?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
x86/speculation/mds: Fix documentation typo
[linux/fpc-iii.git] / net / openvswitch / datapath.c
blob363dd904733ddf23caefafec764c402243d9ace8
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 unsigned int ovs_net_id __read_mostly;
62
63 static struct genl_family dp_packet_genl_family;
64 static struct genl_family dp_flow_genl_family;
65 static struct genl_family dp_datapath_genl_family;
66
67 static const struct nla_policy flow_policy[];
68
69 static const struct genl_multicast_group ovs_dp_flow_multicast_group = {
70 .name = OVS_FLOW_MCGROUP,
71 };
72
73 static const struct genl_multicast_group ovs_dp_datapath_multicast_group = {
74 .name = OVS_DATAPATH_MCGROUP,
75 };
76
77 static const struct genl_multicast_group ovs_dp_vport_multicast_group = {
78 .name = OVS_VPORT_MCGROUP,
79 };
80
81 /* Check if need to build a reply message.
82 * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
83 static bool ovs_must_notify(struct genl_family *family, struct genl_info *info,
84 unsigned int group)
85 {
86 return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
87 genl_has_listeners(family, genl_info_net(info), group);
88 }
89
90 static void ovs_notify(struct genl_family *family,
91 struct sk_buff *skb, struct genl_info *info)
92 {
93 genl_notify(family, skb, info, 0, GFP_KERNEL);
94 }
95
96 /**
97 * DOC: Locking:
98 *
99 * All writes e.g. Writes to device state (add/remove datapath, port, set
100 * operations on vports, etc.), Writes to other state (flow table
101 * modifications, set miscellaneous datapath parameters, etc.) are protected
102 * by ovs_lock.
103 *
104 * Reads are protected by RCU.
105 *
106 * There are a few special cases (mostly stats) that have their own
107 * synchronization but they nest under all of above and don't interact with
108 * each other.
109 *
110 * The RTNL lock nests inside ovs_mutex.
111 */
112
113 static DEFINE_MUTEX(ovs_mutex);
114
115 void ovs_lock(void)
116 {
117 mutex_lock(&ovs_mutex);
118 }
119
120 void ovs_unlock(void)
121 {
122 mutex_unlock(&ovs_mutex);
123 }
124
125 #ifdef CONFIG_LOCKDEP
126 int lockdep_ovsl_is_held(void)
127 {
128 if (debug_locks)
129 return lockdep_is_held(&ovs_mutex);
130 else
131 return 1;
132 }
133 #endif
134
135 static struct vport *new_vport(const struct vport_parms *);
136 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
137 const struct sw_flow_key *,
138 const struct dp_upcall_info *,
139 uint32_t cutlen);
140 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
141 const struct sw_flow_key *,
142 const struct dp_upcall_info *,
143 uint32_t cutlen);
144
145 /* Must be called with rcu_read_lock. */
146 static struct datapath *get_dp_rcu(struct net *net, int dp_ifindex)
147 {
148 struct net_device *dev = dev_get_by_index_rcu(net, dp_ifindex);
149
150 if (dev) {
151 struct vport *vport = ovs_internal_dev_get_vport(dev);
152 if (vport)
153 return vport->dp;
154 }
155
156 return NULL;
157 }
158
159 /* The caller must hold either ovs_mutex or rcu_read_lock to keep the
160 * returned dp pointer valid.
161 */
162 static inline struct datapath *get_dp(struct net *net, int dp_ifindex)
163 {
164 struct datapath *dp;
165
166 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_ovsl_is_held());
167 rcu_read_lock();
168 dp = get_dp_rcu(net, dp_ifindex);
169 rcu_read_unlock();
170
171 return dp;
172 }
173
174 /* Must be called with rcu_read_lock or ovs_mutex. */
175 const char *ovs_dp_name(const struct datapath *dp)
176 {
177 struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
178 return ovs_vport_name(vport);
179 }
180
181 static int get_dpifindex(const struct datapath *dp)
182 {
183 struct vport *local;
184 int ifindex;
185
186 rcu_read_lock();
187
188 local = ovs_vport_rcu(dp, OVSP_LOCAL);
189 if (local)
190 ifindex = local->dev->ifindex;
191 else
192 ifindex = 0;
193
194 rcu_read_unlock();
195
196 return ifindex;
197 }
198
199 static void destroy_dp_rcu(struct rcu_head *rcu)
200 {
201 struct datapath *dp = container_of(rcu, struct datapath, rcu);
202
203 ovs_flow_tbl_destroy(&dp->table);
204 free_percpu(dp->stats_percpu);
205 kfree(dp->ports);
206 kfree(dp);
207 }
208
209 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
210 u16 port_no)
211 {
212 return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
213 }
214
215 /* Called with ovs_mutex or RCU read lock. */
216 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
217 {
218 struct vport *vport;
219 struct hlist_head *head;
220
221 head = vport_hash_bucket(dp, port_no);
222 hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
223 if (vport->port_no == port_no)
224 return vport;
225 }
226 return NULL;
227 }
228
229 /* Called with ovs_mutex. */
230 static struct vport *new_vport(const struct vport_parms *parms)
231 {
232 struct vport *vport;
233
234 vport = ovs_vport_add(parms);
235 if (!IS_ERR(vport)) {
236 struct datapath *dp = parms->dp;
237 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
238
239 hlist_add_head_rcu(&vport->dp_hash_node, head);
240 }
241 return vport;
242 }
243
244 void ovs_dp_detach_port(struct vport *p)
245 {
246 ASSERT_OVSL();
247
248 /* First drop references to device. */
249 hlist_del_rcu(&p->dp_hash_node);
250
251 /* Then destroy it. */
252 ovs_vport_del(p);
253 }
254
255 /* Must be called with rcu_read_lock. */
256 void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
257 {
258 const struct vport *p = OVS_CB(skb)->input_vport;
259 struct datapath *dp = p->dp;
260 struct sw_flow *flow;
261 struct sw_flow_actions *sf_acts;
262 struct dp_stats_percpu *stats;
263 u64 *stats_counter;
264 u32 n_mask_hit;
265
266 stats = this_cpu_ptr(dp->stats_percpu);
267
268 /* Look up flow. */
269 flow = ovs_flow_tbl_lookup_stats(&dp->table, key, &n_mask_hit);
270 if (unlikely(!flow)) {
271 struct dp_upcall_info upcall;
272 int error;
273
274 memset(&upcall, 0, sizeof(upcall));
275 upcall.cmd = OVS_PACKET_CMD_MISS;
276 upcall.portid = ovs_vport_find_upcall_portid(p, skb);
277 upcall.mru = OVS_CB(skb)->mru;
278 error = ovs_dp_upcall(dp, skb, key, &upcall, 0);
279 if (unlikely(error))
280 kfree_skb(skb);
281 else
282 consume_skb(skb);
283 stats_counter = &stats->n_missed;
284 goto out;
285 }
286
287 ovs_flow_stats_update(flow, key->tp.flags, skb);
288 sf_acts = rcu_dereference(flow->sf_acts);
289 ovs_execute_actions(dp, skb, sf_acts, key);
290
291 stats_counter = &stats->n_hit;
292
293 out:
294 /* Update datapath statistics. */
295 u64_stats_update_begin(&stats->syncp);
296 (*stats_counter)++;
297 stats->n_mask_hit += n_mask_hit;
298 u64_stats_update_end(&stats->syncp);
299 }
300
301 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
302 const struct sw_flow_key *key,
303 const struct dp_upcall_info *upcall_info,
304 uint32_t cutlen)
305 {
306 struct dp_stats_percpu *stats;
307 int err;
308
309 if (upcall_info->portid == 0) {
310 err = -ENOTCONN;
311 goto err;
312 }
313
314 if (!skb_is_gso(skb))
315 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
316 else
317 err = queue_gso_packets(dp, skb, key, upcall_info, cutlen);
318 if (err)
319 goto err;
320
321 return 0;
322
323 err:
324 stats = this_cpu_ptr(dp->stats_percpu);
325
326 u64_stats_update_begin(&stats->syncp);
327 stats->n_lost++;
328 u64_stats_update_end(&stats->syncp);
329
330 return err;
331 }
332
333 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
334 const struct sw_flow_key *key,
335 const struct dp_upcall_info *upcall_info,
336 uint32_t cutlen)
337 {
338 unsigned int gso_type = skb_shinfo(skb)->gso_type;
339 struct sw_flow_key later_key;
340 struct sk_buff *segs, *nskb;
341 int err;
342
343 BUILD_BUG_ON(sizeof(*OVS_CB(skb)) > SKB_SGO_CB_OFFSET);
344 segs = __skb_gso_segment(skb, NETIF_F_SG, false);
345 if (IS_ERR(segs))
346 return PTR_ERR(segs);
347 if (segs == NULL)
348 return -EINVAL;
349
350 if (gso_type & SKB_GSO_UDP) {
351 /* The initial flow key extracted by ovs_flow_key_extract()
352 * in this case is for a first fragment, so we need to
353 * properly mark later fragments.
354 */
355 later_key = *key;
356 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
357 }
358
359 /* Queue all of the segments. */
360 skb = segs;
361 do {
362 if (gso_type & SKB_GSO_UDP && skb != segs)
363 key = &later_key;
364
365 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
366 if (err)
367 break;
368
369 } while ((skb = skb->next));
370
371 /* Free all of the segments. */
372 skb = segs;
373 do {
374 nskb = skb->next;
375 if (err)
376 kfree_skb(skb);
377 else
378 consume_skb(skb);
379 } while ((skb = nskb));
380 return err;
381 }
382
383 static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
384 unsigned int hdrlen, int actions_attrlen)
385 {
386 size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
387 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
388 + nla_total_size(ovs_key_attr_size()) /* OVS_PACKET_ATTR_KEY */
389 + nla_total_size(sizeof(unsigned int)); /* OVS_PACKET_ATTR_LEN */
390
391 /* OVS_PACKET_ATTR_USERDATA */
392 if (upcall_info->userdata)
393 size += NLA_ALIGN(upcall_info->userdata->nla_len);
394
395 /* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
396 if (upcall_info->egress_tun_info)
397 size += nla_total_size(ovs_tun_key_attr_size());
398
399 /* OVS_PACKET_ATTR_ACTIONS */
400 if (upcall_info->actions_len)
401 size += nla_total_size(actions_attrlen);
402
403 /* OVS_PACKET_ATTR_MRU */
404 if (upcall_info->mru)
405 size += nla_total_size(sizeof(upcall_info->mru));
406
407 return size;
408 }
409
410 static void pad_packet(struct datapath *dp, struct sk_buff *skb)
411 {
412 if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
413 size_t plen = NLA_ALIGN(skb->len) - skb->len;
414
415 if (plen > 0)
416 skb_put_zero(skb, plen);
417 }
418 }
419
420 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
421 const struct sw_flow_key *key,
422 const struct dp_upcall_info *upcall_info,
423 uint32_t cutlen)
424 {
425 struct ovs_header *upcall;
426 struct sk_buff *nskb = NULL;
427 struct sk_buff *user_skb = NULL; /* to be queued to userspace */
428 struct nlattr *nla;
429 size_t len;
430 unsigned int hlen;
431 int err, dp_ifindex;
432
433 dp_ifindex = get_dpifindex(dp);
434 if (!dp_ifindex)
435 return -ENODEV;
436
437 if (skb_vlan_tag_present(skb)) {
438 nskb = skb_clone(skb, GFP_ATOMIC);
439 if (!nskb)
440 return -ENOMEM;
441
442 nskb = __vlan_hwaccel_push_inside(nskb);
443 if (!nskb)
444 return -ENOMEM;
445
446 skb = nskb;
447 }
448
449 if (nla_attr_size(skb->len) > USHRT_MAX) {
450 err = -EFBIG;
451 goto out;
452 }
453
454 /* Complete checksum if needed */
455 if (skb->ip_summed == CHECKSUM_PARTIAL &&
456 (err = skb_csum_hwoffload_help(skb, 0)))
457 goto out;
458
459 /* Older versions of OVS user space enforce alignment of the last
460 * Netlink attribute to NLA_ALIGNTO which would require extensive
461 * padding logic. Only perform zerocopy if padding is not required.
462 */
463 if (dp->user_features & OVS_DP_F_UNALIGNED)
464 hlen = skb_zerocopy_headlen(skb);
465 else
466 hlen = skb->len;
467
468 len = upcall_msg_size(upcall_info, hlen - cutlen,
469 OVS_CB(skb)->acts_origlen);
470 user_skb = genlmsg_new(len, GFP_ATOMIC);
471 if (!user_skb) {
472 err = -ENOMEM;
473 goto out;
474 }
475
476 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
477 0, upcall_info->cmd);
478 upcall->dp_ifindex = dp_ifindex;
479
480 err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
481 BUG_ON(err);
482
483 if (upcall_info->userdata)
484 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
485 nla_len(upcall_info->userdata),
486 nla_data(upcall_info->userdata));
487
488 if (upcall_info->egress_tun_info) {
489 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_EGRESS_TUN_KEY);
490 err = ovs_nla_put_tunnel_info(user_skb,
491 upcall_info->egress_tun_info);
492 BUG_ON(err);
493 nla_nest_end(user_skb, nla);
494 }
495
496 if (upcall_info->actions_len) {
497 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_ACTIONS);
498 err = ovs_nla_put_actions(upcall_info->actions,
499 upcall_info->actions_len,
500 user_skb);
501 if (!err)
502 nla_nest_end(user_skb, nla);
503 else
504 nla_nest_cancel(user_skb, nla);
505 }
506
507 /* Add OVS_PACKET_ATTR_MRU */
508 if (upcall_info->mru) {
509 if (nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU,
510 upcall_info->mru)) {
511 err = -ENOBUFS;
512 goto out;
513 }
514 pad_packet(dp, user_skb);
515 }
516
517 /* Add OVS_PACKET_ATTR_LEN when packet is truncated */
518 if (cutlen > 0) {
519 if (nla_put_u32(user_skb, OVS_PACKET_ATTR_LEN,
520 skb->len)) {
521 err = -ENOBUFS;
522 goto out;
523 }
524 pad_packet(dp, user_skb);
525 }
526
527 /* Only reserve room for attribute header, packet data is added
528 * in skb_zerocopy() */
529 if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
530 err = -ENOBUFS;
531 goto out;
532 }
533 nla->nla_len = nla_attr_size(skb->len - cutlen);
534
535 err = skb_zerocopy(user_skb, skb, skb->len - cutlen, hlen);
536 if (err)
537 goto out;
538
539 /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
540 pad_packet(dp, user_skb);
541
542 ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
543
544 err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
545 user_skb = NULL;
546 out:
547 if (err)
548 skb_tx_error(skb);
549 kfree_skb(user_skb);
550 kfree_skb(nskb);
551 return err;
552 }
553
554 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
555 {
556 struct ovs_header *ovs_header = info->userhdr;
557 struct net *net = sock_net(skb->sk);
558 struct nlattr **a = info->attrs;
559 struct sw_flow_actions *acts;
560 struct sk_buff *packet;
561 struct sw_flow *flow;
562 struct sw_flow_actions *sf_acts;
563 struct datapath *dp;
564 struct vport *input_vport;
565 u16 mru = 0;
566 int len;
567 int err;
568 bool log = !a[OVS_PACKET_ATTR_PROBE];
569
570 err = -EINVAL;
571 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
572 !a[OVS_PACKET_ATTR_ACTIONS])
573 goto err;
574
575 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
576 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
577 err = -ENOMEM;
578 if (!packet)
579 goto err;
580 skb_reserve(packet, NET_IP_ALIGN);
581
582 nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
583
584 /* Set packet's mru */
585 if (a[OVS_PACKET_ATTR_MRU]) {
586 mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
587 packet->ignore_df = 1;
588 }
589 OVS_CB(packet)->mru = mru;
590
591 /* Build an sw_flow for sending this packet. */
592 flow = ovs_flow_alloc();
593 err = PTR_ERR(flow);
594 if (IS_ERR(flow))
595 goto err_kfree_skb;
596
597 err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
598 packet, &flow->key, log);
599 if (err)
600 goto err_flow_free;
601
602 err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
603 &flow->key, &acts, log);
604 if (err)
605 goto err_flow_free;
606
607 rcu_assign_pointer(flow->sf_acts, acts);
608 packet->priority = flow->key.phy.priority;
609 packet->mark = flow->key.phy.skb_mark;
610
611 rcu_read_lock();
612 dp = get_dp_rcu(net, ovs_header->dp_ifindex);
613 err = -ENODEV;
614 if (!dp)
615 goto err_unlock;
616
617 input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
618 if (!input_vport)
619 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
620
621 if (!input_vport)
622 goto err_unlock;
623
624 packet->dev = input_vport->dev;
625 OVS_CB(packet)->input_vport = input_vport;
626 sf_acts = rcu_dereference(flow->sf_acts);
627
628 local_bh_disable();
629 err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
630 local_bh_enable();
631 rcu_read_unlock();
632
633 ovs_flow_free(flow, false);
634 return err;
635
636 err_unlock:
637 rcu_read_unlock();
638 err_flow_free:
639 ovs_flow_free(flow, false);
640 err_kfree_skb:
641 kfree_skb(packet);
642 err:
643 return err;
644 }
645
646 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
647 [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
648 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
649 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
650 [OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
651 [OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
652 };
653
654 static const struct genl_ops dp_packet_genl_ops[] = {
655 { .cmd = OVS_PACKET_CMD_EXECUTE,
656 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
657 .policy = packet_policy,
658 .doit = ovs_packet_cmd_execute
659 }
660 };
661
662 static struct genl_family dp_packet_genl_family __ro_after_init = {
663 .hdrsize = sizeof(struct ovs_header),
664 .name = OVS_PACKET_FAMILY,
665 .version = OVS_PACKET_VERSION,
666 .maxattr = OVS_PACKET_ATTR_MAX,
667 .netnsok = true,
668 .parallel_ops = true,
669 .ops = dp_packet_genl_ops,
670 .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
671 .module = THIS_MODULE,
672 };
673
674 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
675 struct ovs_dp_megaflow_stats *mega_stats)
676 {
677 int i;
678
679 memset(mega_stats, 0, sizeof(*mega_stats));
680
681 stats->n_flows = ovs_flow_tbl_count(&dp->table);
682 mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
683
684 stats->n_hit = stats->n_missed = stats->n_lost = 0;
685
686 for_each_possible_cpu(i) {
687 const struct dp_stats_percpu *percpu_stats;
688 struct dp_stats_percpu local_stats;
689 unsigned int start;
690
691 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
692
693 do {
694 start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
695 local_stats = *percpu_stats;
696 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
697
698 stats->n_hit += local_stats.n_hit;
699 stats->n_missed += local_stats.n_missed;
700 stats->n_lost += local_stats.n_lost;
701 mega_stats->n_mask_hit += local_stats.n_mask_hit;
702 }
703 }
704
705 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
706 {
707 return ovs_identifier_is_ufid(sfid) &&
708 !(ufid_flags & OVS_UFID_F_OMIT_KEY);
709 }
710
711 static bool should_fill_mask(uint32_t ufid_flags)
712 {
713 return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
714 }
715
716 static bool should_fill_actions(uint32_t ufid_flags)
717 {
718 return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
719 }
720
721 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
722 const struct sw_flow_id *sfid,
723 uint32_t ufid_flags)
724 {
725 size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
726
727 /* OVS_FLOW_ATTR_UFID */
728 if (sfid && ovs_identifier_is_ufid(sfid))
729 len += nla_total_size(sfid->ufid_len);
730
731 /* OVS_FLOW_ATTR_KEY */
732 if (!sfid || should_fill_key(sfid, ufid_flags))
733 len += nla_total_size(ovs_key_attr_size());
734
735 /* OVS_FLOW_ATTR_MASK */
736 if (should_fill_mask(ufid_flags))
737 len += nla_total_size(ovs_key_attr_size());
738
739 /* OVS_FLOW_ATTR_ACTIONS */
740 if (should_fill_actions(ufid_flags))
741 len += nla_total_size(acts->orig_len);
742
743 return len
744 + nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
745 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
746 + nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */
747 }
748
749 /* Called with ovs_mutex or RCU read lock. */
750 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
751 struct sk_buff *skb)
752 {
753 struct ovs_flow_stats stats;
754 __be16 tcp_flags;
755 unsigned long used;
756
757 ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
758
759 if (used &&
760 nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
761 OVS_FLOW_ATTR_PAD))
762 return -EMSGSIZE;
763
764 if (stats.n_packets &&
765 nla_put_64bit(skb, OVS_FLOW_ATTR_STATS,
766 sizeof(struct ovs_flow_stats), &stats,
767 OVS_FLOW_ATTR_PAD))
768 return -EMSGSIZE;
769
770 if ((u8)ntohs(tcp_flags) &&
771 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
772 return -EMSGSIZE;
773
774 return 0;
775 }
776
777 /* Called with ovs_mutex or RCU read lock. */
778 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
779 struct sk_buff *skb, int skb_orig_len)
780 {
781 struct nlattr *start;
782 int err;
783
784 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
785 * this is the first flow to be dumped into 'skb'. This is unusual for
786 * Netlink but individual action lists can be longer than
787 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
788 * The userspace caller can always fetch the actions separately if it
789 * really wants them. (Most userspace callers in fact don't care.)
790 *
791 * This can only fail for dump operations because the skb is always
792 * properly sized for single flows.
793 */
794 start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
795 if (start) {
796 const struct sw_flow_actions *sf_acts;
797
798 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
799 err = ovs_nla_put_actions(sf_acts->actions,
800 sf_acts->actions_len, skb);
801
802 if (!err)
803 nla_nest_end(skb, start);
804 else {
805 if (skb_orig_len)
806 return err;
807
808 nla_nest_cancel(skb, start);
809 }
810 } else if (skb_orig_len) {
811 return -EMSGSIZE;
812 }
813
814 return 0;
815 }
816
817 /* Called with ovs_mutex or RCU read lock. */
818 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
819 struct sk_buff *skb, u32 portid,
820 u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
821 {
822 const int skb_orig_len = skb->len;
823 struct ovs_header *ovs_header;
824 int err;
825
826 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
827 flags, cmd);
828 if (!ovs_header)
829 return -EMSGSIZE;
830
831 ovs_header->dp_ifindex = dp_ifindex;
832
833 err = ovs_nla_put_identifier(flow, skb);
834 if (err)
835 goto error;
836
837 if (should_fill_key(&flow->id, ufid_flags)) {
838 err = ovs_nla_put_masked_key(flow, skb);
839 if (err)
840 goto error;
841 }
842
843 if (should_fill_mask(ufid_flags)) {
844 err = ovs_nla_put_mask(flow, skb);
845 if (err)
846 goto error;
847 }
848
849 err = ovs_flow_cmd_fill_stats(flow, skb);
850 if (err)
851 goto error;
852
853 if (should_fill_actions(ufid_flags)) {
854 err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
855 if (err)
856 goto error;
857 }
858
859 genlmsg_end(skb, ovs_header);
860 return 0;
861
862 error:
863 genlmsg_cancel(skb, ovs_header);
864 return err;
865 }
866
867 /* May not be called with RCU read lock. */
868 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
869 const struct sw_flow_id *sfid,
870 struct genl_info *info,
871 bool always,
872 uint32_t ufid_flags)
873 {
874 struct sk_buff *skb;
875 size_t len;
876
877 if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
878 return NULL;
879
880 len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
881 skb = genlmsg_new(len, GFP_KERNEL);
882 if (!skb)
883 return ERR_PTR(-ENOMEM);
884
885 return skb;
886 }
887
888 /* Called with ovs_mutex. */
889 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
890 int dp_ifindex,
891 struct genl_info *info, u8 cmd,
892 bool always, u32 ufid_flags)
893 {
894 struct sk_buff *skb;
895 int retval;
896
897 skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
898 &flow->id, info, always, ufid_flags);
899 if (IS_ERR_OR_NULL(skb))
900 return skb;
901
902 retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
903 info->snd_portid, info->snd_seq, 0,
904 cmd, ufid_flags);
905 BUG_ON(retval < 0);
906 return skb;
907 }
908
909 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
910 {
911 struct net *net = sock_net(skb->sk);
912 struct nlattr **a = info->attrs;
913 struct ovs_header *ovs_header = info->userhdr;
914 struct sw_flow *flow = NULL, *new_flow;
915 struct sw_flow_mask mask;
916 struct sk_buff *reply;
917 struct datapath *dp;
918 struct sw_flow_actions *acts;
919 struct sw_flow_match match;
920 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
921 int error;
922 bool log = !a[OVS_FLOW_ATTR_PROBE];
923
924 /* Must have key and actions. */
925 error = -EINVAL;
926 if (!a[OVS_FLOW_ATTR_KEY]) {
927 OVS_NLERR(log, "Flow key attr not present in new flow.");
928 goto error;
929 }
930 if (!a[OVS_FLOW_ATTR_ACTIONS]) {
931 OVS_NLERR(log, "Flow actions attr not present in new flow.");
932 goto error;
933 }
934
935 /* Most of the time we need to allocate a new flow, do it before
936 * locking.
937 */
938 new_flow = ovs_flow_alloc();
939 if (IS_ERR(new_flow)) {
940 error = PTR_ERR(new_flow);
941 goto error;
942 }
943
944 /* Extract key. */
945 ovs_match_init(&match, &new_flow->key, false, &mask);
946 error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
947 a[OVS_FLOW_ATTR_MASK], log);
948 if (error)
949 goto err_kfree_flow;
950
951 /* Extract flow identifier. */
952 error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
953 &new_flow->key, log);
954 if (error)
955 goto err_kfree_flow;
956
957 /* unmasked key is needed to match when ufid is not used. */
958 if (ovs_identifier_is_key(&new_flow->id))
959 match.key = new_flow->id.unmasked_key;
960
961 ovs_flow_mask_key(&new_flow->key, &new_flow->key, true, &mask);
962
963 /* Validate actions. */
964 error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
965 &new_flow->key, &acts, log);
966 if (error) {
967 OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
968 goto err_kfree_flow;
969 }
970
971 reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
972 ufid_flags);
973 if (IS_ERR(reply)) {
974 error = PTR_ERR(reply);
975 goto err_kfree_acts;
976 }
977
978 ovs_lock();
979 dp = get_dp(net, ovs_header->dp_ifindex);
980 if (unlikely(!dp)) {
981 error = -ENODEV;
982 goto err_unlock_ovs;
983 }
984
985 /* Check if this is a duplicate flow */
986 if (ovs_identifier_is_ufid(&new_flow->id))
987 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
988 if (!flow)
989 flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->key);
990 if (likely(!flow)) {
991 rcu_assign_pointer(new_flow->sf_acts, acts);
992
993 /* Put flow in bucket. */
994 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
995 if (unlikely(error)) {
996 acts = NULL;
997 goto err_unlock_ovs;
998 }
999
1000 if (unlikely(reply)) {
1001 error = ovs_flow_cmd_fill_info(new_flow,
1002 ovs_header->dp_ifindex,
1003 reply, info->snd_portid,
1004 info->snd_seq, 0,
1005 OVS_FLOW_CMD_NEW,
1006 ufid_flags);
1007 BUG_ON(error < 0);
1008 }
1009 ovs_unlock();
1010 } else {
1011 struct sw_flow_actions *old_acts;
1012
1013 /* Bail out if we're not allowed to modify an existing flow.
1014 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1015 * because Generic Netlink treats the latter as a dump
1016 * request. We also accept NLM_F_EXCL in case that bug ever
1017 * gets fixed.
1018 */
1019 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1020 | NLM_F_EXCL))) {
1021 error = -EEXIST;
1022 goto err_unlock_ovs;
1023 }
1024 /* The flow identifier has to be the same for flow updates.
1025 * Look for any overlapping flow.
1026 */
1027 if (unlikely(!ovs_flow_cmp(flow, &match))) {
1028 if (ovs_identifier_is_key(&flow->id))
1029 flow = ovs_flow_tbl_lookup_exact(&dp->table,
1030 &match);
1031 else /* UFID matches but key is different */
1032 flow = NULL;
1033 if (!flow) {
1034 error = -ENOENT;
1035 goto err_unlock_ovs;
1036 }
1037 }
1038 /* Update actions. */
1039 old_acts = ovsl_dereference(flow->sf_acts);
1040 rcu_assign_pointer(flow->sf_acts, acts);
1041
1042 if (unlikely(reply)) {
1043 error = ovs_flow_cmd_fill_info(flow,
1044 ovs_header->dp_ifindex,
1045 reply, info->snd_portid,
1046 info->snd_seq, 0,
1047 OVS_FLOW_CMD_NEW,
1048 ufid_flags);
1049 BUG_ON(error < 0);
1050 }
1051 ovs_unlock();
1052
1053 ovs_nla_free_flow_actions_rcu(old_acts);
1054 ovs_flow_free(new_flow, false);
1055 }
1056
1057 if (reply)
1058 ovs_notify(&dp_flow_genl_family, reply, info);
1059 return 0;
1060
1061 err_unlock_ovs:
1062 ovs_unlock();
1063 kfree_skb(reply);
1064 err_kfree_acts:
1065 ovs_nla_free_flow_actions(acts);
1066 err_kfree_flow:
1067 ovs_flow_free(new_flow, false);
1068 error:
1069 return error;
1070 }
1071
1072 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1073 static struct sw_flow_actions *get_flow_actions(struct net *net,
1074 const struct nlattr *a,
1075 const struct sw_flow_key *key,
1076 const struct sw_flow_mask *mask,
1077 bool log)
1078 {
1079 struct sw_flow_actions *acts;
1080 struct sw_flow_key masked_key;
1081 int error;
1082
1083 ovs_flow_mask_key(&masked_key, key, true, mask);
1084 error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1085 if (error) {
1086 OVS_NLERR(log,
1087 "Actions may not be safe on all matching packets");
1088 return ERR_PTR(error);
1089 }
1090
1091 return acts;
1092 }
1093
1094 /* Factor out match-init and action-copy to avoid
1095 * "Wframe-larger-than=1024" warning. Because mask is only
1096 * used to get actions, we new a function to save some
1097 * stack space.
1098 *
1099 * If there are not key and action attrs, we return 0
1100 * directly. In the case, the caller will also not use the
1101 * match as before. If there is action attr, we try to get
1102 * actions and save them to *acts. Before returning from
1103 * the function, we reset the match->mask pointer. Because
1104 * we should not to return match object with dangling reference
1105 * to mask.
1106 * */
1107 static int ovs_nla_init_match_and_action(struct net *net,
1108 struct sw_flow_match *match,
1109 struct sw_flow_key *key,
1110 struct nlattr **a,
1111 struct sw_flow_actions **acts,
1112 bool log)
1113 {
1114 struct sw_flow_mask mask;
1115 int error = 0;
1116
1117 if (a[OVS_FLOW_ATTR_KEY]) {
1118 ovs_match_init(match, key, true, &mask);
1119 error = ovs_nla_get_match(net, match, a[OVS_FLOW_ATTR_KEY],
1120 a[OVS_FLOW_ATTR_MASK], log);
1121 if (error)
1122 goto error;
1123 }
1124
1125 if (a[OVS_FLOW_ATTR_ACTIONS]) {
1126 if (!a[OVS_FLOW_ATTR_KEY]) {
1127 OVS_NLERR(log,
1128 "Flow key attribute not present in set flow.");
1129 error = -EINVAL;
1130 goto error;
1131 }
1132
1133 *acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], key,
1134 &mask, log);
1135 if (IS_ERR(*acts)) {
1136 error = PTR_ERR(*acts);
1137 goto error;
1138 }
1139 }
1140
1141 /* On success, error is 0. */
1142 error:
1143 match->mask = NULL;
1144 return error;
1145 }
1146
1147 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1148 {
1149 struct net *net = sock_net(skb->sk);
1150 struct nlattr **a = info->attrs;
1151 struct ovs_header *ovs_header = info->userhdr;
1152 struct sw_flow_key key;
1153 struct sw_flow *flow;
1154 struct sk_buff *reply = NULL;
1155 struct datapath *dp;
1156 struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1157 struct sw_flow_match match;
1158 struct sw_flow_id sfid;
1159 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1160 int error = 0;
1161 bool log = !a[OVS_FLOW_ATTR_PROBE];
1162 bool ufid_present;
1163
1164 ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1165 if (!a[OVS_FLOW_ATTR_KEY] && !ufid_present) {
1166 OVS_NLERR(log,
1167 "Flow set message rejected, Key attribute missing.");
1168 return -EINVAL;
1169 }
1170
1171 error = ovs_nla_init_match_and_action(net, &match, &key, a,
1172 &acts, log);
1173 if (error)
1174 goto error;
1175
1176 if (acts) {
1177 /* Can allocate before locking if have acts. */
1178 reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1179 ufid_flags);
1180 if (IS_ERR(reply)) {
1181 error = PTR_ERR(reply);
1182 goto err_kfree_acts;
1183 }
1184 }
1185
1186 ovs_lock();
1187 dp = get_dp(net, ovs_header->dp_ifindex);
1188 if (unlikely(!dp)) {
1189 error = -ENODEV;
1190 goto err_unlock_ovs;
1191 }
1192 /* Check that the flow exists. */
1193 if (ufid_present)
1194 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1195 else
1196 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1197 if (unlikely(!flow)) {
1198 error = -ENOENT;
1199 goto err_unlock_ovs;
1200 }
1201
1202 /* Update actions, if present. */
1203 if (likely(acts)) {
1204 old_acts = ovsl_dereference(flow->sf_acts);
1205 rcu_assign_pointer(flow->sf_acts, acts);
1206
1207 if (unlikely(reply)) {
1208 error = ovs_flow_cmd_fill_info(flow,
1209 ovs_header->dp_ifindex,
1210 reply, info->snd_portid,
1211 info->snd_seq, 0,
1212 OVS_FLOW_CMD_NEW,
1213 ufid_flags);
1214 BUG_ON(error < 0);
1215 }
1216 } else {
1217 /* Could not alloc without acts before locking. */
1218 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1219 info, OVS_FLOW_CMD_NEW, false,
1220 ufid_flags);
1221
1222 if (IS_ERR(reply)) {
1223 error = PTR_ERR(reply);
1224 goto err_unlock_ovs;
1225 }
1226 }
1227
1228 /* Clear stats. */
1229 if (a[OVS_FLOW_ATTR_CLEAR])
1230 ovs_flow_stats_clear(flow);
1231 ovs_unlock();
1232
1233 if (reply)
1234 ovs_notify(&dp_flow_genl_family, reply, info);
1235 if (old_acts)
1236 ovs_nla_free_flow_actions_rcu(old_acts);
1237
1238 return 0;
1239
1240 err_unlock_ovs:
1241 ovs_unlock();
1242 kfree_skb(reply);
1243 err_kfree_acts:
1244 ovs_nla_free_flow_actions(acts);
1245 error:
1246 return error;
1247 }
1248
1249 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1250 {
1251 struct nlattr **a = info->attrs;
1252 struct ovs_header *ovs_header = info->userhdr;
1253 struct net *net = sock_net(skb->sk);
1254 struct sw_flow_key key;
1255 struct sk_buff *reply;
1256 struct sw_flow *flow;
1257 struct datapath *dp;
1258 struct sw_flow_match match;
1259 struct sw_flow_id ufid;
1260 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1261 int err = 0;
1262 bool log = !a[OVS_FLOW_ATTR_PROBE];
1263 bool ufid_present;
1264
1265 ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1266 if (a[OVS_FLOW_ATTR_KEY]) {
1267 ovs_match_init(&match, &key, true, NULL);
1268 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1269 log);
1270 } else if (!ufid_present) {
1271 OVS_NLERR(log,
1272 "Flow get message rejected, Key attribute missing.");
1273 err = -EINVAL;
1274 }
1275 if (err)
1276 return err;
1277
1278 ovs_lock();
1279 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1280 if (!dp) {
1281 err = -ENODEV;
1282 goto unlock;
1283 }
1284
1285 if (ufid_present)
1286 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1287 else
1288 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1289 if (!flow) {
1290 err = -ENOENT;
1291 goto unlock;
1292 }
1293
1294 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1295 OVS_FLOW_CMD_NEW, true, ufid_flags);
1296 if (IS_ERR(reply)) {
1297 err = PTR_ERR(reply);
1298 goto unlock;
1299 }
1300
1301 ovs_unlock();
1302 return genlmsg_reply(reply, info);
1303 unlock:
1304 ovs_unlock();
1305 return err;
1306 }
1307
1308 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1309 {
1310 struct nlattr **a = info->attrs;
1311 struct ovs_header *ovs_header = info->userhdr;
1312 struct net *net = sock_net(skb->sk);
1313 struct sw_flow_key key;
1314 struct sk_buff *reply;
1315 struct sw_flow *flow = NULL;
1316 struct datapath *dp;
1317 struct sw_flow_match match;
1318 struct sw_flow_id ufid;
1319 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1320 int err;
1321 bool log = !a[OVS_FLOW_ATTR_PROBE];
1322 bool ufid_present;
1323
1324 ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1325 if (a[OVS_FLOW_ATTR_KEY]) {
1326 ovs_match_init(&match, &key, true, NULL);
1327 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1328 NULL, log);
1329 if (unlikely(err))
1330 return err;
1331 }
1332
1333 ovs_lock();
1334 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1335 if (unlikely(!dp)) {
1336 err = -ENODEV;
1337 goto unlock;
1338 }
1339
1340 if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1341 err = ovs_flow_tbl_flush(&dp->table);
1342 goto unlock;
1343 }
1344
1345 if (ufid_present)
1346 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1347 else
1348 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1349 if (unlikely(!flow)) {
1350 err = -ENOENT;
1351 goto unlock;
1352 }
1353
1354 ovs_flow_tbl_remove(&dp->table, flow);
1355 ovs_unlock();
1356
1357 reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1358 &flow->id, info, false, ufid_flags);
1359 if (likely(reply)) {
1360 if (likely(!IS_ERR(reply))) {
1361 rcu_read_lock(); /*To keep RCU checker happy. */
1362 err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1363 reply, info->snd_portid,
1364 info->snd_seq, 0,
1365 OVS_FLOW_CMD_DEL,
1366 ufid_flags);
1367 rcu_read_unlock();
1368 BUG_ON(err < 0);
1369
1370 ovs_notify(&dp_flow_genl_family, reply, info);
1371 } else {
1372 netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
1373 }
1374 }
1375
1376 ovs_flow_free(flow, true);
1377 return 0;
1378 unlock:
1379 ovs_unlock();
1380 return err;
1381 }
1382
1383 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1384 {
1385 struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1386 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1387 struct table_instance *ti;
1388 struct datapath *dp;
1389 u32 ufid_flags;
1390 int err;
1391
1392 err = genlmsg_parse(cb->nlh, &dp_flow_genl_family, a,
1393 OVS_FLOW_ATTR_MAX, flow_policy, NULL);
1394 if (err)
1395 return err;
1396 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1397
1398 rcu_read_lock();
1399 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1400 if (!dp) {
1401 rcu_read_unlock();
1402 return -ENODEV;
1403 }
1404
1405 ti = rcu_dereference(dp->table.ti);
1406 for (;;) {
1407 struct sw_flow *flow;
1408 u32 bucket, obj;
1409
1410 bucket = cb->args[0];
1411 obj = cb->args[1];
1412 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1413 if (!flow)
1414 break;
1415
1416 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1417 NETLINK_CB(cb->skb).portid,
1418 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1419 OVS_FLOW_CMD_NEW, ufid_flags) < 0)
1420 break;
1421
1422 cb->args[0] = bucket;
1423 cb->args[1] = obj;
1424 }
1425 rcu_read_unlock();
1426 return skb->len;
1427 }
1428
1429 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1430 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1431 [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1432 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1433 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1434 [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1435 [OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1436 [OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1437 };
1438
1439 static const struct genl_ops dp_flow_genl_ops[] = {
1440 { .cmd = OVS_FLOW_CMD_NEW,
1441 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1442 .policy = flow_policy,
1443 .doit = ovs_flow_cmd_new
1444 },
1445 { .cmd = OVS_FLOW_CMD_DEL,
1446 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1447 .policy = flow_policy,
1448 .doit = ovs_flow_cmd_del
1449 },
1450 { .cmd = OVS_FLOW_CMD_GET,
1451 .flags = 0, /* OK for unprivileged users. */
1452 .policy = flow_policy,
1453 .doit = ovs_flow_cmd_get,
1454 .dumpit = ovs_flow_cmd_dump
1455 },
1456 { .cmd = OVS_FLOW_CMD_SET,
1457 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1458 .policy = flow_policy,
1459 .doit = ovs_flow_cmd_set,
1460 },
1461 };
1462
1463 static struct genl_family dp_flow_genl_family __ro_after_init = {
1464 .hdrsize = sizeof(struct ovs_header),
1465 .name = OVS_FLOW_FAMILY,
1466 .version = OVS_FLOW_VERSION,
1467 .maxattr = OVS_FLOW_ATTR_MAX,
1468 .netnsok = true,
1469 .parallel_ops = true,
1470 .ops = dp_flow_genl_ops,
1471 .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1472 .mcgrps = &ovs_dp_flow_multicast_group,
1473 .n_mcgrps = 1,
1474 .module = THIS_MODULE,
1475 };
1476
1477 static size_t ovs_dp_cmd_msg_size(void)
1478 {
1479 size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1480
1481 msgsize += nla_total_size(IFNAMSIZ);
1482 msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1483 msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
1484 msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1485
1486 return msgsize;
1487 }
1488
1489 /* Called with ovs_mutex. */
1490 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1491 u32 portid, u32 seq, u32 flags, u8 cmd)
1492 {
1493 struct ovs_header *ovs_header;
1494 struct ovs_dp_stats dp_stats;
1495 struct ovs_dp_megaflow_stats dp_megaflow_stats;
1496 int err;
1497
1498 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1499 flags, cmd);
1500 if (!ovs_header)
1501 goto error;
1502
1503 ovs_header->dp_ifindex = get_dpifindex(dp);
1504
1505 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1506 if (err)
1507 goto nla_put_failure;
1508
1509 get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1510 if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1511 &dp_stats, OVS_DP_ATTR_PAD))
1512 goto nla_put_failure;
1513
1514 if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1515 sizeof(struct ovs_dp_megaflow_stats),
1516 &dp_megaflow_stats, OVS_DP_ATTR_PAD))
1517 goto nla_put_failure;
1518
1519 if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1520 goto nla_put_failure;
1521
1522 genlmsg_end(skb, ovs_header);
1523 return 0;
1524
1525 nla_put_failure:
1526 genlmsg_cancel(skb, ovs_header);
1527 error:
1528 return -EMSGSIZE;
1529 }
1530
1531 static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1532 {
1533 return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1534 }
1535
1536 /* Called with rcu_read_lock or ovs_mutex. */
1537 static struct datapath *lookup_datapath(struct net *net,
1538 const struct ovs_header *ovs_header,
1539 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1540 {
1541 struct datapath *dp;
1542
1543 if (!a[OVS_DP_ATTR_NAME])
1544 dp = get_dp(net, ovs_header->dp_ifindex);
1545 else {
1546 struct vport *vport;
1547
1548 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1549 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1550 }
1551 return dp ? dp : ERR_PTR(-ENODEV);
1552 }
1553
1554 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1555 {
1556 struct datapath *dp;
1557
1558 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1559 if (IS_ERR(dp))
1560 return;
1561
1562 WARN(dp->user_features, "Dropping previously announced user features\n");
1563 dp->user_features = 0;
1564 }
1565
1566 static void ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1567 {
1568 if (a[OVS_DP_ATTR_USER_FEATURES])
1569 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1570 }
1571
1572 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1573 {
1574 struct nlattr **a = info->attrs;
1575 struct vport_parms parms;
1576 struct sk_buff *reply;
1577 struct datapath *dp;
1578 struct vport *vport;
1579 struct ovs_net *ovs_net;
1580 int err, i;
1581
1582 err = -EINVAL;
1583 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1584 goto err;
1585
1586 reply = ovs_dp_cmd_alloc_info();
1587 if (!reply)
1588 return -ENOMEM;
1589
1590 err = -ENOMEM;
1591 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1592 if (dp == NULL)
1593 goto err_free_reply;
1594
1595 ovs_dp_set_net(dp, sock_net(skb->sk));
1596
1597 /* Allocate table. */
1598 err = ovs_flow_tbl_init(&dp->table);
1599 if (err)
1600 goto err_free_dp;
1601
1602 dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1603 if (!dp->stats_percpu) {
1604 err = -ENOMEM;
1605 goto err_destroy_table;
1606 }
1607
1608 dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1609 GFP_KERNEL);
1610 if (!dp->ports) {
1611 err = -ENOMEM;
1612 goto err_destroy_percpu;
1613 }
1614
1615 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1616 INIT_HLIST_HEAD(&dp->ports[i]);
1617
1618 /* Set up our datapath device. */
1619 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1620 parms.type = OVS_VPORT_TYPE_INTERNAL;
1621 parms.options = NULL;
1622 parms.dp = dp;
1623 parms.port_no = OVSP_LOCAL;
1624 parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1625
1626 ovs_dp_change(dp, a);
1627
1628 /* So far only local changes have been made, now need the lock. */
1629 ovs_lock();
1630
1631 vport = new_vport(&parms);
1632 if (IS_ERR(vport)) {
1633 err = PTR_ERR(vport);
1634 if (err == -EBUSY)
1635 err = -EEXIST;
1636
1637 if (err == -EEXIST) {
1638 /* An outdated user space instance that does not understand
1639 * the concept of user_features has attempted to create a new
1640 * datapath and is likely to reuse it. Drop all user features.
1641 */
1642 if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1643 ovs_dp_reset_user_features(skb, info);
1644 }
1645
1646 goto err_destroy_ports_array;
1647 }
1648
1649 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1650 info->snd_seq, 0, OVS_DP_CMD_NEW);
1651 BUG_ON(err < 0);
1652
1653 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1654 list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1655
1656 ovs_unlock();
1657
1658 ovs_notify(&dp_datapath_genl_family, reply, info);
1659 return 0;
1660
1661 err_destroy_ports_array:
1662 ovs_unlock();
1663 kfree(dp->ports);
1664 err_destroy_percpu:
1665 free_percpu(dp->stats_percpu);
1666 err_destroy_table:
1667 ovs_flow_tbl_destroy(&dp->table);
1668 err_free_dp:
1669 kfree(dp);
1670 err_free_reply:
1671 kfree_skb(reply);
1672 err:
1673 return err;
1674 }
1675
1676 /* Called with ovs_mutex. */
1677 static void __dp_destroy(struct datapath *dp)
1678 {
1679 int i;
1680
1681 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1682 struct vport *vport;
1683 struct hlist_node *n;
1684
1685 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1686 if (vport->port_no != OVSP_LOCAL)
1687 ovs_dp_detach_port(vport);
1688 }
1689
1690 list_del_rcu(&dp->list_node);
1691
1692 /* OVSP_LOCAL is datapath internal port. We need to make sure that
1693 * all ports in datapath are destroyed first before freeing datapath.
1694 */
1695 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1696
1697 /* RCU destroy the flow table */
1698 call_rcu(&dp->rcu, destroy_dp_rcu);
1699 }
1700
1701 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1702 {
1703 struct sk_buff *reply;
1704 struct datapath *dp;
1705 int err;
1706
1707 reply = ovs_dp_cmd_alloc_info();
1708 if (!reply)
1709 return -ENOMEM;
1710
1711 ovs_lock();
1712 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1713 err = PTR_ERR(dp);
1714 if (IS_ERR(dp))
1715 goto err_unlock_free;
1716
1717 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1718 info->snd_seq, 0, OVS_DP_CMD_DEL);
1719 BUG_ON(err < 0);
1720
1721 __dp_destroy(dp);
1722 ovs_unlock();
1723
1724 ovs_notify(&dp_datapath_genl_family, reply, info);
1725
1726 return 0;
1727
1728 err_unlock_free:
1729 ovs_unlock();
1730 kfree_skb(reply);
1731 return err;
1732 }
1733
1734 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1735 {
1736 struct sk_buff *reply;
1737 struct datapath *dp;
1738 int err;
1739
1740 reply = ovs_dp_cmd_alloc_info();
1741 if (!reply)
1742 return -ENOMEM;
1743
1744 ovs_lock();
1745 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1746 err = PTR_ERR(dp);
1747 if (IS_ERR(dp))
1748 goto err_unlock_free;
1749
1750 ovs_dp_change(dp, info->attrs);
1751
1752 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1753 info->snd_seq, 0, OVS_DP_CMD_NEW);
1754 BUG_ON(err < 0);
1755
1756 ovs_unlock();
1757 ovs_notify(&dp_datapath_genl_family, reply, info);
1758
1759 return 0;
1760
1761 err_unlock_free:
1762 ovs_unlock();
1763 kfree_skb(reply);
1764 return err;
1765 }
1766
1767 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1768 {
1769 struct sk_buff *reply;
1770 struct datapath *dp;
1771 int err;
1772
1773 reply = ovs_dp_cmd_alloc_info();
1774 if (!reply)
1775 return -ENOMEM;
1776
1777 ovs_lock();
1778 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1779 if (IS_ERR(dp)) {
1780 err = PTR_ERR(dp);
1781 goto err_unlock_free;
1782 }
1783 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1784 info->snd_seq, 0, OVS_DP_CMD_NEW);
1785 BUG_ON(err < 0);
1786 ovs_unlock();
1787
1788 return genlmsg_reply(reply, info);
1789
1790 err_unlock_free:
1791 ovs_unlock();
1792 kfree_skb(reply);
1793 return err;
1794 }
1795
1796 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1797 {
1798 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1799 struct datapath *dp;
1800 int skip = cb->args[0];
1801 int i = 0;
1802
1803 ovs_lock();
1804 list_for_each_entry(dp, &ovs_net->dps, list_node) {
1805 if (i >= skip &&
1806 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1807 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1808 OVS_DP_CMD_NEW) < 0)
1809 break;
1810 i++;
1811 }
1812 ovs_unlock();
1813
1814 cb->args[0] = i;
1815
1816 return skb->len;
1817 }
1818
1819 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1820 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1821 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1822 [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1823 };
1824
1825 static const struct genl_ops dp_datapath_genl_ops[] = {
1826 { .cmd = OVS_DP_CMD_NEW,
1827 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1828 .policy = datapath_policy,
1829 .doit = ovs_dp_cmd_new
1830 },
1831 { .cmd = OVS_DP_CMD_DEL,
1832 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1833 .policy = datapath_policy,
1834 .doit = ovs_dp_cmd_del
1835 },
1836 { .cmd = OVS_DP_CMD_GET,
1837 .flags = 0, /* OK for unprivileged users. */
1838 .policy = datapath_policy,
1839 .doit = ovs_dp_cmd_get,
1840 .dumpit = ovs_dp_cmd_dump
1841 },
1842 { .cmd = OVS_DP_CMD_SET,
1843 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1844 .policy = datapath_policy,
1845 .doit = ovs_dp_cmd_set,
1846 },
1847 };
1848
1849 static struct genl_family dp_datapath_genl_family __ro_after_init = {
1850 .hdrsize = sizeof(struct ovs_header),
1851 .name = OVS_DATAPATH_FAMILY,
1852 .version = OVS_DATAPATH_VERSION,
1853 .maxattr = OVS_DP_ATTR_MAX,
1854 .netnsok = true,
1855 .parallel_ops = true,
1856 .ops = dp_datapath_genl_ops,
1857 .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1858 .mcgrps = &ovs_dp_datapath_multicast_group,
1859 .n_mcgrps = 1,
1860 .module = THIS_MODULE,
1861 };
1862
1863 /* Called with ovs_mutex or RCU read lock. */
1864 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1865 u32 portid, u32 seq, u32 flags, u8 cmd)
1866 {
1867 struct ovs_header *ovs_header;
1868 struct ovs_vport_stats vport_stats;
1869 int err;
1870
1871 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1872 flags, cmd);
1873 if (!ovs_header)
1874 return -EMSGSIZE;
1875
1876 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1877
1878 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1879 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1880 nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1881 ovs_vport_name(vport)))
1882 goto nla_put_failure;
1883
1884 ovs_vport_get_stats(vport, &vport_stats);
1885 if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
1886 sizeof(struct ovs_vport_stats), &vport_stats,
1887 OVS_VPORT_ATTR_PAD))
1888 goto nla_put_failure;
1889
1890 if (ovs_vport_get_upcall_portids(vport, skb))
1891 goto nla_put_failure;
1892
1893 err = ovs_vport_get_options(vport, skb);
1894 if (err == -EMSGSIZE)
1895 goto error;
1896
1897 genlmsg_end(skb, ovs_header);
1898 return 0;
1899
1900 nla_put_failure:
1901 err = -EMSGSIZE;
1902 error:
1903 genlmsg_cancel(skb, ovs_header);
1904 return err;
1905 }
1906
1907 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1908 {
1909 return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1910 }
1911
1912 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1913 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1914 u32 seq, u8 cmd)
1915 {
1916 struct sk_buff *skb;
1917 int retval;
1918
1919 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1920 if (!skb)
1921 return ERR_PTR(-ENOMEM);
1922
1923 retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1924 BUG_ON(retval < 0);
1925
1926 return skb;
1927 }
1928
1929 /* Called with ovs_mutex or RCU read lock. */
1930 static struct vport *lookup_vport(struct net *net,
1931 const struct ovs_header *ovs_header,
1932 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1933 {
1934 struct datapath *dp;
1935 struct vport *vport;
1936
1937 if (a[OVS_VPORT_ATTR_NAME]) {
1938 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1939 if (!vport)
1940 return ERR_PTR(-ENODEV);
1941 if (ovs_header->dp_ifindex &&
1942 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1943 return ERR_PTR(-ENODEV);
1944 return vport;
1945 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1946 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1947
1948 if (port_no >= DP_MAX_PORTS)
1949 return ERR_PTR(-EFBIG);
1950
1951 dp = get_dp(net, ovs_header->dp_ifindex);
1952 if (!dp)
1953 return ERR_PTR(-ENODEV);
1954
1955 vport = ovs_vport_ovsl_rcu(dp, port_no);
1956 if (!vport)
1957 return ERR_PTR(-ENODEV);
1958 return vport;
1959 } else
1960 return ERR_PTR(-EINVAL);
1961 }
1962
1963 /* Called with ovs_mutex */
1964 static void update_headroom(struct datapath *dp)
1965 {
1966 unsigned dev_headroom, max_headroom = 0;
1967 struct net_device *dev;
1968 struct vport *vport;
1969 int i;
1970
1971 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1972 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1973 dev = vport->dev;
1974 dev_headroom = netdev_get_fwd_headroom(dev);
1975 if (dev_headroom > max_headroom)
1976 max_headroom = dev_headroom;
1977 }
1978 }
1979
1980 dp->max_headroom = max_headroom;
1981 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1982 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node)
1983 netdev_set_rx_headroom(vport->dev, max_headroom);
1984 }
1985
1986 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1987 {
1988 struct nlattr **a = info->attrs;
1989 struct ovs_header *ovs_header = info->userhdr;
1990 struct vport_parms parms;
1991 struct sk_buff *reply;
1992 struct vport *vport;
1993 struct datapath *dp;
1994 u32 port_no;
1995 int err;
1996
1997 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1998 !a[OVS_VPORT_ATTR_UPCALL_PID])
1999 return -EINVAL;
2000
2001 port_no = a[OVS_VPORT_ATTR_PORT_NO]
2002 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
2003 if (port_no >= DP_MAX_PORTS)
2004 return -EFBIG;
2005
2006 reply = ovs_vport_cmd_alloc_info();
2007 if (!reply)
2008 return -ENOMEM;
2009
2010 ovs_lock();
2011 restart:
2012 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2013 err = -ENODEV;
2014 if (!dp)
2015 goto exit_unlock_free;
2016
2017 if (port_no) {
2018 vport = ovs_vport_ovsl(dp, port_no);
2019 err = -EBUSY;
2020 if (vport)
2021 goto exit_unlock_free;
2022 } else {
2023 for (port_no = 1; ; port_no++) {
2024 if (port_no >= DP_MAX_PORTS) {
2025 err = -EFBIG;
2026 goto exit_unlock_free;
2027 }
2028 vport = ovs_vport_ovsl(dp, port_no);
2029 if (!vport)
2030 break;
2031 }
2032 }
2033
2034 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2035 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2036 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2037 parms.dp = dp;
2038 parms.port_no = port_no;
2039 parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
2040
2041 vport = new_vport(&parms);
2042 err = PTR_ERR(vport);
2043 if (IS_ERR(vport)) {
2044 if (err == -EAGAIN)
2045 goto restart;
2046 goto exit_unlock_free;
2047 }
2048
2049 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2050 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2051
2052 if (netdev_get_fwd_headroom(vport->dev) > dp->max_headroom)
2053 update_headroom(dp);
2054 else
2055 netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2056
2057 BUG_ON(err < 0);
2058 ovs_unlock();
2059
2060 ovs_notify(&dp_vport_genl_family, reply, info);
2061 return 0;
2062
2063 exit_unlock_free:
2064 ovs_unlock();
2065 kfree_skb(reply);
2066 return err;
2067 }
2068
2069 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2070 {
2071 struct nlattr **a = info->attrs;
2072 struct sk_buff *reply;
2073 struct vport *vport;
2074 int err;
2075
2076 reply = ovs_vport_cmd_alloc_info();
2077 if (!reply)
2078 return -ENOMEM;
2079
2080 ovs_lock();
2081 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2082 err = PTR_ERR(vport);
2083 if (IS_ERR(vport))
2084 goto exit_unlock_free;
2085
2086 if (a[OVS_VPORT_ATTR_TYPE] &&
2087 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2088 err = -EINVAL;
2089 goto exit_unlock_free;
2090 }
2091
2092 if (a[OVS_VPORT_ATTR_OPTIONS]) {
2093 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2094 if (err)
2095 goto exit_unlock_free;
2096 }
2097
2098
2099 if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2100 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2101
2102 err = ovs_vport_set_upcall_portids(vport, ids);
2103 if (err)
2104 goto exit_unlock_free;
2105 }
2106
2107 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2108 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2109 BUG_ON(err < 0);
2110
2111 ovs_unlock();
2112 ovs_notify(&dp_vport_genl_family, reply, info);
2113 return 0;
2114
2115 exit_unlock_free:
2116 ovs_unlock();
2117 kfree_skb(reply);
2118 return err;
2119 }
2120
2121 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2122 {
2123 bool must_update_headroom = false;
2124 struct nlattr **a = info->attrs;
2125 struct sk_buff *reply;
2126 struct datapath *dp;
2127 struct vport *vport;
2128 int err;
2129
2130 reply = ovs_vport_cmd_alloc_info();
2131 if (!reply)
2132 return -ENOMEM;
2133
2134 ovs_lock();
2135 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2136 err = PTR_ERR(vport);
2137 if (IS_ERR(vport))
2138 goto exit_unlock_free;
2139
2140 if (vport->port_no == OVSP_LOCAL) {
2141 err = -EINVAL;
2142 goto exit_unlock_free;
2143 }
2144
2145 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2146 info->snd_seq, 0, OVS_VPORT_CMD_DEL);
2147 BUG_ON(err < 0);
2148
2149 /* the vport deletion may trigger dp headroom update */
2150 dp = vport->dp;
2151 if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2152 must_update_headroom = true;
2153 netdev_reset_rx_headroom(vport->dev);
2154 ovs_dp_detach_port(vport);
2155
2156 if (must_update_headroom)
2157 update_headroom(dp);
2158 ovs_unlock();
2159
2160 ovs_notify(&dp_vport_genl_family, reply, info);
2161 return 0;
2162
2163 exit_unlock_free:
2164 ovs_unlock();
2165 kfree_skb(reply);
2166 return err;
2167 }
2168
2169 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2170 {
2171 struct nlattr **a = info->attrs;
2172 struct ovs_header *ovs_header = info->userhdr;
2173 struct sk_buff *reply;
2174 struct vport *vport;
2175 int err;
2176
2177 reply = ovs_vport_cmd_alloc_info();
2178 if (!reply)
2179 return -ENOMEM;
2180
2181 rcu_read_lock();
2182 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2183 err = PTR_ERR(vport);
2184 if (IS_ERR(vport))
2185 goto exit_unlock_free;
2186 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2187 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2188 BUG_ON(err < 0);
2189 rcu_read_unlock();
2190
2191 return genlmsg_reply(reply, info);
2192
2193 exit_unlock_free:
2194 rcu_read_unlock();
2195 kfree_skb(reply);
2196 return err;
2197 }
2198
2199 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2200 {
2201 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2202 struct datapath *dp;
2203 int bucket = cb->args[0], skip = cb->args[1];
2204 int i, j = 0;
2205
2206 rcu_read_lock();
2207 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2208 if (!dp) {
2209 rcu_read_unlock();
2210 return -ENODEV;
2211 }
2212 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2213 struct vport *vport;
2214
2215 j = 0;
2216 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2217 if (j >= skip &&
2218 ovs_vport_cmd_fill_info(vport, skb,
2219 NETLINK_CB(cb->skb).portid,
2220 cb->nlh->nlmsg_seq,
2221 NLM_F_MULTI,
2222 OVS_VPORT_CMD_NEW) < 0)
2223 goto out;
2224
2225 j++;
2226 }
2227 skip = 0;
2228 }
2229 out:
2230 rcu_read_unlock();
2231
2232 cb->args[0] = i;
2233 cb->args[1] = j;
2234
2235 return skb->len;
2236 }
2237
2238 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2239 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2240 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2241 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2242 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2243 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
2244 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2245 };
2246
2247 static const struct genl_ops dp_vport_genl_ops[] = {
2248 { .cmd = OVS_VPORT_CMD_NEW,
2249 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2250 .policy = vport_policy,
2251 .doit = ovs_vport_cmd_new
2252 },
2253 { .cmd = OVS_VPORT_CMD_DEL,
2254 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2255 .policy = vport_policy,
2256 .doit = ovs_vport_cmd_del
2257 },
2258 { .cmd = OVS_VPORT_CMD_GET,
2259 .flags = 0, /* OK for unprivileged users. */
2260 .policy = vport_policy,
2261 .doit = ovs_vport_cmd_get,
2262 .dumpit = ovs_vport_cmd_dump
2263 },
2264 { .cmd = OVS_VPORT_CMD_SET,
2265 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2266 .policy = vport_policy,
2267 .doit = ovs_vport_cmd_set,
2268 },
2269 };
2270
2271 struct genl_family dp_vport_genl_family __ro_after_init = {
2272 .hdrsize = sizeof(struct ovs_header),
2273 .name = OVS_VPORT_FAMILY,
2274 .version = OVS_VPORT_VERSION,
2275 .maxattr = OVS_VPORT_ATTR_MAX,
2276 .netnsok = true,
2277 .parallel_ops = true,
2278 .ops = dp_vport_genl_ops,
2279 .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2280 .mcgrps = &ovs_dp_vport_multicast_group,
2281 .n_mcgrps = 1,
2282 .module = THIS_MODULE,
2283 };
2284
2285 static struct genl_family * const dp_genl_families[] = {
2286 &dp_datapath_genl_family,
2287 &dp_vport_genl_family,
2288 &dp_flow_genl_family,
2289 &dp_packet_genl_family,
2290 };
2291
2292 static void dp_unregister_genl(int n_families)
2293 {
2294 int i;
2295
2296 for (i = 0; i < n_families; i++)
2297 genl_unregister_family(dp_genl_families[i]);
2298 }
2299
2300 static int __init dp_register_genl(void)
2301 {
2302 int err;
2303 int i;
2304
2305 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2306
2307 err = genl_register_family(dp_genl_families[i]);
2308 if (err)
2309 goto error;
2310 }
2311
2312 return 0;
2313
2314 error:
2315 dp_unregister_genl(i);
2316 return err;
2317 }
2318
2319 static int __net_init ovs_init_net(struct net *net)
2320 {
2321 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2322
2323 INIT_LIST_HEAD(&ovs_net->dps);
2324 INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2325 ovs_ct_init(net);
2326 return 0;
2327 }
2328
2329 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2330 struct list_head *head)
2331 {
2332 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2333 struct datapath *dp;
2334
2335 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2336 int i;
2337
2338 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2339 struct vport *vport;
2340
2341 hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2342 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2343 continue;
2344
2345 if (dev_net(vport->dev) == dnet)
2346 list_add(&vport->detach_list, head);
2347 }
2348 }
2349 }
2350 }
2351
2352 static void __net_exit ovs_exit_net(struct net *dnet)
2353 {
2354 struct datapath *dp, *dp_next;
2355 struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2356 struct vport *vport, *vport_next;
2357 struct net *net;
2358 LIST_HEAD(head);
2359
2360 ovs_ct_exit(dnet);
2361 ovs_lock();
2362 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2363 __dp_destroy(dp);
2364
2365 rtnl_lock();
2366 for_each_net(net)
2367 list_vports_from_net(net, dnet, &head);
2368 rtnl_unlock();
2369
2370 /* Detach all vports from given namespace. */
2371 list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2372 list_del(&vport->detach_list);
2373 ovs_dp_detach_port(vport);
2374 }
2375
2376 ovs_unlock();
2377
2378 cancel_work_sync(&ovs_net->dp_notify_work);
2379 }
2380
2381 static struct pernet_operations ovs_net_ops = {
2382 .init = ovs_init_net,
2383 .exit = ovs_exit_net,
2384 .id = &ovs_net_id,
2385 .size = sizeof(struct ovs_net),
2386 };
2387
2388 static int __init dp_init(void)
2389 {
2390 int err;
2391
2392 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2393
2394 pr_info("Open vSwitch switching datapath\n");
2395
2396 err = action_fifos_init();
2397 if (err)
2398 goto error;
2399
2400 err = ovs_internal_dev_rtnl_link_register();
2401 if (err)
2402 goto error_action_fifos_exit;
2403
2404 err = ovs_flow_init();
2405 if (err)
2406 goto error_unreg_rtnl_link;
2407
2408 err = ovs_vport_init();
2409 if (err)
2410 goto error_flow_exit;
2411
2412 err = register_pernet_device(&ovs_net_ops);
2413 if (err)
2414 goto error_vport_exit;
2415
2416 err = register_netdevice_notifier(&ovs_dp_device_notifier);
2417 if (err)
2418 goto error_netns_exit;
2419
2420 err = ovs_netdev_init();
2421 if (err)
2422 goto error_unreg_notifier;
2423
2424 err = dp_register_genl();
2425 if (err < 0)
2426 goto error_unreg_netdev;
2427
2428 return 0;
2429
2430 error_unreg_netdev:
2431 ovs_netdev_exit();
2432 error_unreg_notifier:
2433 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2434 error_netns_exit:
2435 unregister_pernet_device(&ovs_net_ops);
2436 error_vport_exit:
2437 ovs_vport_exit();
2438 error_flow_exit:
2439 ovs_flow_exit();
2440 error_unreg_rtnl_link:
2441 ovs_internal_dev_rtnl_link_unregister();
2442 error_action_fifos_exit:
2443 action_fifos_exit();
2444 error:
2445 return err;
2446 }
2447
2448 static void dp_cleanup(void)
2449 {
2450 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2451 ovs_netdev_exit();
2452 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2453 unregister_pernet_device(&ovs_net_ops);
2454 rcu_barrier();
2455 ovs_vport_exit();
2456 ovs_flow_exit();
2457 ovs_internal_dev_rtnl_link_unregister();
2458 action_fifos_exit();
2459 }
2460
2461 module_init(dp_init);
2462 module_exit(dp_cleanup);
2463
2464 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2465 MODULE_LICENSE("GPL");
2466 MODULE_ALIAS_GENL_FAMILY(OVS_DATAPATH_FAMILY);
2467 MODULE_ALIAS_GENL_FAMILY(OVS_VPORT_FAMILY);
2468 MODULE_ALIAS_GENL_FAMILY(OVS_FLOW_FAMILY);
2469 MODULE_ALIAS_GENL_FAMILY(OVS_PACKET_FAMILY);
Linux with added FPC-III support