicast(len, &info, GFP_ATOMIC);
441 if (!user_skb) {
442 err = -ENOMEM;
443 goto out;
444 }
445
446 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
447 0, upcall_info->cmd);
448 upcall->dp_ifindex = dp_ifindex;
449
450 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
451 ovs_nla_put_flow(upcall_info->key, upcall_info->key, user_skb);
452 nla_nest_end(user_skb, nla);
453
454 if (upcall_info->userdata)
455 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
456 nla_len(upcall_info->userdata),
457 nla_data(upcall_info->userdata));
458
459 /* Only reserve room for attribute header, packet data is added
460 * in skb_zerocopy() */
461 if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
462 err = -ENOBUFS;
463 goto out;
464 }
465 nla->nla_len = nla_attr_size(skb->len);
466
467 skb_zerocopy(user_skb, skb, skb->len, hlen);
468
469 /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
470 if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
471 size_t plen = NLA_ALIGN(user_skb->len) - user_skb->len;
472
473 if (plen > 0)
474 memset(skb_put(user_skb, plen), 0, plen);
475 }
476
477 ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
478
479 err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
480 out:
481 kfree_skb(nskb);
482 return err;
483 }
484
485 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
486 {
487 struct ovs_header *ovs_header = info->userhdr;
488 struct nlattr **a = info->attrs;
489 struct sw_flow_actions *acts;
490 struct sk_buff *packet;
491 struct sw_flow *flow;
492 struct datapath *dp;
493 struct ethhdr *eth;
494 int len;
495 int err;
496
497 err = -EINVAL;
498 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
499 !a[OVS_PACKET_ATTR_ACTIONS])
500 goto err;
501
502 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
503 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
504 err = -ENOMEM;
505 if (!packet)
506 goto err;
507 skb_reserve(packet, NET_IP_ALIGN);
508
509 nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
510
511 skb_reset_mac_header(packet);
512 eth = eth_hdr(packet);
513
514 /* Normally, setting the skb 'protocol' field would be handled by a
515 * call to eth_type_trans(), but it assumes there's a sending
516 * device, which we may not have. */
517 if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
518 packet->protocol = eth->h_proto;
519 else
520 packet->protocol = htons(ETH_P_802_2);
521
522 /* Build an sw_flow for sending this packet. */
523 flow = ovs_flow_alloc(false);
524 err = PTR_ERR(flow);
525 if (IS_ERR(flow))
526 goto err_kfree_skb;
527
528 err = ovs_flow_extract(packet, -1, &flow->key);
529 if (err)
530 goto err_flow_free;
531
532 err = ovs_nla_get_flow_metadata(flow, a[OVS_PACKET_ATTR_KEY]);
533 if (err)
534 goto err_flow_free;
535 acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_PACKET_ATTR_ACTIONS]));
536 err = PTR_ERR(acts);
537 if (IS_ERR(acts))
538 goto err_flow_free;
539
540 err = ovs_nla_copy_actions(a[OVS_PACKET_ATTR_ACTIONS],
541 &flow->key, 0, &acts);
542 rcu_assign_pointer(flow->sf_acts, acts);
543 if (err)
544 goto err_flow_free;
545
546 OVS_CB(packet)->flow = flow;
547 OVS_CB(packet)->pkt_key = &flow->key;
548 packet->priority = flow->key.phy.priority;
549 packet->mark = flow->key.phy.skb_mark;
550
551 rcu_read_lock();
552 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
553 err = -ENODEV;
554 if (!dp)
555 goto err_unlock;
556
557 local_bh_disable();
558 err = ovs_execute_actions(dp, packet);
559 local_bh_enable();
560 rcu_read_unlock();
561
562 ovs_flow_free(flow, false);
563 return err;
564
565 err_unlock:
566 rcu_read_unlock();
567 err_flow_free:
568 ovs_flow_free(flow, false);
569 err_kfree_skb:
570 kfree_skb(packet);
571 err:
572 return err;
573 }
574
575 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
576 [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
577 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
578 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
579 };
580
581 static const struct genl_ops dp_packet_genl_ops[] = {
582 { .cmd = OVS_PACKET_CMD_EXECUTE,
583 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
584 .policy = packet_policy,
585 .doit = ovs_packet_cmd_execute
586 }
587 };
588
589 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats,
590 struct ovs_dp_megaflow_stats *mega_stats)
591 {
592 int i;
593
594 memset(mega_stats, 0, sizeof(*mega_stats));
595
596 stats->n_flows = ovs_flow_tbl_count(&dp->table);
597 mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
598
599 stats->n_hit = stats->n_missed = stats->n_lost = 0;
600
601 for_each_possible_cpu(i) {
602 const struct dp_stats_percpu *percpu_stats;
603 struct dp_stats_percpu local_stats;
604 unsigned int start;
605
606 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
607
608 do {
609 start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
610 local_stats = *percpu_stats;
611 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
612
613 stats->n_hit += local_stats.n_hit;
614 stats->n_missed += local_stats.n_missed;
615 stats->n_lost += local_stats.n_lost;
616 mega_stats->n_mask_hit += local_stats.n_mask_hit;
617 }
618 }
619
620 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
621 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
622 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
623 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
624 };
625
626 static struct genl_family dp_flow_genl_family = {
627 .id = GENL_ID_GENERATE,
628 .hdrsize = sizeof(struct ovs_header),
629 .name = OVS_FLOW_FAMILY,
630 .version = OVS_FLOW_VERSION,
631 .maxattr = OVS_FLOW_ATTR_MAX,
632 .netnsok = true,
633 .parallel_ops = true,
634 };
635
636 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
637 .name = OVS_FLOW_MCGROUP
638 };
639
640 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
641 {
642 return NLMSG_ALIGN(sizeof(struct ovs_header))
643 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
644 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_MASK */
645 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
646 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
647 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
648 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
649 }
650
651 /* Called with ovs_mutex. */
652 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
653 struct sk_buff *skb, u32 portid,
654 u32 seq, u32 flags, u8 cmd)
655 {
656 const int skb_orig_len = skb->len;
657 struct nlattr *start;
658 struct ovs_flow_stats stats;
659 __be16 tcp_flags;
660 unsigned long used;
661 struct ovs_header *ovs_header;
662 struct nlattr *nla;
663 int err;
664
665 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
666 if (!ovs_header)
667 return -EMSGSIZE;
668
669 ovs_header->dp_ifindex = get_dpifindex(dp);
670
671 /* Fill flow key. */
672 nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
673 if (!nla)
674 goto nla_put_failure;
675
676 err = ovs_nla_put_flow(&flow->unmasked_key, &flow->unmasked_key, skb);
677 if (err)
678 goto error;
679 nla_nest_end(skb, nla);
680
681 nla = nla_nest_start(skb, OVS_FLOW_ATTR_MASK);
682 if (!nla)
683 goto nla_put_failure;
684
685 err = ovs_nla_put_flow(&flow->key, &flow->mask->key, skb);
686 if (err)
687 goto error;
688
689 nla_nest_end(skb, nla);
690
691 ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
692 if (used &&
693 nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
694 goto nla_put_failure;
695
696 if (stats.n_packets &&
697 nla_put(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats))
698 goto nla_put_failure;
699
700 if ((u8)ntohs(tcp_flags) &&
701 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
702 goto nla_put_failure;
703
704 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
705 * this is the first flow to be dumped into 'skb'. This is unusual for
706 * Netlink but individual action lists can be longer than
707 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
708 * The userspace caller can always fetch the actions separately if it
709 * really wants them. (Most userspace callers in fact don't care.)
710 *
711 * This can only fail for dump operations because the skb is always
712 * properly sized for single flows.
713 */
714 start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
715 if (start) {
716 const struct sw_flow_actions *sf_acts;
717
718 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
719
720 err = ovs_nla_put_actions(sf_acts->actions,
721 sf_acts->actions_len, skb);
722 if (!err)
723 nla_nest_end(skb, start);
724 else {
725 if (skb_orig_len)
726 goto error;
727
728 nla_nest_cancel(skb, start);
729 }
730 } else if (skb_orig_len)
731 goto nla_put_failure;
732
733 return genlmsg_end(skb, ovs_header);
734
735 nla_put_failure:
736 err = -EMSGSIZE;
737 error:
738 genlmsg_cancel(skb, ovs_header);
739 return err;
740 }
741
742 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow,
743 struct genl_info *info)
744 {
745 size_t len;
746
747 len = ovs_flow_cmd_msg_size(ovsl_dereference(flow->sf_acts));
748
749 return genlmsg_new_unicast(len, info, GFP_KERNEL);
750 }
751
752 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
753 struct datapath *dp,
754 struct genl_info *info,
755 u8 cmd)
756 {
757 struct sk_buff *skb;
758 int retval;
759
760 skb = ovs_flow_cmd_alloc_info(flow, info);
761 if (!skb)
762 return ERR_PTR(-ENOMEM);
763
764 retval = ovs_flow_cmd_fill_info(flow, dp, skb, info->snd_portid,
765 info->snd_seq, 0, cmd);
766 BUG_ON(retval < 0);
767 return skb;
768 }
769
770 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
771 {
772 struct nlattr **a = info->attrs;
773 struct ovs_header *ovs_header = info->userhdr;
774 struct sw_flow_key key, masked_key;
775 struct sw_flow *flow = NULL;
776 struct sw_flow_mask mask;
777 struct sk_buff *reply;
778 struct datapath *dp;
779 struct sw_flow_actions *acts = NULL;
780 struct sw_flow_match match;
781 bool exact_5tuple;
782 int error;
783
784 /* Extract key. */
785 error = -EINVAL;
786 if (!a[OVS_FLOW_ATTR_KEY])
787 goto error;
788
789 ovs_match_init(&match, &key, &mask);
790 error = ovs_nla_get_match(&match, &exact_5tuple,
791 a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
792 if (error)
793 goto error;
794
795 /* Validate actions. */
796 if (a[OVS_FLOW_ATTR_ACTIONS]) {
797 acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
798 error = PTR_ERR(acts);
799 if (IS_ERR(acts))
800 goto error;
801
802 ovs_flow_mask_key(&masked_key, &key, &mask);
803 error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS],
804 &masked_key, 0, &acts);
805 if (error) {
806 OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
807 goto err_kfree;
808 }
809 } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
810 error = -EINVAL;
811 goto error;
812 }
813
814 ovs_lock();
815 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
816 error = -ENODEV;
817 if (!dp)
818 goto err_unlock_ovs;
819
820 /* Check if this is a duplicate flow */
821 flow = ovs_flow_tbl_lookup(&dp->table, &key);
822 if (!flow) {
823 /* Bail out if we're not allowed to create a new flow. */
824 error = -ENOENT;
825 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
826 goto err_unlock_ovs;
827
828 /* Allocate flow. */
829 flow = ovs_flow_alloc(!exact_5tuple);
830 if (IS_ERR(flow)) {
831 error = PTR_ERR(flow);
832 goto err_unlock_ovs;
833 }
834
835 flow->key = masked_key;
836 flow->unmasked_key = key;
837 rcu_assign_pointer(flow->sf_acts, acts);
838
839 /* Put flow in bucket. */
840 error = ovs_flow_tbl_insert(&dp->table, flow, &mask);
841 if (error) {
842 acts = NULL;
843 goto err_flow_free;
844 }
845
846 reply = ovs_flow_cmd_build_info(flow, dp, info, OVS_FLOW_CMD_NEW);
847 } else {
848 /* We found a matching flow. */
849 struct sw_flow_actions *old_acts;
850
851 /* Bail out if we're not allowed to modify an existing flow.
852 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
853 * because Generic Netlink treats the latter as a dump
854 * request. We also accept NLM_F_EXCL in case that bug ever
855 * gets fixed.
856 */
857 error = -EEXIST;
858 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
859 info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
860 goto err_unlock_ovs;
861
862 /* The unmasked key has to be the same for flow updates. */
863 if (!ovs_flow_cmp_unmasked_key(flow, &match))
864 goto err_unlock_ovs;
865
866 /* Update actions. */
867 old_acts = ovsl_dereference(flow->sf_acts);
868 rcu_assign_pointer(flow->sf_acts, acts);
869 ovs_nla_free_flow_actions(old_acts);
870
871 reply = ovs_flow_cmd_build_info(flow, dp, info, OVS_FLOW_CMD_NEW);
872
873 /* Clear stats. */
874 if (a[OVS_FLOW_ATTR_CLEAR])
875 ovs_flow_stats_clear(flow);
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, NULL, 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, OVS_FLOW_CMD_NEW);
931 if (IS_ERR(reply)) {
932 err = PTR_ERR(reply);
933 goto unlock;
934 }
935
936 ovs_unlock();
937 return genlmsg_reply(reply, info);
938 unlock:
939 ovs_unlock();
940 return err;
941 }
942
943 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
944 {
945 struct nlattr **a = info->attrs;
946 struct ovs_header *ovs_header = info->userhdr;
947 struct sw_flow_key key;
948 struct sk_buff *reply;
949 struct sw_flow *flow;
950 struct datapath *dp;
951 struct sw_flow_match match;
952 int err;
953
954 ovs_lock();
955 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
956 if (!dp) {
957 err = -ENODEV;
958 goto unlock;
959 }
960
961 if (!a[OVS_FLOW_ATTR_KEY]) {
962 err = ovs_flow_tbl_flush(&dp->table);
963 goto unlock;
964 }
965
966 ovs_match_init(&match, &key, NULL);
967 err = ovs_nla_get_match(&match, NULL, a[OVS_FLOW_ATTR_KEY], NULL);
968 if (err)
969 goto unlock;
970
971 flow = ovs_flow_tbl_lookup(&dp->table, &key);
972 if (!flow || !ovs_flow_cmp_unmasked_key(flow, &match)) {
973 err = -ENOENT;
974 goto unlock;
975 }
976
977 reply = ovs_flow_cmd_alloc_info(flow, info);
978 if (!reply) {
979 err = -ENOMEM;
980 goto unlock;
981 }
982
983 ovs_flow_tbl_remove(&dp->table, flow);
984
985 err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_portid,
986 info->snd_seq, 0, OVS_FLOW_CMD_DEL);
987 BUG_ON(err < 0);
988
989 ovs_flow_free(flow, true);
990 ovs_unlock();
991
992 ovs_notify(&dp_flow_genl_family, reply, info);
993 return 0;
994 unlock:
995 ovs_unlock();
996 return err;
997 }
998
999 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1000 {
1001 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1002 struct table_instance *ti;
1003 struct datapath *dp;
1004
1005 rcu_read_lock();
1006 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1007 if (!dp) {
1008 rcu_read_unlock();
1009 return -ENODEV;
1010 }
1011
1012 ti = rcu_dereference(dp->table.ti);
1013 for (;;) {
1014 struct sw_flow *flow;
1015 u32 bucket, obj;
1016
1017 bucket = cb->args[0];
1018 obj = cb->args[1];
1019 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1020 if (!flow)
1021 break;
1022
1023 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1024 NETLINK_CB(cb->skb).portid,
1025 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1026 OVS_FLOW_CMD_NEW) < 0)
1027 break;
1028
1029 cb->args[0] = bucket;
1030 cb->args[1] = obj;
1031 }
1032 rcu_read_unlock();
1033 return skb->len;
1034 }
1035
1036 static const struct genl_ops dp_flow_genl_ops[] = {
1037 { .cmd = OVS_FLOW_CMD_NEW,
1038 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1039 .policy = flow_policy,
1040 .doit = ovs_flow_cmd_new_or_set
1041 },
1042 { .cmd = OVS_FLOW_CMD_DEL,
1043 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1044 .policy = flow_policy,
1045 .doit = ovs_flow_cmd_del
1046 },
1047 { .cmd = OVS_FLOW_CMD_GET,
1048 .flags = 0, /* OK for unprivileged users. */
1049 .policy = flow_policy,
1050 .doit = ovs_flow_cmd_get,
1051 .dumpit = ovs_flow_cmd_dump
1052 },
1053 { .cmd = OVS_FLOW_CMD_SET,
1054 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1055 .policy = flow_policy,
1056 .doit = ovs_flow_cmd_new_or_set,
1057 },
1058 };
1059
1060 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1061 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1062 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1063 [OVS_DP_ATTR_USER_FEATURES] = { .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 msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1088
1089 return msgsize;
1090 }
1091
1092 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1093 u32 portid, u32 seq, u32 flags, u8 cmd)
1094 {
1095 struct ovs_header *ovs_header;
1096 struct ovs_dp_stats dp_stats;
1097 struct ovs_dp_megaflow_stats dp_megaflow_stats;
1098 int err;
1099
1100 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1101 flags, cmd);
1102 if (!ovs_header)
1103 goto error;
1104
1105 ovs_header->dp_ifindex = get_dpifindex(dp);
1106
1107 rcu_read_lock();
1108 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1109 rcu_read_unlock();
1110 if (err)
1111 goto nla_put_failure;
1112
1113 get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1114 if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1115 &dp_stats))
1116 goto nla_put_failure;
1117
1118 if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1119 sizeof(struct ovs_dp_megaflow_stats),
1120 &dp_megaflow_stats))
1121 goto nla_put_failure;
1122
1123 if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1124 goto nla_put_failure;
1125
1126 return genlmsg_end(skb, ovs_header);
1127
1128 nla_put_failure:
1129 genlmsg_cancel(skb, ovs_header);
1130 error:
1131 return -EMSGSIZE;
1132 }
1133
1134 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp,
1135 struct genl_info *info, u8 cmd)
1136 {
1137 struct sk_buff *skb;
1138 int retval;
1139
1140 skb = genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info, GFP_KERNEL);
1141 if (!skb)
1142 return ERR_PTR(-ENOMEM);
1143
1144 retval = ovs_dp_cmd_fill_info(dp, skb, info->snd_portid, info->snd_seq, 0, cmd);
1145 if (retval < 0) {
1146 kfree_skb(skb);
1147 return ERR_PTR(retval);
1148 }
1149 return skb;
1150 }
1151
1152 /* Called with ovs_mutex. */
1153 static struct datapath *lookup_datapath(struct net *net,
1154 struct ovs_header *ovs_header,
1155 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1156 {
1157 struct datapath *dp;
1158
1159 if (!a[OVS_DP_ATTR_NAME])
1160 dp = get_dp(net, ovs_header->dp_ifindex);
1161 else {
1162 struct vport *vport;
1163
1164 rcu_read_lock();
1165 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1166 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1167 rcu_read_unlock();
1168 }
1169 return dp ? dp : ERR_PTR(-ENODEV);
1170 }
1171
1172 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1173 {
1174 struct datapath *dp;
1175
1176 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1177 if (!dp)
1178 return;
1179
1180 WARN(dp->user_features, "Dropping previously announced user features\n");
1181 dp->user_features = 0;
1182 }
1183
1184 static void ovs_dp_change(struct datapath *dp, struct nlattr **a)
1185 {
1186 if (a[OVS_DP_ATTR_USER_FEATURES])
1187 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1188 }
1189
1190 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1191 {
1192 struct nlattr **a = info->attrs;
1193 struct vport_parms parms;
1194 struct sk_buff *reply;
1195 struct datapath *dp;
1196 struct vport *vport;
1197 struct ovs_net *ovs_net;
1198 int err, i;
1199
1200 err = -EINVAL;
1201 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1202 goto err;
1203
1204 ovs_lock();
1205
1206 err = -ENOMEM;
1207 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1208 if (dp == NULL)
1209 goto err_unlock_ovs;
1210
1211 ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1212
1213 /* Allocate table. */
1214 err = ovs_flow_tbl_init(&dp->table);
1215 if (err)
1216 goto err_free_dp;
1217
1218 dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1219 if (!dp->stats_percpu) {
1220 err = -ENOMEM;
1221 goto err_destroy_table;
1222 }
1223
1224 dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1225 GFP_KERNEL);
1226 if (!dp->ports) {
1227 err = -ENOMEM;
1228 goto err_destroy_percpu;
1229 }
1230
1231 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1232 INIT_HLIST_HEAD(&dp->ports[i]);
1233
1234 /* Set up our datapath device. */
1235 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1236 parms.type = OVS_VPORT_TYPE_INTERNAL;
1237 parms.options = NULL;
1238 parms.dp = dp;
1239 parms.port_no = OVSP_LOCAL;
1240 parms.upcall_portid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1241
1242 ovs_dp_change(dp, a);
1243
1244 vport = new_vport(&parms);
1245 if (IS_ERR(vport)) {
1246 err = PTR_ERR(vport);
1247 if (err == -EBUSY)
1248 err = -EEXIST;
1249
1250 if (err == -EEXIST) {
1251 /* An outdated user space instance that does not understand
1252 * the concept of user_features has attempted to create a new
1253 * datapath and is likely to reuse it. Drop all user features.
1254 */
1255 if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1256 ovs_dp_reset_user_features(skb, info);
1257 }
1258
1259 goto err_destroy_ports_array;
1260 }
1261
1262 reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_NEW);
1263 err = PTR_ERR(reply);
1264 if (IS_ERR(reply))
1265 goto err_destroy_local_port;
1266
1267 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1268 list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1269
1270 ovs_unlock();
1271
1272 ovs_notify(&dp_datapath_genl_family, reply, info);
1273 return 0;
1274
1275 err_destroy_local_port:
1276 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1277 err_destroy_ports_array:
1278 kfree(dp->ports);
1279 err_destroy_percpu:
1280 free_percpu(dp->stats_percpu);
1281 err_destroy_table:
1282 ovs_flow_tbl_destroy(&dp->table, false);
1283 err_free_dp:
1284 release_net(ovs_dp_get_net(dp));
1285 kfree(dp);
1286 err_unlock_ovs:
1287 ovs_unlock();
1288 err:
1289 return err;
1290 }
1291
1292 /* Called with ovs_mutex. */
1293 static void __dp_destroy(struct datapath *dp)
1294 {
1295 int i;
1296
1297 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1298 struct vport *vport;
1299 struct hlist_node *n;
1300
1301 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1302 if (vport->port_no != OVSP_LOCAL)
1303 ovs_dp_detach_port(vport);
1304 }
1305
1306 list_del_rcu(&dp->list_node);
1307
1308 /* OVSP_LOCAL is datapath internal port. We need to make sure that
1309 * all ports in datapath are destroyed first before freeing datapath.
1310 */
1311 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1312
1313 /* RCU destroy the flow table */
1314 ovs_flow_tbl_destroy(&dp->table, true);
1315
1316 call_rcu(&dp->rcu, destroy_dp_rcu);
1317 }
1318
1319 static int ovs_dp_cmd_del(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, OVS_DP_CMD_DEL);
1332 err = PTR_ERR(reply);
1333 if (IS_ERR(reply))
1334 goto unlock;
1335
1336 __dp_destroy(dp);
1337 ovs_unlock();
1338
1339 ovs_notify(&dp_datapath_genl_family, reply, info);
1340
1341 return 0;
1342 unlock:
1343 ovs_unlock();
1344 return err;
1345 }
1346
1347 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1348 {
1349 struct sk_buff *reply;
1350 struct datapath *dp;
1351 int err;
1352
1353 ovs_lock();
1354 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1355 err = PTR_ERR(dp);
1356 if (IS_ERR(dp))
1357 goto unlock;
1358
1359 ovs_dp_change(dp, info->attrs);
1360
1361 reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_NEW);
1362 if (IS_ERR(reply)) {
1363 err = PTR_ERR(reply);
1364 genl_set_err(&dp_datapath_genl_family, sock_net(skb->sk), 0,
1365 0, err);
1366 err = 0;
1367 goto unlock;
1368 }
1369
1370 ovs_unlock();
1371 ovs_notify(&dp_datapath_genl_family, reply, info);
1372
1373 return 0;
1374 unlock:
1375 ovs_unlock();
1376 return err;
1377 }
1378
1379 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1380 {
1381 struct sk_buff *reply;
1382 struct datapath *dp;
1383 int err;
1384
1385 ovs_lock();
1386 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1387 if (IS_ERR(dp)) {
1388 err = PTR_ERR(dp);
1389 goto unlock;
1390 }
1391
1392 reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_NEW);
1393 if (IS_ERR(reply)) {
1394 err = PTR_ERR(reply);
1395 goto unlock;
1396 }
1397
1398 ovs_unlock();
1399 return genlmsg_reply(reply, info);
1400
1401 unlock:
1402 ovs_unlock();
1403 return err;
1404 }
1405
1406 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1407 {
1408 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1409 struct datapath *dp;
1410 int skip = cb->args[0];
1411 int i = 0;
1412
1413 rcu_read_lock();
1414 list_for_each_entry_rcu(dp, &ovs_net->dps, list_node) {
1415 if (i >= skip &&
1416 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1417 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1418 OVS_DP_CMD_NEW) < 0)
1419 break;
1420 i++;
1421 }
1422 rcu_read_unlock();
1423
1424 cb->args[0] = i;
1425
1426 return skb->len;
1427 }
1428
1429 static const struct genl_ops dp_datapath_genl_ops[] = {
1430 { .cmd = OVS_DP_CMD_NEW,
1431 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1432 .policy = datapath_policy,
1433 .doit = ovs_dp_cmd_new
1434 },
1435 { .cmd = OVS_DP_CMD_DEL,
1436 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1437 .policy = datapath_policy,
1438 .doit = ovs_dp_cmd_del
1439 },
1440 { .cmd = OVS_DP_CMD_GET,
1441 .flags = 0, /* OK for unprivileged users. */
1442 .policy = datapath_policy,
1443 .doit = ovs_dp_cmd_get,
1444 .dumpit = ovs_dp_cmd_dump
1445 },
1446 { .cmd = OVS_DP_CMD_SET,
1447 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1448 .policy = datapath_policy,
1449 .doit = ovs_dp_cmd_set,
1450 },
1451 };
1452
1453 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1454 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1455 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1456 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1457 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1458 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1459 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1460 };
1461
1462 struct genl_family dp_vport_genl_family = {
1463 .id = GENL_ID_GENERATE,
1464 .hdrsize = sizeof(struct ovs_header),
1465 .name = OVS_VPORT_FAMILY,
1466 .version = OVS_VPORT_VERSION,
1467 .maxattr = OVS_VPORT_ATTR_MAX,
1468 .netnsok = true,
1469 .parallel_ops = true,
1470 };
1471
1472 static struct genl_multicast_group ovs_dp_vport_multicast_group = {
1473 .name = OVS_VPORT_MCGROUP
1474 };
1475
1476 /* Called with ovs_mutex or RCU read lock. */
1477 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1478 u32 portid, u32 seq, u32 flags, u8 cmd)
1479 {
1480 struct ovs_header *ovs_header;
1481 struct ovs_vport_stats vport_stats;
1482 int err;
1483
1484 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1485 flags, cmd);
1486 if (!ovs_header)
1487 return -EMSGSIZE;
1488
1489 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1490
1491 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1492 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1493 nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1494 nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_portid))
1495 goto nla_put_failure;
1496
1497 ovs_vport_get_stats(vport, &vport_stats);
1498 if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1499 &vport_stats))
1500 goto nla_put_failure;
1501
1502 err = ovs_vport_get_options(vport, skb);
1503 if (err == -EMSGSIZE)
1504 goto error;
1505
1506 return genlmsg_end(skb, ovs_header);
1507
1508 nla_put_failure:
1509 err = -EMSGSIZE;
1510 error:
1511 genlmsg_cancel(skb, ovs_header);
1512 return err;
1513 }
1514
1515 /* Called with ovs_mutex or RCU read lock. */
1516 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1517 u32 seq, u8 cmd)
1518 {
1519 struct sk_buff *skb;
1520 int retval;
1521
1522 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1523 if (!skb)
1524 return ERR_PTR(-ENOMEM);
1525
1526 retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1527 BUG_ON(retval < 0);
1528
1529 return skb;
1530 }
1531
1532 /* Called with ovs_mutex or RCU read lock. */
1533 static struct vport *lookup_vport(struct net *net,
1534 struct ovs_header *ovs_header,
1535 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1536 {
1537 struct datapath *dp;
1538 struct vport *vport;
1539
1540 if (a[OVS_VPORT_ATTR_NAME]) {
1541 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1542 if (!vport)
1543 return ERR_PTR(-ENODEV);
1544 if (ovs_header->dp_ifindex &&
1545 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1546 return ERR_PTR(-ENODEV);
1547 return vport;
1548 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1549 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1550
1551 if (port_no >= DP_MAX_PORTS)
1552 return ERR_PTR(-EFBIG);
1553
1554 dp = get_dp(net, ovs_header->dp_ifindex);
1555 if (!dp)
1556 return ERR_PTR(-ENODEV);
1557
1558 vport = ovs_vport_ovsl_rcu(dp, port_no);
1559 if (!vport)
1560 return ERR_PTR(-ENODEV);
1561 return vport;
1562 } else
1563 return ERR_PTR(-EINVAL);
1564 }
1565
1566 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1567 {
1568 struct nlattr **a = info->attrs;
1569 struct ovs_header *ovs_header = info->userhdr;
1570 struct vport_parms parms;
1571 struct sk_buff *reply;
1572 struct vport *vport;
1573 struct datapath *dp;
1574 u32 port_no;
1575 int err;
1576
1577 err = -EINVAL;
1578 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1579 !a[OVS_VPORT_ATTR_UPCALL_PID])
1580 goto exit;
1581
1582 ovs_lock();
1583 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1584 err = -ENODEV;
1585 if (!dp)
1586 goto exit_unlock;
1587
1588 if (a[OVS_VPORT_ATTR_PORT_NO]) {
1589 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1590
1591 err = -EFBIG;
1592 if (port_no >= DP_MAX_PORTS)
1593 goto exit_unlock;
1594
1595 vport = ovs_vport_ovsl(dp, port_no);
1596 err = -EBUSY;
1597 if (vport)
1598 goto exit_unlock;
1599 } else {
1600 for (port_no = 1; ; port_no++) {
1601 if (port_no >= DP_MAX_PORTS) {
1602 err = -EFBIG;
1603 goto exit_unlock;
1604 }
1605 vport = ovs_vport_ovsl(dp, port_no);
1606 if (!vport)
1607 break;
1608 }
1609 }
1610
1611 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1612 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1613 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1614 parms.dp = dp;
1615 parms.port_no = port_no;
1616 parms.upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1617
1618 vport = new_vport(&parms);
1619 err = PTR_ERR(vport);
1620 if (IS_ERR(vport))
1621 goto exit_unlock;
1622
1623 err = 0;
1624 reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
1625 OVS_VPORT_CMD_NEW);
1626 if (IS_ERR(reply)) {
1627 err = PTR_ERR(reply);
1628 ovs_dp_detach_port(vport);
1629 goto exit_unlock;
1630 }
1631
1632 ovs_notify(&dp_vport_genl_family, reply, info);
1633
1634 exit_unlock:
1635 ovs_unlock();
1636 exit:
1637 return err;
1638 }
1639
1640 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1641 {
1642 struct nlattr **a = info->attrs;
1643 struct sk_buff *reply;
1644 struct vport *vport;
1645 int err;
1646
1647 ovs_lock();
1648 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1649 err = PTR_ERR(vport);
1650 if (IS_ERR(vport))
1651 goto exit_unlock;
1652
1653 if (a[OVS_VPORT_ATTR_TYPE] &&
1654 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
1655 err = -EINVAL;
1656 goto exit_unlock;
1657 }
1658
1659 reply = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1660 if (!reply) {
1661 err = -ENOMEM;
1662 goto exit_unlock;
1663 }
1664
1665 if (a[OVS_VPORT_ATTR_OPTIONS]) {
1666 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1667 if (err)
1668 goto exit_free;
1669 }
1670
1671 if (a[OVS_VPORT_ATTR_UPCALL_PID])
1672 vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1673
1674 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1675 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1676 BUG_ON(err < 0);
1677
1678 ovs_unlock();
1679 ovs_notify(&dp_vport_genl_family, reply, info);
1680 return 0;
1681
1682 exit_free:
1683 kfree_skb(reply);
1684 exit_unlock:
1685 ovs_unlock();
1686 return err;
1687 }
1688
1689 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1690 {
1691 struct nlattr **a = info->attrs;
1692 struct sk_buff *reply;
1693 struct vport *vport;
1694 int err;
1695
1696 ovs_lock();
1697 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1698 err = PTR_ERR(vport);
1699 if (IS_ERR(vport))
1700 goto exit_unlock;
1701
1702 if (vport->port_no == OVSP_LOCAL) {
1703 err = -EINVAL;
1704 goto exit_unlock;
1705 }
1706
1707 reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
1708 info->snd_seq, OVS_VPORT_CMD_DEL);
1709 err = PTR_ERR(reply);
1710 if (IS_ERR(reply))
1711 goto exit_unlock;
1712
1713 err = 0;
1714 ovs_dp_detach_port(vport);
1715
1716 ovs_notify(&dp_vport_genl_family, reply, info);
1717
1718 exit_unlock:
1719 ovs_unlock();
1720 return err;
1721 }
1722
1723 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1724 {
1725 struct nlattr **a = info->attrs;
1726 struct ovs_header *ovs_header = info->userhdr;
1727 struct sk_buff *reply;
1728 struct vport *vport;
1729 int err;
1730
1731 rcu_read_lock();
1732 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1733 err = PTR_ERR(vport);
1734 if (IS_ERR(vport))
1735 goto exit_unlock;
1736
1737 reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
1738 info->snd_seq, OVS_VPORT_CMD_NEW);
1739 err = PTR_ERR(reply);
1740 if (IS_ERR(reply))
1741 goto exit_unlock;
1742
1743 rcu_read_unlock();
1744
1745 return genlmsg_reply(reply, info);
1746
1747 exit_unlock:
1748 rcu_read_unlock();
1749 return err;
1750 }
1751
1752 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1753 {
1754 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1755 struct datapath *dp;
1756 int bucket = cb->args[0], skip = cb->args[1];
1757 int i, j = 0;
1758
1759 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1760 if (!dp)
1761 return -ENODEV;
1762
1763 rcu_read_lock();
1764 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
1765 struct vport *vport;
1766
1767 j = 0;
1768 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1769 if (j >= skip &&
1770 ovs_vport_cmd_fill_info(vport, skb,
1771 NETLINK_CB(cb->skb).portid,
1772 cb->nlh->nlmsg_seq,
1773 NLM_F_MULTI,
1774 OVS_VPORT_CMD_NEW) < 0)
1775 goto out;
1776
1777 j++;
1778 }
1779 skip = 0;
1780 }
1781 out:
1782 rcu_read_unlock();
1783
1784 cb->args[0] = i;
1785 cb->args[1] = j;
1786
1787 return skb->len;
1788 }
1789
1790 static const struct genl_ops dp_vport_genl_ops[] = {
1791 { .cmd = OVS_VPORT_CMD_NEW,
1792 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1793 .policy = vport_policy,
1794 .doit = ovs_vport_cmd_new
1795 },
1796 { .cmd = OVS_VPORT_CMD_DEL,
1797 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1798 .policy = vport_policy,
1799 .doit = ovs_vport_cmd_del
1800 },
1801 { .cmd = OVS_VPORT_CMD_GET,
1802 .flags = 0, /* OK for unprivileged users. */
1803 .policy = vport_policy,
1804 .doit = ovs_vport_cmd_get,
1805 .dumpit = ovs_vport_cmd_dump
1806 },
1807 { .cmd = OVS_VPORT_CMD_SET,
1808 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1809 .policy = vport_policy,
1810 .doit = ovs_vport_cmd_set,
1811 },
1812 };
1813
1814 struct genl_family_and_ops {
1815 struct genl_family *family;
1816 const struct genl_ops *ops;
1817 int n_ops;
1818 const struct genl_multicast_group *group;
1819 };
1820
1821 static const struct genl_family_and_ops dp_genl_families[] = {
1822 { &dp_datapath_genl_family,
1823 dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1824 &ovs_dp_datapath_multicast_group },
1825 { &dp_vport_genl_family,
1826 dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1827 &ovs_dp_vport_multicast_group },
1828 { &dp_flow_genl_family,
1829 dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1830 &ovs_dp_flow_multicast_group },
1831 { &dp_packet_genl_family,
1832 dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1833 NULL },
1834 };
1835
1836 static void dp_unregister_genl(int n_families)
1837 {
1838 int i;
1839
1840 for (i = 0; i < n_families; i++)
1841 genl_unregister_family(dp_genl_families[i].family);
1842 }
1843
1844 static int dp_register_genl(void)
1845 {
1846 int n_registered;
1847 int err;
1848 int i;
1849
1850 n_registered = 0;
1851 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
1852 const struct genl_family_and_ops *f = &dp_genl_families[i];
1853
1854 f->family->ops = f->ops;
1855 f->family->n_ops = f->n_ops;
1856 f->family->mcgrps = f->group;
1857 f->family->n_mcgrps = f->group ? 1 : 0;
1858 err = genl_register_family(f->family);
1859 if (err)
1860 goto error;
1861 n_registered++;
1862 }
1863
1864 return 0;
1865
1866 error:
1867 dp_unregister_genl(n_registered);
1868 return err;
1869 }
1870
1871 static int __net_init ovs_init_net(struct net *net)
1872 {
1873 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1874
1875 INIT_LIST_HEAD(&ovs_net->dps);
1876 INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
1877 return 0;
1878 }
1879
1880 static void __net_exit ovs_exit_net(struct net *net)
1881 {
1882 struct datapath *dp, *dp_next;
1883 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1884
1885 ovs_lock();
1886 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
1887 __dp_destroy(dp);
1888 ovs_unlock();
1889
1890 cancel_work_sync(&ovs_net->dp_notify_work);
1891 }
1892
1893 static struct pernet_operations ovs_net_ops = {
1894 .init = ovs_init_net,
1895 .exit = ovs_exit_net,
1896 .id = &ovs_net_id,
1897 .size = sizeof(struct ovs_net),
1898 };
1899
1900 static int __init dp_init(void)
1901 {
1902 int err;
1903
1904 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
1905
1906 pr_info("Open vSwitch switching datapath\n");
1907
1908 err = ovs_flow_init();
1909 if (err)
1910 goto error;
1911
1912 err = ovs_vport_init();
1913 if (err)
1914 goto error_flow_exit;
1915
1916 err = register_pernet_device(&ovs_net_ops);
1917 if (err)
1918 goto error_vport_exit;
1919
1920 err = register_netdevice_notifier(&ovs_dp_device_notifier);
1921 if (err)
1922 goto error_netns_exit;
1923
1924 err = dp_register_genl();
1925 if (err < 0)
1926 goto error_unreg_notifier;
1927
1928 return 0;
1929
1930 error_unreg_notifier:
1931 unregister_netdevice_notifier(&ovs_dp_device_notifier);
1932 error_netns_exit:
1933 unregister_pernet_device(&ovs_net_ops);
1934 error_vport_exit:
1935 ovs_vport_exit();
1936 error_flow_exit:
1937 ovs_flow_exit();
1938 error:
1939 return err;
1940 }
1941
1942 static void dp_cleanup(void)
1943 {
1944 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
1945 unregister_netdevice_notifier(&ovs_dp_device_notifier);
1946 unregister_pernet_device(&ovs_net_ops);
1947 rcu_barrier();
1948 ovs_vport_exit();
1949 ovs_flow_exit();
1950 }
1951
1952 module_init(dp_init);
1953 module_exit(dp_cleanup);
1954
1955 MODULE_DESCRIPTION("Open vSwitch switching datapath");
1956 MODULE_LICENSE("GPL");
Linux kernel - Deliverables
This page took 0.188352 seconds and 6 git commands to generate.