1 /*
2 * Copyright (c) 2007-2014 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/openvswitch.h>
48 #include <linux/rculist.h>
49 #include <linux/dmi.h>
50 #include <net/genetlink.h>
51 #include <net/net_namespace.h>
52 #include <net/netns/generic.h>
53
54 #include "datapath.h"
55 #include "flow.h"
56 #include "flow_table.h"
57 #include "flow_netlink.h"
58 #include "vport-internal_dev.h"
59 #include "vport-netdev.h"
60
61 int ovs_net_id __read_mostly;
62 EXPORT_SYMBOL_GPL(ovs_net_id);
63
64 static struct genl_family dp_packet_genl_family;
65 static struct genl_family dp_flow_genl_family;
66 static struct genl_family dp_datapath_genl_family;
67
68 static const struct nla_policy flow_policy[];
69
70 static const struct genl_multicast_group ovs_dp_flow_multicast_group = {
71 .name = OVS_FLOW_MCGROUP,
72 };
73
74 static const struct genl_multicast_group ovs_dp_datapath_multicast_group = {
75 .name = OVS_DATAPATH_MCGROUP,
76 };
77
78 static const struct genl_multicast_group ovs_dp_vport_multicast_group = {
79 .name = OVS_VPORT_MCGROUP,
80 };
81
82 /* Check if need to build a reply message.
83 * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
ovs_must_notify(struct genl_family * family,struct genl_info * info,unsigned int group)84 static bool ovs_must_notify(struct genl_family *family, struct genl_info *info,
85 unsigned int group)
86 {
87 return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
88 genl_has_listeners(family, genl_info_net(info), group);
89 }
90
ovs_notify(struct genl_family * family,struct sk_buff * skb,struct genl_info * info)91 static void ovs_notify(struct genl_family *family,
92 struct sk_buff *skb, struct genl_info *info)
93 {
94 genl_notify(family, skb, info, 0, GFP_KERNEL);
95 }
96
97 /**
98 * DOC: Locking:
99 *
100 * All writes e.g. Writes to device state (add/remove datapath, port, set
101 * operations on vports, etc.), Writes to other state (flow table
102 * modifications, set miscellaneous datapath parameters, etc.) are protected
103 * by ovs_lock.
104 *
105 * Reads are protected by RCU.
106 *
107 * There are a few special cases (mostly stats) that have their own
108 * synchronization but they nest under all of above and don't interact with
109 * each other.
110 *
111 * The RTNL lock nests inside ovs_mutex.
112 */
113
114 static DEFINE_MUTEX(ovs_mutex);
115
ovs_lock(void)116 void ovs_lock(void)
117 {
118 mutex_lock(&ovs_mutex);
119 }
120
ovs_unlock(void)121 void ovs_unlock(void)
122 {
123 mutex_unlock(&ovs_mutex);
124 }
125
126 #ifdef CONFIG_LOCKDEP
lockdep_ovsl_is_held(void)127 int lockdep_ovsl_is_held(void)
128 {
129 if (debug_locks)
130 return lockdep_is_held(&ovs_mutex);
131 else
132 return 1;
133 }
134 EXPORT_SYMBOL_GPL(lockdep_ovsl_is_held);
135 #endif
136
137 static struct vport *new_vport(const struct vport_parms *);
138 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
139 const struct sw_flow_key *,
140 const struct dp_upcall_info *,
141 uint32_t cutlen);
142 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
143 const struct sw_flow_key *,
144 const struct dp_upcall_info *,
145 uint32_t cutlen);
146
147 /* Must be called with rcu_read_lock. */
get_dp_rcu(struct net * net,int dp_ifindex)148 static struct datapath *get_dp_rcu(struct net *net, int dp_ifindex)
149 {
150 struct net_device *dev = dev_get_by_index_rcu(net, dp_ifindex);
151
152 if (dev) {
153 struct vport *vport = ovs_internal_dev_get_vport(dev);
154 if (vport)
155 return vport->dp;
156 }
157
158 return NULL;
159 }
160
161 /* The caller must hold either ovs_mutex or rcu_read_lock to keep the
162 * returned dp pointer valid.
163 */
get_dp(struct net * net,int dp_ifindex)164 static inline struct datapath *get_dp(struct net *net, int dp_ifindex)
165 {
166 struct datapath *dp;
167
168 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_ovsl_is_held());
169 rcu_read_lock();
170 dp = get_dp_rcu(net, dp_ifindex);
171 rcu_read_unlock();
172
173 return dp;
174 }
175
176 /* Must be called with rcu_read_lock or ovs_mutex. */
ovs_dp_name(const struct datapath * dp)177 const char *ovs_dp_name(const struct datapath *dp)
178 {
179 struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
180 return ovs_vport_name(vport);
181 }
182
get_dpifindex(const struct datapath * dp)183 static int get_dpifindex(const struct datapath *dp)
184 {
185 struct vport *local;
186 int ifindex;
187
188 rcu_read_lock();
189
190 local = ovs_vport_rcu(dp, OVSP_LOCAL);
191 if (local)
192 ifindex = local->dev->ifindex;
193 else
194 ifindex = 0;
195
196 rcu_read_unlock();
197
198 return ifindex;
199 }
200
destroy_dp_rcu(struct rcu_head * rcu)201 static void destroy_dp_rcu(struct rcu_head *rcu)
202 {
203 struct datapath *dp = container_of(rcu, struct datapath, rcu);
204
205 ovs_flow_tbl_destroy(&dp->table);
206 free_percpu(dp->stats_percpu);
207 kfree(dp->ports);
208 kfree(dp);
209 }
210
vport_hash_bucket(const struct datapath * dp,u16 port_no)211 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
212 u16 port_no)
213 {
214 return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
215 }
216
217 /* Called with ovs_mutex or RCU read lock. */
ovs_lookup_vport(const struct datapath * dp,u16 port_no)218 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
219 {
220 struct vport *vport;
221 struct hlist_head *head;
222
223 head = vport_hash_bucket(dp, port_no);
224 hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
225 if (vport->port_no == port_no)
226 return vport;
227 }
228 return NULL;
229 }
230
231 /* Called with ovs_mutex. */
new_vport(const struct vport_parms * parms)232 static struct vport *new_vport(const struct vport_parms *parms)
233 {
234 struct vport *vport;
235
236 vport = ovs_vport_add(parms);
237 if (!IS_ERR(vport)) {
238 struct datapath *dp = parms->dp;
239 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
240
241 hlist_add_head_rcu(&vport->dp_hash_node, head);
242 }
243 return vport;
244 }
245
ovs_dp_detach_port(struct vport * p)246 void ovs_dp_detach_port(struct vport *p)
247 {
248 ASSERT_OVSL();
249
250 /* First drop references to device. */
251 hlist_del_rcu(&p->dp_hash_node);
252
253 /* Then destroy it. */
254 ovs_vport_del(p);
255 }
256
257 /* Must be called with rcu_read_lock. */
ovs_dp_process_packet(struct sk_buff * skb,struct sw_flow_key * key)258 void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
259 {
260 const struct vport *p = OVS_CB(skb)->input_vport;
261 struct datapath *dp = p->dp;
262 struct sw_flow *flow;
263 struct sw_flow_actions *sf_acts;
264 struct dp_stats_percpu *stats;
265 u64 *stats_counter;
266 u32 n_mask_hit;
267
268 stats = this_cpu_ptr(dp->stats_percpu);
269
270 /* Look up flow. */
271 flow = ovs_flow_tbl_lookup_stats(&dp->table, key, &n_mask_hit);
272 if (unlikely(!flow)) {
273 struct dp_upcall_info upcall;
274 int error;
275
276 memset(&upcall, 0, sizeof(upcall));
277 upcall.cmd = OVS_PACKET_CMD_MISS;
278 upcall.portid = ovs_vport_find_upcall_portid(p, skb);
279 upcall.mru = OVS_CB(skb)->mru;
280 error = ovs_dp_upcall(dp, skb, key, &upcall, 0);
281 if (unlikely(error))
282 kfree_skb(skb);
283 else
284 consume_skb(skb);
285 stats_counter = &stats->n_missed;
286 goto out;
287 }
288
289 ovs_flow_stats_update(flow, key->tp.flags, skb);
290 sf_acts = rcu_dereference(flow->sf_acts);
291 ovs_execute_actions(dp, skb, sf_acts, key);
292
293 stats_counter = &stats->n_hit;
294
295 out:
296 /* Update datapath statistics. */
297 u64_stats_update_begin(&stats->syncp);
298 (*stats_counter)++;
299 stats->n_mask_hit += n_mask_hit;
300 u64_stats_update_end(&stats->syncp);
301 }
302
ovs_dp_upcall(struct datapath * dp,struct sk_buff * skb,const struct sw_flow_key * key,const struct dp_upcall_info * upcall_info,uint32_t cutlen)303 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
304 const struct sw_flow_key *key,
305 const struct dp_upcall_info *upcall_info,
306 uint32_t cutlen)
307 {
308 struct dp_stats_percpu *stats;
309 int err;
310
311 if (upcall_info->portid == 0) {
312 err = -ENOTCONN;
313 goto err;
314 }
315
316 if (!skb_is_gso(skb))
317 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
318 else
319 err = queue_gso_packets(dp, skb, key, upcall_info, cutlen);
320 if (err)
321 goto err;
322
323 return 0;
324
325 err:
326 stats = this_cpu_ptr(dp->stats_percpu);
327
328 u64_stats_update_begin(&stats->syncp);
329 stats->n_lost++;
330 u64_stats_update_end(&stats->syncp);
331
332 return err;
333 }
334
queue_gso_packets(struct datapath * dp,struct sk_buff * skb,const struct sw_flow_key * key,const struct dp_upcall_info * upcall_info,uint32_t cutlen)335 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
336 const struct sw_flow_key *key,
337 const struct dp_upcall_info *upcall_info,
338 uint32_t cutlen)
339 {
340 unsigned short gso_type = skb_shinfo(skb)->gso_type;
341 struct sw_flow_key later_key;
342 struct sk_buff *segs, *nskb;
343 int err;
344
345 BUILD_BUG_ON(sizeof(*OVS_CB(skb)) > SKB_SGO_CB_OFFSET);
346 segs = __skb_gso_segment(skb, NETIF_F_SG, false);
347 if (IS_ERR(segs))
348 return PTR_ERR(segs);
349 if (segs == NULL)
350 return -EINVAL;
351
352 if (gso_type & SKB_GSO_UDP) {
353 /* The initial flow key extracted by ovs_flow_key_extract()
354 * in this case is for a first fragment, so we need to
355 * properly mark later fragments.
356 */
357 later_key = *key;
358 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
359 }
360
361 /* Queue all of the segments. */
362 skb = segs;
363 do {
364 if (gso_type & SKB_GSO_UDP && skb != segs)
365 key = &later_key;
366
367 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
368 if (err)
369 break;
370
371 } while ((skb = skb->next));
372
373 /* Free all of the segments. */
374 skb = segs;
375 do {
376 nskb = skb->next;
377 if (err)
378 kfree_skb(skb);
379 else
380 consume_skb(skb);
381 } while ((skb = nskb));
382 return err;
383 }
384
upcall_msg_size(const struct dp_upcall_info * upcall_info,unsigned int hdrlen,int actions_attrlen)385 static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
386 unsigned int hdrlen, int actions_attrlen)
387 {
388 size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
389 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
390 + nla_total_size(ovs_key_attr_size()) /* OVS_PACKET_ATTR_KEY */
391 + nla_total_size(sizeof(unsigned int)); /* OVS_PACKET_ATTR_LEN */
392
393 /* OVS_PACKET_ATTR_USERDATA */
394 if (upcall_info->userdata)
395 size += NLA_ALIGN(upcall_info->userdata->nla_len);
396
397 /* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
398 if (upcall_info->egress_tun_info)
399 size += nla_total_size(ovs_tun_key_attr_size());
400
401 /* OVS_PACKET_ATTR_ACTIONS */
402 if (upcall_info->actions_len)
403 size += nla_total_size(actions_attrlen);
404
405 /* OVS_PACKET_ATTR_MRU */
406 if (upcall_info->mru)
407 size += nla_total_size(sizeof(upcall_info->mru));
408
409 return size;
410 }
411
pad_packet(struct datapath * dp,struct sk_buff * skb)412 static void pad_packet(struct datapath *dp, struct sk_buff *skb)
413 {
414 if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
415 size_t plen = NLA_ALIGN(skb->len) - skb->len;
416
417 if (plen > 0)
418 memset(skb_put(skb, plen), 0, plen);
419 }
420 }
421
queue_userspace_packet(struct datapath * dp,struct sk_buff * skb,const struct sw_flow_key * key,const struct dp_upcall_info * upcall_info,uint32_t cutlen)422 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
423 const struct sw_flow_key *key,
424 const struct dp_upcall_info *upcall_info,
425 uint32_t cutlen)
426 {
427 struct ovs_header *upcall;
428 struct sk_buff *nskb = NULL;
429 struct sk_buff *user_skb = NULL; /* to be queued to userspace */
430 struct nlattr *nla;
431 size_t len;
432 unsigned int hlen;
433 int err, dp_ifindex;
434
435 dp_ifindex = get_dpifindex(dp);
436 if (!dp_ifindex)
437 return -ENODEV;
438
439 if (skb_vlan_tag_present(skb)) {
440 nskb = skb_clone(skb, GFP_ATOMIC);
441 if (!nskb)
442 return -ENOMEM;
443
444 nskb = __vlan_hwaccel_push_inside(nskb);
445 if (!nskb)
446 return -ENOMEM;
447
448 skb = nskb;
449 }
450
451 if (nla_attr_size(skb->len) > USHRT_MAX) {
452 err = -EFBIG;
453 goto out;
454 }
455
456 /* Complete checksum if needed */
457 if (skb->ip_summed == CHECKSUM_PARTIAL &&
458 (err = skb_checksum_help(skb)))
459 goto out;
460
461 /* Older versions of OVS user space enforce alignment of the last
462 * Netlink attribute to NLA_ALIGNTO which would require extensive
463 * padding logic. Only perform zerocopy if padding is not required.
464 */
465 if (dp->user_features & OVS_DP_F_UNALIGNED)
466 hlen = skb_zerocopy_headlen(skb);
467 else
468 hlen = skb->len;
469
470 len = upcall_msg_size(upcall_info, hlen - cutlen,
471 OVS_CB(skb)->acts_origlen);
472 user_skb = genlmsg_new(len, GFP_ATOMIC);
473 if (!user_skb) {
474 err = -ENOMEM;
475 goto out;
476 }
477
478 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
479 0, upcall_info->cmd);
480 upcall->dp_ifindex = dp_ifindex;
481
482 err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
483 BUG_ON(err);
484
485 if (upcall_info->userdata)
486 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
487 nla_len(upcall_info->userdata),
488 nla_data(upcall_info->userdata));
489
490 if (upcall_info->egress_tun_info) {
491 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_EGRESS_TUN_KEY);
492 err = ovs_nla_put_tunnel_info(user_skb,
493 upcall_info->egress_tun_info);
494 BUG_ON(err);
495 nla_nest_end(user_skb, nla);
496 }
497
498 if (upcall_info->actions_len) {
499 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_ACTIONS);
500 err = ovs_nla_put_actions(upcall_info->actions,
501 upcall_info->actions_len,
502 user_skb);
503 if (!err)
504 nla_nest_end(user_skb, nla);
505 else
506 nla_nest_cancel(user_skb, nla);
507 }
508
509 /* Add OVS_PACKET_ATTR_MRU */
510 if (upcall_info->mru) {
511 if (nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU,
512 upcall_info->mru)) {
513 err = -ENOBUFS;
514 goto out;
515 }
516 pad_packet(dp, user_skb);
517 }
518
519 /* Add OVS_PACKET_ATTR_LEN when packet is truncated */
520 if (cutlen > 0) {
521 if (nla_put_u32(user_skb, OVS_PACKET_ATTR_LEN,
522 skb->len)) {
523 err = -ENOBUFS;
524 goto out;
525 }
526 pad_packet(dp, user_skb);
527 }
528
529 /* Only reserve room for attribute header, packet data is added
530 * in skb_zerocopy() */
531 if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
532 err = -ENOBUFS;
533 goto out;
534 }
535 nla->nla_len = nla_attr_size(skb->len - cutlen);
536
537 err = skb_zerocopy(user_skb, skb, skb->len - cutlen, hlen);
538 if (err)
539 goto out;
540
541 /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
542 pad_packet(dp, user_skb);
543
544 ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
545
546 err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
547 user_skb = NULL;
548 out:
549 if (err)
550 skb_tx_error(skb);
551 kfree_skb(user_skb);
552 kfree_skb(nskb);
553 return err;
554 }
555
ovs_packet_cmd_execute(struct sk_buff * skb,struct genl_info * info)556 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
557 {
558 struct ovs_header *ovs_header = info->userhdr;
559 struct net *net = sock_net(skb->sk);
560 struct nlattr **a = info->attrs;
561 struct sw_flow_actions *acts;
562 struct sk_buff *packet;
563 struct sw_flow *flow;
564 struct sw_flow_actions *sf_acts;
565 struct datapath *dp;
566 struct ethhdr *eth;
567 struct vport *input_vport;
568 u16 mru = 0;
569 int len;
570 int err;
571 bool log = !a[OVS_PACKET_ATTR_PROBE];
572
573 err = -EINVAL;
574 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
575 !a[OVS_PACKET_ATTR_ACTIONS])
576 goto err;
577
578 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
579 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
580 err = -ENOMEM;
581 if (!packet)
582 goto err;
583 skb_reserve(packet, NET_IP_ALIGN);
584
585 nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
586
587 skb_reset_mac_header(packet);
588 eth = eth_hdr(packet);
589
590 /* Normally, setting the skb 'protocol' field would be handled by a
591 * call to eth_type_trans(), but it assumes there's a sending
592 * device, which we may not have. */
593 if (eth_proto_is_802_3(eth->h_proto))
594 packet->protocol = eth->h_proto;
595 else
596 packet->protocol = htons(ETH_P_802_2);
597
598 /* Set packet's mru */
599 if (a[OVS_PACKET_ATTR_MRU]) {
600 mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
601 packet->ignore_df = 1;
602 }
603 OVS_CB(packet)->mru = mru;
604
605 /* Build an sw_flow for sending this packet. */
606 flow = ovs_flow_alloc();
607 err = PTR_ERR(flow);
608 if (IS_ERR(flow))
609 goto err_kfree_skb;
610
611 err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
612 packet, &flow->key, log);
613 if (err)
614 goto err_flow_free;
615
616 err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
617 &flow->key, &acts, log);
618 if (err)
619 goto err_flow_free;
620
621 rcu_assign_pointer(flow->sf_acts, acts);
622 packet->priority = flow->key.phy.priority;
623 packet->mark = flow->key.phy.skb_mark;
624
625 rcu_read_lock();
626 dp = get_dp_rcu(net, ovs_header->dp_ifindex);
627 err = -ENODEV;
628 if (!dp)
629 goto err_unlock;
630
631 input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
632 if (!input_vport)
633 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
634
635 if (!input_vport)
636 goto err_unlock;
637
638 packet->dev = input_vport->dev;
639 OVS_CB(packet)->input_vport = input_vport;
640 sf_acts = rcu_dereference(flow->sf_acts);
641
642 local_bh_disable();
643 err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
644 local_bh_enable();
645 rcu_read_unlock();
646
647 ovs_flow_free(flow, false);
648 return err;
649
650 err_unlock:
651 rcu_read_unlock();
652 err_flow_free:
653 ovs_flow_free(flow, false);
654 err_kfree_skb:
655 kfree_skb(packet);
656 err:
657 return err;
658 }
659
660 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
661 [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
662 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
663 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
664 [OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
665 [OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
666 };
667
668 static const struct genl_ops dp_packet_genl_ops[] = {
669 { .cmd = OVS_PACKET_CMD_EXECUTE,
670 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
671 .policy = packet_policy,
672 .doit = ovs_packet_cmd_execute
673 }
674 };
675
676 static struct genl_family dp_packet_genl_family = {
677 .id = GENL_ID_GENERATE,
678 .hdrsize = sizeof(struct ovs_header),
679 .name = OVS_PACKET_FAMILY,
680 .version = OVS_PACKET_VERSION,
681 .maxattr = OVS_PACKET_ATTR_MAX,
682 .netnsok = true,
683 .parallel_ops = true,
684 .ops = dp_packet_genl_ops,
685 .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
686 };
687
get_dp_stats(const struct datapath * dp,struct ovs_dp_stats * stats,struct ovs_dp_megaflow_stats * mega_stats)688 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
689 struct ovs_dp_megaflow_stats *mega_stats)
690 {
691 int i;
692
693 memset(mega_stats, 0, sizeof(*mega_stats));
694
695 stats->n_flows = ovs_flow_tbl_count(&dp->table);
696 mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
697
698 stats->n_hit = stats->n_missed = stats->n_lost = 0;
699
700 for_each_possible_cpu(i) {
701 const struct dp_stats_percpu *percpu_stats;
702 struct dp_stats_percpu local_stats;
703 unsigned int start;
704
705 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
706
707 do {
708 start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
709 local_stats = *percpu_stats;
710 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
711
712 stats->n_hit += local_stats.n_hit;
713 stats->n_missed += local_stats.n_missed;
714 stats->n_lost += local_stats.n_lost;
715 mega_stats->n_mask_hit += local_stats.n_mask_hit;
716 }
717 }
718
should_fill_key(const struct sw_flow_id * sfid,uint32_t ufid_flags)719 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
720 {
721 return ovs_identifier_is_ufid(sfid) &&
722 !(ufid_flags & OVS_UFID_F_OMIT_KEY);
723 }
724
should_fill_mask(uint32_t ufid_flags)725 static bool should_fill_mask(uint32_t ufid_flags)
726 {
727 return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
728 }
729
should_fill_actions(uint32_t ufid_flags)730 static bool should_fill_actions(uint32_t ufid_flags)
731 {
732 return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
733 }
734
ovs_flow_cmd_msg_size(const struct sw_flow_actions * acts,const struct sw_flow_id * sfid,uint32_t ufid_flags)735 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
736 const struct sw_flow_id *sfid,
737 uint32_t ufid_flags)
738 {
739 size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
740
741 /* OVS_FLOW_ATTR_UFID */
742 if (sfid && ovs_identifier_is_ufid(sfid))
743 len += nla_total_size(sfid->ufid_len);
744
745 /* OVS_FLOW_ATTR_KEY */
746 if (!sfid || should_fill_key(sfid, ufid_flags))
747 len += nla_total_size(ovs_key_attr_size());
748
749 /* OVS_FLOW_ATTR_MASK */
750 if (should_fill_mask(ufid_flags))
751 len += nla_total_size(ovs_key_attr_size());
752
753 /* OVS_FLOW_ATTR_ACTIONS */
754 if (should_fill_actions(ufid_flags))
755 len += nla_total_size(acts->orig_len);
756
757 return len
758 + nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
759 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
760 + nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */
761 }
762
763 /* Called with ovs_mutex or RCU read lock. */
ovs_flow_cmd_fill_stats(const struct sw_flow * flow,struct sk_buff * skb)764 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
765 struct sk_buff *skb)
766 {
767 struct ovs_flow_stats stats;
768 __be16 tcp_flags;
769 unsigned long used;
770
771 ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
772
773 if (used &&
774 nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
775 OVS_FLOW_ATTR_PAD))
776 return -EMSGSIZE;
777
778 if (stats.n_packets &&
779 nla_put_64bit(skb, OVS_FLOW_ATTR_STATS,
780 sizeof(struct ovs_flow_stats), &stats,
781 OVS_FLOW_ATTR_PAD))
782 return -EMSGSIZE;
783
784 if ((u8)ntohs(tcp_flags) &&
785 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
786 return -EMSGSIZE;
787
788 return 0;
789 }
790
791 /* Called with ovs_mutex or RCU read lock. */
ovs_flow_cmd_fill_actions(const struct sw_flow * flow,struct sk_buff * skb,int skb_orig_len)792 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
793 struct sk_buff *skb, int skb_orig_len)
794 {
795 struct nlattr *start;
796 int err;
797
798 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
799 * this is the first flow to be dumped into 'skb'. This is unusual for
800 * Netlink but individual action lists can be longer than
801 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
802 * The userspace caller can always fetch the actions separately if it
803 * really wants them. (Most userspace callers in fact don't care.)
804 *
805 * This can only fail for dump operations because the skb is always
806 * properly sized for single flows.
807 */
808 start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
809 if (start) {
810 const struct sw_flow_actions *sf_acts;
811
812 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
813 err = ovs_nla_put_actions(sf_acts->actions,
814 sf_acts->actions_len, skb);
815
816 if (!err)
817 nla_nest_end(skb, start);
818 else {
819 if (skb_orig_len)
820 return err;
821
822 nla_nest_cancel(skb, start);
823 }
824 } else if (skb_orig_len) {
825 return -EMSGSIZE;
826 }
827
828 return 0;
829 }
830
831 /* Called with ovs_mutex or RCU read lock. */
ovs_flow_cmd_fill_info(const struct sw_flow * flow,int dp_ifindex,struct sk_buff * skb,u32 portid,u32 seq,u32 flags,u8 cmd,u32 ufid_flags)832 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
833 struct sk_buff *skb, u32 portid,
834 u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
835 {
836 const int skb_orig_len = skb->len;
837 struct ovs_header *ovs_header;
838 int err;
839
840 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
841 flags, cmd);
842 if (!ovs_header)
843 return -EMSGSIZE;
844
845 ovs_header->dp_ifindex = dp_ifindex;
846
847 err = ovs_nla_put_identifier(flow, skb);
848 if (err)
849 goto error;
850
851 if (should_fill_key(&flow->id, ufid_flags)) {
852 err = ovs_nla_put_masked_key(flow, skb);
853 if (err)
854 goto error;
855 }
856
857 if (should_fill_mask(ufid_flags)) {
858 err = ovs_nla_put_mask(flow, skb);
859 if (err)
860 goto error;
861 }
862
863 err = ovs_flow_cmd_fill_stats(flow, skb);
864 if (err)
865 goto error;
866
867 if (should_fill_actions(ufid_flags)) {
868 err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
869 if (err)
870 goto error;
871 }
872
873 genlmsg_end(skb, ovs_header);
874 return 0;
875
876 error:
877 genlmsg_cancel(skb, ovs_header);
878 return err;
879 }
880
881 /* May not be called with RCU read lock. */
ovs_flow_cmd_alloc_info(const struct sw_flow_actions * acts,const struct sw_flow_id * sfid,struct genl_info * info,bool always,uint32_t ufid_flags)882 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
883 const struct sw_flow_id *sfid,
884 struct genl_info *info,
885 bool always,
886 uint32_t ufid_flags)
887 {
888 struct sk_buff *skb;
889 size_t len;
890
891 if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
892 return NULL;
893
894 len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
895 skb = genlmsg_new(len, GFP_KERNEL);
896 if (!skb)
897 return ERR_PTR(-ENOMEM);
898
899 return skb;
900 }
901
902 /* Called with ovs_mutex. */
ovs_flow_cmd_build_info(const struct sw_flow * flow,int dp_ifindex,struct genl_info * info,u8 cmd,bool always,u32 ufid_flags)903 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
904 int dp_ifindex,
905 struct genl_info *info, u8 cmd,
906 bool always, u32 ufid_flags)
907 {
908 struct sk_buff *skb;
909 int retval;
910
911 skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
912 &flow->id, info, always, ufid_flags);
913 if (IS_ERR_OR_NULL(skb))
914 return skb;
915
916 retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
917 info->snd_portid, info->snd_seq, 0,
918 cmd, ufid_flags);
919 BUG_ON(retval < 0);
920 return skb;
921 }
922
ovs_flow_cmd_new(struct sk_buff * skb,struct genl_info * info)923 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
924 {
925 struct net *net = sock_net(skb->sk);
926 struct nlattr **a = info->attrs;
927 struct ovs_header *ovs_header = info->userhdr;
928 struct sw_flow *flow = NULL, *new_flow;
929 struct sw_flow_mask mask;
930 struct sk_buff *reply;
931 struct datapath *dp;
932 struct sw_flow_actions *acts;
933 struct sw_flow_match match;
934 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
935 int error;
936 bool log = !a[OVS_FLOW_ATTR_PROBE];
937
938 /* Must have key and actions. */
939 error = -EINVAL;
940 if (!a[OVS_FLOW_ATTR_KEY]) {
941 OVS_NLERR(log, "Flow key attr not present in new flow.");
942 goto error;
943 }
944 if (!a[OVS_FLOW_ATTR_ACTIONS]) {
945 OVS_NLERR(log, "Flow actions attr not present in new flow.");
946 goto error;
947 }
948
949 /* Most of the time we need to allocate a new flow, do it before
950 * locking.
951 */
952 new_flow = ovs_flow_alloc();
953 if (IS_ERR(new_flow)) {
954 error = PTR_ERR(new_flow);
955 goto error;
956 }
957
958 /* Extract key. */
959 ovs_match_init(&match, &new_flow->key, false, &mask);
960 error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
961 a[OVS_FLOW_ATTR_MASK], log);
962 if (error)
963 goto err_kfree_flow;
964
965 /* Extract flow identifier. */
966 error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
967 &new_flow->key, log);
968 if (error)
969 goto err_kfree_flow;
970
971 /* unmasked key is needed to match when ufid is not used. */
972 if (ovs_identifier_is_key(&new_flow->id))
973 match.key = new_flow->id.unmasked_key;
974
975 ovs_flow_mask_key(&new_flow->key, &new_flow->key, true, &mask);
976
977 /* Validate actions. */
978 error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
979 &new_flow->key, &acts, log);
980 if (error) {
981 OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
982 goto err_kfree_flow;
983 }
984
985 reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
986 ufid_flags);
987 if (IS_ERR(reply)) {
988 error = PTR_ERR(reply);
989 goto err_kfree_acts;
990 }
991
992 ovs_lock();
993 dp = get_dp(net, ovs_header->dp_ifindex);
994 if (unlikely(!dp)) {
995 error = -ENODEV;
996 goto err_unlock_ovs;
997 }
998
999 /* Check if this is a duplicate flow */
1000 if (ovs_identifier_is_ufid(&new_flow->id))
1001 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
1002 if (!flow)
1003 flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->key);
1004 if (likely(!flow)) {
1005 rcu_assign_pointer(new_flow->sf_acts, acts);
1006
1007 /* Put flow in bucket. */
1008 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
1009 if (unlikely(error)) {
1010 acts = NULL;
1011 goto err_unlock_ovs;
1012 }
1013
1014 if (unlikely(reply)) {
1015 error = ovs_flow_cmd_fill_info(new_flow,
1016 ovs_header->dp_ifindex,
1017 reply, info->snd_portid,
1018 info->snd_seq, 0,
1019 OVS_FLOW_CMD_NEW,
1020 ufid_flags);
1021 BUG_ON(error < 0);
1022 }
1023 ovs_unlock();
1024 } else {
1025 struct sw_flow_actions *old_acts;
1026
1027 /* Bail out if we're not allowed to modify an existing flow.
1028 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1029 * because Generic Netlink treats the latter as a dump
1030 * request. We also accept NLM_F_EXCL in case that bug ever
1031 * gets fixed.
1032 */
1033 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1034 | NLM_F_EXCL))) {
1035 error = -EEXIST;
1036 goto err_unlock_ovs;
1037 }
1038 /* The flow identifier has to be the same for flow updates.
1039 * Look for any overlapping flow.
1040 */
1041 if (unlikely(!ovs_flow_cmp(flow, &match))) {
1042 if (ovs_identifier_is_key(&flow->id))
1043 flow = ovs_flow_tbl_lookup_exact(&dp->table,
1044 &match);
1045 else /* UFID matches but key is different */
1046 flow = NULL;
1047 if (!flow) {
1048 error = -ENOENT;
1049 goto err_unlock_ovs;
1050 }
1051 }
1052 /* Update actions. */
1053 old_acts = ovsl_dereference(flow->sf_acts);
1054 rcu_assign_pointer(flow->sf_acts, acts);
1055
1056 if (unlikely(reply)) {
1057 error = ovs_flow_cmd_fill_info(flow,
1058 ovs_header->dp_ifindex,
1059 reply, info->snd_portid,
1060 info->snd_seq, 0,
1061 OVS_FLOW_CMD_NEW,
1062 ufid_flags);
1063 BUG_ON(error < 0);
1064 }
1065 ovs_unlock();
1066
1067 ovs_nla_free_flow_actions_rcu(old_acts);
1068 ovs_flow_free(new_flow, false);
1069 }
1070
1071 if (reply)
1072 ovs_notify(&dp_flow_genl_family, reply, info);
1073 return 0;
1074
1075 err_unlock_ovs:
1076 ovs_unlock();
1077 kfree_skb(reply);
1078 err_kfree_acts:
1079 ovs_nla_free_flow_actions(acts);
1080 err_kfree_flow:
1081 ovs_flow_free(new_flow, false);
1082 error:
1083 return error;
1084 }
1085
1086 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
get_flow_actions(struct net * net,const struct nlattr * a,const struct sw_flow_key * key,const struct sw_flow_mask * mask,bool log)1087 static struct sw_flow_actions *get_flow_actions(struct net *net,
1088 const struct nlattr *a,
1089 const struct sw_flow_key *key,
1090 const struct sw_flow_mask *mask,
1091 bool log)
1092 {
1093 struct sw_flow_actions *acts;
1094 struct sw_flow_key masked_key;
1095 int error;
1096
1097 ovs_flow_mask_key(&masked_key, key, true, mask);
1098 error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1099 if (error) {
1100 OVS_NLERR(log,
1101 "Actions may not be safe on all matching packets");
1102 return ERR_PTR(error);
1103 }
1104
1105 return acts;
1106 }
1107
ovs_flow_cmd_set(struct sk_buff * skb,struct genl_info * info)1108 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1109 {
1110 struct net *net = sock_net(skb->sk);
1111 struct nlattr **a = info->attrs;
1112 struct ovs_header *ovs_header = info->userhdr;
1113 struct sw_flow_key key;
1114 struct sw_flow *flow;
1115 struct sw_flow_mask mask;
1116 struct sk_buff *reply = NULL;
1117 struct datapath *dp;
1118 struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1119 struct sw_flow_match match;
1120 struct sw_flow_id sfid;
1121 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1122 int error = 0;
1123 bool log = !a[OVS_FLOW_ATTR_PROBE];
1124 bool ufid_present;
1125
1126 ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1127 if (a[OVS_FLOW_ATTR_KEY]) {
1128 ovs_match_init(&match, &key, true, &mask);
1129 error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1130 a[OVS_FLOW_ATTR_MASK], log);
1131 } else if (!ufid_present) {
1132 OVS_NLERR(log,
1133 "Flow set message rejected, Key attribute missing.");
1134 error = -EINVAL;
1135 }
1136 if (error)
1137 goto error;
1138
1139 /* Validate actions. */
1140 if (a[OVS_FLOW_ATTR_ACTIONS]) {
1141 if (!a[OVS_FLOW_ATTR_KEY]) {
1142 OVS_NLERR(log,
1143 "Flow key attribute not present in set flow.");
1144 error = -EINVAL;
1145 goto error;
1146 }
1147
1148 acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], &key,
1149 &mask, log);
1150 if (IS_ERR(acts)) {
1151 error = PTR_ERR(acts);
1152 goto error;
1153 }
1154
1155 /* Can allocate before locking if have acts. */
1156 reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1157 ufid_flags);
1158 if (IS_ERR(reply)) {
1159 error = PTR_ERR(reply);
1160 goto err_kfree_acts;
1161 }
1162 }
1163
1164 ovs_lock();
1165 dp = get_dp(net, ovs_header->dp_ifindex);
1166 if (unlikely(!dp)) {
1167 error = -ENODEV;
1168 goto err_unlock_ovs;
1169 }
1170 /* Check that the flow exists. */
1171 if (ufid_present)
1172 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1173 else
1174 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1175 if (unlikely(!flow)) {
1176 error = -ENOENT;
1177 goto err_unlock_ovs;
1178 }
1179
1180 /* Update actions, if present. */
1181 if (likely(acts)) {
1182 old_acts = ovsl_dereference(flow->sf_acts);
1183 rcu_assign_pointer(flow->sf_acts, acts);
1184
1185 if (unlikely(reply)) {
1186 error = ovs_flow_cmd_fill_info(flow,
1187 ovs_header->dp_ifindex,
1188 reply, info->snd_portid,
1189 info->snd_seq, 0,
1190 OVS_FLOW_CMD_NEW,
1191 ufid_flags);
1192 BUG_ON(error < 0);
1193 }
1194 } else {
1195 /* Could not alloc without acts before locking. */
1196 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1197 info, OVS_FLOW_CMD_NEW, false,
1198 ufid_flags);
1199
1200 if (IS_ERR(reply)) {
1201 error = PTR_ERR(reply);
1202 goto err_unlock_ovs;
1203 }
1204 }
1205
1206 /* Clear stats. */
1207 if (a[OVS_FLOW_ATTR_CLEAR])
1208 ovs_flow_stats_clear(flow);
1209 ovs_unlock();
1210
1211 if (reply)
1212 ovs_notify(&dp_flow_genl_family, reply, info);
1213 if (old_acts)
1214 ovs_nla_free_flow_actions_rcu(old_acts);
1215
1216 return 0;
1217
1218 err_unlock_ovs:
1219 ovs_unlock();
1220 kfree_skb(reply);
1221 err_kfree_acts:
1222 ovs_nla_free_flow_actions(acts);
1223 error:
1224 return error;
1225 }
1226
ovs_flow_cmd_get(struct sk_buff * skb,struct genl_info * info)1227 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1228 {
1229 struct nlattr **a = info->attrs;
1230 struct ovs_header *ovs_header = info->userhdr;
1231 struct net *net = sock_net(skb->sk);
1232 struct sw_flow_key key;
1233 struct sk_buff *reply;
1234 struct sw_flow *flow;
1235 struct datapath *dp;
1236 struct sw_flow_match match;
1237 struct sw_flow_id ufid;
1238 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1239 int err = 0;
1240 bool log = !a[OVS_FLOW_ATTR_PROBE];
1241 bool ufid_present;
1242
1243 ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1244 if (a[OVS_FLOW_ATTR_KEY]) {
1245 ovs_match_init(&match, &key, true, NULL);
1246 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1247 log);
1248 } else if (!ufid_present) {
1249 OVS_NLERR(log,
1250 "Flow get message rejected, Key attribute missing.");
1251 err = -EINVAL;
1252 }
1253 if (err)
1254 return err;
1255
1256 ovs_lock();
1257 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1258 if (!dp) {
1259 err = -ENODEV;
1260 goto unlock;
1261 }
1262
1263 if (ufid_present)
1264 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1265 else
1266 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1267 if (!flow) {
1268 err = -ENOENT;
1269 goto unlock;
1270 }
1271
1272 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1273 OVS_FLOW_CMD_NEW, true, ufid_flags);
1274 if (IS_ERR(reply)) {
1275 err = PTR_ERR(reply);
1276 goto unlock;
1277 }
1278
1279 ovs_unlock();
1280 return genlmsg_reply(reply, info);
1281 unlock:
1282 ovs_unlock();
1283 return err;
1284 }
1285
ovs_flow_cmd_del(struct sk_buff * skb,struct genl_info * info)1286 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1287 {
1288 struct nlattr **a = info->attrs;
1289 struct ovs_header *ovs_header = info->userhdr;
1290 struct net *net = sock_net(skb->sk);
1291 struct sw_flow_key key;
1292 struct sk_buff *reply;
1293 struct sw_flow *flow = NULL;
1294 struct datapath *dp;
1295 struct sw_flow_match match;
1296 struct sw_flow_id ufid;
1297 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1298 int err;
1299 bool log = !a[OVS_FLOW_ATTR_PROBE];
1300 bool ufid_present;
1301
1302 ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1303 if (a[OVS_FLOW_ATTR_KEY]) {
1304 ovs_match_init(&match, &key, true, NULL);
1305 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1306 NULL, log);
1307 if (unlikely(err))
1308 return err;
1309 }
1310
1311 ovs_lock();
1312 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1313 if (unlikely(!dp)) {
1314 err = -ENODEV;
1315 goto unlock;
1316 }
1317
1318 if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1319 err = ovs_flow_tbl_flush(&dp->table);
1320 goto unlock;
1321 }
1322
1323 if (ufid_present)
1324 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1325 else
1326 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1327 if (unlikely(!flow)) {
1328 err = -ENOENT;
1329 goto unlock;
1330 }
1331
1332 ovs_flow_tbl_remove(&dp->table, flow);
1333 ovs_unlock();
1334
1335 reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1336 &flow->id, info, false, ufid_flags);
1337 if (likely(reply)) {
1338 if (likely(!IS_ERR(reply))) {
1339 rcu_read_lock(); /*To keep RCU checker happy. */
1340 err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1341 reply, info->snd_portid,
1342 info->snd_seq, 0,
1343 OVS_FLOW_CMD_DEL,
1344 ufid_flags);
1345 rcu_read_unlock();
1346 BUG_ON(err < 0);
1347
1348 ovs_notify(&dp_flow_genl_family, reply, info);
1349 } else {
1350 netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
1351 }
1352 }
1353
1354 ovs_flow_free(flow, true);
1355 return 0;
1356 unlock:
1357 ovs_unlock();
1358 return err;
1359 }
1360
ovs_flow_cmd_dump(struct sk_buff * skb,struct netlink_callback * cb)1361 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1362 {
1363 struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1364 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1365 struct table_instance *ti;
1366 struct datapath *dp;
1367 u32 ufid_flags;
1368 int err;
1369
1370 err = genlmsg_parse(cb->nlh, &dp_flow_genl_family, a,
1371 OVS_FLOW_ATTR_MAX, flow_policy);
1372 if (err)
1373 return err;
1374 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1375
1376 rcu_read_lock();
1377 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1378 if (!dp) {
1379 rcu_read_unlock();
1380 return -ENODEV;
1381 }
1382
1383 ti = rcu_dereference(dp->table.ti);
1384 for (;;) {
1385 struct sw_flow *flow;
1386 u32 bucket, obj;
1387
1388 bucket = cb->args[0];
1389 obj = cb->args[1];
1390 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1391 if (!flow)
1392 break;
1393
1394 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1395 NETLINK_CB(cb->skb).portid,
1396 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1397 OVS_FLOW_CMD_NEW, ufid_flags) < 0)
1398 break;
1399
1400 cb->args[0] = bucket;
1401 cb->args[1] = obj;
1402 }
1403 rcu_read_unlock();
1404 return skb->len;
1405 }
1406
1407 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1408 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1409 [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1410 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1411 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1412 [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1413 [OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1414 [OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1415 };
1416
1417 static const struct genl_ops dp_flow_genl_ops[] = {
1418 { .cmd = OVS_FLOW_CMD_NEW,
1419 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1420 .policy = flow_policy,
1421 .doit = ovs_flow_cmd_new
1422 },
1423 { .cmd = OVS_FLOW_CMD_DEL,
1424 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1425 .policy = flow_policy,
1426 .doit = ovs_flow_cmd_del
1427 },
1428 { .cmd = OVS_FLOW_CMD_GET,
1429 .flags = 0, /* OK for unprivileged users. */
1430 .policy = flow_policy,
1431 .doit = ovs_flow_cmd_get,
1432 .dumpit = ovs_flow_cmd_dump
1433 },
1434 { .cmd = OVS_FLOW_CMD_SET,
1435 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1436 .policy = flow_policy,
1437 .doit = ovs_flow_cmd_set,
1438 },
1439 };
1440
1441 static struct genl_family dp_flow_genl_family = {
1442 .id = GENL_ID_GENERATE,
1443 .hdrsize = sizeof(struct ovs_header),
1444 .name = OVS_FLOW_FAMILY,
1445 .version = OVS_FLOW_VERSION,
1446 .maxattr = OVS_FLOW_ATTR_MAX,
1447 .netnsok = true,
1448 .parallel_ops = true,
1449 .ops = dp_flow_genl_ops,
1450 .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1451 .mcgrps = &ovs_dp_flow_multicast_group,
1452 .n_mcgrps = 1,
1453 };
1454
ovs_dp_cmd_msg_size(void)1455 static size_t ovs_dp_cmd_msg_size(void)
1456 {
1457 size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1458
1459 msgsize += nla_total_size(IFNAMSIZ);
1460 msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1461 msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
1462 msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1463
1464 return msgsize;
1465 }
1466
1467 /* Called with ovs_mutex. */
ovs_dp_cmd_fill_info(struct datapath * dp,struct sk_buff * skb,u32 portid,u32 seq,u32 flags,u8 cmd)1468 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1469 u32 portid, u32 seq, u32 flags, u8 cmd)
1470 {
1471 struct ovs_header *ovs_header;
1472 struct ovs_dp_stats dp_stats;
1473 struct ovs_dp_megaflow_stats dp_megaflow_stats;
1474 int err;
1475
1476 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1477 flags, cmd);
1478 if (!ovs_header)
1479 goto error;
1480
1481 ovs_header->dp_ifindex = get_dpifindex(dp);
1482
1483 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1484 if (err)
1485 goto nla_put_failure;
1486
1487 get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1488 if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1489 &dp_stats, OVS_DP_ATTR_PAD))
1490 goto nla_put_failure;
1491
1492 if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1493 sizeof(struct ovs_dp_megaflow_stats),
1494 &dp_megaflow_stats, OVS_DP_ATTR_PAD))
1495 goto nla_put_failure;
1496
1497 if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1498 goto nla_put_failure;
1499
1500 genlmsg_end(skb, ovs_header);
1501 return 0;
1502
1503 nla_put_failure:
1504 genlmsg_cancel(skb, ovs_header);
1505 error:
1506 return -EMSGSIZE;
1507 }
1508
ovs_dp_cmd_alloc_info(void)1509 static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1510 {
1511 return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1512 }
1513
1514 /* Called with rcu_read_lock or ovs_mutex. */
lookup_datapath(struct net * net,const struct ovs_header * ovs_header,struct nlattr * a[OVS_DP_ATTR_MAX+1])1515 static struct datapath *lookup_datapath(struct net *net,
1516 const struct ovs_header *ovs_header,
1517 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1518 {
1519 struct datapath *dp;
1520
1521 if (!a[OVS_DP_ATTR_NAME])
1522 dp = get_dp(net, ovs_header->dp_ifindex);
1523 else {
1524 struct vport *vport;
1525
1526 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1527 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1528 }
1529 return dp ? dp : ERR_PTR(-ENODEV);
1530 }
1531
ovs_dp_reset_user_features(struct sk_buff * skb,struct genl_info * info)1532 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1533 {
1534 struct datapath *dp;
1535
1536 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1537 if (IS_ERR(dp))
1538 return;
1539
1540 WARN(dp->user_features, "Dropping previously announced user features\n");
1541 dp->user_features = 0;
1542 }
1543
ovs_dp_change(struct datapath * dp,struct nlattr * a[])1544 static void ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1545 {
1546 if (a[OVS_DP_ATTR_USER_FEATURES])
1547 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1548 }
1549
ovs_dp_cmd_new(struct sk_buff * skb,struct genl_info * info)1550 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1551 {
1552 struct nlattr **a = info->attrs;
1553 struct vport_parms parms;
1554 struct sk_buff *reply;
1555 struct datapath *dp;
1556 struct vport *vport;
1557 struct ovs_net *ovs_net;
1558 int err, i;
1559
1560 err = -EINVAL;
1561 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1562 goto err;
1563
1564 reply = ovs_dp_cmd_alloc_info();
1565 if (!reply)
1566 return -ENOMEM;
1567
1568 err = -ENOMEM;
1569 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1570 if (dp == NULL)
1571 goto err_free_reply;
1572
1573 ovs_dp_set_net(dp, sock_net(skb->sk));
1574
1575 /* Allocate table. */
1576 err = ovs_flow_tbl_init(&dp->table);
1577 if (err)
1578 goto err_free_dp;
1579
1580 dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1581 if (!dp->stats_percpu) {
1582 err = -ENOMEM;
1583 goto err_destroy_table;
1584 }
1585
1586 dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1587 GFP_KERNEL);
1588 if (!dp->ports) {
1589 err = -ENOMEM;
1590 goto err_destroy_percpu;
1591 }
1592
1593 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1594 INIT_HLIST_HEAD(&dp->ports[i]);
1595
1596 /* Set up our datapath device. */
1597 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1598 parms.type = OVS_VPORT_TYPE_INTERNAL;
1599 parms.options = NULL;
1600 parms.dp = dp;
1601 parms.port_no = OVSP_LOCAL;
1602 parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1603
1604 ovs_dp_change(dp, a);
1605
1606 /* So far only local changes have been made, now need the lock. */
1607 ovs_lock();
1608
1609 vport = new_vport(&parms);
1610 if (IS_ERR(vport)) {
1611 err = PTR_ERR(vport);
1612 if (err == -EBUSY)
1613 err = -EEXIST;
1614
1615 if (err == -EEXIST) {
1616 /* An outdated user space instance that does not understand
1617 * the concept of user_features has attempted to create a new
1618 * datapath and is likely to reuse it. Drop all user features.
1619 */
1620 if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1621 ovs_dp_reset_user_features(skb, info);
1622 }
1623
1624 goto err_destroy_ports_array;
1625 }
1626
1627 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1628 info->snd_seq, 0, OVS_DP_CMD_NEW);
1629 BUG_ON(err < 0);
1630
1631 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1632 list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1633
1634 ovs_unlock();
1635
1636 ovs_notify(&dp_datapath_genl_family, reply, info);
1637 return 0;
1638
1639 err_destroy_ports_array:
1640 ovs_unlock();
1641 kfree(dp->ports);
1642 err_destroy_percpu:
1643 free_percpu(dp->stats_percpu);
1644 err_destroy_table:
1645 ovs_flow_tbl_destroy(&dp->table);
1646 err_free_dp:
1647 kfree(dp);
1648 err_free_reply:
1649 kfree_skb(reply);
1650 err:
1651 return err;
1652 }
1653
1654 /* Called with ovs_mutex. */
__dp_destroy(struct datapath * dp)1655 static void __dp_destroy(struct datapath *dp)
1656 {
1657 int i;
1658
1659 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1660 struct vport *vport;
1661 struct hlist_node *n;
1662
1663 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1664 if (vport->port_no != OVSP_LOCAL)
1665 ovs_dp_detach_port(vport);
1666 }
1667
1668 list_del_rcu(&dp->list_node);
1669
1670 /* OVSP_LOCAL is datapath internal port. We need to make sure that
1671 * all ports in datapath are destroyed first before freeing datapath.
1672 */
1673 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1674
1675 /* RCU destroy the flow table */
1676 call_rcu(&dp->rcu, destroy_dp_rcu);
1677 }
1678
ovs_dp_cmd_del(struct sk_buff * skb,struct genl_info * info)1679 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1680 {
1681 struct sk_buff *reply;
1682 struct datapath *dp;
1683 int err;
1684
1685 reply = ovs_dp_cmd_alloc_info();
1686 if (!reply)
1687 return -ENOMEM;
1688
1689 ovs_lock();
1690 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1691 err = PTR_ERR(dp);
1692 if (IS_ERR(dp))
1693 goto err_unlock_free;
1694
1695 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1696 info->snd_seq, 0, OVS_DP_CMD_DEL);
1697 BUG_ON(err < 0);
1698
1699 __dp_destroy(dp);
1700 ovs_unlock();
1701
1702 ovs_notify(&dp_datapath_genl_family, reply, info);
1703
1704 return 0;
1705
1706 err_unlock_free:
1707 ovs_unlock();
1708 kfree_skb(reply);
1709 return err;
1710 }
1711
ovs_dp_cmd_set(struct sk_buff * skb,struct genl_info * info)1712 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1713 {
1714 struct sk_buff *reply;
1715 struct datapath *dp;
1716 int err;
1717
1718 reply = ovs_dp_cmd_alloc_info();
1719 if (!reply)
1720 return -ENOMEM;
1721
1722 ovs_lock();
1723 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1724 err = PTR_ERR(dp);
1725 if (IS_ERR(dp))
1726 goto err_unlock_free;
1727
1728 ovs_dp_change(dp, info->attrs);
1729
1730 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1731 info->snd_seq, 0, OVS_DP_CMD_NEW);
1732 BUG_ON(err < 0);
1733
1734 ovs_unlock();
1735 ovs_notify(&dp_datapath_genl_family, reply, info);
1736
1737 return 0;
1738
1739 err_unlock_free:
1740 ovs_unlock();
1741 kfree_skb(reply);
1742 return err;
1743 }
1744
ovs_dp_cmd_get(struct sk_buff * skb,struct genl_info * info)1745 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1746 {
1747 struct sk_buff *reply;
1748 struct datapath *dp;
1749 int err;
1750
1751 reply = ovs_dp_cmd_alloc_info();
1752 if (!reply)
1753 return -ENOMEM;
1754
1755 ovs_lock();
1756 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1757 if (IS_ERR(dp)) {
1758 err = PTR_ERR(dp);
1759 goto err_unlock_free;
1760 }
1761 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1762 info->snd_seq, 0, OVS_DP_CMD_NEW);
1763 BUG_ON(err < 0);
1764 ovs_unlock();
1765
1766 return genlmsg_reply(reply, info);
1767
1768 err_unlock_free:
1769 ovs_unlock();
1770 kfree_skb(reply);
1771 return err;
1772 }
1773
ovs_dp_cmd_dump(struct sk_buff * skb,struct netlink_callback * cb)1774 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1775 {
1776 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1777 struct datapath *dp;
1778 int skip = cb->args[0];
1779 int i = 0;
1780
1781 ovs_lock();
1782 list_for_each_entry(dp, &ovs_net->dps, list_node) {
1783 if (i >= skip &&
1784 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1785 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1786 OVS_DP_CMD_NEW) < 0)
1787 break;
1788 i++;
1789 }
1790 ovs_unlock();
1791
1792 cb->args[0] = i;
1793
1794 return skb->len;
1795 }
1796
1797 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1798 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1799 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1800 [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1801 };
1802
1803 static const struct genl_ops dp_datapath_genl_ops[] = {
1804 { .cmd = OVS_DP_CMD_NEW,
1805 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1806 .policy = datapath_policy,
1807 .doit = ovs_dp_cmd_new
1808 },
1809 { .cmd = OVS_DP_CMD_DEL,
1810 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1811 .policy = datapath_policy,
1812 .doit = ovs_dp_cmd_del
1813 },
1814 { .cmd = OVS_DP_CMD_GET,
1815 .flags = 0, /* OK for unprivileged users. */
1816 .policy = datapath_policy,
1817 .doit = ovs_dp_cmd_get,
1818 .dumpit = ovs_dp_cmd_dump
1819 },
1820 { .cmd = OVS_DP_CMD_SET,
1821 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1822 .policy = datapath_policy,
1823 .doit = ovs_dp_cmd_set,
1824 },
1825 };
1826
1827 static struct genl_family dp_datapath_genl_family = {
1828 .id = GENL_ID_GENERATE,
1829 .hdrsize = sizeof(struct ovs_header),
1830 .name = OVS_DATAPATH_FAMILY,
1831 .version = OVS_DATAPATH_VERSION,
1832 .maxattr = OVS_DP_ATTR_MAX,
1833 .netnsok = true,
1834 .parallel_ops = true,
1835 .ops = dp_datapath_genl_ops,
1836 .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1837 .mcgrps = &ovs_dp_datapath_multicast_group,
1838 .n_mcgrps = 1,
1839 };
1840
1841 /* Called with ovs_mutex or RCU read lock. */
ovs_vport_cmd_fill_info(struct vport * vport,struct sk_buff * skb,u32 portid,u32 seq,u32 flags,u8 cmd)1842 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1843 u32 portid, u32 seq, u32 flags, u8 cmd)
1844 {
1845 struct ovs_header *ovs_header;
1846 struct ovs_vport_stats vport_stats;
1847 int err;
1848
1849 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1850 flags, cmd);
1851 if (!ovs_header)
1852 return -EMSGSIZE;
1853
1854 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1855
1856 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1857 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1858 nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1859 ovs_vport_name(vport)))
1860 goto nla_put_failure;
1861
1862 ovs_vport_get_stats(vport, &vport_stats);
1863 if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
1864 sizeof(struct ovs_vport_stats), &vport_stats,
1865 OVS_VPORT_ATTR_PAD))
1866 goto nla_put_failure;
1867
1868 if (ovs_vport_get_upcall_portids(vport, skb))
1869 goto nla_put_failure;
1870
1871 err = ovs_vport_get_options(vport, skb);
1872 if (err == -EMSGSIZE)
1873 goto error;
1874
1875 genlmsg_end(skb, ovs_header);
1876 return 0;
1877
1878 nla_put_failure:
1879 err = -EMSGSIZE;
1880 error:
1881 genlmsg_cancel(skb, ovs_header);
1882 return err;
1883 }
1884
ovs_vport_cmd_alloc_info(void)1885 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1886 {
1887 return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1888 }
1889
1890 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
ovs_vport_cmd_build_info(struct vport * vport,u32 portid,u32 seq,u8 cmd)1891 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1892 u32 seq, u8 cmd)
1893 {
1894 struct sk_buff *skb;
1895 int retval;
1896
1897 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1898 if (!skb)
1899 return ERR_PTR(-ENOMEM);
1900
1901 retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1902 BUG_ON(retval < 0);
1903
1904 return skb;
1905 }
1906
1907 /* Called with ovs_mutex or RCU read lock. */
lookup_vport(struct net * net,const struct ovs_header * ovs_header,struct nlattr * a[OVS_VPORT_ATTR_MAX+1])1908 static struct vport *lookup_vport(struct net *net,
1909 const struct ovs_header *ovs_header,
1910 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1911 {
1912 struct datapath *dp;
1913 struct vport *vport;
1914
1915 if (a[OVS_VPORT_ATTR_NAME]) {
1916 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1917 if (!vport)
1918 return ERR_PTR(-ENODEV);
1919 if (ovs_header->dp_ifindex &&
1920 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1921 return ERR_PTR(-ENODEV);
1922 return vport;
1923 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1924 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1925
1926 if (port_no >= DP_MAX_PORTS)
1927 return ERR_PTR(-EFBIG);
1928
1929 dp = get_dp(net, ovs_header->dp_ifindex);
1930 if (!dp)
1931 return ERR_PTR(-ENODEV);
1932
1933 vport = ovs_vport_ovsl_rcu(dp, port_no);
1934 if (!vport)
1935 return ERR_PTR(-ENODEV);
1936 return vport;
1937 } else
1938 return ERR_PTR(-EINVAL);
1939 }
1940
1941 /* Called with ovs_mutex */
update_headroom(struct datapath * dp)1942 static void update_headroom(struct datapath *dp)
1943 {
1944 unsigned dev_headroom, max_headroom = 0;
1945 struct net_device *dev;
1946 struct vport *vport;
1947 int i;
1948
1949 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1950 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1951 dev = vport->dev;
1952 dev_headroom = netdev_get_fwd_headroom(dev);
1953 if (dev_headroom > max_headroom)
1954 max_headroom = dev_headroom;
1955 }
1956 }
1957
1958 dp->max_headroom = max_headroom;
1959 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1960 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node)
1961 netdev_set_rx_headroom(vport->dev, max_headroom);
1962 }
1963
ovs_vport_cmd_new(struct sk_buff * skb,struct genl_info * info)1964 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1965 {
1966 struct nlattr **a = info->attrs;
1967 struct ovs_header *ovs_header = info->userhdr;
1968 struct vport_parms parms;
1969 struct sk_buff *reply;
1970 struct vport *vport;
1971 struct datapath *dp;
1972 u32 port_no;
1973 int err;
1974
1975 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1976 !a[OVS_VPORT_ATTR_UPCALL_PID])
1977 return -EINVAL;
1978
1979 port_no = a[OVS_VPORT_ATTR_PORT_NO]
1980 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
1981 if (port_no >= DP_MAX_PORTS)
1982 return -EFBIG;
1983
1984 reply = ovs_vport_cmd_alloc_info();
1985 if (!reply)
1986 return -ENOMEM;
1987
1988 ovs_lock();
1989 restart:
1990 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1991 err = -ENODEV;
1992 if (!dp)
1993 goto exit_unlock_free;
1994
1995 if (port_no) {
1996 vport = ovs_vport_ovsl(dp, port_no);
1997 err = -EBUSY;
1998 if (vport)
1999 goto exit_unlock_free;
2000 } else {
2001 for (port_no = 1; ; port_no++) {
2002 if (port_no >= DP_MAX_PORTS) {
2003 err = -EFBIG;
2004 goto exit_unlock_free;
2005 }
2006 vport = ovs_vport_ovsl(dp, port_no);
2007 if (!vport)
2008 break;
2009 }
2010 }
2011
2012 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2013 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2014 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2015 parms.dp = dp;
2016 parms.port_no = port_no;
2017 parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
2018
2019 vport = new_vport(&parms);
2020 err = PTR_ERR(vport);
2021 if (IS_ERR(vport)) {
2022 if (err == -EAGAIN)
2023 goto restart;
2024 goto exit_unlock_free;
2025 }
2026
2027 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2028 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2029
2030 if (netdev_get_fwd_headroom(vport->dev) > dp->max_headroom)
2031 update_headroom(dp);
2032 else
2033 netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2034
2035 BUG_ON(err < 0);
2036 ovs_unlock();
2037
2038 ovs_notify(&dp_vport_genl_family, reply, info);
2039 return 0;
2040
2041 exit_unlock_free:
2042 ovs_unlock();
2043 kfree_skb(reply);
2044 return err;
2045 }
2046
ovs_vport_cmd_set(struct sk_buff * skb,struct genl_info * info)2047 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2048 {
2049 struct nlattr **a = info->attrs;
2050 struct sk_buff *reply;
2051 struct vport *vport;
2052 int err;
2053
2054 reply = ovs_vport_cmd_alloc_info();
2055 if (!reply)
2056 return -ENOMEM;
2057
2058 ovs_lock();
2059 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2060 err = PTR_ERR(vport);
2061 if (IS_ERR(vport))
2062 goto exit_unlock_free;
2063
2064 if (a[OVS_VPORT_ATTR_TYPE] &&
2065 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2066 err = -EINVAL;
2067 goto exit_unlock_free;
2068 }
2069
2070 if (a[OVS_VPORT_ATTR_OPTIONS]) {
2071 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2072 if (err)
2073 goto exit_unlock_free;
2074 }
2075
2076
2077 if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2078 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2079
2080 err = ovs_vport_set_upcall_portids(vport, ids);
2081 if (err)
2082 goto exit_unlock_free;
2083 }
2084
2085 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2086 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2087 BUG_ON(err < 0);
2088
2089 ovs_unlock();
2090 ovs_notify(&dp_vport_genl_family, reply, info);
2091 return 0;
2092
2093 exit_unlock_free:
2094 ovs_unlock();
2095 kfree_skb(reply);
2096 return err;
2097 }
2098
ovs_vport_cmd_del(struct sk_buff * skb,struct genl_info * info)2099 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2100 {
2101 bool must_update_headroom = false;
2102 struct nlattr **a = info->attrs;
2103 struct sk_buff *reply;
2104 struct datapath *dp;
2105 struct vport *vport;
2106 int err;
2107
2108 reply = ovs_vport_cmd_alloc_info();
2109 if (!reply)
2110 return -ENOMEM;
2111
2112 ovs_lock();
2113 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2114 err = PTR_ERR(vport);
2115 if (IS_ERR(vport))
2116 goto exit_unlock_free;
2117
2118 if (vport->port_no == OVSP_LOCAL) {
2119 err = -EINVAL;
2120 goto exit_unlock_free;
2121 }
2122
2123 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2124 info->snd_seq, 0, OVS_VPORT_CMD_DEL);
2125 BUG_ON(err < 0);
2126
2127 /* the vport deletion may trigger dp headroom update */
2128 dp = vport->dp;
2129 if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2130 must_update_headroom = true;
2131 netdev_reset_rx_headroom(vport->dev);
2132 ovs_dp_detach_port(vport);
2133
2134 if (must_update_headroom)
2135 update_headroom(dp);
2136 ovs_unlock();
2137
2138 ovs_notify(&dp_vport_genl_family, reply, info);
2139 return 0;
2140
2141 exit_unlock_free:
2142 ovs_unlock();
2143 kfree_skb(reply);
2144 return err;
2145 }
2146
ovs_vport_cmd_get(struct sk_buff * skb,struct genl_info * info)2147 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2148 {
2149 struct nlattr **a = info->attrs;
2150 struct ovs_header *ovs_header = info->userhdr;
2151 struct sk_buff *reply;
2152 struct vport *vport;
2153 int err;
2154
2155 reply = ovs_vport_cmd_alloc_info();
2156 if (!reply)
2157 return -ENOMEM;
2158
2159 rcu_read_lock();
2160 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2161 err = PTR_ERR(vport);
2162 if (IS_ERR(vport))
2163 goto exit_unlock_free;
2164 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2165 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2166 BUG_ON(err < 0);
2167 rcu_read_unlock();
2168
2169 return genlmsg_reply(reply, info);
2170
2171 exit_unlock_free:
2172 rcu_read_unlock();
2173 kfree_skb(reply);
2174 return err;
2175 }
2176
ovs_vport_cmd_dump(struct sk_buff * skb,struct netlink_callback * cb)2177 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2178 {
2179 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2180 struct datapath *dp;
2181 int bucket = cb->args[0], skip = cb->args[1];
2182 int i, j = 0;
2183
2184 rcu_read_lock();
2185 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2186 if (!dp) {
2187 rcu_read_unlock();
2188 return -ENODEV;
2189 }
2190 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2191 struct vport *vport;
2192
2193 j = 0;
2194 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2195 if (j >= skip &&
2196 ovs_vport_cmd_fill_info(vport, skb,
2197 NETLINK_CB(cb->skb).portid,
2198 cb->nlh->nlmsg_seq,
2199 NLM_F_MULTI,
2200 OVS_VPORT_CMD_NEW) < 0)
2201 goto out;
2202
2203 j++;
2204 }
2205 skip = 0;
2206 }
2207 out:
2208 rcu_read_unlock();
2209
2210 cb->args[0] = i;
2211 cb->args[1] = j;
2212
2213 return skb->len;
2214 }
2215
2216 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2217 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2218 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2219 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2220 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2221 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
2222 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2223 };
2224
2225 static const struct genl_ops dp_vport_genl_ops[] = {
2226 { .cmd = OVS_VPORT_CMD_NEW,
2227 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2228 .policy = vport_policy,
2229 .doit = ovs_vport_cmd_new
2230 },
2231 { .cmd = OVS_VPORT_CMD_DEL,
2232 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2233 .policy = vport_policy,
2234 .doit = ovs_vport_cmd_del
2235 },
2236 { .cmd = OVS_VPORT_CMD_GET,
2237 .flags = 0, /* OK for unprivileged users. */
2238 .policy = vport_policy,
2239 .doit = ovs_vport_cmd_get,
2240 .dumpit = ovs_vport_cmd_dump
2241 },
2242 { .cmd = OVS_VPORT_CMD_SET,
2243 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2244 .policy = vport_policy,
2245 .doit = ovs_vport_cmd_set,
2246 },
2247 };
2248
2249 struct genl_family dp_vport_genl_family = {
2250 .id = GENL_ID_GENERATE,
2251 .hdrsize = sizeof(struct ovs_header),
2252 .name = OVS_VPORT_FAMILY,
2253 .version = OVS_VPORT_VERSION,
2254 .maxattr = OVS_VPORT_ATTR_MAX,
2255 .netnsok = true,
2256 .parallel_ops = true,
2257 .ops = dp_vport_genl_ops,
2258 .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2259 .mcgrps = &ovs_dp_vport_multicast_group,
2260 .n_mcgrps = 1,
2261 };
2262
2263 static struct genl_family * const dp_genl_families[] = {
2264 &dp_datapath_genl_family,
2265 &dp_vport_genl_family,
2266 &dp_flow_genl_family,
2267 &dp_packet_genl_family,
2268 };
2269
dp_unregister_genl(int n_families)2270 static void dp_unregister_genl(int n_families)
2271 {
2272 int i;
2273
2274 for (i = 0; i < n_families; i++)
2275 genl_unregister_family(dp_genl_families[i]);
2276 }
2277
dp_register_genl(void)2278 static int dp_register_genl(void)
2279 {
2280 int err;
2281 int i;
2282
2283 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2284
2285 err = genl_register_family(dp_genl_families[i]);
2286 if (err)
2287 goto error;
2288 }
2289
2290 return 0;
2291
2292 error:
2293 dp_unregister_genl(i);
2294 return err;
2295 }
2296
ovs_init_net(struct net * net)2297 static int __net_init ovs_init_net(struct net *net)
2298 {
2299 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2300
2301 INIT_LIST_HEAD(&ovs_net->dps);
2302 INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2303 ovs_ct_init(net);
2304 return 0;
2305 }
2306
list_vports_from_net(struct net * net,struct net * dnet,struct list_head * head)2307 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2308 struct list_head *head)
2309 {
2310 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2311 struct datapath *dp;
2312
2313 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2314 int i;
2315
2316 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2317 struct vport *vport;
2318
2319 hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2320 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2321 continue;
2322
2323 if (dev_net(vport->dev) == dnet)
2324 list_add(&vport->detach_list, head);
2325 }
2326 }
2327 }
2328 }
2329
ovs_exit_net(struct net * dnet)2330 static void __net_exit ovs_exit_net(struct net *dnet)
2331 {
2332 struct datapath *dp, *dp_next;
2333 struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2334 struct vport *vport, *vport_next;
2335 struct net *net;
2336 LIST_HEAD(head);
2337
2338 ovs_ct_exit(dnet);
2339 ovs_lock();
2340 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2341 __dp_destroy(dp);
2342
2343 rtnl_lock();
2344 for_each_net(net)
2345 list_vports_from_net(net, dnet, &head);
2346 rtnl_unlock();
2347
2348 /* Detach all vports from given namespace. */
2349 list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2350 list_del(&vport->detach_list);
2351 ovs_dp_detach_port(vport);
2352 }
2353
2354 ovs_unlock();
2355
2356 cancel_work_sync(&ovs_net->dp_notify_work);
2357 }
2358
2359 static struct pernet_operations ovs_net_ops = {
2360 .init = ovs_init_net,
2361 .exit = ovs_exit_net,
2362 .id = &ovs_net_id,
2363 .size = sizeof(struct ovs_net),
2364 };
2365
dp_init(void)2366 static int __init dp_init(void)
2367 {
2368 int err;
2369
2370 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2371
2372 pr_info("Open vSwitch switching datapath\n");
2373
2374 err = action_fifos_init();
2375 if (err)
2376 goto error;
2377
2378 err = ovs_internal_dev_rtnl_link_register();
2379 if (err)
2380 goto error_action_fifos_exit;
2381
2382 err = ovs_flow_init();
2383 if (err)
2384 goto error_unreg_rtnl_link;
2385
2386 err = ovs_vport_init();
2387 if (err)
2388 goto error_flow_exit;
2389
2390 err = register_pernet_device(&ovs_net_ops);
2391 if (err)
2392 goto error_vport_exit;
2393
2394 err = register_netdevice_notifier(&ovs_dp_device_notifier);
2395 if (err)
2396 goto error_netns_exit;
2397
2398 err = ovs_netdev_init();
2399 if (err)
2400 goto error_unreg_notifier;
2401
2402 err = dp_register_genl();
2403 if (err < 0)
2404 goto error_unreg_netdev;
2405
2406 return 0;
2407
2408 error_unreg_netdev:
2409 ovs_netdev_exit();
2410 error_unreg_notifier:
2411 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2412 error_netns_exit:
2413 unregister_pernet_device(&ovs_net_ops);
2414 error_vport_exit:
2415 ovs_vport_exit();
2416 error_flow_exit:
2417 ovs_flow_exit();
2418 error_unreg_rtnl_link:
2419 ovs_internal_dev_rtnl_link_unregister();
2420 error_action_fifos_exit:
2421 action_fifos_exit();
2422 error:
2423 return err;
2424 }
2425
dp_cleanup(void)2426 static void dp_cleanup(void)
2427 {
2428 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2429 ovs_netdev_exit();
2430 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2431 unregister_pernet_device(&ovs_net_ops);
2432 rcu_barrier();
2433 ovs_vport_exit();
2434 ovs_flow_exit();
2435 ovs_internal_dev_rtnl_link_unregister();
2436 action_fifos_exit();
2437 }
2438
2439 module_init(dp_init);
2440 module_exit(dp_cleanup);
2441
2442 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2443 MODULE_LICENSE("GPL");
2444 MODULE_ALIAS_GENL_FAMILY(OVS_DATAPATH_FAMILY);
2445 MODULE_ALIAS_GENL_FAMILY(OVS_VPORT_FAMILY);
2446 MODULE_ALIAS_GENL_FAMILY(OVS_FLOW_FAMILY);
2447 MODULE_ALIAS_GENL_FAMILY(OVS_PACKET_FAMILY);
2448