octeontx2-pf: Fix error return code in otx2_probe()
[linux/fpc-iii.git] / net / openvswitch / datapath.c
blob94b024534987a02428e7bd579e531958ca4b190b
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * Copyright (c) 2007-2014 Nicira, Inc.
4 */
6 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
8 #include <linux/init.h>
9 #include <linux/module.h>
10 #include <linux/if_arp.h>
11 #include <linux/if_vlan.h>
12 #include <linux/in.h>
13 #include <linux/ip.h>
14 #include <linux/jhash.h>
15 #include <linux/delay.h>
16 #include <linux/time.h>
17 #include <linux/etherdevice.h>
18 #include <linux/genetlink.h>
19 #include <linux/kernel.h>
20 #include <linux/kthread.h>
21 #include <linux/mutex.h>
22 #include <linux/percpu.h>
23 #include <linux/rcupdate.h>
24 #include <linux/tcp.h>
25 #include <linux/udp.h>
26 #include <linux/ethtool.h>
27 #include <linux/wait.h>
28 #include <asm/div64.h>
29 #include <linux/highmem.h>
30 #include <linux/netfilter_bridge.h>
31 #include <linux/netfilter_ipv4.h>
32 #include <linux/inetdevice.h>
33 #include <linux/list.h>
34 #include <linux/openvswitch.h>
35 #include <linux/rculist.h>
36 #include <linux/dmi.h>
37 #include <net/genetlink.h>
38 #include <net/net_namespace.h>
39 #include <net/netns/generic.h>
41 #include "datapath.h"
42 #include "flow.h"
43 #include "flow_table.h"
44 #include "flow_netlink.h"
45 #include "meter.h"
46 #include "vport-internal_dev.h"
47 #include "vport-netdev.h"
49 unsigned int ovs_net_id __read_mostly;
51 static struct genl_family dp_packet_genl_family;
52 static struct genl_family dp_flow_genl_family;
53 static struct genl_family dp_datapath_genl_family;
55 static const struct nla_policy flow_policy[];
57 static const struct genl_multicast_group ovs_dp_flow_multicast_group = {
58 .name = OVS_FLOW_MCGROUP,
61 static const struct genl_multicast_group ovs_dp_datapath_multicast_group = {
62 .name = OVS_DATAPATH_MCGROUP,
65 static const struct genl_multicast_group ovs_dp_vport_multicast_group = {
66 .name = OVS_VPORT_MCGROUP,
69 /* Check if need to build a reply message.
70 * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
71 static bool ovs_must_notify(struct genl_family *family, struct genl_info *info,
72 unsigned int group)
74 return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
75 genl_has_listeners(family, genl_info_net(info), group);
78 static void ovs_notify(struct genl_family *family,
79 struct sk_buff *skb, struct genl_info *info)
81 genl_notify(family, skb, info, 0, GFP_KERNEL);
84 /**
85 * DOC: Locking:
87 * All writes e.g. Writes to device state (add/remove datapath, port, set
88 * operations on vports, etc.), Writes to other state (flow table
89 * modifications, set miscellaneous datapath parameters, etc.) are protected
90 * by ovs_lock.
92 * Reads are protected by RCU.
94 * There are a few special cases (mostly stats) that have their own
95 * synchronization but they nest under all of above and don't interact with
96 * each other.
98 * The RTNL lock nests inside ovs_mutex.
101 static DEFINE_MUTEX(ovs_mutex);
103 void ovs_lock(void)
105 mutex_lock(&ovs_mutex);
108 void ovs_unlock(void)
110 mutex_unlock(&ovs_mutex);
113 #ifdef CONFIG_LOCKDEP
114 int lockdep_ovsl_is_held(void)
116 if (debug_locks)
117 return lockdep_is_held(&ovs_mutex);
118 else
119 return 1;
121 #endif
123 static struct vport *new_vport(const struct vport_parms *);
124 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
125 const struct sw_flow_key *,
126 const struct dp_upcall_info *,
127 uint32_t cutlen);
128 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
129 const struct sw_flow_key *,
130 const struct dp_upcall_info *,
131 uint32_t cutlen);
133 /* Must be called with rcu_read_lock or ovs_mutex. */
134 const char *ovs_dp_name(const struct datapath *dp)
136 struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
137 return ovs_vport_name(vport);
140 static int get_dpifindex(const struct datapath *dp)
142 struct vport *local;
143 int ifindex;
145 rcu_read_lock();
147 local = ovs_vport_rcu(dp, OVSP_LOCAL);
148 if (local)
149 ifindex = local->dev->ifindex;
150 else
151 ifindex = 0;
153 rcu_read_unlock();
155 return ifindex;
158 static void destroy_dp_rcu(struct rcu_head *rcu)
160 struct datapath *dp = container_of(rcu, struct datapath, rcu);
162 ovs_flow_tbl_destroy(&dp->table);
163 free_percpu(dp->stats_percpu);
164 kfree(dp->ports);
165 ovs_meters_exit(dp);
166 kfree(dp);
169 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
170 u16 port_no)
172 return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
175 /* Called with ovs_mutex or RCU read lock. */
176 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
178 struct vport *vport;
179 struct hlist_head *head;
181 head = vport_hash_bucket(dp, port_no);
182 hlist_for_each_entry_rcu(vport, head, dp_hash_node,
183 lockdep_ovsl_is_held()) {
184 if (vport->port_no == port_no)
185 return vport;
187 return NULL;
190 /* Called with ovs_mutex. */
191 static struct vport *new_vport(const struct vport_parms *parms)
193 struct vport *vport;
195 vport = ovs_vport_add(parms);
196 if (!IS_ERR(vport)) {
197 struct datapath *dp = parms->dp;
198 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
200 hlist_add_head_rcu(&vport->dp_hash_node, head);
202 return vport;
205 void ovs_dp_detach_port(struct vport *p)
207 ASSERT_OVSL();
209 /* First drop references to device. */
210 hlist_del_rcu(&p->dp_hash_node);
212 /* Then destroy it. */
213 ovs_vport_del(p);
216 /* Must be called with rcu_read_lock. */
217 void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
219 const struct vport *p = OVS_CB(skb)->input_vport;
220 struct datapath *dp = p->dp;
221 struct sw_flow *flow;
222 struct sw_flow_actions *sf_acts;
223 struct dp_stats_percpu *stats;
224 u64 *stats_counter;
225 u32 n_mask_hit;
226 int error;
228 stats = this_cpu_ptr(dp->stats_percpu);
230 /* Look up flow. */
231 flow = ovs_flow_tbl_lookup_stats(&dp->table, key, skb_get_hash(skb),
232 &n_mask_hit);
233 if (unlikely(!flow)) {
234 struct dp_upcall_info upcall;
236 memset(&upcall, 0, sizeof(upcall));
237 upcall.cmd = OVS_PACKET_CMD_MISS;
238 upcall.portid = ovs_vport_find_upcall_portid(p, skb);
239 upcall.mru = OVS_CB(skb)->mru;
240 error = ovs_dp_upcall(dp, skb, key, &upcall, 0);
241 if (unlikely(error))
242 kfree_skb(skb);
243 else
244 consume_skb(skb);
245 stats_counter = &stats->n_missed;
246 goto out;
249 ovs_flow_stats_update(flow, key->tp.flags, skb);
250 sf_acts = rcu_dereference(flow->sf_acts);
251 error = ovs_execute_actions(dp, skb, sf_acts, key);
252 if (unlikely(error))
253 net_dbg_ratelimited("ovs: action execution error on datapath %s: %d\n",
254 ovs_dp_name(dp), error);
256 stats_counter = &stats->n_hit;
258 out:
259 /* Update datapath statistics. */
260 u64_stats_update_begin(&stats->syncp);
261 (*stats_counter)++;
262 stats->n_mask_hit += n_mask_hit;
263 u64_stats_update_end(&stats->syncp);
266 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
267 const struct sw_flow_key *key,
268 const struct dp_upcall_info *upcall_info,
269 uint32_t cutlen)
271 struct dp_stats_percpu *stats;
272 int err;
274 if (upcall_info->portid == 0) {
275 err = -ENOTCONN;
276 goto err;
279 if (!skb_is_gso(skb))
280 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
281 else
282 err = queue_gso_packets(dp, skb, key, upcall_info, cutlen);
283 if (err)
284 goto err;
286 return 0;
288 err:
289 stats = this_cpu_ptr(dp->stats_percpu);
291 u64_stats_update_begin(&stats->syncp);
292 stats->n_lost++;
293 u64_stats_update_end(&stats->syncp);
295 return err;
298 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
299 const struct sw_flow_key *key,
300 const struct dp_upcall_info *upcall_info,
301 uint32_t cutlen)
303 unsigned int gso_type = skb_shinfo(skb)->gso_type;
304 struct sw_flow_key later_key;
305 struct sk_buff *segs, *nskb;
306 int err;
308 BUILD_BUG_ON(sizeof(*OVS_CB(skb)) > SKB_GSO_CB_OFFSET);
309 segs = __skb_gso_segment(skb, NETIF_F_SG, false);
310 if (IS_ERR(segs))
311 return PTR_ERR(segs);
312 if (segs == NULL)
313 return -EINVAL;
315 if (gso_type & SKB_GSO_UDP) {
316 /* The initial flow key extracted by ovs_flow_key_extract()
317 * in this case is for a first fragment, so we need to
318 * properly mark later fragments.
320 later_key = *key;
321 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
324 /* Queue all of the segments. */
325 skb_list_walk_safe(segs, skb, nskb) {
326 if (gso_type & SKB_GSO_UDP && skb != segs)
327 key = &later_key;
329 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
330 if (err)
331 break;
335 /* Free all of the segments. */
336 skb_list_walk_safe(segs, skb, nskb) {
337 if (err)
338 kfree_skb(skb);
339 else
340 consume_skb(skb);
342 return err;
345 static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
346 unsigned int hdrlen, int actions_attrlen)
348 size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
349 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
350 + nla_total_size(ovs_key_attr_size()) /* OVS_PACKET_ATTR_KEY */
351 + nla_total_size(sizeof(unsigned int)) /* OVS_PACKET_ATTR_LEN */
352 + nla_total_size(sizeof(u64)); /* OVS_PACKET_ATTR_HASH */
354 /* OVS_PACKET_ATTR_USERDATA */
355 if (upcall_info->userdata)
356 size += NLA_ALIGN(upcall_info->userdata->nla_len);
358 /* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
359 if (upcall_info->egress_tun_info)
360 size += nla_total_size(ovs_tun_key_attr_size());
362 /* OVS_PACKET_ATTR_ACTIONS */
363 if (upcall_info->actions_len)
364 size += nla_total_size(actions_attrlen);
366 /* OVS_PACKET_ATTR_MRU */
367 if (upcall_info->mru)
368 size += nla_total_size(sizeof(upcall_info->mru));
370 return size;
373 static void pad_packet(struct datapath *dp, struct sk_buff *skb)
375 if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
376 size_t plen = NLA_ALIGN(skb->len) - skb->len;
378 if (plen > 0)
379 skb_put_zero(skb, plen);
383 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
384 const struct sw_flow_key *key,
385 const struct dp_upcall_info *upcall_info,
386 uint32_t cutlen)
388 struct ovs_header *upcall;
389 struct sk_buff *nskb = NULL;
390 struct sk_buff *user_skb = NULL; /* to be queued to userspace */
391 struct nlattr *nla;
392 size_t len;
393 unsigned int hlen;
394 int err, dp_ifindex;
395 u64 hash;
397 dp_ifindex = get_dpifindex(dp);
398 if (!dp_ifindex)
399 return -ENODEV;
401 if (skb_vlan_tag_present(skb)) {
402 nskb = skb_clone(skb, GFP_ATOMIC);
403 if (!nskb)
404 return -ENOMEM;
406 nskb = __vlan_hwaccel_push_inside(nskb);
407 if (!nskb)
408 return -ENOMEM;
410 skb = nskb;
413 if (nla_attr_size(skb->len) > USHRT_MAX) {
414 err = -EFBIG;
415 goto out;
418 /* Complete checksum if needed */
419 if (skb->ip_summed == CHECKSUM_PARTIAL &&
420 (err = skb_csum_hwoffload_help(skb, 0)))
421 goto out;
423 /* Older versions of OVS user space enforce alignment of the last
424 * Netlink attribute to NLA_ALIGNTO which would require extensive
425 * padding logic. Only perform zerocopy if padding is not required.
427 if (dp->user_features & OVS_DP_F_UNALIGNED)
428 hlen = skb_zerocopy_headlen(skb);
429 else
430 hlen = skb->len;
432 len = upcall_msg_size(upcall_info, hlen - cutlen,
433 OVS_CB(skb)->acts_origlen);
434 user_skb = genlmsg_new(len, GFP_ATOMIC);
435 if (!user_skb) {
436 err = -ENOMEM;
437 goto out;
440 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
441 0, upcall_info->cmd);
442 if (!upcall) {
443 err = -EINVAL;
444 goto out;
446 upcall->dp_ifindex = dp_ifindex;
448 err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
449 if (err)
450 goto out;
452 if (upcall_info->userdata)
453 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
454 nla_len(upcall_info->userdata),
455 nla_data(upcall_info->userdata));
457 if (upcall_info->egress_tun_info) {
458 nla = nla_nest_start_noflag(user_skb,
459 OVS_PACKET_ATTR_EGRESS_TUN_KEY);
460 if (!nla) {
461 err = -EMSGSIZE;
462 goto out;
464 err = ovs_nla_put_tunnel_info(user_skb,
465 upcall_info->egress_tun_info);
466 if (err)
467 goto out;
469 nla_nest_end(user_skb, nla);
472 if (upcall_info->actions_len) {
473 nla = nla_nest_start_noflag(user_skb, OVS_PACKET_ATTR_ACTIONS);
474 if (!nla) {
475 err = -EMSGSIZE;
476 goto out;
478 err = ovs_nla_put_actions(upcall_info->actions,
479 upcall_info->actions_len,
480 user_skb);
481 if (!err)
482 nla_nest_end(user_skb, nla);
483 else
484 nla_nest_cancel(user_skb, nla);
487 /* Add OVS_PACKET_ATTR_MRU */
488 if (upcall_info->mru &&
489 nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU, upcall_info->mru)) {
490 err = -ENOBUFS;
491 goto out;
494 /* Add OVS_PACKET_ATTR_LEN when packet is truncated */
495 if (cutlen > 0 &&
496 nla_put_u32(user_skb, OVS_PACKET_ATTR_LEN, skb->len)) {
497 err = -ENOBUFS;
498 goto out;
501 /* Add OVS_PACKET_ATTR_HASH */
502 hash = skb_get_hash_raw(skb);
503 if (skb->sw_hash)
504 hash |= OVS_PACKET_HASH_SW_BIT;
506 if (skb->l4_hash)
507 hash |= OVS_PACKET_HASH_L4_BIT;
509 if (nla_put(user_skb, OVS_PACKET_ATTR_HASH, sizeof (u64), &hash)) {
510 err = -ENOBUFS;
511 goto out;
514 /* Only reserve room for attribute header, packet data is added
515 * in skb_zerocopy() */
516 if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
517 err = -ENOBUFS;
518 goto out;
520 nla->nla_len = nla_attr_size(skb->len - cutlen);
522 err = skb_zerocopy(user_skb, skb, skb->len - cutlen, hlen);
523 if (err)
524 goto out;
526 /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
527 pad_packet(dp, user_skb);
529 ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
531 err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
532 user_skb = NULL;
533 out:
534 if (err)
535 skb_tx_error(skb);
536 kfree_skb(user_skb);
537 kfree_skb(nskb);
538 return err;
541 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
543 struct ovs_header *ovs_header = info->userhdr;
544 struct net *net = sock_net(skb->sk);
545 struct nlattr **a = info->attrs;
546 struct sw_flow_actions *acts;
547 struct sk_buff *packet;
548 struct sw_flow *flow;
549 struct sw_flow_actions *sf_acts;
550 struct datapath *dp;
551 struct vport *input_vport;
552 u16 mru = 0;
553 u64 hash;
554 int len;
555 int err;
556 bool log = !a[OVS_PACKET_ATTR_PROBE];
558 err = -EINVAL;
559 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
560 !a[OVS_PACKET_ATTR_ACTIONS])
561 goto err;
563 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
564 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
565 err = -ENOMEM;
566 if (!packet)
567 goto err;
568 skb_reserve(packet, NET_IP_ALIGN);
570 nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
572 /* Set packet's mru */
573 if (a[OVS_PACKET_ATTR_MRU]) {
574 mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
575 packet->ignore_df = 1;
577 OVS_CB(packet)->mru = mru;
579 if (a[OVS_PACKET_ATTR_HASH]) {
580 hash = nla_get_u64(a[OVS_PACKET_ATTR_HASH]);
582 __skb_set_hash(packet, hash & 0xFFFFFFFFULL,
583 !!(hash & OVS_PACKET_HASH_SW_BIT),
584 !!(hash & OVS_PACKET_HASH_L4_BIT));
587 /* Build an sw_flow for sending this packet. */
588 flow = ovs_flow_alloc();
589 err = PTR_ERR(flow);
590 if (IS_ERR(flow))
591 goto err_kfree_skb;
593 err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
594 packet, &flow->key, log);
595 if (err)
596 goto err_flow_free;
598 err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
599 &flow->key, &acts, log);
600 if (err)
601 goto err_flow_free;
603 rcu_assign_pointer(flow->sf_acts, acts);
604 packet->priority = flow->key.phy.priority;
605 packet->mark = flow->key.phy.skb_mark;
607 rcu_read_lock();
608 dp = get_dp_rcu(net, ovs_header->dp_ifindex);
609 err = -ENODEV;
610 if (!dp)
611 goto err_unlock;
613 input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
614 if (!input_vport)
615 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
617 if (!input_vport)
618 goto err_unlock;
620 packet->dev = input_vport->dev;
621 OVS_CB(packet)->input_vport = input_vport;
622 sf_acts = rcu_dereference(flow->sf_acts);
624 local_bh_disable();
625 err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
626 local_bh_enable();
627 rcu_read_unlock();
629 ovs_flow_free(flow, false);
630 return err;
632 err_unlock:
633 rcu_read_unlock();
634 err_flow_free:
635 ovs_flow_free(flow, false);
636 err_kfree_skb:
637 kfree_skb(packet);
638 err:
639 return err;
642 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
643 [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
644 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
645 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
646 [OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
647 [OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
648 [OVS_PACKET_ATTR_HASH] = { .type = NLA_U64 },
651 static const struct genl_ops dp_packet_genl_ops[] = {
652 { .cmd = OVS_PACKET_CMD_EXECUTE,
653 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
654 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
655 .doit = ovs_packet_cmd_execute
659 static struct genl_family dp_packet_genl_family __ro_after_init = {
660 .hdrsize = sizeof(struct ovs_header),
661 .name = OVS_PACKET_FAMILY,
662 .version = OVS_PACKET_VERSION,
663 .maxattr = OVS_PACKET_ATTR_MAX,
664 .policy = packet_policy,
665 .netnsok = true,
666 .parallel_ops = true,
667 .ops = dp_packet_genl_ops,
668 .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
669 .module = THIS_MODULE,
672 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
673 struct ovs_dp_megaflow_stats *mega_stats)
675 int i;
677 memset(mega_stats, 0, sizeof(*mega_stats));
679 stats->n_flows = ovs_flow_tbl_count(&dp->table);
680 mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
682 stats->n_hit = stats->n_missed = stats->n_lost = 0;
684 for_each_possible_cpu(i) {
685 const struct dp_stats_percpu *percpu_stats;
686 struct dp_stats_percpu local_stats;
687 unsigned int start;
689 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
691 do {
692 start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
693 local_stats = *percpu_stats;
694 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
696 stats->n_hit += local_stats.n_hit;
697 stats->n_missed += local_stats.n_missed;
698 stats->n_lost += local_stats.n_lost;
699 mega_stats->n_mask_hit += local_stats.n_mask_hit;
703 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
705 return ovs_identifier_is_ufid(sfid) &&
706 !(ufid_flags & OVS_UFID_F_OMIT_KEY);
709 static bool should_fill_mask(uint32_t ufid_flags)
711 return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
714 static bool should_fill_actions(uint32_t ufid_flags)
716 return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
719 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
720 const struct sw_flow_id *sfid,
721 uint32_t ufid_flags)
723 size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
725 /* OVS_FLOW_ATTR_UFID, or unmasked flow key as fallback
726 * see ovs_nla_put_identifier()
728 if (sfid && ovs_identifier_is_ufid(sfid))
729 len += nla_total_size(sfid->ufid_len);
730 else
731 len += nla_total_size(ovs_key_attr_size());
733 /* OVS_FLOW_ATTR_KEY */
734 if (!sfid || should_fill_key(sfid, ufid_flags))
735 len += nla_total_size(ovs_key_attr_size());
737 /* OVS_FLOW_ATTR_MASK */
738 if (should_fill_mask(ufid_flags))
739 len += nla_total_size(ovs_key_attr_size());
741 /* OVS_FLOW_ATTR_ACTIONS */
742 if (should_fill_actions(ufid_flags))
743 len += nla_total_size(acts->orig_len);
745 return len
746 + nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
747 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
748 + nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */
751 /* Called with ovs_mutex or RCU read lock. */
752 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
753 struct sk_buff *skb)
755 struct ovs_flow_stats stats;
756 __be16 tcp_flags;
757 unsigned long used;
759 ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
761 if (used &&
762 nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
763 OVS_FLOW_ATTR_PAD))
764 return -EMSGSIZE;
766 if (stats.n_packets &&
767 nla_put_64bit(skb, OVS_FLOW_ATTR_STATS,
768 sizeof(struct ovs_flow_stats), &stats,
769 OVS_FLOW_ATTR_PAD))
770 return -EMSGSIZE;
772 if ((u8)ntohs(tcp_flags) &&
773 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
774 return -EMSGSIZE;
776 return 0;
779 /* Called with ovs_mutex or RCU read lock. */
780 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
781 struct sk_buff *skb, int skb_orig_len)
783 struct nlattr *start;
784 int err;
786 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
787 * this is the first flow to be dumped into 'skb'. This is unusual for
788 * Netlink but individual action lists can be longer than
789 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
790 * The userspace caller can always fetch the actions separately if it
791 * really wants them. (Most userspace callers in fact don't care.)
793 * This can only fail for dump operations because the skb is always
794 * properly sized for single flows.
796 start = nla_nest_start_noflag(skb, OVS_FLOW_ATTR_ACTIONS);
797 if (start) {
798 const struct sw_flow_actions *sf_acts;
800 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
801 err = ovs_nla_put_actions(sf_acts->actions,
802 sf_acts->actions_len, skb);
804 if (!err)
805 nla_nest_end(skb, start);
806 else {
807 if (skb_orig_len)
808 return err;
810 nla_nest_cancel(skb, start);
812 } else if (skb_orig_len) {
813 return -EMSGSIZE;
816 return 0;
819 /* Called with ovs_mutex or RCU read lock. */
820 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
821 struct sk_buff *skb, u32 portid,
822 u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
824 const int skb_orig_len = skb->len;
825 struct ovs_header *ovs_header;
826 int err;
828 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
829 flags, cmd);
830 if (!ovs_header)
831 return -EMSGSIZE;
833 ovs_header->dp_ifindex = dp_ifindex;
835 err = ovs_nla_put_identifier(flow, skb);
836 if (err)
837 goto error;
839 if (should_fill_key(&flow->id, ufid_flags)) {
840 err = ovs_nla_put_masked_key(flow, skb);
841 if (err)
842 goto error;
845 if (should_fill_mask(ufid_flags)) {
846 err = ovs_nla_put_mask(flow, skb);
847 if (err)
848 goto error;
851 err = ovs_flow_cmd_fill_stats(flow, skb);
852 if (err)
853 goto error;
855 if (should_fill_actions(ufid_flags)) {
856 err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
857 if (err)
858 goto error;
861 genlmsg_end(skb, ovs_header);
862 return 0;
864 error:
865 genlmsg_cancel(skb, ovs_header);
866 return err;
869 /* May not be called with RCU read lock. */
870 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
871 const struct sw_flow_id *sfid,
872 struct genl_info *info,
873 bool always,
874 uint32_t ufid_flags)
876 struct sk_buff *skb;
877 size_t len;
879 if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
880 return NULL;
882 len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
883 skb = genlmsg_new(len, GFP_KERNEL);
884 if (!skb)
885 return ERR_PTR(-ENOMEM);
887 return skb;
890 /* Called with ovs_mutex. */
891 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
892 int dp_ifindex,
893 struct genl_info *info, u8 cmd,
894 bool always, u32 ufid_flags)
896 struct sk_buff *skb;
897 int retval;
899 skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
900 &flow->id, info, always, ufid_flags);
901 if (IS_ERR_OR_NULL(skb))
902 return skb;
904 retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
905 info->snd_portid, info->snd_seq, 0,
906 cmd, ufid_flags);
907 if (WARN_ON_ONCE(retval < 0)) {
908 kfree_skb(skb);
909 skb = ERR_PTR(retval);
911 return skb;
914 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
916 struct net *net = sock_net(skb->sk);
917 struct nlattr **a = info->attrs;
918 struct ovs_header *ovs_header = info->userhdr;
919 struct sw_flow *flow = NULL, *new_flow;
920 struct sw_flow_mask mask;
921 struct sk_buff *reply;
922 struct datapath *dp;
923 struct sw_flow_actions *acts;
924 struct sw_flow_match match;
925 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
926 int error;
927 bool log = !a[OVS_FLOW_ATTR_PROBE];
929 /* Must have key and actions. */
930 error = -EINVAL;
931 if (!a[OVS_FLOW_ATTR_KEY]) {
932 OVS_NLERR(log, "Flow key attr not present in new flow.");
933 goto error;
935 if (!a[OVS_FLOW_ATTR_ACTIONS]) {
936 OVS_NLERR(log, "Flow actions attr not present in new flow.");
937 goto error;
940 /* Most of the time we need to allocate a new flow, do it before
941 * locking.
943 new_flow = ovs_flow_alloc();
944 if (IS_ERR(new_flow)) {
945 error = PTR_ERR(new_flow);
946 goto error;
949 /* Extract key. */
950 ovs_match_init(&match, &new_flow->key, false, &mask);
951 error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
952 a[OVS_FLOW_ATTR_MASK], log);
953 if (error)
954 goto err_kfree_flow;
956 /* Extract flow identifier. */
957 error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
958 &new_flow->key, log);
959 if (error)
960 goto err_kfree_flow;
962 /* unmasked key is needed to match when ufid is not used. */
963 if (ovs_identifier_is_key(&new_flow->id))
964 match.key = new_flow->id.unmasked_key;
966 ovs_flow_mask_key(&new_flow->key, &new_flow->key, true, &mask);
968 /* Validate actions. */
969 error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
970 &new_flow->key, &acts, log);
971 if (error) {
972 OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
973 goto err_kfree_flow;
976 reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
977 ufid_flags);
978 if (IS_ERR(reply)) {
979 error = PTR_ERR(reply);
980 goto err_kfree_acts;
983 ovs_lock();
984 dp = get_dp(net, ovs_header->dp_ifindex);
985 if (unlikely(!dp)) {
986 error = -ENODEV;
987 goto err_unlock_ovs;
990 /* Check if this is a duplicate flow */
991 if (ovs_identifier_is_ufid(&new_flow->id))
992 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
993 if (!flow)
994 flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->key);
995 if (likely(!flow)) {
996 rcu_assign_pointer(new_flow->sf_acts, acts);
998 /* Put flow in bucket. */
999 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
1000 if (unlikely(error)) {
1001 acts = NULL;
1002 goto err_unlock_ovs;
1005 if (unlikely(reply)) {
1006 error = ovs_flow_cmd_fill_info(new_flow,
1007 ovs_header->dp_ifindex,
1008 reply, info->snd_portid,
1009 info->snd_seq, 0,
1010 OVS_FLOW_CMD_NEW,
1011 ufid_flags);
1012 BUG_ON(error < 0);
1014 ovs_unlock();
1015 } else {
1016 struct sw_flow_actions *old_acts;
1018 /* Bail out if we're not allowed to modify an existing flow.
1019 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1020 * because Generic Netlink treats the latter as a dump
1021 * request. We also accept NLM_F_EXCL in case that bug ever
1022 * gets fixed.
1024 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1025 | NLM_F_EXCL))) {
1026 error = -EEXIST;
1027 goto err_unlock_ovs;
1029 /* The flow identifier has to be the same for flow updates.
1030 * Look for any overlapping flow.
1032 if (unlikely(!ovs_flow_cmp(flow, &match))) {
1033 if (ovs_identifier_is_key(&flow->id))
1034 flow = ovs_flow_tbl_lookup_exact(&dp->table,
1035 &match);
1036 else /* UFID matches but key is different */
1037 flow = NULL;
1038 if (!flow) {
1039 error = -ENOENT;
1040 goto err_unlock_ovs;
1043 /* Update actions. */
1044 old_acts = ovsl_dereference(flow->sf_acts);
1045 rcu_assign_pointer(flow->sf_acts, acts);
1047 if (unlikely(reply)) {
1048 error = ovs_flow_cmd_fill_info(flow,
1049 ovs_header->dp_ifindex,
1050 reply, info->snd_portid,
1051 info->snd_seq, 0,
1052 OVS_FLOW_CMD_NEW,
1053 ufid_flags);
1054 BUG_ON(error < 0);
1056 ovs_unlock();
1058 ovs_nla_free_flow_actions_rcu(old_acts);
1059 ovs_flow_free(new_flow, false);
1062 if (reply)
1063 ovs_notify(&dp_flow_genl_family, reply, info);
1064 return 0;
1066 err_unlock_ovs:
1067 ovs_unlock();
1068 kfree_skb(reply);
1069 err_kfree_acts:
1070 ovs_nla_free_flow_actions(acts);
1071 err_kfree_flow:
1072 ovs_flow_free(new_flow, false);
1073 error:
1074 return error;
1077 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1078 static noinline_for_stack struct sw_flow_actions *get_flow_actions(struct net *net,
1079 const struct nlattr *a,
1080 const struct sw_flow_key *key,
1081 const struct sw_flow_mask *mask,
1082 bool log)
1084 struct sw_flow_actions *acts;
1085 struct sw_flow_key masked_key;
1086 int error;
1088 ovs_flow_mask_key(&masked_key, key, true, mask);
1089 error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1090 if (error) {
1091 OVS_NLERR(log,
1092 "Actions may not be safe on all matching packets");
1093 return ERR_PTR(error);
1096 return acts;
1099 /* Factor out match-init and action-copy to avoid
1100 * "Wframe-larger-than=1024" warning. Because mask is only
1101 * used to get actions, we new a function to save some
1102 * stack space.
1104 * If there are not key and action attrs, we return 0
1105 * directly. In the case, the caller will also not use the
1106 * match as before. If there is action attr, we try to get
1107 * actions and save them to *acts. Before returning from
1108 * the function, we reset the match->mask pointer. Because
1109 * we should not to return match object with dangling reference
1110 * to mask.
1111 * */
1112 static noinline_for_stack int
1113 ovs_nla_init_match_and_action(struct net *net,
1114 struct sw_flow_match *match,
1115 struct sw_flow_key *key,
1116 struct nlattr **a,
1117 struct sw_flow_actions **acts,
1118 bool log)
1120 struct sw_flow_mask mask;
1121 int error = 0;
1123 if (a[OVS_FLOW_ATTR_KEY]) {
1124 ovs_match_init(match, key, true, &mask);
1125 error = ovs_nla_get_match(net, match, a[OVS_FLOW_ATTR_KEY],
1126 a[OVS_FLOW_ATTR_MASK], log);
1127 if (error)
1128 goto error;
1131 if (a[OVS_FLOW_ATTR_ACTIONS]) {
1132 if (!a[OVS_FLOW_ATTR_KEY]) {
1133 OVS_NLERR(log,
1134 "Flow key attribute not present in set flow.");
1135 error = -EINVAL;
1136 goto error;
1139 *acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], key,
1140 &mask, log);
1141 if (IS_ERR(*acts)) {
1142 error = PTR_ERR(*acts);
1143 goto error;
1147 /* On success, error is 0. */
1148 error:
1149 match->mask = NULL;
1150 return error;
1153 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1155 struct net *net = sock_net(skb->sk);
1156 struct nlattr **a = info->attrs;
1157 struct ovs_header *ovs_header = info->userhdr;
1158 struct sw_flow_key key;
1159 struct sw_flow *flow;
1160 struct sk_buff *reply = NULL;
1161 struct datapath *dp;
1162 struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1163 struct sw_flow_match match;
1164 struct sw_flow_id sfid;
1165 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1166 int error = 0;
1167 bool log = !a[OVS_FLOW_ATTR_PROBE];
1168 bool ufid_present;
1170 ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1171 if (!a[OVS_FLOW_ATTR_KEY] && !ufid_present) {
1172 OVS_NLERR(log,
1173 "Flow set message rejected, Key attribute missing.");
1174 return -EINVAL;
1177 error = ovs_nla_init_match_and_action(net, &match, &key, a,
1178 &acts, log);
1179 if (error)
1180 goto error;
1182 if (acts) {
1183 /* Can allocate before locking if have acts. */
1184 reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1185 ufid_flags);
1186 if (IS_ERR(reply)) {
1187 error = PTR_ERR(reply);
1188 goto err_kfree_acts;
1192 ovs_lock();
1193 dp = get_dp(net, ovs_header->dp_ifindex);
1194 if (unlikely(!dp)) {
1195 error = -ENODEV;
1196 goto err_unlock_ovs;
1198 /* Check that the flow exists. */
1199 if (ufid_present)
1200 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1201 else
1202 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1203 if (unlikely(!flow)) {
1204 error = -ENOENT;
1205 goto err_unlock_ovs;
1208 /* Update actions, if present. */
1209 if (likely(acts)) {
1210 old_acts = ovsl_dereference(flow->sf_acts);
1211 rcu_assign_pointer(flow->sf_acts, acts);
1213 if (unlikely(reply)) {
1214 error = ovs_flow_cmd_fill_info(flow,
1215 ovs_header->dp_ifindex,
1216 reply, info->snd_portid,
1217 info->snd_seq, 0,
1218 OVS_FLOW_CMD_SET,
1219 ufid_flags);
1220 BUG_ON(error < 0);
1222 } else {
1223 /* Could not alloc without acts before locking. */
1224 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1225 info, OVS_FLOW_CMD_SET, false,
1226 ufid_flags);
1228 if (IS_ERR(reply)) {
1229 error = PTR_ERR(reply);
1230 goto err_unlock_ovs;
1234 /* Clear stats. */
1235 if (a[OVS_FLOW_ATTR_CLEAR])
1236 ovs_flow_stats_clear(flow);
1237 ovs_unlock();
1239 if (reply)
1240 ovs_notify(&dp_flow_genl_family, reply, info);
1241 if (old_acts)
1242 ovs_nla_free_flow_actions_rcu(old_acts);
1244 return 0;
1246 err_unlock_ovs:
1247 ovs_unlock();
1248 kfree_skb(reply);
1249 err_kfree_acts:
1250 ovs_nla_free_flow_actions(acts);
1251 error:
1252 return error;
1255 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1257 struct nlattr **a = info->attrs;
1258 struct ovs_header *ovs_header = info->userhdr;
1259 struct net *net = sock_net(skb->sk);
1260 struct sw_flow_key key;
1261 struct sk_buff *reply;
1262 struct sw_flow *flow;
1263 struct datapath *dp;
1264 struct sw_flow_match match;
1265 struct sw_flow_id ufid;
1266 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1267 int err = 0;
1268 bool log = !a[OVS_FLOW_ATTR_PROBE];
1269 bool ufid_present;
1271 ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1272 if (a[OVS_FLOW_ATTR_KEY]) {
1273 ovs_match_init(&match, &key, true, NULL);
1274 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1275 log);
1276 } else if (!ufid_present) {
1277 OVS_NLERR(log,
1278 "Flow get message rejected, Key attribute missing.");
1279 err = -EINVAL;
1281 if (err)
1282 return err;
1284 ovs_lock();
1285 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1286 if (!dp) {
1287 err = -ENODEV;
1288 goto unlock;
1291 if (ufid_present)
1292 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1293 else
1294 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1295 if (!flow) {
1296 err = -ENOENT;
1297 goto unlock;
1300 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1301 OVS_FLOW_CMD_GET, true, ufid_flags);
1302 if (IS_ERR(reply)) {
1303 err = PTR_ERR(reply);
1304 goto unlock;
1307 ovs_unlock();
1308 return genlmsg_reply(reply, info);
1309 unlock:
1310 ovs_unlock();
1311 return err;
1314 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1316 struct nlattr **a = info->attrs;
1317 struct ovs_header *ovs_header = info->userhdr;
1318 struct net *net = sock_net(skb->sk);
1319 struct sw_flow_key key;
1320 struct sk_buff *reply;
1321 struct sw_flow *flow = NULL;
1322 struct datapath *dp;
1323 struct sw_flow_match match;
1324 struct sw_flow_id ufid;
1325 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1326 int err;
1327 bool log = !a[OVS_FLOW_ATTR_PROBE];
1328 bool ufid_present;
1330 ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1331 if (a[OVS_FLOW_ATTR_KEY]) {
1332 ovs_match_init(&match, &key, true, NULL);
1333 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1334 NULL, log);
1335 if (unlikely(err))
1336 return err;
1339 ovs_lock();
1340 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1341 if (unlikely(!dp)) {
1342 err = -ENODEV;
1343 goto unlock;
1346 if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1347 err = ovs_flow_tbl_flush(&dp->table);
1348 goto unlock;
1351 if (ufid_present)
1352 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1353 else
1354 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1355 if (unlikely(!flow)) {
1356 err = -ENOENT;
1357 goto unlock;
1360 ovs_flow_tbl_remove(&dp->table, flow);
1361 ovs_unlock();
1363 reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1364 &flow->id, info, false, ufid_flags);
1365 if (likely(reply)) {
1366 if (!IS_ERR(reply)) {
1367 rcu_read_lock(); /*To keep RCU checker happy. */
1368 err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1369 reply, info->snd_portid,
1370 info->snd_seq, 0,
1371 OVS_FLOW_CMD_DEL,
1372 ufid_flags);
1373 rcu_read_unlock();
1374 if (WARN_ON_ONCE(err < 0)) {
1375 kfree_skb(reply);
1376 goto out_free;
1379 ovs_notify(&dp_flow_genl_family, reply, info);
1380 } else {
1381 netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
1385 out_free:
1386 ovs_flow_free(flow, true);
1387 return 0;
1388 unlock:
1389 ovs_unlock();
1390 return err;
1393 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1395 struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1396 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1397 struct table_instance *ti;
1398 struct datapath *dp;
1399 u32 ufid_flags;
1400 int err;
1402 err = genlmsg_parse_deprecated(cb->nlh, &dp_flow_genl_family, a,
1403 OVS_FLOW_ATTR_MAX, flow_policy, NULL);
1404 if (err)
1405 return err;
1406 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1408 rcu_read_lock();
1409 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1410 if (!dp) {
1411 rcu_read_unlock();
1412 return -ENODEV;
1415 ti = rcu_dereference(dp->table.ti);
1416 for (;;) {
1417 struct sw_flow *flow;
1418 u32 bucket, obj;
1420 bucket = cb->args[0];
1421 obj = cb->args[1];
1422 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1423 if (!flow)
1424 break;
1426 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1427 NETLINK_CB(cb->skb).portid,
1428 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1429 OVS_FLOW_CMD_GET, ufid_flags) < 0)
1430 break;
1432 cb->args[0] = bucket;
1433 cb->args[1] = obj;
1435 rcu_read_unlock();
1436 return skb->len;
1439 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1440 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1441 [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1442 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1443 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1444 [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1445 [OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1446 [OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1449 static const struct genl_ops dp_flow_genl_ops[] = {
1450 { .cmd = OVS_FLOW_CMD_NEW,
1451 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1452 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1453 .doit = ovs_flow_cmd_new
1455 { .cmd = OVS_FLOW_CMD_DEL,
1456 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1457 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1458 .doit = ovs_flow_cmd_del
1460 { .cmd = OVS_FLOW_CMD_GET,
1461 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1462 .flags = 0, /* OK for unprivileged users. */
1463 .doit = ovs_flow_cmd_get,
1464 .dumpit = ovs_flow_cmd_dump
1466 { .cmd = OVS_FLOW_CMD_SET,
1467 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1468 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1469 .doit = ovs_flow_cmd_set,
1473 static struct genl_family dp_flow_genl_family __ro_after_init = {
1474 .hdrsize = sizeof(struct ovs_header),
1475 .name = OVS_FLOW_FAMILY,
1476 .version = OVS_FLOW_VERSION,
1477 .maxattr = OVS_FLOW_ATTR_MAX,
1478 .policy = flow_policy,
1479 .netnsok = true,
1480 .parallel_ops = true,
1481 .ops = dp_flow_genl_ops,
1482 .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1483 .mcgrps = &ovs_dp_flow_multicast_group,
1484 .n_mcgrps = 1,
1485 .module = THIS_MODULE,
1488 static size_t ovs_dp_cmd_msg_size(void)
1490 size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1492 msgsize += nla_total_size(IFNAMSIZ);
1493 msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1494 msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
1495 msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1497 return msgsize;
1500 /* Called with ovs_mutex. */
1501 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1502 u32 portid, u32 seq, u32 flags, u8 cmd)
1504 struct ovs_header *ovs_header;
1505 struct ovs_dp_stats dp_stats;
1506 struct ovs_dp_megaflow_stats dp_megaflow_stats;
1507 int err;
1509 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1510 flags, cmd);
1511 if (!ovs_header)
1512 goto error;
1514 ovs_header->dp_ifindex = get_dpifindex(dp);
1516 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1517 if (err)
1518 goto nla_put_failure;
1520 get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1521 if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1522 &dp_stats, OVS_DP_ATTR_PAD))
1523 goto nla_put_failure;
1525 if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1526 sizeof(struct ovs_dp_megaflow_stats),
1527 &dp_megaflow_stats, OVS_DP_ATTR_PAD))
1528 goto nla_put_failure;
1530 if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1531 goto nla_put_failure;
1533 genlmsg_end(skb, ovs_header);
1534 return 0;
1536 nla_put_failure:
1537 genlmsg_cancel(skb, ovs_header);
1538 error:
1539 return -EMSGSIZE;
1542 static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1544 return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1547 /* Called with rcu_read_lock or ovs_mutex. */
1548 static struct datapath *lookup_datapath(struct net *net,
1549 const struct ovs_header *ovs_header,
1550 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1552 struct datapath *dp;
1554 if (!a[OVS_DP_ATTR_NAME])
1555 dp = get_dp(net, ovs_header->dp_ifindex);
1556 else {
1557 struct vport *vport;
1559 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1560 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1562 return dp ? dp : ERR_PTR(-ENODEV);
1565 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1567 struct datapath *dp;
1569 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1570 if (IS_ERR(dp))
1571 return;
1573 WARN(dp->user_features, "Dropping previously announced user features\n");
1574 dp->user_features = 0;
1577 DEFINE_STATIC_KEY_FALSE(tc_recirc_sharing_support);
1579 static int ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1581 u32 user_features = 0;
1583 if (a[OVS_DP_ATTR_USER_FEATURES]) {
1584 user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1586 if (user_features & ~(OVS_DP_F_VPORT_PIDS |
1587 OVS_DP_F_UNALIGNED |
1588 OVS_DP_F_TC_RECIRC_SHARING))
1589 return -EOPNOTSUPP;
1591 #if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1592 if (user_features & OVS_DP_F_TC_RECIRC_SHARING)
1593 return -EOPNOTSUPP;
1594 #endif
1597 dp->user_features = user_features;
1599 if (dp->user_features & OVS_DP_F_TC_RECIRC_SHARING)
1600 static_branch_enable(&tc_recirc_sharing_support);
1601 else
1602 static_branch_disable(&tc_recirc_sharing_support);
1604 return 0;
1607 static int ovs_dp_stats_init(struct datapath *dp)
1609 dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1610 if (!dp->stats_percpu)
1611 return -ENOMEM;
1613 return 0;
1616 static int ovs_dp_vport_init(struct datapath *dp)
1618 int i;
1620 dp->ports = kmalloc_array(DP_VPORT_HASH_BUCKETS,
1621 sizeof(struct hlist_head),
1622 GFP_KERNEL);
1623 if (!dp->ports)
1624 return -ENOMEM;
1626 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1627 INIT_HLIST_HEAD(&dp->ports[i]);
1629 return 0;
1632 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1634 struct nlattr **a = info->attrs;
1635 struct vport_parms parms;
1636 struct sk_buff *reply;
1637 struct datapath *dp;
1638 struct vport *vport;
1639 struct ovs_net *ovs_net;
1640 int err;
1642 err = -EINVAL;
1643 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1644 goto err;
1646 reply = ovs_dp_cmd_alloc_info();
1647 if (!reply)
1648 return -ENOMEM;
1650 err = -ENOMEM;
1651 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1652 if (dp == NULL)
1653 goto err_destroy_reply;
1655 ovs_dp_set_net(dp, sock_net(skb->sk));
1657 /* Allocate table. */
1658 err = ovs_flow_tbl_init(&dp->table);
1659 if (err)
1660 goto err_destroy_dp;
1662 err = ovs_dp_stats_init(dp);
1663 if (err)
1664 goto err_destroy_table;
1666 err = ovs_dp_vport_init(dp);
1667 if (err)
1668 goto err_destroy_stats;
1670 err = ovs_meters_init(dp);
1671 if (err)
1672 goto err_destroy_ports;
1674 /* Set up our datapath device. */
1675 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1676 parms.type = OVS_VPORT_TYPE_INTERNAL;
1677 parms.options = NULL;
1678 parms.dp = dp;
1679 parms.port_no = OVSP_LOCAL;
1680 parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1682 err = ovs_dp_change(dp, a);
1683 if (err)
1684 goto err_destroy_meters;
1686 /* So far only local changes have been made, now need the lock. */
1687 ovs_lock();
1689 vport = new_vport(&parms);
1690 if (IS_ERR(vport)) {
1691 err = PTR_ERR(vport);
1692 if (err == -EBUSY)
1693 err = -EEXIST;
1695 if (err == -EEXIST) {
1696 /* An outdated user space instance that does not understand
1697 * the concept of user_features has attempted to create a new
1698 * datapath and is likely to reuse it. Drop all user features.
1700 if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1701 ovs_dp_reset_user_features(skb, info);
1704 ovs_unlock();
1705 goto err_destroy_meters;
1708 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1709 info->snd_seq, 0, OVS_DP_CMD_NEW);
1710 BUG_ON(err < 0);
1712 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1713 list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1715 ovs_unlock();
1717 ovs_notify(&dp_datapath_genl_family, reply, info);
1718 return 0;
1720 err_destroy_meters:
1721 ovs_meters_exit(dp);
1722 err_destroy_ports:
1723 kfree(dp->ports);
1724 err_destroy_stats:
1725 free_percpu(dp->stats_percpu);
1726 err_destroy_table:
1727 ovs_flow_tbl_destroy(&dp->table);
1728 err_destroy_dp:
1729 kfree(dp);
1730 err_destroy_reply:
1731 kfree_skb(reply);
1732 err:
1733 return err;
1736 /* Called with ovs_mutex. */
1737 static void __dp_destroy(struct datapath *dp)
1739 int i;
1741 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1742 struct vport *vport;
1743 struct hlist_node *n;
1745 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1746 if (vport->port_no != OVSP_LOCAL)
1747 ovs_dp_detach_port(vport);
1750 list_del_rcu(&dp->list_node);
1752 /* OVSP_LOCAL is datapath internal port. We need to make sure that
1753 * all ports in datapath are destroyed first before freeing datapath.
1755 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1757 /* RCU destroy the flow table */
1758 call_rcu(&dp->rcu, destroy_dp_rcu);
1761 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1763 struct sk_buff *reply;
1764 struct datapath *dp;
1765 int err;
1767 reply = ovs_dp_cmd_alloc_info();
1768 if (!reply)
1769 return -ENOMEM;
1771 ovs_lock();
1772 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1773 err = PTR_ERR(dp);
1774 if (IS_ERR(dp))
1775 goto err_unlock_free;
1777 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1778 info->snd_seq, 0, OVS_DP_CMD_DEL);
1779 BUG_ON(err < 0);
1781 __dp_destroy(dp);
1782 ovs_unlock();
1784 ovs_notify(&dp_datapath_genl_family, reply, info);
1786 return 0;
1788 err_unlock_free:
1789 ovs_unlock();
1790 kfree_skb(reply);
1791 return err;
1794 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1796 struct sk_buff *reply;
1797 struct datapath *dp;
1798 int err;
1800 reply = ovs_dp_cmd_alloc_info();
1801 if (!reply)
1802 return -ENOMEM;
1804 ovs_lock();
1805 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1806 err = PTR_ERR(dp);
1807 if (IS_ERR(dp))
1808 goto err_unlock_free;
1810 err = ovs_dp_change(dp, info->attrs);
1811 if (err)
1812 goto err_unlock_free;
1814 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1815 info->snd_seq, 0, OVS_DP_CMD_SET);
1816 BUG_ON(err < 0);
1818 ovs_unlock();
1819 ovs_notify(&dp_datapath_genl_family, reply, info);
1821 return 0;
1823 err_unlock_free:
1824 ovs_unlock();
1825 kfree_skb(reply);
1826 return err;
1829 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1831 struct sk_buff *reply;
1832 struct datapath *dp;
1833 int err;
1835 reply = ovs_dp_cmd_alloc_info();
1836 if (!reply)
1837 return -ENOMEM;
1839 ovs_lock();
1840 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1841 if (IS_ERR(dp)) {
1842 err = PTR_ERR(dp);
1843 goto err_unlock_free;
1845 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1846 info->snd_seq, 0, OVS_DP_CMD_GET);
1847 BUG_ON(err < 0);
1848 ovs_unlock();
1850 return genlmsg_reply(reply, info);
1852 err_unlock_free:
1853 ovs_unlock();
1854 kfree_skb(reply);
1855 return err;
1858 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1860 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1861 struct datapath *dp;
1862 int skip = cb->args[0];
1863 int i = 0;
1865 ovs_lock();
1866 list_for_each_entry(dp, &ovs_net->dps, list_node) {
1867 if (i >= skip &&
1868 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1869 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1870 OVS_DP_CMD_GET) < 0)
1871 break;
1872 i++;
1874 ovs_unlock();
1876 cb->args[0] = i;
1878 return skb->len;
1881 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1882 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1883 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1884 [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1887 static const struct genl_ops dp_datapath_genl_ops[] = {
1888 { .cmd = OVS_DP_CMD_NEW,
1889 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1890 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1891 .doit = ovs_dp_cmd_new
1893 { .cmd = OVS_DP_CMD_DEL,
1894 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1895 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1896 .doit = ovs_dp_cmd_del
1898 { .cmd = OVS_DP_CMD_GET,
1899 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1900 .flags = 0, /* OK for unprivileged users. */
1901 .doit = ovs_dp_cmd_get,
1902 .dumpit = ovs_dp_cmd_dump
1904 { .cmd = OVS_DP_CMD_SET,
1905 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1906 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1907 .doit = ovs_dp_cmd_set,
1911 static struct genl_family dp_datapath_genl_family __ro_after_init = {
1912 .hdrsize = sizeof(struct ovs_header),
1913 .name = OVS_DATAPATH_FAMILY,
1914 .version = OVS_DATAPATH_VERSION,
1915 .maxattr = OVS_DP_ATTR_MAX,
1916 .policy = datapath_policy,
1917 .netnsok = true,
1918 .parallel_ops = true,
1919 .ops = dp_datapath_genl_ops,
1920 .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1921 .mcgrps = &ovs_dp_datapath_multicast_group,
1922 .n_mcgrps = 1,
1923 .module = THIS_MODULE,
1926 /* Called with ovs_mutex or RCU read lock. */
1927 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1928 struct net *net, u32 portid, u32 seq,
1929 u32 flags, u8 cmd, gfp_t gfp)
1931 struct ovs_header *ovs_header;
1932 struct ovs_vport_stats vport_stats;
1933 int err;
1935 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1936 flags, cmd);
1937 if (!ovs_header)
1938 return -EMSGSIZE;
1940 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1942 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1943 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1944 nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1945 ovs_vport_name(vport)) ||
1946 nla_put_u32(skb, OVS_VPORT_ATTR_IFINDEX, vport->dev->ifindex))
1947 goto nla_put_failure;
1949 if (!net_eq(net, dev_net(vport->dev))) {
1950 int id = peernet2id_alloc(net, dev_net(vport->dev), gfp);
1952 if (nla_put_s32(skb, OVS_VPORT_ATTR_NETNSID, id))
1953 goto nla_put_failure;
1956 ovs_vport_get_stats(vport, &vport_stats);
1957 if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
1958 sizeof(struct ovs_vport_stats), &vport_stats,
1959 OVS_VPORT_ATTR_PAD))
1960 goto nla_put_failure;
1962 if (ovs_vport_get_upcall_portids(vport, skb))
1963 goto nla_put_failure;
1965 err = ovs_vport_get_options(vport, skb);
1966 if (err == -EMSGSIZE)
1967 goto error;
1969 genlmsg_end(skb, ovs_header);
1970 return 0;
1972 nla_put_failure:
1973 err = -EMSGSIZE;
1974 error:
1975 genlmsg_cancel(skb, ovs_header);
1976 return err;
1979 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1981 return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1984 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1985 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, struct net *net,
1986 u32 portid, u32 seq, u8 cmd)
1988 struct sk_buff *skb;
1989 int retval;
1991 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1992 if (!skb)
1993 return ERR_PTR(-ENOMEM);
1995 retval = ovs_vport_cmd_fill_info(vport, skb, net, portid, seq, 0, cmd,
1996 GFP_KERNEL);
1997 BUG_ON(retval < 0);
1999 return skb;
2002 /* Called with ovs_mutex or RCU read lock. */
2003 static struct vport *lookup_vport(struct net *net,
2004 const struct ovs_header *ovs_header,
2005 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
2007 struct datapath *dp;
2008 struct vport *vport;
2010 if (a[OVS_VPORT_ATTR_IFINDEX])
2011 return ERR_PTR(-EOPNOTSUPP);
2012 if (a[OVS_VPORT_ATTR_NAME]) {
2013 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
2014 if (!vport)
2015 return ERR_PTR(-ENODEV);
2016 if (ovs_header->dp_ifindex &&
2017 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
2018 return ERR_PTR(-ENODEV);
2019 return vport;
2020 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
2021 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
2023 if (port_no >= DP_MAX_PORTS)
2024 return ERR_PTR(-EFBIG);
2026 dp = get_dp(net, ovs_header->dp_ifindex);
2027 if (!dp)
2028 return ERR_PTR(-ENODEV);
2030 vport = ovs_vport_ovsl_rcu(dp, port_no);
2031 if (!vport)
2032 return ERR_PTR(-ENODEV);
2033 return vport;
2034 } else
2035 return ERR_PTR(-EINVAL);
2039 static unsigned int ovs_get_max_headroom(struct datapath *dp)
2041 unsigned int dev_headroom, max_headroom = 0;
2042 struct net_device *dev;
2043 struct vport *vport;
2044 int i;
2046 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2047 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node,
2048 lockdep_ovsl_is_held()) {
2049 dev = vport->dev;
2050 dev_headroom = netdev_get_fwd_headroom(dev);
2051 if (dev_headroom > max_headroom)
2052 max_headroom = dev_headroom;
2056 return max_headroom;
2059 /* Called with ovs_mutex */
2060 static void ovs_update_headroom(struct datapath *dp, unsigned int new_headroom)
2062 struct vport *vport;
2063 int i;
2065 dp->max_headroom = new_headroom;
2066 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
2067 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node,
2068 lockdep_ovsl_is_held())
2069 netdev_set_rx_headroom(vport->dev, new_headroom);
2072 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
2074 struct nlattr **a = info->attrs;
2075 struct ovs_header *ovs_header = info->userhdr;
2076 struct vport_parms parms;
2077 struct sk_buff *reply;
2078 struct vport *vport;
2079 struct datapath *dp;
2080 unsigned int new_headroom;
2081 u32 port_no;
2082 int err;
2084 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
2085 !a[OVS_VPORT_ATTR_UPCALL_PID])
2086 return -EINVAL;
2087 if (a[OVS_VPORT_ATTR_IFINDEX])
2088 return -EOPNOTSUPP;
2090 port_no = a[OVS_VPORT_ATTR_PORT_NO]
2091 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
2092 if (port_no >= DP_MAX_PORTS)
2093 return -EFBIG;
2095 reply = ovs_vport_cmd_alloc_info();
2096 if (!reply)
2097 return -ENOMEM;
2099 ovs_lock();
2100 restart:
2101 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2102 err = -ENODEV;
2103 if (!dp)
2104 goto exit_unlock_free;
2106 if (port_no) {
2107 vport = ovs_vport_ovsl(dp, port_no);
2108 err = -EBUSY;
2109 if (vport)
2110 goto exit_unlock_free;
2111 } else {
2112 for (port_no = 1; ; port_no++) {
2113 if (port_no >= DP_MAX_PORTS) {
2114 err = -EFBIG;
2115 goto exit_unlock_free;
2117 vport = ovs_vport_ovsl(dp, port_no);
2118 if (!vport)
2119 break;
2123 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2124 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2125 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2126 parms.dp = dp;
2127 parms.port_no = port_no;
2128 parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
2130 vport = new_vport(&parms);
2131 err = PTR_ERR(vport);
2132 if (IS_ERR(vport)) {
2133 if (err == -EAGAIN)
2134 goto restart;
2135 goto exit_unlock_free;
2138 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2139 info->snd_portid, info->snd_seq, 0,
2140 OVS_VPORT_CMD_NEW, GFP_KERNEL);
2142 new_headroom = netdev_get_fwd_headroom(vport->dev);
2144 if (new_headroom > dp->max_headroom)
2145 ovs_update_headroom(dp, new_headroom);
2146 else
2147 netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2149 BUG_ON(err < 0);
2150 ovs_unlock();
2152 ovs_notify(&dp_vport_genl_family, reply, info);
2153 return 0;
2155 exit_unlock_free:
2156 ovs_unlock();
2157 kfree_skb(reply);
2158 return err;
2161 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2163 struct nlattr **a = info->attrs;
2164 struct sk_buff *reply;
2165 struct vport *vport;
2166 int err;
2168 reply = ovs_vport_cmd_alloc_info();
2169 if (!reply)
2170 return -ENOMEM;
2172 ovs_lock();
2173 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2174 err = PTR_ERR(vport);
2175 if (IS_ERR(vport))
2176 goto exit_unlock_free;
2178 if (a[OVS_VPORT_ATTR_TYPE] &&
2179 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2180 err = -EINVAL;
2181 goto exit_unlock_free;
2184 if (a[OVS_VPORT_ATTR_OPTIONS]) {
2185 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2186 if (err)
2187 goto exit_unlock_free;
2191 if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2192 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2194 err = ovs_vport_set_upcall_portids(vport, ids);
2195 if (err)
2196 goto exit_unlock_free;
2199 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2200 info->snd_portid, info->snd_seq, 0,
2201 OVS_VPORT_CMD_SET, GFP_KERNEL);
2202 BUG_ON(err < 0);
2204 ovs_unlock();
2205 ovs_notify(&dp_vport_genl_family, reply, info);
2206 return 0;
2208 exit_unlock_free:
2209 ovs_unlock();
2210 kfree_skb(reply);
2211 return err;
2214 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2216 bool update_headroom = false;
2217 struct nlattr **a = info->attrs;
2218 struct sk_buff *reply;
2219 struct datapath *dp;
2220 struct vport *vport;
2221 unsigned int new_headroom;
2222 int err;
2224 reply = ovs_vport_cmd_alloc_info();
2225 if (!reply)
2226 return -ENOMEM;
2228 ovs_lock();
2229 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2230 err = PTR_ERR(vport);
2231 if (IS_ERR(vport))
2232 goto exit_unlock_free;
2234 if (vport->port_no == OVSP_LOCAL) {
2235 err = -EINVAL;
2236 goto exit_unlock_free;
2239 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2240 info->snd_portid, info->snd_seq, 0,
2241 OVS_VPORT_CMD_DEL, GFP_KERNEL);
2242 BUG_ON(err < 0);
2244 /* the vport deletion may trigger dp headroom update */
2245 dp = vport->dp;
2246 if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2247 update_headroom = true;
2249 netdev_reset_rx_headroom(vport->dev);
2250 ovs_dp_detach_port(vport);
2252 if (update_headroom) {
2253 new_headroom = ovs_get_max_headroom(dp);
2255 if (new_headroom < dp->max_headroom)
2256 ovs_update_headroom(dp, new_headroom);
2258 ovs_unlock();
2260 ovs_notify(&dp_vport_genl_family, reply, info);
2261 return 0;
2263 exit_unlock_free:
2264 ovs_unlock();
2265 kfree_skb(reply);
2266 return err;
2269 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2271 struct nlattr **a = info->attrs;
2272 struct ovs_header *ovs_header = info->userhdr;
2273 struct sk_buff *reply;
2274 struct vport *vport;
2275 int err;
2277 reply = ovs_vport_cmd_alloc_info();
2278 if (!reply)
2279 return -ENOMEM;
2281 rcu_read_lock();
2282 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2283 err = PTR_ERR(vport);
2284 if (IS_ERR(vport))
2285 goto exit_unlock_free;
2286 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2287 info->snd_portid, info->snd_seq, 0,
2288 OVS_VPORT_CMD_GET, GFP_ATOMIC);
2289 BUG_ON(err < 0);
2290 rcu_read_unlock();
2292 return genlmsg_reply(reply, info);
2294 exit_unlock_free:
2295 rcu_read_unlock();
2296 kfree_skb(reply);
2297 return err;
2300 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2302 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2303 struct datapath *dp;
2304 int bucket = cb->args[0], skip = cb->args[1];
2305 int i, j = 0;
2307 rcu_read_lock();
2308 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2309 if (!dp) {
2310 rcu_read_unlock();
2311 return -ENODEV;
2313 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2314 struct vport *vport;
2316 j = 0;
2317 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2318 if (j >= skip &&
2319 ovs_vport_cmd_fill_info(vport, skb,
2320 sock_net(skb->sk),
2321 NETLINK_CB(cb->skb).portid,
2322 cb->nlh->nlmsg_seq,
2323 NLM_F_MULTI,
2324 OVS_VPORT_CMD_GET,
2325 GFP_ATOMIC) < 0)
2326 goto out;
2328 j++;
2330 skip = 0;
2332 out:
2333 rcu_read_unlock();
2335 cb->args[0] = i;
2336 cb->args[1] = j;
2338 return skb->len;
2341 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2342 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2343 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2344 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2345 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2346 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_UNSPEC },
2347 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2348 [OVS_VPORT_ATTR_IFINDEX] = { .type = NLA_U32 },
2349 [OVS_VPORT_ATTR_NETNSID] = { .type = NLA_S32 },
2352 static const struct genl_ops dp_vport_genl_ops[] = {
2353 { .cmd = OVS_VPORT_CMD_NEW,
2354 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2355 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2356 .doit = ovs_vport_cmd_new
2358 { .cmd = OVS_VPORT_CMD_DEL,
2359 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2360 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2361 .doit = ovs_vport_cmd_del
2363 { .cmd = OVS_VPORT_CMD_GET,
2364 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2365 .flags = 0, /* OK for unprivileged users. */
2366 .doit = ovs_vport_cmd_get,
2367 .dumpit = ovs_vport_cmd_dump
2369 { .cmd = OVS_VPORT_CMD_SET,
2370 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2371 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2372 .doit = ovs_vport_cmd_set,
2376 struct genl_family dp_vport_genl_family __ro_after_init = {
2377 .hdrsize = sizeof(struct ovs_header),
2378 .name = OVS_VPORT_FAMILY,
2379 .version = OVS_VPORT_VERSION,
2380 .maxattr = OVS_VPORT_ATTR_MAX,
2381 .policy = vport_policy,
2382 .netnsok = true,
2383 .parallel_ops = true,
2384 .ops = dp_vport_genl_ops,
2385 .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2386 .mcgrps = &ovs_dp_vport_multicast_group,
2387 .n_mcgrps = 1,
2388 .module = THIS_MODULE,
2391 static struct genl_family * const dp_genl_families[] = {
2392 &dp_datapath_genl_family,
2393 &dp_vport_genl_family,
2394 &dp_flow_genl_family,
2395 &dp_packet_genl_family,
2396 &dp_meter_genl_family,
2397 #if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2398 &dp_ct_limit_genl_family,
2399 #endif
2402 static void dp_unregister_genl(int n_families)
2404 int i;
2406 for (i = 0; i < n_families; i++)
2407 genl_unregister_family(dp_genl_families[i]);
2410 static int __init dp_register_genl(void)
2412 int err;
2413 int i;
2415 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2417 err = genl_register_family(dp_genl_families[i]);
2418 if (err)
2419 goto error;
2422 return 0;
2424 error:
2425 dp_unregister_genl(i);
2426 return err;
2429 static int __net_init ovs_init_net(struct net *net)
2431 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2433 INIT_LIST_HEAD(&ovs_net->dps);
2434 INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2435 return ovs_ct_init(net);
2438 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2439 struct list_head *head)
2441 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2442 struct datapath *dp;
2444 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2445 int i;
2447 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2448 struct vport *vport;
2450 hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2451 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2452 continue;
2454 if (dev_net(vport->dev) == dnet)
2455 list_add(&vport->detach_list, head);
2461 static void __net_exit ovs_exit_net(struct net *dnet)
2463 struct datapath *dp, *dp_next;
2464 struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2465 struct vport *vport, *vport_next;
2466 struct net *net;
2467 LIST_HEAD(head);
2469 ovs_lock();
2471 ovs_ct_exit(dnet);
2473 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2474 __dp_destroy(dp);
2476 down_read(&net_rwsem);
2477 for_each_net(net)
2478 list_vports_from_net(net, dnet, &head);
2479 up_read(&net_rwsem);
2481 /* Detach all vports from given namespace. */
2482 list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2483 list_del(&vport->detach_list);
2484 ovs_dp_detach_port(vport);
2487 ovs_unlock();
2489 cancel_work_sync(&ovs_net->dp_notify_work);
2492 static struct pernet_operations ovs_net_ops = {
2493 .init = ovs_init_net,
2494 .exit = ovs_exit_net,
2495 .id = &ovs_net_id,
2496 .size = sizeof(struct ovs_net),
2499 static int __init dp_init(void)
2501 int err;
2503 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof_field(struct sk_buff, cb));
2505 pr_info("Open vSwitch switching datapath\n");
2507 err = action_fifos_init();
2508 if (err)
2509 goto error;
2511 err = ovs_internal_dev_rtnl_link_register();
2512 if (err)
2513 goto error_action_fifos_exit;
2515 err = ovs_flow_init();
2516 if (err)
2517 goto error_unreg_rtnl_link;
2519 err = ovs_vport_init();
2520 if (err)
2521 goto error_flow_exit;
2523 err = register_pernet_device(&ovs_net_ops);
2524 if (err)
2525 goto error_vport_exit;
2527 err = register_netdevice_notifier(&ovs_dp_device_notifier);
2528 if (err)
2529 goto error_netns_exit;
2531 err = ovs_netdev_init();
2532 if (err)
2533 goto error_unreg_notifier;
2535 err = dp_register_genl();
2536 if (err < 0)
2537 goto error_unreg_netdev;
2539 return 0;
2541 error_unreg_netdev:
2542 ovs_netdev_exit();
2543 error_unreg_notifier:
2544 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2545 error_netns_exit:
2546 unregister_pernet_device(&ovs_net_ops);
2547 error_vport_exit:
2548 ovs_vport_exit();
2549 error_flow_exit:
2550 ovs_flow_exit();
2551 error_unreg_rtnl_link:
2552 ovs_internal_dev_rtnl_link_unregister();
2553 error_action_fifos_exit:
2554 action_fifos_exit();
2555 error:
2556 return err;
2559 static void dp_cleanup(void)
2561 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2562 ovs_netdev_exit();
2563 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2564 unregister_pernet_device(&ovs_net_ops);
2565 rcu_barrier();
2566 ovs_vport_exit();
2567 ovs_flow_exit();
2568 ovs_internal_dev_rtnl_link_unregister();
2569 action_fifos_exit();
2572 module_init(dp_init);
2573 module_exit(dp_cleanup);
2575 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2576 MODULE_LICENSE("GPL");
2577 MODULE_ALIAS_GENL_FAMILY(OVS_DATAPATH_FAMILY);
2578 MODULE_ALIAS_GENL_FAMILY(OVS_VPORT_FAMILY);
2579 MODULE_ALIAS_GENL_FAMILY(OVS_FLOW_FAMILY);
2580 MODULE_ALIAS_GENL_FAMILY(OVS_PACKET_FAMILY);
2581 MODULE_ALIAS_GENL_FAMILY(OVS_METER_FAMILY);
2582 MODULE_ALIAS_GENL_FAMILY(OVS_CT_LIMIT_FAMILY);