]> git.proxmox.com Git - ovs.git/blob - datapath/datapath.c
projects / ovs.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
compat: Initialize IPv4 reassembly secret timer
[ovs.git] / datapath / datapath.c
1 /*
2 * Copyright (c) 2007-2015 Nicira, Inc.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of version 2 of the GNU General Public
6 * License as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public License
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
16 * 02110-1301, USA
17 */
18
19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
20
21 #include <linux/init.h>
22 #include <linux/module.h>
23 #include <linux/if_arp.h>
24 #include <linux/if_vlan.h>
25 #include <linux/in.h>
26 #include <linux/ip.h>
27 #include <linux/jhash.h>
28 #include <linux/delay.h>
29 #include <linux/time.h>
30 #include <linux/etherdevice.h>
31 #include <linux/genetlink.h>
32 #include <linux/kernel.h>
33 #include <linux/kthread.h>
34 #include <linux/mutex.h>
35 #include <linux/percpu.h>
36 #include <linux/rcupdate.h>
37 #include <linux/tcp.h>
38 #include <linux/udp.h>
39 #include <linux/version.h>
40 #include <linux/ethtool.h>
41 #include <linux/wait.h>
42 #include <asm/div64.h>
43 #include <linux/highmem.h>
44 #include <linux/netfilter_bridge.h>
45 #include <linux/netfilter_ipv4.h>
46 #include <linux/inetdevice.h>
47 #include <linux/list.h>
48 #include <linux/openvswitch.h>
49 #include <linux/rculist.h>
50 #include <linux/dmi.h>
51 #include <net/genetlink.h>
52 #include <net/net_namespace.h>
53 #include <net/netns/generic.h>
54 #include <net/nsh.h>
55
56 #include "datapath.h"
57 #include "conntrack.h"
58 #include "flow.h"
59 #include "flow_table.h"
60 #include "flow_netlink.h"
61 #include "meter.h"
62 #include "gso.h"
63 #include "vport-internal_dev.h"
64 #include "vport-netdev.h"
65
66 unsigned int ovs_net_id __read_mostly;
67
68 static struct genl_family dp_packet_genl_family;
69 static struct genl_family dp_flow_genl_family;
70 static struct genl_family dp_datapath_genl_family;
71
72 static const struct nla_policy flow_policy[];
73
74 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
75 .name = OVS_FLOW_MCGROUP
76 };
77
78 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
79 .name = OVS_DATAPATH_MCGROUP
80 };
81
82 struct genl_multicast_group ovs_dp_vport_multicast_group = {
83 .name = OVS_VPORT_MCGROUP
84 };
85
86 /* Check if need to build a reply message.
87 * OVS userspace sets the NLM_F_ECHO flag if it needs the reply.
88 */
89 static bool ovs_must_notify(struct genl_family *family, struct genl_info *info,
90 unsigned int group)
91 {
92 return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
93 genl_has_listeners(family, genl_info_net(info), group);
94 }
95
96 static void ovs_notify(struct genl_family *family, struct genl_multicast_group *grp,
97 struct sk_buff *skb, struct genl_info *info)
98 {
99 genl_notify(family, skb, info, GROUP_ID(grp), GFP_KERNEL);
100 }
101
102 /**
103 * DOC: Locking:
104 *
105 * All writes e.g. Writes to device state (add/remove datapath, port, set
106 * operations on vports, etc.), Writes to other state (flow table
107 * modifications, set miscellaneous datapath parameters, etc.) are protected
108 * by ovs_lock.
109 *
110 * Reads are protected by RCU.
111 *
112 * There are a few special cases (mostly stats) that have their own
113 * synchronization but they nest under all of above and don't interact with
114 * each other.
115 *
116 * The RTNL lock nests inside ovs_mutex.
117 */
118
119 static DEFINE_MUTEX(ovs_mutex);
120
121 void ovs_lock(void)
122 {
123 mutex_lock(&ovs_mutex);
124 }
125
126 void ovs_unlock(void)
127 {
128 mutex_unlock(&ovs_mutex);
129 }
130
131 #ifdef CONFIG_LOCKDEP
132 int lockdep_ovsl_is_held(void)
133 {
134 if (debug_locks)
135 return lockdep_is_held(&ovs_mutex);
136 else
137 return 1;
138 }
139 #endif
140
141 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
142 const struct sw_flow_key *,
143 const struct dp_upcall_info *,
144 uint32_t cutlen);
145 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
146 const struct sw_flow_key *,
147 const struct dp_upcall_info *,
148 uint32_t cutlen);
149
150 /* Must be called with rcu_read_lock or ovs_mutex. */
151 const char *ovs_dp_name(const struct datapath *dp)
152 {
153 struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
154 return ovs_vport_name(vport);
155 }
156
157 static int get_dpifindex(const struct datapath *dp)
158 {
159 struct vport *local;
160 int ifindex;
161
162 rcu_read_lock();
163
164 local = ovs_vport_rcu(dp, OVSP_LOCAL);
165 if (local)
166 ifindex = local->dev->ifindex;
167 else
168 ifindex = 0;
169
170 rcu_read_unlock();
171
172 return ifindex;
173 }
174
175 static void destroy_dp_rcu(struct rcu_head *rcu)
176 {
177 struct datapath *dp = container_of(rcu, struct datapath, rcu);
178
179 ovs_flow_tbl_destroy(&dp->table);
180 free_percpu(dp->stats_percpu);
181 kfree(dp->ports);
182 ovs_meters_exit(dp);
183 kfree(dp);
184 }
185
186 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
187 u16 port_no)
188 {
189 return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
190 }
191
192 /* Called with ovs_mutex or RCU read lock. */
193 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
194 {
195 struct vport *vport;
196 struct hlist_head *head;
197
198 head = vport_hash_bucket(dp, port_no);
199 hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
200 if (vport->port_no == port_no)
201 return vport;
202 }
203 return NULL;
204 }
205
206 /* Called with ovs_mutex. */
207 static struct vport *new_vport(const struct vport_parms *parms)
208 {
209 struct vport *vport;
210
211 vport = ovs_vport_add(parms);
212 if (!IS_ERR(vport)) {
213 struct datapath *dp = parms->dp;
214 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
215
216 hlist_add_head_rcu(&vport->dp_hash_node, head);
217 }
218 return vport;
219 }
220
221 void ovs_dp_detach_port(struct vport *p)
222 {
223 ASSERT_OVSL();
224
225 /* First drop references to device. */
226 hlist_del_rcu(&p->dp_hash_node);
227
228 /* Then destroy it. */
229 ovs_vport_del(p);
230 }
231
232 /* Must be called with rcu_read_lock. */
233 void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
234 {
235 const struct vport *p = OVS_CB(skb)->input_vport;
236 struct datapath *dp = p->dp;
237 struct sw_flow *flow;
238 struct sw_flow_actions *sf_acts;
239 struct dp_stats_percpu *stats;
240 u64 *stats_counter;
241 u32 n_mask_hit;
242
243 stats = this_cpu_ptr(dp->stats_percpu);
244
245 /* Look up flow. */
246 flow = ovs_flow_tbl_lookup_stats(&dp->table, key, skb_get_hash(skb),
247 &n_mask_hit);
248 if (unlikely(!flow)) {
249 struct dp_upcall_info upcall;
250 int error;
251
252 memset(&upcall, 0, sizeof(upcall));
253 upcall.cmd = OVS_PACKET_CMD_MISS;
254 upcall.portid = ovs_vport_find_upcall_portid(p, skb);
255 upcall.mru = OVS_CB(skb)->mru;
256 error = ovs_dp_upcall(dp, skb, key, &upcall, 0);
257 if (unlikely(error))
258 kfree_skb(skb);
259 else
260 consume_skb(skb);
261 stats_counter = &stats->n_missed;
262 goto out;
263 }
264
265 ovs_flow_stats_update(flow, key->tp.flags, skb);
266 sf_acts = rcu_dereference(flow->sf_acts);
267 ovs_execute_actions(dp, skb, sf_acts, key);
268
269 stats_counter = &stats->n_hit;
270
271 out:
272 /* Update datapath statistics. */
273 u64_stats_update_begin(&stats->syncp);
274 (*stats_counter)++;
275 stats->n_mask_hit += n_mask_hit;
276 u64_stats_update_end(&stats->syncp);
277 }
278
279 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
280 const struct sw_flow_key *key,
281 const struct dp_upcall_info *upcall_info,
282 uint32_t cutlen)
283 {
284 struct dp_stats_percpu *stats;
285 int err;
286
287 if (upcall_info->portid == 0) {
288 err = -ENOTCONN;
289 goto err;
290 }
291
292 if (!skb_is_gso(skb))
293 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
294 else
295 err = queue_gso_packets(dp, skb, key, upcall_info, cutlen);
296 if (err)
297 goto err;
298
299 return 0;
300
301 err:
302 stats = this_cpu_ptr(dp->stats_percpu);
303
304 u64_stats_update_begin(&stats->syncp);
305 stats->n_lost++;
306 u64_stats_update_end(&stats->syncp);
307
308 return err;
309 }
310
311 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
312 const struct sw_flow_key *key,
313 const struct dp_upcall_info *upcall_info,
314 uint32_t cutlen)
315 {
316 #ifdef HAVE_SKB_GSO_UDP
317 unsigned int gso_type = skb_shinfo(skb)->gso_type;
318 struct sw_flow_key later_key;
319 #endif
320 struct sk_buff *segs, *nskb;
321 struct ovs_skb_cb ovs_cb;
322 int err;
323
324 ovs_cb = *OVS_CB(skb);
325 segs = __skb_gso_segment(skb, NETIF_F_SG, false);
326 *OVS_CB(skb) = ovs_cb;
327 if (IS_ERR(segs))
328 return PTR_ERR(segs);
329 if (segs == NULL)
330 return -EINVAL;
331 #ifdef HAVE_SKB_GSO_UDP
332 if (gso_type & SKB_GSO_UDP) {
333 /* The initial flow key extracted by ovs_flow_key_extract()
334 * in this case is for a first fragment, so we need to
335 * properly mark later fragments.
336 */
337 later_key = *key;
338 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
339 }
340 #endif
341 /* Queue all of the segments. */
342 skb = segs;
343 do {
344 *OVS_CB(skb) = ovs_cb;
345 #ifdef HAVE_SKB_GSO_UDP
346 if (gso_type & SKB_GSO_UDP && skb != segs)
347 key = &later_key;
348 #endif
349 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
350 if (err)
351 break;
352
353 } while ((skb = skb->next));
354
355 /* Free all of the segments. */
356 skb = segs;
357 do {
358 nskb = skb->next;
359 if (err)
360 kfree_skb(skb);
361 else
362 consume_skb(skb);
363 } while ((skb = nskb));
364 return err;
365 }
366
367 static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
368 unsigned int hdrlen, int actions_attrlen)
369 {
370 size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
371 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
372 + nla_total_size(ovs_key_attr_size()) /* OVS_PACKET_ATTR_KEY */
373 + nla_total_size(sizeof(unsigned int)); /* OVS_PACKET_ATTR_LEN */
374
375 /* OVS_PACKET_ATTR_USERDATA */
376 if (upcall_info->userdata)
377 size += NLA_ALIGN(upcall_info->userdata->nla_len);
378
379 /* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
380 if (upcall_info->egress_tun_info)
381 size += nla_total_size(ovs_tun_key_attr_size());
382
383 /* OVS_PACKET_ATTR_ACTIONS */
384 if (upcall_info->actions_len)
385 size += nla_total_size(actions_attrlen);
386
387 /* OVS_PACKET_ATTR_MRU */
388 if (upcall_info->mru)
389 size += nla_total_size(sizeof(upcall_info->mru));
390
391 return size;
392 }
393
394 static void pad_packet(struct datapath *dp, struct sk_buff *skb)
395 {
396 if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
397 size_t plen = NLA_ALIGN(skb->len) - skb->len;
398
399 if (plen > 0)
400 skb_put_zero(skb, plen);
401 }
402 }
403
404 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
405 const struct sw_flow_key *key,
406 const struct dp_upcall_info *upcall_info,
407 uint32_t cutlen)
408 {
409 struct ovs_header *upcall;
410 struct sk_buff *nskb = NULL;
411 struct sk_buff *user_skb = NULL; /* to be queued to userspace */
412 struct nlattr *nla;
413 size_t len;
414 unsigned int hlen;
415 int err, dp_ifindex;
416
417 dp_ifindex = get_dpifindex(dp);
418 if (!dp_ifindex)
419 return -ENODEV;
420
421 if (skb_vlan_tag_present(skb)) {
422 nskb = skb_clone(skb, GFP_ATOMIC);
423 if (!nskb)
424 return -ENOMEM;
425
426 nskb = __vlan_hwaccel_push_inside(nskb);
427 if (!nskb)
428 return -ENOMEM;
429
430 skb = nskb;
431 }
432
433 if (nla_attr_size(skb->len) > USHRT_MAX) {
434 err = -EFBIG;
435 goto out;
436 }
437
438 /* Complete checksum if needed */
439 if (skb->ip_summed == CHECKSUM_PARTIAL &&
440 (err = skb_csum_hwoffload_help(skb, 0)))
441 goto out;
442
443 /* Older versions of OVS user space enforce alignment of the last
444 * Netlink attribute to NLA_ALIGNTO which would require extensive
445 * padding logic. Only perform zerocopy if padding is not required.
446 */
447 if (dp->user_features & OVS_DP_F_UNALIGNED)
448 hlen = skb_zerocopy_headlen(skb);
449 else
450 hlen = skb->len;
451
452 len = upcall_msg_size(upcall_info, hlen - cutlen,
453 OVS_CB(skb)->acts_origlen);
454 user_skb = genlmsg_new(len, GFP_ATOMIC);
455 if (!user_skb) {
456 err = -ENOMEM;
457 goto out;
458 }
459
460 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
461 0, upcall_info->cmd);
462 upcall->dp_ifindex = dp_ifindex;
463
464 err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
465 BUG_ON(err);
466
467 if (upcall_info->userdata)
468 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
469 nla_len(upcall_info->userdata),
470 nla_data(upcall_info->userdata));
471
472
473 if (upcall_info->egress_tun_info) {
474 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_EGRESS_TUN_KEY);
475 err = ovs_nla_put_tunnel_info(user_skb,
476 upcall_info->egress_tun_info);
477 BUG_ON(err);
478 nla_nest_end(user_skb, nla);
479 }
480
481 if (upcall_info->actions_len) {
482 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_ACTIONS);
483 err = ovs_nla_put_actions(upcall_info->actions,
484 upcall_info->actions_len,
485 user_skb);
486 if (!err)
487 nla_nest_end(user_skb, nla);
488 else
489 nla_nest_cancel(user_skb, nla);
490 }
491
492 /* Add OVS_PACKET_ATTR_MRU */
493 if (upcall_info->mru) {
494 if (nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU,
495 upcall_info->mru)) {
496 err = -ENOBUFS;
497 goto out;
498 }
499 pad_packet(dp, user_skb);
500 }
501
502 /* Add OVS_PACKET_ATTR_LEN when packet is truncated */
503 if (cutlen > 0) {
504 if (nla_put_u32(user_skb, OVS_PACKET_ATTR_LEN,
505 skb->len)) {
506 err = -ENOBUFS;
507 goto out;
508 }
509 pad_packet(dp, user_skb);
510 }
511
512 /* Only reserve room for attribute header, packet data is added
513 * in skb_zerocopy()
514 */
515 if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
516 err = -ENOBUFS;
517 goto out;
518 }
519 nla->nla_len = nla_attr_size(skb->len - cutlen);
520
521 err = skb_zerocopy(user_skb, skb, skb->len - cutlen, hlen);
522 if (err)
523 goto out;
524
525 /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
526 pad_packet(dp, user_skb);
527
528 ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
529
530 err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
531 user_skb = NULL;
532 out:
533 if (err)
534 skb_tx_error(skb);
535 kfree_skb(user_skb);
536 kfree_skb(nskb);
537 return err;
538 }
539
540 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
541 {
542 struct ovs_header *ovs_header = info->userhdr;
543 struct net *net = sock_net(skb->sk);
544 struct nlattr **a = info->attrs;
545 struct sw_flow_actions *acts;
546 struct sk_buff *packet;
547 struct sw_flow *flow;
548 struct sw_flow_actions *sf_acts;
549 struct datapath *dp;
550 struct vport *input_vport;
551 u16 mru = 0;
552 int len;
553 int err;
554 bool log = !a[OVS_PACKET_ATTR_PROBE];
555
556 err = -EINVAL;
557 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
558 !a[OVS_PACKET_ATTR_ACTIONS])
559 goto err;
560
561 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
562 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
563 err = -ENOMEM;
564 if (!packet)
565 goto err;
566 skb_reserve(packet, NET_IP_ALIGN);
567
568 nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
569
570 /* Set packet's mru */
571 if (a[OVS_PACKET_ATTR_MRU]) {
572 mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
573 packet->ignore_df = 1;
574 }
575 OVS_CB(packet)->mru = mru;
576
577 /* Build an sw_flow for sending this packet. */
578 flow = ovs_flow_alloc();
579 err = PTR_ERR(flow);
580 if (IS_ERR(flow))
581 goto err_kfree_skb;
582
583 err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
584 packet, &flow->key, log);
585 if (err)
586 goto err_flow_free;
587
588 err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
589 &flow->key, &acts, log);
590 if (err)
591 goto err_flow_free;
592
593 rcu_assign_pointer(flow->sf_acts, acts);
594 packet->priority = flow->key.phy.priority;
595 packet->mark = flow->key.phy.skb_mark;
596
597 rcu_read_lock();
598 dp = get_dp_rcu(net, ovs_header->dp_ifindex);
599 err = -ENODEV;
600 if (!dp)
601 goto err_unlock;
602
603 input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
604 if (!input_vport)
605 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
606
607 if (!input_vport)
608 goto err_unlock;
609
610 packet->dev = input_vport->dev;
611 OVS_CB(packet)->input_vport = input_vport;
612 sf_acts = rcu_dereference(flow->sf_acts);
613
614 local_bh_disable();
615 err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
616 local_bh_enable();
617 rcu_read_unlock();
618
619 ovs_flow_free(flow, false);
620 return err;
621
622 err_unlock:
623 rcu_read_unlock();
624 err_flow_free:
625 ovs_flow_free(flow, false);
626 err_kfree_skb:
627 kfree_skb(packet);
628 err:
629 return err;
630 }
631
632 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
633 [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
634 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
635 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
636 [OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
637 [OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
638 };
639
640 static struct genl_ops dp_packet_genl_ops[] = {
641 { .cmd = OVS_PACKET_CMD_EXECUTE,
642 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
643 .policy = packet_policy,
644 .doit = ovs_packet_cmd_execute
645 }
646 };
647
648 static struct genl_family dp_packet_genl_family __ro_after_init = {
649 .hdrsize = sizeof(struct ovs_header),
650 .name = OVS_PACKET_FAMILY,
651 .version = OVS_PACKET_VERSION,
652 .maxattr = OVS_PACKET_ATTR_MAX,
653 .netnsok = true,
654 .parallel_ops = true,
655 .ops = dp_packet_genl_ops,
656 .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
657 .module = THIS_MODULE,
658 };
659
660 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
661 struct ovs_dp_megaflow_stats *mega_stats)
662 {
663 int i;
664
665 memset(mega_stats, 0, sizeof(*mega_stats));
666
667 stats->n_flows = ovs_flow_tbl_count(&dp->table);
668 mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
669
670 stats->n_hit = stats->n_missed = stats->n_lost = 0;
671
672 for_each_possible_cpu(i) {
673 const struct dp_stats_percpu *percpu_stats;
674 struct dp_stats_percpu local_stats;
675 unsigned int start;
676
677 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
678
679 do {
680 start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
681 local_stats = *percpu_stats;
682 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
683
684 stats->n_hit += local_stats.n_hit;
685 stats->n_missed += local_stats.n_missed;
686 stats->n_lost += local_stats.n_lost;
687 mega_stats->n_mask_hit += local_stats.n_mask_hit;
688 }
689 }
690
691 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
692 {
693 return ovs_identifier_is_ufid(sfid) &&
694 !(ufid_flags & OVS_UFID_F_OMIT_KEY);
695 }
696
697 static bool should_fill_mask(uint32_t ufid_flags)
698 {
699 return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
700 }
701
702 static bool should_fill_actions(uint32_t ufid_flags)
703 {
704 return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
705 }
706
707 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
708 const struct sw_flow_id *sfid,
709 uint32_t ufid_flags)
710 {
711 size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
712
713 /* OVS_FLOW_ATTR_UFID */
714 if (sfid && ovs_identifier_is_ufid(sfid))
715 len += nla_total_size(sfid->ufid_len);
716
717 /* OVS_FLOW_ATTR_KEY */
718 if (!sfid || should_fill_key(sfid, ufid_flags))
719 len += nla_total_size(ovs_key_attr_size());
720
721 /* OVS_FLOW_ATTR_MASK */
722 if (should_fill_mask(ufid_flags))
723 len += nla_total_size(ovs_key_attr_size());
724
725 /* OVS_FLOW_ATTR_ACTIONS */
726 if (should_fill_actions(ufid_flags))
727 len += nla_total_size(acts->orig_len);
728
729 return len
730 + nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
731 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
732 + nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */
733 }
734
735 /* Called with ovs_mutex or RCU read lock. */
736 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
737 struct sk_buff *skb)
738 {
739 struct ovs_flow_stats stats;
740 __be16 tcp_flags;
741 unsigned long used;
742
743 ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
744
745 if (used &&
746 nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
747 OVS_FLOW_ATTR_PAD))
748 return -EMSGSIZE;
749
750 if (stats.n_packets &&
751 nla_put_64bit(skb, OVS_FLOW_ATTR_STATS,
752 sizeof(struct ovs_flow_stats), &stats,
753 OVS_FLOW_ATTR_PAD))
754 return -EMSGSIZE;
755
756 if ((u8)ntohs(tcp_flags) &&
757 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
758 return -EMSGSIZE;
759
760 return 0;
761 }
762
763 /* Called with ovs_mutex or RCU read lock. */
764 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
765 struct sk_buff *skb, int skb_orig_len)
766 {
767 struct nlattr *start;
768 int err;
769
770 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
771 * this is the first flow to be dumped into 'skb'. This is unusual for
772 * Netlink but individual action lists can be longer than
773 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
774 * The userspace caller can always fetch the actions separately if it
775 * really wants them. (Most userspace callers in fact don't care.)
776 *
777 * This can only fail for dump operations because the skb is always
778 * properly sized for single flows.
779 */
780 start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
781 if (start) {
782 const struct sw_flow_actions *sf_acts;
783
784 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
785 err = ovs_nla_put_actions(sf_acts->actions,
786 sf_acts->actions_len, skb);
787
788 if (!err)
789 nla_nest_end(skb, start);
790 else {
791 if (skb_orig_len)
792 return err;
793
794 nla_nest_cancel(skb, start);
795 }
796 } else if (skb_orig_len) {
797 return -EMSGSIZE;
798 }
799
800 return 0;
801 }
802
803 /* Called with ovs_mutex or RCU read lock. */
804 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
805 struct sk_buff *skb, u32 portid,
806 u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
807 {
808 const int skb_orig_len = skb->len;
809 struct ovs_header *ovs_header;
810 int err;
811
812 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
813 flags, cmd);
814 if (!ovs_header)
815 return -EMSGSIZE;
816
817 ovs_header->dp_ifindex = dp_ifindex;
818
819 err = ovs_nla_put_identifier(flow, skb);
820 if (err)
821 goto error;
822
823 if (should_fill_key(&flow->id, ufid_flags)) {
824 err = ovs_nla_put_masked_key(flow, skb);
825 if (err)
826 goto error;
827 }
828
829 if (should_fill_mask(ufid_flags)) {
830 err = ovs_nla_put_mask(flow, skb);
831 if (err)
832 goto error;
833 }
834
835 err = ovs_flow_cmd_fill_stats(flow, skb);
836 if (err)
837 goto error;
838
839 if (should_fill_actions(ufid_flags)) {
840 err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
841 if (err)
842 goto error;
843 }
844
845 genlmsg_end(skb, ovs_header);
846 return 0;
847
848 error:
849 genlmsg_cancel(skb, ovs_header);
850 return err;
851 }
852
853 /* May not be called with RCU read lock. */
854 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
855 const struct sw_flow_id *sfid,
856 struct genl_info *info,
857 bool always,
858 uint32_t ufid_flags)
859 {
860 struct sk_buff *skb;
861 size_t len;
862
863 if (!always && !ovs_must_notify(&dp_flow_genl_family, info,
864 GROUP_ID(&ovs_dp_flow_multicast_group)))
865 return NULL;
866
867 len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
868 skb = genlmsg_new(len, GFP_KERNEL);
869 if (!skb)
870 return ERR_PTR(-ENOMEM);
871
872 return skb;
873 }
874
875 /* Called with ovs_mutex. */
876 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
877 int dp_ifindex,
878 struct genl_info *info, u8 cmd,
879 bool always, u32 ufid_flags)
880 {
881 struct sk_buff *skb;
882 int retval;
883
884 skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
885 &flow->id, info, always, ufid_flags);
886 if (IS_ERR_OR_NULL(skb))
887 return skb;
888
889 retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
890 info->snd_portid, info->snd_seq, 0,
891 cmd, ufid_flags);
892 BUG_ON(retval < 0);
893 return skb;
894 }
895
896 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
897 {
898 struct net *net = sock_net(skb->sk);
899 struct nlattr **a = info->attrs;
900 struct ovs_header *ovs_header = info->userhdr;
901 struct sw_flow *flow = NULL, *new_flow;
902 struct sw_flow_mask mask;
903 struct sk_buff *reply;
904 struct datapath *dp;
905 struct sw_flow_actions *acts;
906 struct sw_flow_match match;
907 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
908 int error;
909 bool log = !a[OVS_FLOW_ATTR_PROBE];
910
911 /* Must have key and actions. */
912 error = -EINVAL;
913 if (!a[OVS_FLOW_ATTR_KEY]) {
914 OVS_NLERR(log, "Flow key attr not present in new flow.");
915 goto error;
916 }
917 if (!a[OVS_FLOW_ATTR_ACTIONS]) {
918 OVS_NLERR(log, "Flow actions attr not present in new flow.");
919 goto error;
920 }
921
922 /* Most of the time we need to allocate a new flow, do it before
923 * locking.
924 */
925 new_flow = ovs_flow_alloc();
926 if (IS_ERR(new_flow)) {
927 error = PTR_ERR(new_flow);
928 goto error;
929 }
930
931 /* Extract key. */
932 ovs_match_init(&match, &new_flow->key, false, &mask);
933 error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
934 a[OVS_FLOW_ATTR_MASK], log);
935 if (error)
936 goto err_kfree_flow;
937
938 /* Extract flow identifier. */
939 error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
940 &new_flow->key, log);
941 if (error)
942 goto err_kfree_flow;
943
944 /* unmasked key is needed to match when ufid is not used. */
945 if (ovs_identifier_is_key(&new_flow->id))
946 match.key = new_flow->id.unmasked_key;
947
948 ovs_flow_mask_key(&new_flow->key, &new_flow->key, true, &mask);
949
950 /* Validate actions. */
951 error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
952 &new_flow->key, &acts, log);
953 if (error) {
954 OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
955 goto err_kfree_flow;
956 }
957
958 reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
959 ufid_flags);
960 if (IS_ERR(reply)) {
961 error = PTR_ERR(reply);
962 goto err_kfree_acts;
963 }
964
965 ovs_lock();
966 dp = get_dp(net, ovs_header->dp_ifindex);
967 if (unlikely(!dp)) {
968 error = -ENODEV;
969 goto err_unlock_ovs;
970 }
971
972 /* Check if this is a duplicate flow */
973 if (ovs_identifier_is_ufid(&new_flow->id))
974 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
975 if (!flow)
976 flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->key);
977 if (likely(!flow)) {
978 rcu_assign_pointer(new_flow->sf_acts, acts);
979
980 /* Put flow in bucket. */
981 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
982 if (unlikely(error)) {
983 acts = NULL;
984 goto err_unlock_ovs;
985 }
986
987 if (unlikely(reply)) {
988 error = ovs_flow_cmd_fill_info(new_flow,
989 ovs_header->dp_ifindex,
990 reply, info->snd_portid,
991 info->snd_seq, 0,
992 OVS_FLOW_CMD_NEW,
993 ufid_flags);
994 BUG_ON(error < 0);
995 }
996 ovs_unlock();
997 } else {
998 struct sw_flow_actions *old_acts;
999
1000 /* Bail out if we're not allowed to modify an existing flow.
1001 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1002 * because Generic Netlink treats the latter as a dump
1003 * request. We also accept NLM_F_EXCL in case that bug ever
1004 * gets fixed.
1005 */
1006 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1007 | NLM_F_EXCL))) {
1008 error = -EEXIST;
1009 goto err_unlock_ovs;
1010 }
1011 /* The flow identifier has to be the same for flow updates.
1012 * Look for any overlapping flow.
1013 */
1014 if (unlikely(!ovs_flow_cmp(flow, &match))) {
1015 if (ovs_identifier_is_key(&flow->id))
1016 flow = ovs_flow_tbl_lookup_exact(&dp->table,
1017 &match);
1018 else /* UFID matches but key is different */
1019 flow = NULL;
1020 if (!flow) {
1021 error = -ENOENT;
1022 goto err_unlock_ovs;
1023 }
1024 }
1025 /* Update actions. */
1026 old_acts = ovsl_dereference(flow->sf_acts);
1027 rcu_assign_pointer(flow->sf_acts, acts);
1028
1029 if (unlikely(reply)) {
1030 error = ovs_flow_cmd_fill_info(flow,
1031 ovs_header->dp_ifindex,
1032 reply, info->snd_portid,
1033 info->snd_seq, 0,
1034 OVS_FLOW_CMD_NEW,
1035 ufid_flags);
1036 BUG_ON(error < 0);
1037 }
1038 ovs_unlock();
1039
1040 ovs_nla_free_flow_actions_rcu(old_acts);
1041 ovs_flow_free(new_flow, false);
1042 }
1043
1044 if (reply)
1045 ovs_notify(&dp_flow_genl_family, &ovs_dp_flow_multicast_group, reply, info);
1046 return 0;
1047
1048 err_unlock_ovs:
1049 ovs_unlock();
1050 kfree_skb(reply);
1051 err_kfree_acts:
1052 ovs_nla_free_flow_actions(acts);
1053 err_kfree_flow:
1054 ovs_flow_free(new_flow, false);
1055 error:
1056 return error;
1057 }
1058
1059 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1060 static struct sw_flow_actions *get_flow_actions(struct net *net,
1061 const struct nlattr *a,
1062 const struct sw_flow_key *key,
1063 const struct sw_flow_mask *mask,
1064 bool log)
1065 {
1066 struct sw_flow_actions *acts;
1067 struct sw_flow_key masked_key;
1068 int error;
1069
1070 ovs_flow_mask_key(&masked_key, key, true, mask);
1071 error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1072 if (error) {
1073 OVS_NLERR(log,
1074 "Actions may not be safe on all matching packets");
1075 return ERR_PTR(error);
1076 }
1077
1078 return acts;
1079 }
1080
1081 /* Factor out match-init and action-copy to avoid
1082 * "Wframe-larger-than=1024" warning. Because mask is only
1083 * used to get actions, we new a function to save some
1084 * stack space.
1085 *
1086 * If there are not key and action attrs, we return 0
1087 * directly. In the case, the caller will also not use the
1088 * match as before. If there is action attr, we try to get
1089 * actions and save them to *acts. Before returning from
1090 * the function, we reset the match->mask pointer. Because
1091 * we should not to return match object with dangling reference
1092 * to mask.
1093 * */
1094 static int ovs_nla_init_match_and_action(struct net *net,
1095 struct sw_flow_match *match,
1096 struct sw_flow_key *key,
1097 struct nlattr **a,
1098 struct sw_flow_actions **acts,
1099 bool log)
1100 {
1101 struct sw_flow_mask mask;
1102 int error = 0;
1103
1104 if (a[OVS_FLOW_ATTR_KEY]) {
1105 ovs_match_init(match, key, true, &mask);
1106 error = ovs_nla_get_match(net, match, a[OVS_FLOW_ATTR_KEY],
1107 a[OVS_FLOW_ATTR_MASK], log);
1108 if (error)
1109 goto error;
1110 }
1111
1112 if (a[OVS_FLOW_ATTR_ACTIONS]) {
1113 if (!a[OVS_FLOW_ATTR_KEY]) {
1114 OVS_NLERR(log,
1115 "Flow key attribute not present in set flow.");
1116 error = -EINVAL;
1117 goto error;
1118 }
1119
1120 *acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], key,
1121 &mask, log);
1122 if (IS_ERR(*acts)) {
1123 error = PTR_ERR(*acts);
1124 goto error;
1125 }
1126 }
1127
1128 /* On success, error is 0. */
1129 error:
1130 match->mask = NULL;
1131 return error;
1132 }
1133
1134 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1135 {
1136 struct net *net = sock_net(skb->sk);
1137 struct nlattr **a = info->attrs;
1138 struct ovs_header *ovs_header = info->userhdr;
1139 struct sw_flow_key key;
1140 struct sw_flow *flow;
1141 struct sk_buff *reply = NULL;
1142 struct datapath *dp;
1143 struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1144 struct sw_flow_match match;
1145 struct sw_flow_id sfid;
1146 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1147 int error = 0;
1148 bool log = !a[OVS_FLOW_ATTR_PROBE];
1149 bool ufid_present;
1150
1151 ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1152 if (!a[OVS_FLOW_ATTR_KEY] && !ufid_present) {
1153 OVS_NLERR(log,
1154 "Flow set message rejected, Key attribute missing.");
1155 return -EINVAL;
1156 }
1157
1158 error = ovs_nla_init_match_and_action(net, &match, &key, a,
1159 &acts, log);
1160 if (error)
1161 goto error;
1162
1163 if (acts) {
1164 /* Can allocate before locking if have acts. */
1165 reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1166 ufid_flags);
1167 if (IS_ERR(reply)) {
1168 error = PTR_ERR(reply);
1169 goto err_kfree_acts;
1170 }
1171 }
1172
1173 ovs_lock();
1174 dp = get_dp(net, ovs_header->dp_ifindex);
1175 if (unlikely(!dp)) {
1176 error = -ENODEV;
1177 goto err_unlock_ovs;
1178 }
1179 /* Check that the flow exists. */
1180 if (ufid_present)
1181 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1182 else
1183 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1184 if (unlikely(!flow)) {
1185 error = -ENOENT;
1186 goto err_unlock_ovs;
1187 }
1188
1189 /* Update actions, if present. */
1190 if (likely(acts)) {
1191 old_acts = ovsl_dereference(flow->sf_acts);
1192 rcu_assign_pointer(flow->sf_acts, acts);
1193
1194 if (unlikely(reply)) {
1195 error = ovs_flow_cmd_fill_info(flow,
1196 ovs_header->dp_ifindex,
1197 reply, info->snd_portid,
1198 info->snd_seq, 0,
1199 OVS_FLOW_CMD_NEW,
1200 ufid_flags);
1201 BUG_ON(error < 0);
1202 }
1203 } else {
1204 /* Could not alloc without acts before locking. */
1205 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1206 info, OVS_FLOW_CMD_NEW, false,
1207 ufid_flags);
1208
1209 if (unlikely(IS_ERR(reply))) {
1210 error = PTR_ERR(reply);
1211 goto err_unlock_ovs;
1212 }
1213 }
1214
1215 /* Clear stats. */
1216 if (a[OVS_FLOW_ATTR_CLEAR])
1217 ovs_flow_stats_clear(flow);
1218 ovs_unlock();
1219
1220 if (reply)
1221 ovs_notify(&dp_flow_genl_family, &ovs_dp_flow_multicast_group, reply, info);
1222 if (old_acts)
1223 ovs_nla_free_flow_actions_rcu(old_acts);
1224
1225 return 0;
1226
1227 err_unlock_ovs:
1228 ovs_unlock();
1229 kfree_skb(reply);
1230 err_kfree_acts:
1231 ovs_nla_free_flow_actions(acts);
1232 error:
1233 return error;
1234 }
1235
1236 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1237 {
1238 struct nlattr **a = info->attrs;
1239 struct ovs_header *ovs_header = info->userhdr;
1240 struct net *net = sock_net(skb->sk);
1241 struct sw_flow_key key;
1242 struct sk_buff *reply;
1243 struct sw_flow *flow;
1244 struct datapath *dp;
1245 struct sw_flow_match match;
1246 struct sw_flow_id ufid;
1247 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1248 int err = 0;
1249 bool log = !a[OVS_FLOW_ATTR_PROBE];
1250 bool ufid_present;
1251
1252 ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1253 if (a[OVS_FLOW_ATTR_KEY]) {
1254 ovs_match_init(&match, &key, true, NULL);
1255 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1256 log);
1257 } else if (!ufid_present) {
1258 OVS_NLERR(log,
1259 "Flow get message rejected, Key attribute missing.");
1260 err = -EINVAL;
1261 }
1262 if (err)
1263 return err;
1264
1265 ovs_lock();
1266 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1267 if (!dp) {
1268 err = -ENODEV;
1269 goto unlock;
1270 }
1271
1272 if (ufid_present)
1273 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1274 else
1275 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1276 if (!flow) {
1277 err = -ENOENT;
1278 goto unlock;
1279 }
1280
1281 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1282 OVS_FLOW_CMD_NEW, true, ufid_flags);
1283 if (IS_ERR(reply)) {
1284 err = PTR_ERR(reply);
1285 goto unlock;
1286 }
1287
1288 ovs_unlock();
1289 return genlmsg_reply(reply, info);
1290 unlock:
1291 ovs_unlock();
1292 return err;
1293 }
1294
1295 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1296 {
1297 struct nlattr **a = info->attrs;
1298 struct ovs_header *ovs_header = info->userhdr;
1299 struct net *net = sock_net(skb->sk);
1300 struct sw_flow_key key;
1301 struct sk_buff *reply;
1302 struct sw_flow *flow = NULL;
1303 struct datapath *dp;
1304 struct sw_flow_match match;
1305 struct sw_flow_id ufid;
1306 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1307 int err;
1308 bool log = !a[OVS_FLOW_ATTR_PROBE];
1309 bool ufid_present;
1310
1311 ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1312 if (a[OVS_FLOW_ATTR_KEY]) {
1313 ovs_match_init(&match, &key, true, NULL);
1314 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1315 NULL, log);
1316 if (unlikely(err))
1317 return err;
1318 }
1319
1320 ovs_lock();
1321 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1322 if (unlikely(!dp)) {
1323 err = -ENODEV;
1324 goto unlock;
1325 }
1326
1327 if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1328 err = ovs_flow_tbl_flush(&dp->table);
1329 goto unlock;
1330 }
1331
1332 if (ufid_present)
1333 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1334 else
1335 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1336 if (unlikely(!flow)) {
1337 err = -ENOENT;
1338 goto unlock;
1339 }
1340
1341 ovs_flow_tbl_remove(&dp->table, flow);
1342 ovs_unlock();
1343
1344 reply = ovs_flow_cmd_alloc_info(rcu_dereference_raw(flow->sf_acts),
1345 &flow->id, info, false, ufid_flags);
1346
1347 if (likely(reply)) {
1348 if (likely(!IS_ERR(reply))) {
1349 rcu_read_lock(); /*To keep RCU checker happy. */
1350 err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1351 reply, info->snd_portid,
1352 info->snd_seq, 0,
1353 OVS_FLOW_CMD_DEL,
1354 ufid_flags);
1355 rcu_read_unlock();
1356 BUG_ON(err < 0);
1357 ovs_notify(&dp_flow_genl_family, &ovs_dp_flow_multicast_group, reply, info);
1358 } else {
1359 genl_set_err(&dp_flow_genl_family, sock_net(skb->sk), 0,
1360 GROUP_ID(&ovs_dp_flow_multicast_group), PTR_ERR(reply));
1361
1362 }
1363 }
1364
1365 ovs_flow_free(flow, true);
1366 return 0;
1367 unlock:
1368 ovs_unlock();
1369 return err;
1370 }
1371
1372 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1373 {
1374 struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1375 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1376 struct table_instance *ti;
1377 struct datapath *dp;
1378 u32 ufid_flags;
1379 int err;
1380
1381 err = genlmsg_parse(cb->nlh, &dp_flow_genl_family, a,
1382 OVS_FLOW_ATTR_MAX, flow_policy, NULL);
1383 if (err)
1384 return err;
1385 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1386
1387 rcu_read_lock();
1388 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1389 if (!dp) {
1390 rcu_read_unlock();
1391 return -ENODEV;
1392 }
1393
1394 ti = rcu_dereference(dp->table.ti);
1395 for (;;) {
1396 struct sw_flow *flow;
1397 u32 bucket, obj;
1398
1399 bucket = cb->args[0];
1400 obj = cb->args[1];
1401 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1402 if (!flow)
1403 break;
1404
1405 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1406 NETLINK_CB(cb->skb).portid,
1407 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1408 OVS_FLOW_CMD_NEW, ufid_flags) < 0)
1409 break;
1410
1411 cb->args[0] = bucket;
1412 cb->args[1] = obj;
1413 }
1414 rcu_read_unlock();
1415 return skb->len;
1416 }
1417
1418 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1419 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1420 [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1421 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1422 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1423 [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1424 [OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1425 [OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1426 };
1427
1428 static struct genl_ops dp_flow_genl_ops[] = {
1429 { .cmd = OVS_FLOW_CMD_NEW,
1430 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1431 .policy = flow_policy,
1432 .doit = ovs_flow_cmd_new
1433 },
1434 { .cmd = OVS_FLOW_CMD_DEL,
1435 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1436 .policy = flow_policy,
1437 .doit = ovs_flow_cmd_del
1438 },
1439 { .cmd = OVS_FLOW_CMD_GET,
1440 .flags = 0, /* OK for unprivileged users. */
1441 .policy = flow_policy,
1442 .doit = ovs_flow_cmd_get,
1443 .dumpit = ovs_flow_cmd_dump
1444 },
1445 { .cmd = OVS_FLOW_CMD_SET,
1446 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1447 .policy = flow_policy,
1448 .doit = ovs_flow_cmd_set,
1449 },
1450 };
1451
1452 static struct genl_family dp_flow_genl_family __ro_after_init = {
1453 .hdrsize = sizeof(struct ovs_header),
1454 .name = OVS_FLOW_FAMILY,
1455 .version = OVS_FLOW_VERSION,
1456 .maxattr = OVS_FLOW_ATTR_MAX,
1457 .netnsok = true,
1458 .parallel_ops = true,
1459 .ops = dp_flow_genl_ops,
1460 .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1461 .mcgrps = &ovs_dp_flow_multicast_group,
1462 .n_mcgrps = 1,
1463 .module = THIS_MODULE,
1464 };
1465
1466 static size_t ovs_dp_cmd_msg_size(void)
1467 {
1468 size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1469
1470 msgsize += nla_total_size(IFNAMSIZ);
1471 msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1472 msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
1473 msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1474
1475 return msgsize;
1476 }
1477
1478 /* Called with ovs_mutex. */
1479 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1480 u32 portid, u32 seq, u32 flags, u8 cmd)
1481 {
1482 struct ovs_header *ovs_header;
1483 struct ovs_dp_stats dp_stats;
1484 struct ovs_dp_megaflow_stats dp_megaflow_stats;
1485 int err;
1486
1487 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1488 flags, cmd);
1489 if (!ovs_header)
1490 goto error;
1491
1492 ovs_header->dp_ifindex = get_dpifindex(dp);
1493
1494 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1495 if (err)
1496 goto nla_put_failure;
1497
1498 get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1499 if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1500 &dp_stats, OVS_DP_ATTR_PAD))
1501 goto nla_put_failure;
1502
1503 if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1504 sizeof(struct ovs_dp_megaflow_stats),
1505 &dp_megaflow_stats, OVS_DP_ATTR_PAD))
1506 goto nla_put_failure;
1507
1508 if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1509 goto nla_put_failure;
1510
1511 genlmsg_end(skb, ovs_header);
1512 return 0;
1513
1514 nla_put_failure:
1515 genlmsg_cancel(skb, ovs_header);
1516 error:
1517 return -EMSGSIZE;
1518 }
1519
1520 static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1521 {
1522 return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1523 }
1524
1525 /* Called with rcu_read_lock or ovs_mutex. */
1526 static struct datapath *lookup_datapath(struct net *net,
1527 const struct ovs_header *ovs_header,
1528 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1529 {
1530 struct datapath *dp;
1531
1532 if (!a[OVS_DP_ATTR_NAME])
1533 dp = get_dp(net, ovs_header->dp_ifindex);
1534 else {
1535 struct vport *vport;
1536
1537 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1538 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1539 }
1540 return dp ? dp : ERR_PTR(-ENODEV);
1541 }
1542
1543 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1544 {
1545 struct datapath *dp;
1546
1547 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1548 if (IS_ERR(dp))
1549 return;
1550
1551 WARN(dp->user_features, "Dropping previously announced user features\n");
1552 dp->user_features = 0;
1553 }
1554
1555 static void ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1556 {
1557 if (a[OVS_DP_ATTR_USER_FEATURES])
1558 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1559 }
1560
1561 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1562 {
1563 struct nlattr **a = info->attrs;
1564 struct vport_parms parms;
1565 struct sk_buff *reply;
1566 struct datapath *dp;
1567 struct vport *vport;
1568 struct ovs_net *ovs_net;
1569 int err, i;
1570
1571 err = -EINVAL;
1572 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1573 goto err;
1574
1575 reply = ovs_dp_cmd_alloc_info();
1576 if (!reply)
1577 return -ENOMEM;
1578
1579 err = -ENOMEM;
1580 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1581 if (dp == NULL)
1582 goto err_free_reply;
1583
1584 ovs_dp_set_net(dp, sock_net(skb->sk));
1585
1586 /* Allocate table. */
1587 err = ovs_flow_tbl_init(&dp->table);
1588 if (err)
1589 goto err_free_dp;
1590
1591 dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1592 if (!dp->stats_percpu) {
1593 err = -ENOMEM;
1594 goto err_destroy_table;
1595 }
1596
1597 dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1598 GFP_KERNEL);
1599 if (!dp->ports) {
1600 err = -ENOMEM;
1601 goto err_destroy_percpu;
1602 }
1603
1604 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1605 INIT_HLIST_HEAD(&dp->ports[i]);
1606
1607 err = ovs_meters_init(dp);
1608 if (err)
1609 goto err_destroy_ports_array;
1610
1611 /* Set up our datapath device. */
1612 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1613 parms.type = OVS_VPORT_TYPE_INTERNAL;
1614 parms.options = NULL;
1615 parms.dp = dp;
1616 parms.port_no = OVSP_LOCAL;
1617 parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1618
1619 ovs_dp_change(dp, a);
1620
1621 /* So far only local changes have been made, now need the lock. */
1622 ovs_lock();
1623
1624 vport = new_vport(&parms);
1625 if (IS_ERR(vport)) {
1626 err = PTR_ERR(vport);
1627 if (err == -EBUSY)
1628 err = -EEXIST;
1629
1630 if (err == -EEXIST) {
1631 /* An outdated user space instance that does not understand
1632 * the concept of user_features has attempted to create a new
1633 * datapath and is likely to reuse it. Drop all user features.
1634 */
1635 if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1636 ovs_dp_reset_user_features(skb, info);
1637 }
1638
1639 goto err_destroy_meters;
1640 }
1641
1642 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1643 info->snd_seq, 0, OVS_DP_CMD_NEW);
1644 BUG_ON(err < 0);
1645
1646 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1647 list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1648
1649 ovs_unlock();
1650
1651 ovs_notify(&dp_datapath_genl_family, &ovs_dp_datapath_multicast_group, reply, info);
1652 return 0;
1653
1654 err_destroy_meters:
1655 ovs_unlock();
1656 ovs_meters_exit(dp);
1657 err_destroy_ports_array:
1658 kfree(dp->ports);
1659 err_destroy_percpu:
1660 free_percpu(dp->stats_percpu);
1661 err_destroy_table:
1662 ovs_flow_tbl_destroy(&dp->table);
1663 err_free_dp:
1664 kfree(dp);
1665 err_free_reply:
1666 kfree_skb(reply);
1667 err:
1668 return err;
1669 }
1670
1671 /* Called with ovs_mutex. */
1672 static void __dp_destroy(struct datapath *dp)
1673 {
1674 int i;
1675
1676 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1677 struct vport *vport;
1678 struct hlist_node *n;
1679
1680 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1681 if (vport->port_no != OVSP_LOCAL)
1682 ovs_dp_detach_port(vport);
1683 }
1684
1685 list_del_rcu(&dp->list_node);
1686
1687 /* OVSP_LOCAL is datapath internal port. We need to make sure that
1688 * all ports in datapath are destroyed first before freeing datapath.
1689 */
1690 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1691
1692 /* RCU destroy the flow table */
1693 call_rcu(&dp->rcu, destroy_dp_rcu);
1694 }
1695
1696 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1697 {
1698 struct sk_buff *reply;
1699 struct datapath *dp;
1700 int err;
1701
1702 reply = ovs_dp_cmd_alloc_info();
1703 if (!reply)
1704 return -ENOMEM;
1705
1706 ovs_lock();
1707 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1708 err = PTR_ERR(dp);
1709 if (IS_ERR(dp))
1710 goto err_unlock_free;
1711
1712 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1713 info->snd_seq, 0, OVS_DP_CMD_DEL);
1714 BUG_ON(err < 0);
1715
1716 __dp_destroy(dp);
1717 ovs_unlock();
1718
1719 ovs_notify(&dp_datapath_genl_family, &ovs_dp_datapath_multicast_group, reply, info);
1720 return 0;
1721
1722 err_unlock_free:
1723 ovs_unlock();
1724 kfree_skb(reply);
1725 return err;
1726 }
1727
1728 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1729 {
1730 struct sk_buff *reply;
1731 struct datapath *dp;
1732 int err;
1733
1734 reply = ovs_dp_cmd_alloc_info();
1735 if (!reply)
1736 return -ENOMEM;
1737
1738 ovs_lock();
1739 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1740 err = PTR_ERR(dp);
1741 if (IS_ERR(dp))
1742 goto err_unlock_free;
1743
1744 ovs_dp_change(dp, info->attrs);
1745
1746 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1747 info->snd_seq, 0, OVS_DP_CMD_NEW);
1748 BUG_ON(err < 0);
1749
1750 ovs_unlock();
1751
1752 ovs_notify(&dp_datapath_genl_family, &ovs_dp_datapath_multicast_group, reply, info);
1753 return 0;
1754
1755 err_unlock_free:
1756 ovs_unlock();
1757 kfree_skb(reply);
1758 return err;
1759 }
1760
1761 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1762 {
1763 struct sk_buff *reply;
1764 struct datapath *dp;
1765 int err;
1766
1767 reply = ovs_dp_cmd_alloc_info();
1768 if (!reply)
1769 return -ENOMEM;
1770
1771 ovs_lock();
1772 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1773 if (IS_ERR(dp)) {
1774 err = PTR_ERR(dp);
1775 goto err_unlock_free;
1776 }
1777 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1778 info->snd_seq, 0, OVS_DP_CMD_NEW);
1779 BUG_ON(err < 0);
1780 ovs_unlock();
1781
1782 return genlmsg_reply(reply, info);
1783
1784 err_unlock_free:
1785 ovs_unlock();
1786 kfree_skb(reply);
1787 return err;
1788 }
1789
1790 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1791 {
1792 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1793 struct datapath *dp;
1794 int skip = cb->args[0];
1795 int i = 0;
1796
1797 ovs_lock();
1798 list_for_each_entry(dp, &ovs_net->dps, list_node) {
1799 if (i >= skip &&
1800 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1801 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1802 OVS_DP_CMD_NEW) < 0)
1803 break;
1804 i++;
1805 }
1806 ovs_unlock();
1807
1808 cb->args[0] = i;
1809
1810 return skb->len;
1811 }
1812
1813 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1814 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1815 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1816 [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1817 };
1818
1819 static struct genl_ops dp_datapath_genl_ops[] = {
1820 { .cmd = OVS_DP_CMD_NEW,
1821 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1822 .policy = datapath_policy,
1823 .doit = ovs_dp_cmd_new
1824 },
1825 { .cmd = OVS_DP_CMD_DEL,
1826 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1827 .policy = datapath_policy,
1828 .doit = ovs_dp_cmd_del
1829 },
1830 { .cmd = OVS_DP_CMD_GET,
1831 .flags = 0, /* OK for unprivileged users. */
1832 .policy = datapath_policy,
1833 .doit = ovs_dp_cmd_get,
1834 .dumpit = ovs_dp_cmd_dump
1835 },
1836 { .cmd = OVS_DP_CMD_SET,
1837 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1838 .policy = datapath_policy,
1839 .doit = ovs_dp_cmd_set,
1840 },
1841 };
1842
1843 static struct genl_family dp_datapath_genl_family __ro_after_init = {
1844 .hdrsize = sizeof(struct ovs_header),
1845 .name = OVS_DATAPATH_FAMILY,
1846 .version = OVS_DATAPATH_VERSION,
1847 .maxattr = OVS_DP_ATTR_MAX,
1848 .netnsok = true,
1849 .parallel_ops = true,
1850 .ops = dp_datapath_genl_ops,
1851 .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1852 .mcgrps = &ovs_dp_datapath_multicast_group,
1853 .n_mcgrps = 1,
1854 .module = THIS_MODULE,
1855 };
1856
1857 /* Called with ovs_mutex or RCU read lock. */
1858 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1859 struct net *net, u32 portid, u32 seq,
1860 u32 flags, u8 cmd)
1861 {
1862 struct ovs_header *ovs_header;
1863 struct ovs_vport_stats vport_stats;
1864 int err;
1865
1866 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1867 flags, cmd);
1868 if (!ovs_header)
1869 return -EMSGSIZE;
1870
1871 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1872
1873 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1874 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1875 nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1876 ovs_vport_name(vport)) ||
1877 nla_put_u32(skb, OVS_VPORT_ATTR_IFINDEX, vport->dev->ifindex))
1878 goto nla_put_failure;
1879
1880 #ifdef HAVE_PEERNET2ID_ALLOC
1881 if (!net_eq(net, dev_net(vport->dev))) {
1882 int id = peernet2id_alloc(net, dev_net(vport->dev));
1883
1884 if (nla_put_s32(skb, OVS_VPORT_ATTR_NETNSID, id))
1885 goto nla_put_failure;
1886 }
1887
1888 #endif
1889 ovs_vport_get_stats(vport, &vport_stats);
1890 if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
1891 sizeof(struct ovs_vport_stats), &vport_stats,
1892 OVS_VPORT_ATTR_PAD))
1893 goto nla_put_failure;
1894
1895 if (ovs_vport_get_upcall_portids(vport, skb))
1896 goto nla_put_failure;
1897
1898 err = ovs_vport_get_options(vport, skb);
1899 if (err == -EMSGSIZE)
1900 goto error;
1901
1902 genlmsg_end(skb, ovs_header);
1903 return 0;
1904
1905 nla_put_failure:
1906 err = -EMSGSIZE;
1907 error:
1908 genlmsg_cancel(skb, ovs_header);
1909 return err;
1910 }
1911
1912 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1913 {
1914 return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1915 }
1916
1917 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1918 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, struct net *net,
1919 u32 portid, u32 seq, u8 cmd)
1920 {
1921 struct sk_buff *skb;
1922 int retval;
1923
1924 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1925 if (!skb)
1926 return ERR_PTR(-ENOMEM);
1927
1928 retval = ovs_vport_cmd_fill_info(vport, skb, net, portid, seq, 0, cmd);
1929 BUG_ON(retval < 0);
1930
1931 return skb;
1932 }
1933
1934 /* Called with ovs_mutex or RCU read lock. */
1935 static struct vport *lookup_vport(struct net *net,
1936 const struct ovs_header *ovs_header,
1937 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1938 {
1939 struct datapath *dp;
1940 struct vport *vport;
1941
1942 if (a[OVS_VPORT_ATTR_IFINDEX])
1943 return ERR_PTR(-EOPNOTSUPP);
1944 if (a[OVS_VPORT_ATTR_NAME]) {
1945 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1946 if (!vport)
1947 return ERR_PTR(-ENODEV);
1948 if (ovs_header->dp_ifindex &&
1949 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1950 return ERR_PTR(-ENODEV);
1951 return vport;
1952 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1953 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1954
1955 if (port_no >= DP_MAX_PORTS)
1956 return ERR_PTR(-EFBIG);
1957
1958 dp = get_dp(net, ovs_header->dp_ifindex);
1959 if (!dp)
1960 return ERR_PTR(-ENODEV);
1961
1962 vport = ovs_vport_ovsl_rcu(dp, port_no);
1963 if (!vport)
1964 return ERR_PTR(-ENODEV);
1965 return vport;
1966 } else
1967 return ERR_PTR(-EINVAL);
1968
1969 }
1970
1971 /* Called with ovs_mutex */
1972 static void update_headroom(struct datapath *dp)
1973 {
1974 unsigned dev_headroom, max_headroom = 0;
1975 struct net_device *dev;
1976 struct vport *vport;
1977 int i;
1978
1979 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1980 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1981 dev = vport->dev;
1982 dev_headroom = netdev_get_fwd_headroom(dev);
1983 if (dev_headroom > max_headroom)
1984 max_headroom = dev_headroom;
1985 }
1986 }
1987
1988 dp->max_headroom = max_headroom;
1989 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1990 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node)
1991 netdev_set_rx_headroom(vport->dev, max_headroom);
1992 }
1993
1994 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1995 {
1996 struct nlattr **a = info->attrs;
1997 struct ovs_header *ovs_header = info->userhdr;
1998 struct vport_parms parms;
1999 struct sk_buff *reply;
2000 struct vport *vport;
2001 struct datapath *dp;
2002 u32 port_no;
2003 int err;
2004
2005 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
2006 !a[OVS_VPORT_ATTR_UPCALL_PID])
2007 return -EINVAL;
2008 if (a[OVS_VPORT_ATTR_IFINDEX])
2009 return -EOPNOTSUPP;
2010
2011 port_no = a[OVS_VPORT_ATTR_PORT_NO]
2012 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
2013 if (port_no >= DP_MAX_PORTS)
2014 return -EFBIG;
2015
2016 reply = ovs_vport_cmd_alloc_info();
2017 if (!reply)
2018 return -ENOMEM;
2019
2020 ovs_lock();
2021 restart:
2022 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2023 err = -ENODEV;
2024 if (!dp)
2025 goto exit_unlock_free;
2026
2027 if (port_no) {
2028 vport = ovs_vport_ovsl(dp, port_no);
2029 err = -EBUSY;
2030 if (vport)
2031 goto exit_unlock_free;
2032 } else {
2033 for (port_no = 1; ; port_no++) {
2034 if (port_no >= DP_MAX_PORTS) {
2035 err = -EFBIG;
2036 goto exit_unlock_free;
2037 }
2038 vport = ovs_vport_ovsl(dp, port_no);
2039 if (!vport)
2040 break;
2041 }
2042 }
2043
2044 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2045 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2046 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2047 parms.dp = dp;
2048 parms.port_no = port_no;
2049 parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
2050
2051 vport = new_vport(&parms);
2052 err = PTR_ERR(vport);
2053 if (IS_ERR(vport)) {
2054 if (err == -EAGAIN)
2055 goto restart;
2056 goto exit_unlock_free;
2057 }
2058
2059 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2060 info->snd_portid, info->snd_seq, 0,
2061 OVS_VPORT_CMD_NEW);
2062 BUG_ON(err < 0);
2063
2064 if (netdev_get_fwd_headroom(vport->dev) > dp->max_headroom)
2065 update_headroom(dp);
2066 else
2067 netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2068
2069 ovs_unlock();
2070
2071 ovs_notify(&dp_vport_genl_family, &ovs_dp_vport_multicast_group, reply, info);
2072 return 0;
2073
2074 exit_unlock_free:
2075 ovs_unlock();
2076 kfree_skb(reply);
2077 return err;
2078 }
2079
2080 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2081 {
2082 struct nlattr **a = info->attrs;
2083 struct sk_buff *reply;
2084 struct vport *vport;
2085 int err;
2086
2087 reply = ovs_vport_cmd_alloc_info();
2088 if (!reply)
2089 return -ENOMEM;
2090
2091 ovs_lock();
2092 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2093 err = PTR_ERR(vport);
2094 if (IS_ERR(vport))
2095 goto exit_unlock_free;
2096
2097 if (a[OVS_VPORT_ATTR_TYPE] &&
2098 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2099 err = -EINVAL;
2100 goto exit_unlock_free;
2101 }
2102
2103 if (a[OVS_VPORT_ATTR_OPTIONS]) {
2104 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2105 if (err)
2106 goto exit_unlock_free;
2107 }
2108
2109 if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2110 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2111
2112 err = ovs_vport_set_upcall_portids(vport, ids);
2113 if (err)
2114 goto exit_unlock_free;
2115 }
2116
2117 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2118 info->snd_portid, info->snd_seq, 0,
2119 OVS_VPORT_CMD_NEW);
2120 BUG_ON(err < 0);
2121 ovs_unlock();
2122
2123 ovs_notify(&dp_vport_genl_family, &ovs_dp_vport_multicast_group, reply, info);
2124 return 0;
2125
2126 exit_unlock_free:
2127 ovs_unlock();
2128 kfree_skb(reply);
2129 return err;
2130 }
2131
2132 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2133 {
2134 bool must_update_headroom = false;
2135 struct nlattr **a = info->attrs;
2136 struct sk_buff *reply;
2137 struct datapath *dp;
2138 struct vport *vport;
2139 int err;
2140
2141 reply = ovs_vport_cmd_alloc_info();
2142 if (!reply)
2143 return -ENOMEM;
2144
2145 ovs_lock();
2146 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2147 err = PTR_ERR(vport);
2148 if (IS_ERR(vport))
2149 goto exit_unlock_free;
2150
2151 if (vport->port_no == OVSP_LOCAL) {
2152 err = -EINVAL;
2153 goto exit_unlock_free;
2154 }
2155
2156 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2157 info->snd_portid, info->snd_seq, 0,
2158 OVS_VPORT_CMD_DEL);
2159 BUG_ON(err < 0);
2160
2161 /* the vport deletion may trigger dp headroom update */
2162 dp = vport->dp;
2163 if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2164 must_update_headroom = true;
2165 netdev_reset_rx_headroom(vport->dev);
2166 ovs_dp_detach_port(vport);
2167
2168 if (must_update_headroom)
2169 update_headroom(dp);
2170
2171 ovs_unlock();
2172
2173 ovs_notify(&dp_vport_genl_family, &ovs_dp_vport_multicast_group, reply, info);
2174 return 0;
2175
2176 exit_unlock_free:
2177 ovs_unlock();
2178 kfree_skb(reply);
2179 return err;
2180 }
2181
2182 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2183 {
2184 struct nlattr **a = info->attrs;
2185 struct ovs_header *ovs_header = info->userhdr;
2186 struct sk_buff *reply;
2187 struct vport *vport;
2188 int err;
2189
2190 reply = ovs_vport_cmd_alloc_info();
2191 if (!reply)
2192 return -ENOMEM;
2193
2194 rcu_read_lock();
2195 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2196 err = PTR_ERR(vport);
2197 if (IS_ERR(vport))
2198 goto exit_unlock_free;
2199 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2200 info->snd_portid, info->snd_seq, 0,
2201 OVS_VPORT_CMD_NEW);
2202 BUG_ON(err < 0);
2203 rcu_read_unlock();
2204
2205 return genlmsg_reply(reply, info);
2206
2207 exit_unlock_free:
2208 rcu_read_unlock();
2209 kfree_skb(reply);
2210 return err;
2211 }
2212
2213 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2214 {
2215 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2216 struct datapath *dp;
2217 int bucket = cb->args[0], skip = cb->args[1];
2218 int i, j = 0;
2219
2220 rcu_read_lock();
2221 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2222 if (!dp) {
2223 rcu_read_unlock();
2224 return -ENODEV;
2225 }
2226 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2227 struct vport *vport;
2228
2229 j = 0;
2230 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2231 if (j >= skip &&
2232 ovs_vport_cmd_fill_info(vport, skb,
2233 sock_net(skb->sk),
2234 NETLINK_CB(cb->skb).portid,
2235 cb->nlh->nlmsg_seq,
2236 NLM_F_MULTI,
2237 OVS_VPORT_CMD_NEW) < 0)
2238 goto out;
2239
2240 j++;
2241 }
2242 skip = 0;
2243 }
2244 out:
2245 rcu_read_unlock();
2246
2247 cb->args[0] = i;
2248 cb->args[1] = j;
2249
2250 return skb->len;
2251 }
2252
2253 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2254 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2255 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2256 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2257 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2258 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
2259 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2260 [OVS_VPORT_ATTR_IFINDEX] = { .type = NLA_U32 },
2261 [OVS_VPORT_ATTR_NETNSID] = { .type = NLA_S32 },
2262 };
2263
2264 static struct genl_ops dp_vport_genl_ops[] = {
2265 { .cmd = OVS_VPORT_CMD_NEW,
2266 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2267 .policy = vport_policy,
2268 .doit = ovs_vport_cmd_new
2269 },
2270 { .cmd = OVS_VPORT_CMD_DEL,
2271 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2272 .policy = vport_policy,
2273 .doit = ovs_vport_cmd_del
2274 },
2275 { .cmd = OVS_VPORT_CMD_GET,
2276 .flags = 0, /* OK for unprivileged users. */
2277 .policy = vport_policy,
2278 .doit = ovs_vport_cmd_get,
2279 .dumpit = ovs_vport_cmd_dump
2280 },
2281 { .cmd = OVS_VPORT_CMD_SET,
2282 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2283 .policy = vport_policy,
2284 .doit = ovs_vport_cmd_set,
2285 },
2286 };
2287
2288 struct genl_family dp_vport_genl_family __ro_after_init = {
2289 .hdrsize = sizeof(struct ovs_header),
2290 .name = OVS_VPORT_FAMILY,
2291 .version = OVS_VPORT_VERSION,
2292 .maxattr = OVS_VPORT_ATTR_MAX,
2293 .netnsok = true,
2294 .parallel_ops = true,
2295 .ops = dp_vport_genl_ops,
2296 .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2297 .mcgrps = &ovs_dp_vport_multicast_group,
2298 .n_mcgrps = 1,
2299 .module = THIS_MODULE,
2300 };
2301
2302 static struct genl_family *dp_genl_families[] = {
2303 &dp_datapath_genl_family,
2304 &dp_vport_genl_family,
2305 &dp_flow_genl_family,
2306 &dp_packet_genl_family,
2307 &dp_meter_genl_family,
2308 };
2309
2310 static void dp_unregister_genl(int n_families)
2311 {
2312 int i;
2313
2314 for (i = 0; i < n_families; i++)
2315 genl_unregister_family(dp_genl_families[i]);
2316 }
2317
2318 static int __init dp_register_genl(void)
2319 {
2320 int err;
2321 int i;
2322
2323 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2324
2325 err = genl_register_family(dp_genl_families[i]);
2326 if (err)
2327 goto error;
2328 }
2329
2330 return 0;
2331
2332 error:
2333 dp_unregister_genl(i);
2334 return err;
2335 }
2336
2337 static int __net_init ovs_init_net(struct net *net)
2338 {
2339 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2340
2341 INIT_LIST_HEAD(&ovs_net->dps);
2342 INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2343 ovs_ct_init(net);
2344 ovs_netns_frags_init(net);
2345 ovs_netns_frags6_init(net);
2346 return 0;
2347 }
2348
2349 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2350 struct list_head *head)
2351 {
2352 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2353 struct datapath *dp;
2354
2355 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2356 int i;
2357
2358 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2359 struct vport *vport;
2360
2361 hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2362
2363 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2364 continue;
2365
2366 if (dev_net(vport->dev) == dnet)
2367 list_add(&vport->detach_list, head);
2368 }
2369 }
2370 }
2371 }
2372
2373 static void __net_exit ovs_exit_net(struct net *dnet)
2374 {
2375 struct datapath *dp, *dp_next;
2376 struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2377 struct vport *vport, *vport_next;
2378 struct net *net;
2379 LIST_HEAD(head);
2380
2381 ovs_netns_frags6_exit(dnet);
2382 ovs_netns_frags_exit(dnet);
2383 ovs_ct_exit(dnet);
2384 ovs_lock();
2385 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2386 __dp_destroy(dp);
2387
2388 rtnl_lock();
2389 for_each_net(net)
2390 list_vports_from_net(net, dnet, &head);
2391 rtnl_unlock();
2392
2393 /* Detach all vports from given namespace. */
2394 list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2395 list_del(&vport->detach_list);
2396 ovs_dp_detach_port(vport);
2397 }
2398
2399 ovs_unlock();
2400
2401 cancel_work_sync(&ovs_net->dp_notify_work);
2402 }
2403
2404 static struct pernet_operations ovs_net_ops = {
2405 .init = ovs_init_net,
2406 .exit = ovs_exit_net,
2407 .id = &ovs_net_id,
2408 .size = sizeof(struct ovs_net),
2409 };
2410
2411 static int __init dp_init(void)
2412 {
2413 int err;
2414
2415 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2416
2417 pr_info("Open vSwitch switching datapath %s\n", VERSION);
2418
2419 ovs_nsh_init();
2420 err = action_fifos_init();
2421 if (err)
2422 goto error;
2423
2424 err = ovs_internal_dev_rtnl_link_register();
2425 if (err)
2426 goto error_action_fifos_exit;
2427
2428 err = ovs_flow_init();
2429 if (err)
2430 goto error_unreg_rtnl_link;
2431
2432 err = ovs_vport_init();
2433 if (err)
2434 goto error_flow_exit;
2435
2436 err = register_pernet_device(&ovs_net_ops);
2437 if (err)
2438 goto error_vport_exit;
2439
2440 err = compat_init();
2441 if (err)
2442 goto error_netns_exit;
2443
2444 err = register_netdevice_notifier(&ovs_dp_device_notifier);
2445 if (err)
2446 goto error_compat_exit;
2447
2448 err = ovs_netdev_init();
2449 if (err)
2450 goto error_unreg_notifier;
2451
2452 err = dp_register_genl();
2453 if (err < 0)
2454 goto error_unreg_netdev;
2455
2456 return 0;
2457
2458 error_unreg_netdev:
2459 ovs_netdev_exit();
2460 error_unreg_notifier:
2461 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2462 error_compat_exit:
2463 compat_exit();
2464 error_netns_exit:
2465 unregister_pernet_device(&ovs_net_ops);
2466 error_vport_exit:
2467 ovs_vport_exit();
2468 error_flow_exit:
2469 ovs_flow_exit();
2470 error_unreg_rtnl_link:
2471 ovs_internal_dev_rtnl_link_unregister();
2472 error_action_fifos_exit:
2473 action_fifos_exit();
2474 error:
2475 ovs_nsh_cleanup();
2476 return err;
2477 }
2478
2479 static void dp_cleanup(void)
2480 {
2481 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2482 ovs_netdev_exit();
2483 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2484 compat_exit();
2485 unregister_pernet_device(&ovs_net_ops);
2486 rcu_barrier();
2487 ovs_vport_exit();
2488 ovs_flow_exit();
2489 ovs_internal_dev_rtnl_link_unregister();
2490 action_fifos_exit();
2491 ovs_nsh_cleanup();
2492 }
2493
2494 module_init(dp_init);
2495 module_exit(dp_cleanup);
2496
2497 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2498 MODULE_LICENSE("GPL");
2499 MODULE_VERSION(VERSION);
2500 MODULE_ALIAS_GENL_FAMILY(OVS_DATAPATH_FAMILY);
2501 MODULE_ALIAS_GENL_FAMILY(OVS_VPORT_FAMILY);
2502 MODULE_ALIAS_GENL_FAMILY(OVS_FLOW_FAMILY);
2503 MODULE_ALIAS_GENL_FAMILY(OVS_PACKET_FAMILY);
2504 MODULE_ALIAS_GENL_FAMILY(OVS_METER_FAMILY);
openvswitch packages for PVE