dc9ff936722102264d3fd1631d0f886928ee9b7f
[muen/linux.git] / net / openvswitch / datapath.c
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 "meter.h"
59 #include "vport-internal_dev.h"
60 #include "vport-netdev.h"
61
62 unsigned int ovs_net_id __read_mostly;
63
64 static struct genl_family dp_packet_genl_family;
65 static struct genl_family dp_flow_genl_family;
66 static struct genl_family dp_datapath_genl_family;
67
68 static const struct nla_policy flow_policy[];
69
70 static const struct genl_multicast_group ovs_dp_flow_multicast_group = {
71         .name = OVS_FLOW_MCGROUP,
72 };
73
74 static const struct genl_multicast_group ovs_dp_datapath_multicast_group = {
75         .name = OVS_DATAPATH_MCGROUP,
76 };
77
78 static const struct genl_multicast_group ovs_dp_vport_multicast_group = {
79         .name = OVS_VPORT_MCGROUP,
80 };
81
82 /* Check if need to build a reply message.
83  * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
84 static bool ovs_must_notify(struct genl_family *family, struct genl_info *info,
85                             unsigned int group)
86 {
87         return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
88                genl_has_listeners(family, genl_info_net(info), group);
89 }
90
91 static void ovs_notify(struct genl_family *family,
92                        struct sk_buff *skb, struct genl_info *info)
93 {
94         genl_notify(family, skb, info, 0, GFP_KERNEL);
95 }
96
97 /**
98  * DOC: Locking:
99  *
100  * All writes e.g. Writes to device state (add/remove datapath, port, set
101  * operations on vports, etc.), Writes to other state (flow table
102  * modifications, set miscellaneous datapath parameters, etc.) are protected
103  * by ovs_lock.
104  *
105  * Reads are protected by RCU.
106  *
107  * There are a few special cases (mostly stats) that have their own
108  * synchronization but they nest under all of above and don't interact with
109  * each other.
110  *
111  * The RTNL lock nests inside ovs_mutex.
112  */
113
114 static DEFINE_MUTEX(ovs_mutex);
115
116 void ovs_lock(void)
117 {
118         mutex_lock(&ovs_mutex);
119 }
120
121 void ovs_unlock(void)
122 {
123         mutex_unlock(&ovs_mutex);
124 }
125
126 #ifdef CONFIG_LOCKDEP
127 int lockdep_ovsl_is_held(void)
128 {
129         if (debug_locks)
130                 return lockdep_is_held(&ovs_mutex);
131         else
132                 return 1;
133 }
134 #endif
135
136 static struct vport *new_vport(const struct vport_parms *);
137 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
138                              const struct sw_flow_key *,
139                              const struct dp_upcall_info *,
140                              uint32_t cutlen);
141 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
142                                   const struct sw_flow_key *,
143                                   const struct dp_upcall_info *,
144                                   uint32_t cutlen);
145
146 /* Must be called with rcu_read_lock or ovs_mutex. */
147 const char *ovs_dp_name(const struct datapath *dp)
148 {
149         struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
150         return ovs_vport_name(vport);
151 }
152
153 static int get_dpifindex(const struct datapath *dp)
154 {
155         struct vport *local;
156         int ifindex;
157
158         rcu_read_lock();
159
160         local = ovs_vport_rcu(dp, OVSP_LOCAL);
161         if (local)
162                 ifindex = local->dev->ifindex;
163         else
164                 ifindex = 0;
165
166         rcu_read_unlock();
167
168         return ifindex;
169 }
170
171 static void destroy_dp_rcu(struct rcu_head *rcu)
172 {
173         struct datapath *dp = container_of(rcu, struct datapath, rcu);
174
175         ovs_flow_tbl_destroy(&dp->table);
176         free_percpu(dp->stats_percpu);
177         kfree(dp->ports);
178         ovs_meters_exit(dp);
179         kfree(dp);
180 }
181
182 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
183                                             u16 port_no)
184 {
185         return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
186 }
187
188 /* Called with ovs_mutex or RCU read lock. */
189 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
190 {
191         struct vport *vport;
192         struct hlist_head *head;
193
194         head = vport_hash_bucket(dp, port_no);
195         hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
196                 if (vport->port_no == port_no)
197                         return vport;
198         }
199         return NULL;
200 }
201
202 /* Called with ovs_mutex. */
203 static struct vport *new_vport(const struct vport_parms *parms)
204 {
205         struct vport *vport;
206
207         vport = ovs_vport_add(parms);
208         if (!IS_ERR(vport)) {
209                 struct datapath *dp = parms->dp;
210                 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
211
212                 hlist_add_head_rcu(&vport->dp_hash_node, head);
213         }
214         return vport;
215 }
216
217 void ovs_dp_detach_port(struct vport *p)
218 {
219         ASSERT_OVSL();
220
221         /* First drop references to device. */
222         hlist_del_rcu(&p->dp_hash_node);
223
224         /* Then destroy it. */
225         ovs_vport_del(p);
226 }
227
228 /* Must be called with rcu_read_lock. */
229 void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
230 {
231         const struct vport *p = OVS_CB(skb)->input_vport;
232         struct datapath *dp = p->dp;
233         struct sw_flow *flow;
234         struct sw_flow_actions *sf_acts;
235         struct dp_stats_percpu *stats;
236         u64 *stats_counter;
237         u32 n_mask_hit;
238
239         stats = this_cpu_ptr(dp->stats_percpu);
240
241         /* Look up flow. */
242         flow = ovs_flow_tbl_lookup_stats(&dp->table, key, &n_mask_hit);
243         if (unlikely(!flow)) {
244                 struct dp_upcall_info upcall;
245                 int error;
246
247                 memset(&upcall, 0, sizeof(upcall));
248                 upcall.cmd = OVS_PACKET_CMD_MISS;
249                 upcall.portid = ovs_vport_find_upcall_portid(p, skb);
250                 upcall.mru = OVS_CB(skb)->mru;
251                 error = ovs_dp_upcall(dp, skb, key, &upcall, 0);
252                 if (unlikely(error))
253                         kfree_skb(skb);
254                 else
255                         consume_skb(skb);
256                 stats_counter = &stats->n_missed;
257                 goto out;
258         }
259
260         ovs_flow_stats_update(flow, key->tp.flags, skb);
261         sf_acts = rcu_dereference(flow->sf_acts);
262         ovs_execute_actions(dp, skb, sf_acts, key);
263
264         stats_counter = &stats->n_hit;
265
266 out:
267         /* Update datapath statistics. */
268         u64_stats_update_begin(&stats->syncp);
269         (*stats_counter)++;
270         stats->n_mask_hit += n_mask_hit;
271         u64_stats_update_end(&stats->syncp);
272 }
273
274 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
275                   const struct sw_flow_key *key,
276                   const struct dp_upcall_info *upcall_info,
277                   uint32_t cutlen)
278 {
279         struct dp_stats_percpu *stats;
280         int err;
281
282         if (upcall_info->portid == 0) {
283                 err = -ENOTCONN;
284                 goto err;
285         }
286
287         if (!skb_is_gso(skb))
288                 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
289         else
290                 err = queue_gso_packets(dp, skb, key, upcall_info, cutlen);
291         if (err)
292                 goto err;
293
294         return 0;
295
296 err:
297         stats = this_cpu_ptr(dp->stats_percpu);
298
299         u64_stats_update_begin(&stats->syncp);
300         stats->n_lost++;
301         u64_stats_update_end(&stats->syncp);
302
303         return err;
304 }
305
306 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
307                              const struct sw_flow_key *key,
308                              const struct dp_upcall_info *upcall_info,
309                                  uint32_t cutlen)
310 {
311         unsigned int gso_type = skb_shinfo(skb)->gso_type;
312         struct sw_flow_key later_key;
313         struct sk_buff *segs, *nskb;
314         int err;
315
316         BUILD_BUG_ON(sizeof(*OVS_CB(skb)) > SKB_SGO_CB_OFFSET);
317         segs = __skb_gso_segment(skb, NETIF_F_SG, false);
318         if (IS_ERR(segs))
319                 return PTR_ERR(segs);
320         if (segs == NULL)
321                 return -EINVAL;
322
323         if (gso_type & SKB_GSO_UDP) {
324                 /* The initial flow key extracted by ovs_flow_key_extract()
325                  * in this case is for a first fragment, so we need to
326                  * properly mark later fragments.
327                  */
328                 later_key = *key;
329                 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
330         }
331
332         /* Queue all of the segments. */
333         skb = segs;
334         do {
335                 if (gso_type & SKB_GSO_UDP && skb != segs)
336                         key = &later_key;
337
338                 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
339                 if (err)
340                         break;
341
342         } while ((skb = skb->next));
343
344         /* Free all of the segments. */
345         skb = segs;
346         do {
347                 nskb = skb->next;
348                 if (err)
349                         kfree_skb(skb);
350                 else
351                         consume_skb(skb);
352         } while ((skb = nskb));
353         return err;
354 }
355
356 static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
357                               unsigned int hdrlen, int actions_attrlen)
358 {
359         size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
360                 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
361                 + nla_total_size(ovs_key_attr_size()) /* OVS_PACKET_ATTR_KEY */
362                 + nla_total_size(sizeof(unsigned int)); /* OVS_PACKET_ATTR_LEN */
363
364         /* OVS_PACKET_ATTR_USERDATA */
365         if (upcall_info->userdata)
366                 size += NLA_ALIGN(upcall_info->userdata->nla_len);
367
368         /* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
369         if (upcall_info->egress_tun_info)
370                 size += nla_total_size(ovs_tun_key_attr_size());
371
372         /* OVS_PACKET_ATTR_ACTIONS */
373         if (upcall_info->actions_len)
374                 size += nla_total_size(actions_attrlen);
375
376         /* OVS_PACKET_ATTR_MRU */
377         if (upcall_info->mru)
378                 size += nla_total_size(sizeof(upcall_info->mru));
379
380         return size;
381 }
382
383 static void pad_packet(struct datapath *dp, struct sk_buff *skb)
384 {
385         if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
386                 size_t plen = NLA_ALIGN(skb->len) - skb->len;
387
388                 if (plen > 0)
389                         skb_put_zero(skb, plen);
390         }
391 }
392
393 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
394                                   const struct sw_flow_key *key,
395                                   const struct dp_upcall_info *upcall_info,
396                                   uint32_t cutlen)
397 {
398         struct ovs_header *upcall;
399         struct sk_buff *nskb = NULL;
400         struct sk_buff *user_skb = NULL; /* to be queued to userspace */
401         struct nlattr *nla;
402         size_t len;
403         unsigned int hlen;
404         int err, dp_ifindex;
405
406         dp_ifindex = get_dpifindex(dp);
407         if (!dp_ifindex)
408                 return -ENODEV;
409
410         if (skb_vlan_tag_present(skb)) {
411                 nskb = skb_clone(skb, GFP_ATOMIC);
412                 if (!nskb)
413                         return -ENOMEM;
414
415                 nskb = __vlan_hwaccel_push_inside(nskb);
416                 if (!nskb)
417                         return -ENOMEM;
418
419                 skb = nskb;
420         }
421
422         if (nla_attr_size(skb->len) > USHRT_MAX) {
423                 err = -EFBIG;
424                 goto out;
425         }
426
427         /* Complete checksum if needed */
428         if (skb->ip_summed == CHECKSUM_PARTIAL &&
429             (err = skb_csum_hwoffload_help(skb, 0)))
430                 goto out;
431
432         /* Older versions of OVS user space enforce alignment of the last
433          * Netlink attribute to NLA_ALIGNTO which would require extensive
434          * padding logic. Only perform zerocopy if padding is not required.
435          */
436         if (dp->user_features & OVS_DP_F_UNALIGNED)
437                 hlen = skb_zerocopy_headlen(skb);
438         else
439                 hlen = skb->len;
440
441         len = upcall_msg_size(upcall_info, hlen - cutlen,
442                               OVS_CB(skb)->acts_origlen);
443         user_skb = genlmsg_new(len, GFP_ATOMIC);
444         if (!user_skb) {
445                 err = -ENOMEM;
446                 goto out;
447         }
448
449         upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
450                              0, upcall_info->cmd);
451         if (!upcall) {
452                 err = -EINVAL;
453                 goto out;
454         }
455         upcall->dp_ifindex = dp_ifindex;
456
457         err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
458         if (err)
459                 goto out;
460
461         if (upcall_info->userdata)
462                 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
463                           nla_len(upcall_info->userdata),
464                           nla_data(upcall_info->userdata));
465
466         if (upcall_info->egress_tun_info) {
467                 nla = nla_nest_start_noflag(user_skb,
468                                             OVS_PACKET_ATTR_EGRESS_TUN_KEY);
469                 if (!nla) {
470                         err = -EMSGSIZE;
471                         goto out;
472                 }
473                 err = ovs_nla_put_tunnel_info(user_skb,
474                                               upcall_info->egress_tun_info);
475                 if (err)
476                         goto out;
477
478                 nla_nest_end(user_skb, nla);
479         }
480
481         if (upcall_info->actions_len) {
482                 nla = nla_nest_start_noflag(user_skb, OVS_PACKET_ATTR_ACTIONS);
483                 if (!nla) {
484                         err = -EMSGSIZE;
485                         goto out;
486                 }
487                 err = ovs_nla_put_actions(upcall_info->actions,
488                                           upcall_info->actions_len,
489                                           user_skb);
490                 if (!err)
491                         nla_nest_end(user_skb, nla);
492                 else
493                         nla_nest_cancel(user_skb, nla);
494         }
495
496         /* Add OVS_PACKET_ATTR_MRU */
497         if (upcall_info->mru) {
498                 if (nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU,
499                                 upcall_info->mru)) {
500                         err = -ENOBUFS;
501                         goto out;
502                 }
503                 pad_packet(dp, user_skb);
504         }
505
506         /* Add OVS_PACKET_ATTR_LEN when packet is truncated */
507         if (cutlen > 0) {
508                 if (nla_put_u32(user_skb, OVS_PACKET_ATTR_LEN,
509                                 skb->len)) {
510                         err = -ENOBUFS;
511                         goto out;
512                 }
513                 pad_packet(dp, user_skb);
514         }
515
516         /* Only reserve room for attribute header, packet data is added
517          * in skb_zerocopy() */
518         if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
519                 err = -ENOBUFS;
520                 goto out;
521         }
522         nla->nla_len = nla_attr_size(skb->len - cutlen);
523
524         err = skb_zerocopy(user_skb, skb, skb->len - cutlen, hlen);
525         if (err)
526                 goto out;
527
528         /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
529         pad_packet(dp, user_skb);
530
531         ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
532
533         err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
534         user_skb = NULL;
535 out:
536         if (err)
537                 skb_tx_error(skb);
538         kfree_skb(user_skb);
539         kfree_skb(nskb);
540         return err;
541 }
542
543 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
544 {
545         struct ovs_header *ovs_header = info->userhdr;
546         struct net *net = sock_net(skb->sk);
547         struct nlattr **a = info->attrs;
548         struct sw_flow_actions *acts;
549         struct sk_buff *packet;
550         struct sw_flow *flow;
551         struct sw_flow_actions *sf_acts;
552         struct datapath *dp;
553         struct vport *input_vport;
554         u16 mru = 0;
555         int len;
556         int err;
557         bool log = !a[OVS_PACKET_ATTR_PROBE];
558
559         err = -EINVAL;
560         if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
561             !a[OVS_PACKET_ATTR_ACTIONS])
562                 goto err;
563
564         len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
565         packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
566         err = -ENOMEM;
567         if (!packet)
568                 goto err;
569         skb_reserve(packet, NET_IP_ALIGN);
570
571         nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
572
573         /* Set packet's mru */
574         if (a[OVS_PACKET_ATTR_MRU]) {
575                 mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
576                 packet->ignore_df = 1;
577         }
578         OVS_CB(packet)->mru = mru;
579
580         /* Build an sw_flow for sending this packet. */
581         flow = ovs_flow_alloc();
582         err = PTR_ERR(flow);
583         if (IS_ERR(flow))
584                 goto err_kfree_skb;
585
586         err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
587                                              packet, &flow->key, log);
588         if (err)
589                 goto err_flow_free;
590
591         err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
592                                    &flow->key, &acts, log);
593         if (err)
594                 goto err_flow_free;
595
596         rcu_assign_pointer(flow->sf_acts, acts);
597         packet->priority = flow->key.phy.priority;
598         packet->mark = flow->key.phy.skb_mark;
599
600         rcu_read_lock();
601         dp = get_dp_rcu(net, ovs_header->dp_ifindex);
602         err = -ENODEV;
603         if (!dp)
604                 goto err_unlock;
605
606         input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
607         if (!input_vport)
608                 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
609
610         if (!input_vport)
611                 goto err_unlock;
612
613         packet->dev = input_vport->dev;
614         OVS_CB(packet)->input_vport = input_vport;
615         sf_acts = rcu_dereference(flow->sf_acts);
616
617         local_bh_disable();
618         err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
619         local_bh_enable();
620         rcu_read_unlock();
621
622         ovs_flow_free(flow, false);
623         return err;
624
625 err_unlock:
626         rcu_read_unlock();
627 err_flow_free:
628         ovs_flow_free(flow, false);
629 err_kfree_skb:
630         kfree_skb(packet);
631 err:
632         return err;
633 }
634
635 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
636         [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
637         [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
638         [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
639         [OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
640         [OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
641 };
642
643 static const struct genl_ops dp_packet_genl_ops[] = {
644         { .cmd = OVS_PACKET_CMD_EXECUTE,
645           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
646           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
647           .doit = ovs_packet_cmd_execute
648         }
649 };
650
651 static struct genl_family dp_packet_genl_family __ro_after_init = {
652         .hdrsize = sizeof(struct ovs_header),
653         .name = OVS_PACKET_FAMILY,
654         .version = OVS_PACKET_VERSION,
655         .maxattr = OVS_PACKET_ATTR_MAX,
656         .policy = packet_policy,
657         .netnsok = true,
658         .parallel_ops = true,
659         .ops = dp_packet_genl_ops,
660         .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
661         .module = THIS_MODULE,
662 };
663
664 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
665                          struct ovs_dp_megaflow_stats *mega_stats)
666 {
667         int i;
668
669         memset(mega_stats, 0, sizeof(*mega_stats));
670
671         stats->n_flows = ovs_flow_tbl_count(&dp->table);
672         mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
673
674         stats->n_hit = stats->n_missed = stats->n_lost = 0;
675
676         for_each_possible_cpu(i) {
677                 const struct dp_stats_percpu *percpu_stats;
678                 struct dp_stats_percpu local_stats;
679                 unsigned int start;
680
681                 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
682
683                 do {
684                         start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
685                         local_stats = *percpu_stats;
686                 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
687
688                 stats->n_hit += local_stats.n_hit;
689                 stats->n_missed += local_stats.n_missed;
690                 stats->n_lost += local_stats.n_lost;
691                 mega_stats->n_mask_hit += local_stats.n_mask_hit;
692         }
693 }
694
695 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
696 {
697         return ovs_identifier_is_ufid(sfid) &&
698                !(ufid_flags & OVS_UFID_F_OMIT_KEY);
699 }
700
701 static bool should_fill_mask(uint32_t ufid_flags)
702 {
703         return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
704 }
705
706 static bool should_fill_actions(uint32_t ufid_flags)
707 {
708         return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
709 }
710
711 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
712                                     const struct sw_flow_id *sfid,
713                                     uint32_t ufid_flags)
714 {
715         size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
716
717         /* OVS_FLOW_ATTR_UFID */
718         if (sfid && ovs_identifier_is_ufid(sfid))
719                 len += nla_total_size(sfid->ufid_len);
720
721         /* OVS_FLOW_ATTR_KEY */
722         if (!sfid || should_fill_key(sfid, ufid_flags))
723                 len += nla_total_size(ovs_key_attr_size());
724
725         /* OVS_FLOW_ATTR_MASK */
726         if (should_fill_mask(ufid_flags))
727                 len += nla_total_size(ovs_key_attr_size());
728
729         /* OVS_FLOW_ATTR_ACTIONS */
730         if (should_fill_actions(ufid_flags))
731                 len += nla_total_size(acts->orig_len);
732
733         return len
734                 + nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
735                 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
736                 + nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */
737 }
738
739 /* Called with ovs_mutex or RCU read lock. */
740 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
741                                    struct sk_buff *skb)
742 {
743         struct ovs_flow_stats stats;
744         __be16 tcp_flags;
745         unsigned long used;
746
747         ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
748
749         if (used &&
750             nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
751                               OVS_FLOW_ATTR_PAD))
752                 return -EMSGSIZE;
753
754         if (stats.n_packets &&
755             nla_put_64bit(skb, OVS_FLOW_ATTR_STATS,
756                           sizeof(struct ovs_flow_stats), &stats,
757                           OVS_FLOW_ATTR_PAD))
758                 return -EMSGSIZE;
759
760         if ((u8)ntohs(tcp_flags) &&
761              nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
762                 return -EMSGSIZE;
763
764         return 0;
765 }
766
767 /* Called with ovs_mutex or RCU read lock. */
768 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
769                                      struct sk_buff *skb, int skb_orig_len)
770 {
771         struct nlattr *start;
772         int err;
773
774         /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
775          * this is the first flow to be dumped into 'skb'.  This is unusual for
776          * Netlink but individual action lists can be longer than
777          * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
778          * The userspace caller can always fetch the actions separately if it
779          * really wants them.  (Most userspace callers in fact don't care.)
780          *
781          * This can only fail for dump operations because the skb is always
782          * properly sized for single flows.
783          */
784         start = nla_nest_start_noflag(skb, OVS_FLOW_ATTR_ACTIONS);
785         if (start) {
786                 const struct sw_flow_actions *sf_acts;
787
788                 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
789                 err = ovs_nla_put_actions(sf_acts->actions,
790                                           sf_acts->actions_len, skb);
791
792                 if (!err)
793                         nla_nest_end(skb, start);
794                 else {
795                         if (skb_orig_len)
796                                 return err;
797
798                         nla_nest_cancel(skb, start);
799                 }
800         } else if (skb_orig_len) {
801                 return -EMSGSIZE;
802         }
803
804         return 0;
805 }
806
807 /* Called with ovs_mutex or RCU read lock. */
808 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
809                                   struct sk_buff *skb, u32 portid,
810                                   u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
811 {
812         const int skb_orig_len = skb->len;
813         struct ovs_header *ovs_header;
814         int err;
815
816         ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
817                                  flags, cmd);
818         if (!ovs_header)
819                 return -EMSGSIZE;
820
821         ovs_header->dp_ifindex = dp_ifindex;
822
823         err = ovs_nla_put_identifier(flow, skb);
824         if (err)
825                 goto error;
826
827         if (should_fill_key(&flow->id, ufid_flags)) {
828                 err = ovs_nla_put_masked_key(flow, skb);
829                 if (err)
830                         goto error;
831         }
832
833         if (should_fill_mask(ufid_flags)) {
834                 err = ovs_nla_put_mask(flow, skb);
835                 if (err)
836                         goto error;
837         }
838
839         err = ovs_flow_cmd_fill_stats(flow, skb);
840         if (err)
841                 goto error;
842
843         if (should_fill_actions(ufid_flags)) {
844                 err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
845                 if (err)
846                         goto error;
847         }
848
849         genlmsg_end(skb, ovs_header);
850         return 0;
851
852 error:
853         genlmsg_cancel(skb, ovs_header);
854         return err;
855 }
856
857 /* May not be called with RCU read lock. */
858 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
859                                                const struct sw_flow_id *sfid,
860                                                struct genl_info *info,
861                                                bool always,
862                                                uint32_t ufid_flags)
863 {
864         struct sk_buff *skb;
865         size_t len;
866
867         if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
868                 return NULL;
869
870         len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
871         skb = genlmsg_new(len, GFP_KERNEL);
872         if (!skb)
873                 return ERR_PTR(-ENOMEM);
874
875         return skb;
876 }
877
878 /* Called with ovs_mutex. */
879 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
880                                                int dp_ifindex,
881                                                struct genl_info *info, u8 cmd,
882                                                bool always, u32 ufid_flags)
883 {
884         struct sk_buff *skb;
885         int retval;
886
887         skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
888                                       &flow->id, info, always, ufid_flags);
889         if (IS_ERR_OR_NULL(skb))
890                 return skb;
891
892         retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
893                                         info->snd_portid, info->snd_seq, 0,
894                                         cmd, ufid_flags);
895         BUG_ON(retval < 0);
896         return skb;
897 }
898
899 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
900 {
901         struct net *net = sock_net(skb->sk);
902         struct nlattr **a = info->attrs;
903         struct ovs_header *ovs_header = info->userhdr;
904         struct sw_flow *flow = NULL, *new_flow;
905         struct sw_flow_mask mask;
906         struct sk_buff *reply;
907         struct datapath *dp;
908         struct sw_flow_actions *acts;
909         struct sw_flow_match match;
910         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
911         int error;
912         bool log = !a[OVS_FLOW_ATTR_PROBE];
913
914         /* Must have key and actions. */
915         error = -EINVAL;
916         if (!a[OVS_FLOW_ATTR_KEY]) {
917                 OVS_NLERR(log, "Flow key attr not present in new flow.");
918                 goto error;
919         }
920         if (!a[OVS_FLOW_ATTR_ACTIONS]) {
921                 OVS_NLERR(log, "Flow actions attr not present in new flow.");
922                 goto error;
923         }
924
925         /* Most of the time we need to allocate a new flow, do it before
926          * locking.
927          */
928         new_flow = ovs_flow_alloc();
929         if (IS_ERR(new_flow)) {
930                 error = PTR_ERR(new_flow);
931                 goto error;
932         }
933
934         /* Extract key. */
935         ovs_match_init(&match, &new_flow->key, false, &mask);
936         error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
937                                   a[OVS_FLOW_ATTR_MASK], log);
938         if (error)
939                 goto err_kfree_flow;
940
941         /* Extract flow identifier. */
942         error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
943                                        &new_flow->key, log);
944         if (error)
945                 goto err_kfree_flow;
946
947         /* unmasked key is needed to match when ufid is not used. */
948         if (ovs_identifier_is_key(&new_flow->id))
949                 match.key = new_flow->id.unmasked_key;
950
951         ovs_flow_mask_key(&new_flow->key, &new_flow->key, true, &mask);
952
953         /* Validate actions. */
954         error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
955                                      &new_flow->key, &acts, log);
956         if (error) {
957                 OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
958                 goto err_kfree_flow;
959         }
960
961         reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
962                                         ufid_flags);
963         if (IS_ERR(reply)) {
964                 error = PTR_ERR(reply);
965                 goto err_kfree_acts;
966         }
967
968         ovs_lock();
969         dp = get_dp(net, ovs_header->dp_ifindex);
970         if (unlikely(!dp)) {
971                 error = -ENODEV;
972                 goto err_unlock_ovs;
973         }
974
975         /* Check if this is a duplicate flow */
976         if (ovs_identifier_is_ufid(&new_flow->id))
977                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
978         if (!flow)
979                 flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->key);
980         if (likely(!flow)) {
981                 rcu_assign_pointer(new_flow->sf_acts, acts);
982
983                 /* Put flow in bucket. */
984                 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
985                 if (unlikely(error)) {
986                         acts = NULL;
987                         goto err_unlock_ovs;
988                 }
989
990                 if (unlikely(reply)) {
991                         error = ovs_flow_cmd_fill_info(new_flow,
992                                                        ovs_header->dp_ifindex,
993                                                        reply, info->snd_portid,
994                                                        info->snd_seq, 0,
995                                                        OVS_FLOW_CMD_NEW,
996                                                        ufid_flags);
997                         BUG_ON(error < 0);
998                 }
999                 ovs_unlock();
1000         } else {
1001                 struct sw_flow_actions *old_acts;
1002
1003                 /* Bail out if we're not allowed to modify an existing flow.
1004                  * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1005                  * because Generic Netlink treats the latter as a dump
1006                  * request.  We also accept NLM_F_EXCL in case that bug ever
1007                  * gets fixed.
1008                  */
1009                 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1010                                                          | NLM_F_EXCL))) {
1011                         error = -EEXIST;
1012                         goto err_unlock_ovs;
1013                 }
1014                 /* The flow identifier has to be the same for flow updates.
1015                  * Look for any overlapping flow.
1016                  */
1017                 if (unlikely(!ovs_flow_cmp(flow, &match))) {
1018                         if (ovs_identifier_is_key(&flow->id))
1019                                 flow = ovs_flow_tbl_lookup_exact(&dp->table,
1020                                                                  &match);
1021                         else /* UFID matches but key is different */
1022                                 flow = NULL;
1023                         if (!flow) {
1024                                 error = -ENOENT;
1025                                 goto err_unlock_ovs;
1026                         }
1027                 }
1028                 /* Update actions. */
1029                 old_acts = ovsl_dereference(flow->sf_acts);
1030                 rcu_assign_pointer(flow->sf_acts, acts);
1031
1032                 if (unlikely(reply)) {
1033                         error = ovs_flow_cmd_fill_info(flow,
1034                                                        ovs_header->dp_ifindex,
1035                                                        reply, info->snd_portid,
1036                                                        info->snd_seq, 0,
1037                                                        OVS_FLOW_CMD_NEW,
1038                                                        ufid_flags);
1039                         BUG_ON(error < 0);
1040                 }
1041                 ovs_unlock();
1042
1043                 ovs_nla_free_flow_actions_rcu(old_acts);
1044                 ovs_flow_free(new_flow, false);
1045         }
1046
1047         if (reply)
1048                 ovs_notify(&dp_flow_genl_family, reply, info);
1049         return 0;
1050
1051 err_unlock_ovs:
1052         ovs_unlock();
1053         kfree_skb(reply);
1054 err_kfree_acts:
1055         ovs_nla_free_flow_actions(acts);
1056 err_kfree_flow:
1057         ovs_flow_free(new_flow, false);
1058 error:
1059         return error;
1060 }
1061
1062 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1063 static struct sw_flow_actions *get_flow_actions(struct net *net,
1064                                                 const struct nlattr *a,
1065                                                 const struct sw_flow_key *key,
1066                                                 const struct sw_flow_mask *mask,
1067                                                 bool log)
1068 {
1069         struct sw_flow_actions *acts;
1070         struct sw_flow_key masked_key;
1071         int error;
1072
1073         ovs_flow_mask_key(&masked_key, key, true, mask);
1074         error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1075         if (error) {
1076                 OVS_NLERR(log,
1077                           "Actions may not be safe on all matching packets");
1078                 return ERR_PTR(error);
1079         }
1080
1081         return acts;
1082 }
1083
1084 /* Factor out match-init and action-copy to avoid
1085  * "Wframe-larger-than=1024" warning. Because mask is only
1086  * used to get actions, we new a function to save some
1087  * stack space.
1088  *
1089  * If there are not key and action attrs, we return 0
1090  * directly. In the case, the caller will also not use the
1091  * match as before. If there is action attr, we try to get
1092  * actions and save them to *acts. Before returning from
1093  * the function, we reset the match->mask pointer. Because
1094  * we should not to return match object with dangling reference
1095  * to mask.
1096  * */
1097 static int ovs_nla_init_match_and_action(struct net *net,
1098                                          struct sw_flow_match *match,
1099                                          struct sw_flow_key *key,
1100                                          struct nlattr **a,
1101                                          struct sw_flow_actions **acts,
1102                                          bool log)
1103 {
1104         struct sw_flow_mask mask;
1105         int error = 0;
1106
1107         if (a[OVS_FLOW_ATTR_KEY]) {
1108                 ovs_match_init(match, key, true, &mask);
1109                 error = ovs_nla_get_match(net, match, a[OVS_FLOW_ATTR_KEY],
1110                                           a[OVS_FLOW_ATTR_MASK], log);
1111                 if (error)
1112                         goto error;
1113         }
1114
1115         if (a[OVS_FLOW_ATTR_ACTIONS]) {
1116                 if (!a[OVS_FLOW_ATTR_KEY]) {
1117                         OVS_NLERR(log,
1118                                   "Flow key attribute not present in set flow.");
1119                         error = -EINVAL;
1120                         goto error;
1121                 }
1122
1123                 *acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], key,
1124                                          &mask, log);
1125                 if (IS_ERR(*acts)) {
1126                         error = PTR_ERR(*acts);
1127                         goto error;
1128                 }
1129         }
1130
1131         /* On success, error is 0. */
1132 error:
1133         match->mask = NULL;
1134         return error;
1135 }
1136
1137 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1138 {
1139         struct net *net = sock_net(skb->sk);
1140         struct nlattr **a = info->attrs;
1141         struct ovs_header *ovs_header = info->userhdr;
1142         struct sw_flow_key key;
1143         struct sw_flow *flow;
1144         struct sk_buff *reply = NULL;
1145         struct datapath *dp;
1146         struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1147         struct sw_flow_match match;
1148         struct sw_flow_id sfid;
1149         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1150         int error = 0;
1151         bool log = !a[OVS_FLOW_ATTR_PROBE];
1152         bool ufid_present;
1153
1154         ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1155         if (!a[OVS_FLOW_ATTR_KEY] && !ufid_present) {
1156                 OVS_NLERR(log,
1157                           "Flow set message rejected, Key attribute missing.");
1158                 return -EINVAL;
1159         }
1160
1161         error = ovs_nla_init_match_and_action(net, &match, &key, a,
1162                                               &acts, log);
1163         if (error)
1164                 goto error;
1165
1166         if (acts) {
1167                 /* Can allocate before locking if have acts. */
1168                 reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1169                                                 ufid_flags);
1170                 if (IS_ERR(reply)) {
1171                         error = PTR_ERR(reply);
1172                         goto err_kfree_acts;
1173                 }
1174         }
1175
1176         ovs_lock();
1177         dp = get_dp(net, ovs_header->dp_ifindex);
1178         if (unlikely(!dp)) {
1179                 error = -ENODEV;
1180                 goto err_unlock_ovs;
1181         }
1182         /* Check that the flow exists. */
1183         if (ufid_present)
1184                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1185         else
1186                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1187         if (unlikely(!flow)) {
1188                 error = -ENOENT;
1189                 goto err_unlock_ovs;
1190         }
1191
1192         /* Update actions, if present. */
1193         if (likely(acts)) {
1194                 old_acts = ovsl_dereference(flow->sf_acts);
1195                 rcu_assign_pointer(flow->sf_acts, acts);
1196
1197                 if (unlikely(reply)) {
1198                         error = ovs_flow_cmd_fill_info(flow,
1199                                                        ovs_header->dp_ifindex,
1200                                                        reply, info->snd_portid,
1201                                                        info->snd_seq, 0,
1202                                                        OVS_FLOW_CMD_SET,
1203                                                        ufid_flags);
1204                         BUG_ON(error < 0);
1205                 }
1206         } else {
1207                 /* Could not alloc without acts before locking. */
1208                 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1209                                                 info, OVS_FLOW_CMD_SET, false,
1210                                                 ufid_flags);
1211
1212                 if (IS_ERR(reply)) {
1213                         error = PTR_ERR(reply);
1214                         goto err_unlock_ovs;
1215                 }
1216         }
1217
1218         /* Clear stats. */
1219         if (a[OVS_FLOW_ATTR_CLEAR])
1220                 ovs_flow_stats_clear(flow);
1221         ovs_unlock();
1222
1223         if (reply)
1224                 ovs_notify(&dp_flow_genl_family, reply, info);
1225         if (old_acts)
1226                 ovs_nla_free_flow_actions_rcu(old_acts);
1227
1228         return 0;
1229
1230 err_unlock_ovs:
1231         ovs_unlock();
1232         kfree_skb(reply);
1233 err_kfree_acts:
1234         ovs_nla_free_flow_actions(acts);
1235 error:
1236         return error;
1237 }
1238
1239 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1240 {
1241         struct nlattr **a = info->attrs;
1242         struct ovs_header *ovs_header = info->userhdr;
1243         struct net *net = sock_net(skb->sk);
1244         struct sw_flow_key key;
1245         struct sk_buff *reply;
1246         struct sw_flow *flow;
1247         struct datapath *dp;
1248         struct sw_flow_match match;
1249         struct sw_flow_id ufid;
1250         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1251         int err = 0;
1252         bool log = !a[OVS_FLOW_ATTR_PROBE];
1253         bool ufid_present;
1254
1255         ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1256         if (a[OVS_FLOW_ATTR_KEY]) {
1257                 ovs_match_init(&match, &key, true, NULL);
1258                 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1259                                         log);
1260         } else if (!ufid_present) {
1261                 OVS_NLERR(log,
1262                           "Flow get message rejected, Key attribute missing.");
1263                 err = -EINVAL;
1264         }
1265         if (err)
1266                 return err;
1267
1268         ovs_lock();
1269         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1270         if (!dp) {
1271                 err = -ENODEV;
1272                 goto unlock;
1273         }
1274
1275         if (ufid_present)
1276                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1277         else
1278                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1279         if (!flow) {
1280                 err = -ENOENT;
1281                 goto unlock;
1282         }
1283
1284         reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1285                                         OVS_FLOW_CMD_GET, true, ufid_flags);
1286         if (IS_ERR(reply)) {
1287                 err = PTR_ERR(reply);
1288                 goto unlock;
1289         }
1290
1291         ovs_unlock();
1292         return genlmsg_reply(reply, info);
1293 unlock:
1294         ovs_unlock();
1295         return err;
1296 }
1297
1298 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1299 {
1300         struct nlattr **a = info->attrs;
1301         struct ovs_header *ovs_header = info->userhdr;
1302         struct net *net = sock_net(skb->sk);
1303         struct sw_flow_key key;
1304         struct sk_buff *reply;
1305         struct sw_flow *flow = NULL;
1306         struct datapath *dp;
1307         struct sw_flow_match match;
1308         struct sw_flow_id ufid;
1309         u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1310         int err;
1311         bool log = !a[OVS_FLOW_ATTR_PROBE];
1312         bool ufid_present;
1313
1314         ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1315         if (a[OVS_FLOW_ATTR_KEY]) {
1316                 ovs_match_init(&match, &key, true, NULL);
1317                 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1318                                         NULL, log);
1319                 if (unlikely(err))
1320                         return err;
1321         }
1322
1323         ovs_lock();
1324         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1325         if (unlikely(!dp)) {
1326                 err = -ENODEV;
1327                 goto unlock;
1328         }
1329
1330         if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1331                 err = ovs_flow_tbl_flush(&dp->table);
1332                 goto unlock;
1333         }
1334
1335         if (ufid_present)
1336                 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1337         else
1338                 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1339         if (unlikely(!flow)) {
1340                 err = -ENOENT;
1341                 goto unlock;
1342         }
1343
1344         ovs_flow_tbl_remove(&dp->table, flow);
1345         ovs_unlock();
1346
1347         reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1348                                         &flow->id, info, false, ufid_flags);
1349         if (likely(reply)) {
1350                 if (likely(!IS_ERR(reply))) {
1351                         rcu_read_lock();        /*To keep RCU checker happy. */
1352                         err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1353                                                      reply, info->snd_portid,
1354                                                      info->snd_seq, 0,
1355                                                      OVS_FLOW_CMD_DEL,
1356                                                      ufid_flags);
1357                         rcu_read_unlock();
1358                         BUG_ON(err < 0);
1359
1360                         ovs_notify(&dp_flow_genl_family, reply, info);
1361                 } else {
1362                         netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
1363                 }
1364         }
1365
1366         ovs_flow_free(flow, true);
1367         return 0;
1368 unlock:
1369         ovs_unlock();
1370         return err;
1371 }
1372
1373 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1374 {
1375         struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1376         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1377         struct table_instance *ti;
1378         struct datapath *dp;
1379         u32 ufid_flags;
1380         int err;
1381
1382         err = genlmsg_parse_deprecated(cb->nlh, &dp_flow_genl_family, a,
1383                                        OVS_FLOW_ATTR_MAX, flow_policy, NULL);
1384         if (err)
1385                 return err;
1386         ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1387
1388         rcu_read_lock();
1389         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1390         if (!dp) {
1391                 rcu_read_unlock();
1392                 return -ENODEV;
1393         }
1394
1395         ti = rcu_dereference(dp->table.ti);
1396         for (;;) {
1397                 struct sw_flow *flow;
1398                 u32 bucket, obj;
1399
1400                 bucket = cb->args[0];
1401                 obj = cb->args[1];
1402                 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1403                 if (!flow)
1404                         break;
1405
1406                 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1407                                            NETLINK_CB(cb->skb).portid,
1408                                            cb->nlh->nlmsg_seq, NLM_F_MULTI,
1409                                            OVS_FLOW_CMD_GET, ufid_flags) < 0)
1410                         break;
1411
1412                 cb->args[0] = bucket;
1413                 cb->args[1] = obj;
1414         }
1415         rcu_read_unlock();
1416         return skb->len;
1417 }
1418
1419 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1420         [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1421         [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1422         [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1423         [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1424         [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1425         [OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1426         [OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1427 };
1428
1429 static const struct genl_ops dp_flow_genl_ops[] = {
1430         { .cmd = OVS_FLOW_CMD_NEW,
1431           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1432           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1433           .doit = ovs_flow_cmd_new
1434         },
1435         { .cmd = OVS_FLOW_CMD_DEL,
1436           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1437           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1438           .doit = ovs_flow_cmd_del
1439         },
1440         { .cmd = OVS_FLOW_CMD_GET,
1441           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1442           .flags = 0,               /* OK for unprivileged users. */
1443           .doit = ovs_flow_cmd_get,
1444           .dumpit = ovs_flow_cmd_dump
1445         },
1446         { .cmd = OVS_FLOW_CMD_SET,
1447           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1448           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1449           .doit = ovs_flow_cmd_set,
1450         },
1451 };
1452
1453 static struct genl_family dp_flow_genl_family __ro_after_init = {
1454         .hdrsize = sizeof(struct ovs_header),
1455         .name = OVS_FLOW_FAMILY,
1456         .version = OVS_FLOW_VERSION,
1457         .maxattr = OVS_FLOW_ATTR_MAX,
1458         .policy = flow_policy,
1459         .netnsok = true,
1460         .parallel_ops = true,
1461         .ops = dp_flow_genl_ops,
1462         .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1463         .mcgrps = &ovs_dp_flow_multicast_group,
1464         .n_mcgrps = 1,
1465         .module = THIS_MODULE,
1466 };
1467
1468 static size_t ovs_dp_cmd_msg_size(void)
1469 {
1470         size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1471
1472         msgsize += nla_total_size(IFNAMSIZ);
1473         msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1474         msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
1475         msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1476
1477         return msgsize;
1478 }
1479
1480 /* Called with ovs_mutex. */
1481 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1482                                 u32 portid, u32 seq, u32 flags, u8 cmd)
1483 {
1484         struct ovs_header *ovs_header;
1485         struct ovs_dp_stats dp_stats;
1486         struct ovs_dp_megaflow_stats dp_megaflow_stats;
1487         int err;
1488
1489         ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1490                                    flags, cmd);
1491         if (!ovs_header)
1492                 goto error;
1493
1494         ovs_header->dp_ifindex = get_dpifindex(dp);
1495
1496         err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1497         if (err)
1498                 goto nla_put_failure;
1499
1500         get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1501         if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1502                           &dp_stats, OVS_DP_ATTR_PAD))
1503                 goto nla_put_failure;
1504
1505         if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1506                           sizeof(struct ovs_dp_megaflow_stats),
1507                           &dp_megaflow_stats, OVS_DP_ATTR_PAD))
1508                 goto nla_put_failure;
1509
1510         if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1511                 goto nla_put_failure;
1512
1513         genlmsg_end(skb, ovs_header);
1514         return 0;
1515
1516 nla_put_failure:
1517         genlmsg_cancel(skb, ovs_header);
1518 error:
1519         return -EMSGSIZE;
1520 }
1521
1522 static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1523 {
1524         return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1525 }
1526
1527 /* Called with rcu_read_lock or ovs_mutex. */
1528 static struct datapath *lookup_datapath(struct net *net,
1529                                         const struct ovs_header *ovs_header,
1530                                         struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1531 {
1532         struct datapath *dp;
1533
1534         if (!a[OVS_DP_ATTR_NAME])
1535                 dp = get_dp(net, ovs_header->dp_ifindex);
1536         else {
1537                 struct vport *vport;
1538
1539                 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1540                 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1541         }
1542         return dp ? dp : ERR_PTR(-ENODEV);
1543 }
1544
1545 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1546 {
1547         struct datapath *dp;
1548
1549         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1550         if (IS_ERR(dp))
1551                 return;
1552
1553         WARN(dp->user_features, "Dropping previously announced user features\n");
1554         dp->user_features = 0;
1555 }
1556
1557 static void ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1558 {
1559         if (a[OVS_DP_ATTR_USER_FEATURES])
1560                 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1561 }
1562
1563 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1564 {
1565         struct nlattr **a = info->attrs;
1566         struct vport_parms parms;
1567         struct sk_buff *reply;
1568         struct datapath *dp;
1569         struct vport *vport;
1570         struct ovs_net *ovs_net;
1571         int err, i;
1572
1573         err = -EINVAL;
1574         if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1575                 goto err;
1576
1577         reply = ovs_dp_cmd_alloc_info();
1578         if (!reply)
1579                 return -ENOMEM;
1580
1581         err = -ENOMEM;
1582         dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1583         if (dp == NULL)
1584                 goto err_free_reply;
1585
1586         ovs_dp_set_net(dp, sock_net(skb->sk));
1587
1588         /* Allocate table. */
1589         err = ovs_flow_tbl_init(&dp->table);
1590         if (err)
1591                 goto err_free_dp;
1592
1593         dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1594         if (!dp->stats_percpu) {
1595                 err = -ENOMEM;
1596                 goto err_destroy_table;
1597         }
1598
1599         dp->ports = kmalloc_array(DP_VPORT_HASH_BUCKETS,
1600                                   sizeof(struct hlist_head),
1601                                   GFP_KERNEL);
1602         if (!dp->ports) {
1603                 err = -ENOMEM;
1604                 goto err_destroy_percpu;
1605         }
1606
1607         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1608                 INIT_HLIST_HEAD(&dp->ports[i]);
1609
1610         err = ovs_meters_init(dp);
1611         if (err)
1612                 goto err_destroy_ports_array;
1613
1614         /* Set up our datapath device. */
1615         parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1616         parms.type = OVS_VPORT_TYPE_INTERNAL;
1617         parms.options = NULL;
1618         parms.dp = dp;
1619         parms.port_no = OVSP_LOCAL;
1620         parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1621
1622         ovs_dp_change(dp, a);
1623
1624         /* So far only local changes have been made, now need the lock. */
1625         ovs_lock();
1626
1627         vport = new_vport(&parms);
1628         if (IS_ERR(vport)) {
1629                 err = PTR_ERR(vport);
1630                 if (err == -EBUSY)
1631                         err = -EEXIST;
1632
1633                 if (err == -EEXIST) {
1634                         /* An outdated user space instance that does not understand
1635                          * the concept of user_features has attempted to create a new
1636                          * datapath and is likely to reuse it. Drop all user features.
1637                          */
1638                         if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1639                                 ovs_dp_reset_user_features(skb, info);
1640                 }
1641
1642                 goto err_destroy_meters;
1643         }
1644
1645         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1646                                    info->snd_seq, 0, OVS_DP_CMD_NEW);
1647         BUG_ON(err < 0);
1648
1649         ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1650         list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1651
1652         ovs_unlock();
1653
1654         ovs_notify(&dp_datapath_genl_family, reply, info);
1655         return 0;
1656
1657 err_destroy_meters:
1658         ovs_unlock();
1659         ovs_meters_exit(dp);
1660 err_destroy_ports_array:
1661         kfree(dp->ports);
1662 err_destroy_percpu:
1663         free_percpu(dp->stats_percpu);
1664 err_destroy_table:
1665         ovs_flow_tbl_destroy(&dp->table);
1666 err_free_dp:
1667         kfree(dp);
1668 err_free_reply:
1669         kfree_skb(reply);
1670 err:
1671         return err;
1672 }
1673
1674 /* Called with ovs_mutex. */
1675 static void __dp_destroy(struct datapath *dp)
1676 {
1677         int i;
1678
1679         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1680                 struct vport *vport;
1681                 struct hlist_node *n;
1682
1683                 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1684                         if (vport->port_no != OVSP_LOCAL)
1685                                 ovs_dp_detach_port(vport);
1686         }
1687
1688         list_del_rcu(&dp->list_node);
1689
1690         /* OVSP_LOCAL is datapath internal port. We need to make sure that
1691          * all ports in datapath are destroyed first before freeing datapath.
1692          */
1693         ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1694
1695         /* RCU destroy the flow table */
1696         call_rcu(&dp->rcu, destroy_dp_rcu);
1697 }
1698
1699 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1700 {
1701         struct sk_buff *reply;
1702         struct datapath *dp;
1703         int err;
1704
1705         reply = ovs_dp_cmd_alloc_info();
1706         if (!reply)
1707                 return -ENOMEM;
1708
1709         ovs_lock();
1710         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1711         err = PTR_ERR(dp);
1712         if (IS_ERR(dp))
1713                 goto err_unlock_free;
1714
1715         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1716                                    info->snd_seq, 0, OVS_DP_CMD_DEL);
1717         BUG_ON(err < 0);
1718
1719         __dp_destroy(dp);
1720         ovs_unlock();
1721
1722         ovs_notify(&dp_datapath_genl_family, reply, info);
1723
1724         return 0;
1725
1726 err_unlock_free:
1727         ovs_unlock();
1728         kfree_skb(reply);
1729         return err;
1730 }
1731
1732 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1733 {
1734         struct sk_buff *reply;
1735         struct datapath *dp;
1736         int err;
1737
1738         reply = ovs_dp_cmd_alloc_info();
1739         if (!reply)
1740                 return -ENOMEM;
1741
1742         ovs_lock();
1743         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1744         err = PTR_ERR(dp);
1745         if (IS_ERR(dp))
1746                 goto err_unlock_free;
1747
1748         ovs_dp_change(dp, info->attrs);
1749
1750         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1751                                    info->snd_seq, 0, OVS_DP_CMD_SET);
1752         BUG_ON(err < 0);
1753
1754         ovs_unlock();
1755         ovs_notify(&dp_datapath_genl_family, reply, info);
1756
1757         return 0;
1758
1759 err_unlock_free:
1760         ovs_unlock();
1761         kfree_skb(reply);
1762         return err;
1763 }
1764
1765 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1766 {
1767         struct sk_buff *reply;
1768         struct datapath *dp;
1769         int err;
1770
1771         reply = ovs_dp_cmd_alloc_info();
1772         if (!reply)
1773                 return -ENOMEM;
1774
1775         ovs_lock();
1776         dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1777         if (IS_ERR(dp)) {
1778                 err = PTR_ERR(dp);
1779                 goto err_unlock_free;
1780         }
1781         err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1782                                    info->snd_seq, 0, OVS_DP_CMD_GET);
1783         BUG_ON(err < 0);
1784         ovs_unlock();
1785
1786         return genlmsg_reply(reply, info);
1787
1788 err_unlock_free:
1789         ovs_unlock();
1790         kfree_skb(reply);
1791         return err;
1792 }
1793
1794 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1795 {
1796         struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1797         struct datapath *dp;
1798         int skip = cb->args[0];
1799         int i = 0;
1800
1801         ovs_lock();
1802         list_for_each_entry(dp, &ovs_net->dps, list_node) {
1803                 if (i >= skip &&
1804                     ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1805                                          cb->nlh->nlmsg_seq, NLM_F_MULTI,
1806                                          OVS_DP_CMD_GET) < 0)
1807                         break;
1808                 i++;
1809         }
1810         ovs_unlock();
1811
1812         cb->args[0] = i;
1813
1814         return skb->len;
1815 }
1816
1817 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1818         [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1819         [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1820         [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1821 };
1822
1823 static const struct genl_ops dp_datapath_genl_ops[] = {
1824         { .cmd = OVS_DP_CMD_NEW,
1825           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1826           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1827           .doit = ovs_dp_cmd_new
1828         },
1829         { .cmd = OVS_DP_CMD_DEL,
1830           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1831           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1832           .doit = ovs_dp_cmd_del
1833         },
1834         { .cmd = OVS_DP_CMD_GET,
1835           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1836           .flags = 0,               /* OK for unprivileged users. */
1837           .doit = ovs_dp_cmd_get,
1838           .dumpit = ovs_dp_cmd_dump
1839         },
1840         { .cmd = OVS_DP_CMD_SET,
1841           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1842           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1843           .doit = ovs_dp_cmd_set,
1844         },
1845 };
1846
1847 static struct genl_family dp_datapath_genl_family __ro_after_init = {
1848         .hdrsize = sizeof(struct ovs_header),
1849         .name = OVS_DATAPATH_FAMILY,
1850         .version = OVS_DATAPATH_VERSION,
1851         .maxattr = OVS_DP_ATTR_MAX,
1852         .policy = datapath_policy,
1853         .netnsok = true,
1854         .parallel_ops = true,
1855         .ops = dp_datapath_genl_ops,
1856         .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1857         .mcgrps = &ovs_dp_datapath_multicast_group,
1858         .n_mcgrps = 1,
1859         .module = THIS_MODULE,
1860 };
1861
1862 /* Called with ovs_mutex or RCU read lock. */
1863 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1864                                    struct net *net, u32 portid, u32 seq,
1865                                    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             nla_put_u32(skb, OVS_VPORT_ATTR_IFINDEX, vport->dev->ifindex))
1883                 goto nla_put_failure;
1884
1885         if (!net_eq(net, dev_net(vport->dev))) {
1886                 int id = peernet2id_alloc(net, dev_net(vport->dev));
1887
1888                 if (nla_put_s32(skb, OVS_VPORT_ATTR_NETNSID, id))
1889                         goto nla_put_failure;
1890         }
1891
1892         ovs_vport_get_stats(vport, &vport_stats);
1893         if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
1894                           sizeof(struct ovs_vport_stats), &vport_stats,
1895                           OVS_VPORT_ATTR_PAD))
1896                 goto nla_put_failure;
1897
1898         if (ovs_vport_get_upcall_portids(vport, skb))
1899                 goto nla_put_failure;
1900
1901         err = ovs_vport_get_options(vport, skb);
1902         if (err == -EMSGSIZE)
1903                 goto error;
1904
1905         genlmsg_end(skb, ovs_header);
1906         return 0;
1907
1908 nla_put_failure:
1909         err = -EMSGSIZE;
1910 error:
1911         genlmsg_cancel(skb, ovs_header);
1912         return err;
1913 }
1914
1915 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1916 {
1917         return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1918 }
1919
1920 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1921 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, struct net *net,
1922                                          u32 portid, u32 seq, u8 cmd)
1923 {
1924         struct sk_buff *skb;
1925         int retval;
1926
1927         skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1928         if (!skb)
1929                 return ERR_PTR(-ENOMEM);
1930
1931         retval = ovs_vport_cmd_fill_info(vport, skb, net, portid, seq, 0, cmd);
1932         BUG_ON(retval < 0);
1933
1934         return skb;
1935 }
1936
1937 /* Called with ovs_mutex or RCU read lock. */
1938 static struct vport *lookup_vport(struct net *net,
1939                                   const struct ovs_header *ovs_header,
1940                                   struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1941 {
1942         struct datapath *dp;
1943         struct vport *vport;
1944
1945         if (a[OVS_VPORT_ATTR_IFINDEX])
1946                 return ERR_PTR(-EOPNOTSUPP);
1947         if (a[OVS_VPORT_ATTR_NAME]) {
1948                 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1949                 if (!vport)
1950                         return ERR_PTR(-ENODEV);
1951                 if (ovs_header->dp_ifindex &&
1952                     ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1953                         return ERR_PTR(-ENODEV);
1954                 return vport;
1955         } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1956                 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1957
1958                 if (port_no >= DP_MAX_PORTS)
1959                         return ERR_PTR(-EFBIG);
1960
1961                 dp = get_dp(net, ovs_header->dp_ifindex);
1962                 if (!dp)
1963                         return ERR_PTR(-ENODEV);
1964
1965                 vport = ovs_vport_ovsl_rcu(dp, port_no);
1966                 if (!vport)
1967                         return ERR_PTR(-ENODEV);
1968                 return vport;
1969         } else
1970                 return ERR_PTR(-EINVAL);
1971
1972 }
1973
1974 /* Called with ovs_mutex */
1975 static void update_headroom(struct datapath *dp)
1976 {
1977         unsigned dev_headroom, max_headroom = 0;
1978         struct net_device *dev;
1979         struct vport *vport;
1980         int i;
1981
1982         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1983                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1984                         dev = vport->dev;
1985                         dev_headroom = netdev_get_fwd_headroom(dev);
1986                         if (dev_headroom > max_headroom)
1987                                 max_headroom = dev_headroom;
1988                 }
1989         }
1990
1991         dp->max_headroom = max_headroom;
1992         for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1993                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node)
1994                         netdev_set_rx_headroom(vport->dev, max_headroom);
1995 }
1996
1997 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1998 {
1999         struct nlattr **a = info->attrs;
2000         struct ovs_header *ovs_header = info->userhdr;
2001         struct vport_parms parms;
2002         struct sk_buff *reply;
2003         struct vport *vport;
2004         struct datapath *dp;
2005         u32 port_no;
2006         int err;
2007
2008         if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
2009             !a[OVS_VPORT_ATTR_UPCALL_PID])
2010                 return -EINVAL;
2011         if (a[OVS_VPORT_ATTR_IFINDEX])
2012                 return -EOPNOTSUPP;
2013
2014         port_no = a[OVS_VPORT_ATTR_PORT_NO]
2015                 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
2016         if (port_no >= DP_MAX_PORTS)
2017                 return -EFBIG;
2018
2019         reply = ovs_vport_cmd_alloc_info();
2020         if (!reply)
2021                 return -ENOMEM;
2022
2023         ovs_lock();
2024 restart:
2025         dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2026         err = -ENODEV;
2027         if (!dp)
2028                 goto exit_unlock_free;
2029
2030         if (port_no) {
2031                 vport = ovs_vport_ovsl(dp, port_no);
2032                 err = -EBUSY;
2033                 if (vport)
2034                         goto exit_unlock_free;
2035         } else {
2036                 for (port_no = 1; ; port_no++) {
2037                         if (port_no >= DP_MAX_PORTS) {
2038                                 err = -EFBIG;
2039                                 goto exit_unlock_free;
2040                         }
2041                         vport = ovs_vport_ovsl(dp, port_no);
2042                         if (!vport)
2043                                 break;
2044                 }
2045         }
2046
2047         parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2048         parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2049         parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2050         parms.dp = dp;
2051         parms.port_no = port_no;
2052         parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
2053
2054         vport = new_vport(&parms);
2055         err = PTR_ERR(vport);
2056         if (IS_ERR(vport)) {
2057                 if (err == -EAGAIN)
2058                         goto restart;
2059                 goto exit_unlock_free;
2060         }
2061
2062         err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2063                                       info->snd_portid, info->snd_seq, 0,
2064                                       OVS_VPORT_CMD_NEW);
2065
2066         if (netdev_get_fwd_headroom(vport->dev) > dp->max_headroom)
2067                 update_headroom(dp);
2068         else
2069                 netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2070
2071         BUG_ON(err < 0);
2072         ovs_unlock();
2073
2074         ovs_notify(&dp_vport_genl_family, reply, info);
2075         return 0;
2076
2077 exit_unlock_free:
2078         ovs_unlock();
2079         kfree_skb(reply);
2080         return err;
2081 }
2082
2083 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2084 {
2085         struct nlattr **a = info->attrs;
2086         struct sk_buff *reply;
2087         struct vport *vport;
2088         int err;
2089
2090         reply = ovs_vport_cmd_alloc_info();
2091         if (!reply)
2092                 return -ENOMEM;
2093
2094         ovs_lock();
2095         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2096         err = PTR_ERR(vport);
2097         if (IS_ERR(vport))
2098                 goto exit_unlock_free;
2099
2100         if (a[OVS_VPORT_ATTR_TYPE] &&
2101             nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2102                 err = -EINVAL;
2103                 goto exit_unlock_free;
2104         }
2105
2106         if (a[OVS_VPORT_ATTR_OPTIONS]) {
2107                 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2108                 if (err)
2109                         goto exit_unlock_free;
2110         }
2111
2112
2113         if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2114                 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2115
2116                 err = ovs_vport_set_upcall_portids(vport, ids);
2117                 if (err)
2118                         goto exit_unlock_free;
2119         }
2120
2121         err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2122                                       info->snd_portid, info->snd_seq, 0,
2123                                       OVS_VPORT_CMD_SET);
2124         BUG_ON(err < 0);
2125
2126         ovs_unlock();
2127         ovs_notify(&dp_vport_genl_family, reply, info);
2128         return 0;
2129
2130 exit_unlock_free:
2131         ovs_unlock();
2132         kfree_skb(reply);
2133         return err;
2134 }
2135
2136 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2137 {
2138         bool must_update_headroom = false;
2139         struct nlattr **a = info->attrs;
2140         struct sk_buff *reply;
2141         struct datapath *dp;
2142         struct vport *vport;
2143         int err;
2144
2145         reply = ovs_vport_cmd_alloc_info();
2146         if (!reply)
2147                 return -ENOMEM;
2148
2149         ovs_lock();
2150         vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2151         err = PTR_ERR(vport);
2152         if (IS_ERR(vport))
2153                 goto exit_unlock_free;
2154
2155         if (vport->port_no == OVSP_LOCAL) {
2156                 err = -EINVAL;
2157                 goto exit_unlock_free;
2158         }
2159
2160         err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2161                                       info->snd_portid, info->snd_seq, 0,
2162                                       OVS_VPORT_CMD_DEL);
2163         BUG_ON(err < 0);
2164
2165         /* the vport deletion may trigger dp headroom update */
2166         dp = vport->dp;
2167         if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2168                 must_update_headroom = true;
2169         netdev_reset_rx_headroom(vport->dev);
2170         ovs_dp_detach_port(vport);
2171
2172         if (must_update_headroom)
2173                 update_headroom(dp);
2174         ovs_unlock();
2175
2176         ovs_notify(&dp_vport_genl_family, reply, info);
2177         return 0;
2178
2179 exit_unlock_free:
2180         ovs_unlock();
2181         kfree_skb(reply);
2182         return err;
2183 }
2184
2185 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2186 {
2187         struct nlattr **a = info->attrs;
2188         struct ovs_header *ovs_header = info->userhdr;
2189         struct sk_buff *reply;
2190         struct vport *vport;
2191         int err;
2192
2193         reply = ovs_vport_cmd_alloc_info();
2194         if (!reply)
2195                 return -ENOMEM;
2196
2197         rcu_read_lock();
2198         vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2199         err = PTR_ERR(vport);
2200         if (IS_ERR(vport))
2201                 goto exit_unlock_free;
2202         err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2203                                       info->snd_portid, info->snd_seq, 0,
2204                                       OVS_VPORT_CMD_GET);
2205         BUG_ON(err < 0);
2206         rcu_read_unlock();
2207
2208         return genlmsg_reply(reply, info);
2209
2210 exit_unlock_free:
2211         rcu_read_unlock();
2212         kfree_skb(reply);
2213         return err;
2214 }
2215
2216 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2217 {
2218         struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2219         struct datapath *dp;
2220         int bucket = cb->args[0], skip = cb->args[1];
2221         int i, j = 0;
2222
2223         rcu_read_lock();
2224         dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2225         if (!dp) {
2226                 rcu_read_unlock();
2227                 return -ENODEV;
2228         }
2229         for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2230                 struct vport *vport;
2231
2232                 j = 0;
2233                 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2234                         if (j >= skip &&
2235                             ovs_vport_cmd_fill_info(vport, skb,
2236                                                     sock_net(skb->sk),
2237                                                     NETLINK_CB(cb->skb).portid,
2238                                                     cb->nlh->nlmsg_seq,
2239                                                     NLM_F_MULTI,
2240                                                     OVS_VPORT_CMD_GET) < 0)
2241                                 goto out;
2242
2243                         j++;
2244                 }
2245                 skip = 0;
2246         }
2247 out:
2248         rcu_read_unlock();
2249
2250         cb->args[0] = i;
2251         cb->args[1] = j;
2252
2253         return skb->len;
2254 }
2255
2256 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2257         [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2258         [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2259         [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2260         [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2261         [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
2262         [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2263         [OVS_VPORT_ATTR_IFINDEX] = { .type = NLA_U32 },
2264         [OVS_VPORT_ATTR_NETNSID] = { .type = NLA_S32 },
2265 };
2266
2267 static const struct genl_ops dp_vport_genl_ops[] = {
2268         { .cmd = OVS_VPORT_CMD_NEW,
2269           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2270           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2271           .doit = ovs_vport_cmd_new
2272         },
2273         { .cmd = OVS_VPORT_CMD_DEL,
2274           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2275           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2276           .doit = ovs_vport_cmd_del
2277         },
2278         { .cmd = OVS_VPORT_CMD_GET,
2279           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2280           .flags = 0,               /* OK for unprivileged users. */
2281           .doit = ovs_vport_cmd_get,
2282           .dumpit = ovs_vport_cmd_dump
2283         },
2284         { .cmd = OVS_VPORT_CMD_SET,
2285           .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2286           .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2287           .doit = ovs_vport_cmd_set,
2288         },
2289 };
2290
2291 struct genl_family dp_vport_genl_family __ro_after_init = {
2292         .hdrsize = sizeof(struct ovs_header),
2293         .name = OVS_VPORT_FAMILY,
2294         .version = OVS_VPORT_VERSION,
2295         .maxattr = OVS_VPORT_ATTR_MAX,
2296         .policy = vport_policy,
2297         .netnsok = true,
2298         .parallel_ops = true,
2299         .ops = dp_vport_genl_ops,
2300         .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2301         .mcgrps = &ovs_dp_vport_multicast_group,
2302         .n_mcgrps = 1,
2303         .module = THIS_MODULE,
2304 };
2305
2306 static struct genl_family * const dp_genl_families[] = {
2307         &dp_datapath_genl_family,
2308         &dp_vport_genl_family,
2309         &dp_flow_genl_family,
2310         &dp_packet_genl_family,
2311         &dp_meter_genl_family,
2312 #if     IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2313         &dp_ct_limit_genl_family,
2314 #endif
2315 };
2316
2317 static void dp_unregister_genl(int n_families)
2318 {
2319         int i;
2320
2321         for (i = 0; i < n_families; i++)
2322                 genl_unregister_family(dp_genl_families[i]);
2323 }
2324
2325 static int __init dp_register_genl(void)
2326 {
2327         int err;
2328         int i;
2329
2330         for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2331
2332                 err = genl_register_family(dp_genl_families[i]);
2333                 if (err)
2334                         goto error;
2335         }
2336
2337         return 0;
2338
2339 error:
2340         dp_unregister_genl(i);
2341         return err;
2342 }
2343
2344 static int __net_init ovs_init_net(struct net *net)
2345 {
2346         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2347
2348         INIT_LIST_HEAD(&ovs_net->dps);
2349         INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2350         return ovs_ct_init(net);
2351 }
2352
2353 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2354                                             struct list_head *head)
2355 {
2356         struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2357         struct datapath *dp;
2358
2359         list_for_each_entry(dp, &ovs_net->dps, list_node) {
2360                 int i;
2361
2362                 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2363                         struct vport *vport;
2364
2365                         hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2366                                 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2367                                         continue;
2368
2369                                 if (dev_net(vport->dev) == dnet)
2370                                         list_add(&vport->detach_list, head);
2371                         }
2372                 }
2373         }
2374 }
2375
2376 static void __net_exit ovs_exit_net(struct net *dnet)
2377 {
2378         struct datapath *dp, *dp_next;
2379         struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2380         struct vport *vport, *vport_next;
2381         struct net *net;
2382         LIST_HEAD(head);
2383
2384         ovs_ct_exit(dnet);
2385         ovs_lock();
2386         list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2387                 __dp_destroy(dp);
2388
2389         down_read(&net_rwsem);
2390         for_each_net(net)
2391                 list_vports_from_net(net, dnet, &head);
2392         up_read(&net_rwsem);
2393
2394         /* Detach all vports from given namespace. */
2395         list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2396                 list_del(&vport->detach_list);
2397                 ovs_dp_detach_port(vport);
2398         }
2399
2400         ovs_unlock();
2401
2402         cancel_work_sync(&ovs_net->dp_notify_work);
2403 }
2404
2405 static struct pernet_operations ovs_net_ops = {
2406         .init = ovs_init_net,
2407         .exit = ovs_exit_net,
2408         .id   = &ovs_net_id,
2409         .size = sizeof(struct ovs_net),
2410 };
2411
2412 static int __init dp_init(void)
2413 {
2414         int err;
2415
2416         BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2417
2418         pr_info("Open vSwitch switching datapath\n");
2419
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 = register_netdevice_notifier(&ovs_dp_device_notifier);
2441         if (err)
2442                 goto error_netns_exit;
2443
2444         err = ovs_netdev_init();
2445         if (err)
2446                 goto error_unreg_notifier;
2447
2448         err = dp_register_genl();
2449         if (err < 0)
2450                 goto error_unreg_netdev;
2451
2452         return 0;
2453
2454 error_unreg_netdev:
2455         ovs_netdev_exit();
2456 error_unreg_notifier:
2457         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2458 error_netns_exit:
2459         unregister_pernet_device(&ovs_net_ops);
2460 error_vport_exit:
2461         ovs_vport_exit();
2462 error_flow_exit:
2463         ovs_flow_exit();
2464 error_unreg_rtnl_link:
2465         ovs_internal_dev_rtnl_link_unregister();
2466 error_action_fifos_exit:
2467         action_fifos_exit();
2468 error:
2469         return err;
2470 }
2471
2472 static void dp_cleanup(void)
2473 {
2474         dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2475         ovs_netdev_exit();
2476         unregister_netdevice_notifier(&ovs_dp_device_notifier);
2477         unregister_pernet_device(&ovs_net_ops);
2478         rcu_barrier();
2479         ovs_vport_exit();
2480         ovs_flow_exit();
2481         ovs_internal_dev_rtnl_link_unregister();
2482         action_fifos_exit();
2483 }
2484
2485 module_init(dp_init);
2486 module_exit(dp_cleanup);
2487
2488 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2489 MODULE_LICENSE("GPL");
2490 MODULE_ALIAS_GENL_FAMILY(OVS_DATAPATH_FAMILY);
2491 MODULE_ALIAS_GENL_FAMILY(OVS_VPORT_FAMILY);
2492 MODULE_ALIAS_GENL_FAMILY(OVS_FLOW_FAMILY);
2493 MODULE_ALIAS_GENL_FAMILY(OVS_PACKET_FAMILY);
2494 MODULE_ALIAS_GENL_FAMILY(OVS_METER_FAMILY);
2495 MODULE_ALIAS_GENL_FAMILY(OVS_CT_LIMIT_FAMILY);