2 * Copyright (c) 2007-2013 Nicira, Inc.
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.
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.
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
19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
21 #include <linux/init.h>
22 #include <linux/module.h>
23 #include <linux/if_arp.h>
24 #include <linux/if_vlan.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/lockdep.h>
48 #include <linux/openvswitch.h>
49 #include <linux/rculist.h>
50 #include <linux/dmi.h>
51 #include <linux/workqueue.h>
52 #include <net/genetlink.h>
53 #include <net/net_namespace.h>
54 #include <net/netns/generic.h>
58 #include "flow_table.h"
59 #include "flow_netlink.h"
60 #include "vport-internal_dev.h"
61 #include "vport-netdev.h"
63 int ovs_net_id __read_mostly
;
65 static void ovs_notify(struct genl_family
*family
,
66 struct sk_buff
*skb
, struct genl_info
*info
)
68 genl_notify(family
, skb
, genl_info_net(info
), info
->snd_portid
,
69 0, info
->nlhdr
, GFP_KERNEL
);
75 * All writes e.g. Writes to device state (add/remove datapath, port, set
76 * operations on vports, etc.), Writes to other state (flow table
77 * modifications, set miscellaneous datapath parameters, etc.) are protected
80 * Reads are protected by RCU.
82 * There are a few special cases (mostly stats) that have their own
83 * synchronization but they nest under all of above and don't interact with
86 * The RTNL lock nests inside ovs_mutex.
89 static DEFINE_MUTEX(ovs_mutex
);
93 mutex_lock(&ovs_mutex
);
98 mutex_unlock(&ovs_mutex
);
101 #ifdef CONFIG_LOCKDEP
102 int lockdep_ovsl_is_held(void)
105 return lockdep_is_held(&ovs_mutex
);
111 static struct vport
*new_vport(const struct vport_parms
*);
112 static int queue_gso_packets(struct datapath
*dp
, struct sk_buff
*,
113 const struct dp_upcall_info
*);
114 static int queue_userspace_packet(struct datapath
*dp
, struct sk_buff
*,
115 const struct dp_upcall_info
*);
117 /* Must be called with rcu_read_lock or ovs_mutex. */
118 static struct datapath
*get_dp(struct net
*net
, int dp_ifindex
)
120 struct datapath
*dp
= NULL
;
121 struct net_device
*dev
;
124 dev
= dev_get_by_index_rcu(net
, dp_ifindex
);
126 struct vport
*vport
= ovs_internal_dev_get_vport(dev
);
135 /* Must be called with rcu_read_lock or ovs_mutex. */
136 static const char *ovs_dp_name(const struct datapath
*dp
)
138 struct vport
*vport
= ovs_vport_ovsl_rcu(dp
, OVSP_LOCAL
);
139 return vport
->ops
->get_name(vport
);
142 static int get_dpifindex(struct datapath
*dp
)
149 local
= ovs_vport_rcu(dp
, OVSP_LOCAL
);
151 ifindex
= netdev_vport_priv(local
)->dev
->ifindex
;
160 static void destroy_dp_rcu(struct rcu_head
*rcu
)
162 struct datapath
*dp
= container_of(rcu
, struct datapath
, rcu
);
164 free_percpu(dp
->stats_percpu
);
165 release_net(ovs_dp_get_net(dp
));
170 static struct hlist_head
*vport_hash_bucket(const struct datapath
*dp
,
173 return &dp
->ports
[port_no
& (DP_VPORT_HASH_BUCKETS
- 1)];
176 struct vport
*ovs_lookup_vport(const struct datapath
*dp
, u16 port_no
)
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 if (vport
->port_no
== port_no
)
189 /* Called with ovs_mutex. */
190 static struct vport
*new_vport(const struct vport_parms
*parms
)
194 vport
= ovs_vport_add(parms
);
195 if (!IS_ERR(vport
)) {
196 struct datapath
*dp
= parms
->dp
;
197 struct hlist_head
*head
= vport_hash_bucket(dp
, vport
->port_no
);
199 hlist_add_head_rcu(&vport
->dp_hash_node
, head
);
204 void ovs_dp_detach_port(struct vport
*p
)
208 /* First drop references to device. */
209 hlist_del_rcu(&p
->dp_hash_node
);
211 /* Then destroy it. */
215 /* Must be called with rcu_read_lock. */
216 void ovs_dp_process_received_packet(struct vport
*p
, struct sk_buff
*skb
)
218 struct datapath
*dp
= p
->dp
;
219 struct sw_flow
*flow
;
220 struct dp_stats_percpu
*stats
;
221 struct sw_flow_key key
;
226 stats
= this_cpu_ptr(dp
->stats_percpu
);
228 /* Extract flow from 'skb' into 'key'. */
229 error
= ovs_flow_extract(skb
, p
->port_no
, &key
);
230 if (unlikely(error
)) {
236 flow
= ovs_flow_tbl_lookup_stats(&dp
->table
, &key
, &n_mask_hit
);
237 if (unlikely(!flow
)) {
238 struct dp_upcall_info upcall
;
240 upcall
.cmd
= OVS_PACKET_CMD_MISS
;
242 upcall
.userdata
= NULL
;
243 upcall
.portid
= p
->upcall_portid
;
244 ovs_dp_upcall(dp
, skb
, &upcall
);
246 stats_counter
= &stats
->n_missed
;
250 OVS_CB(skb
)->flow
= flow
;
251 OVS_CB(skb
)->pkt_key
= &key
;
253 ovs_flow_stats_update(OVS_CB(skb
)->flow
, skb
);
254 ovs_execute_actions(dp
, skb
);
255 stats_counter
= &stats
->n_hit
;
258 /* Update datapath statistics. */
259 u64_stats_update_begin(&stats
->syncp
);
261 stats
->n_mask_hit
+= n_mask_hit
;
262 u64_stats_update_end(&stats
->syncp
);
265 static struct genl_family dp_packet_genl_family
= {
266 .id
= GENL_ID_GENERATE
,
267 .hdrsize
= sizeof(struct ovs_header
),
268 .name
= OVS_PACKET_FAMILY
,
269 .version
= OVS_PACKET_VERSION
,
270 .maxattr
= OVS_PACKET_ATTR_MAX
,
272 .parallel_ops
= true,
275 int ovs_dp_upcall(struct datapath
*dp
, struct sk_buff
*skb
,
276 const struct dp_upcall_info
*upcall_info
)
278 struct dp_stats_percpu
*stats
;
281 if (upcall_info
->portid
== 0) {
286 if (!skb_is_gso(skb
))
287 err
= queue_userspace_packet(dp
, skb
, upcall_info
);
289 err
= queue_gso_packets(dp
, skb
, upcall_info
);
296 stats
= this_cpu_ptr(dp
->stats_percpu
);
298 u64_stats_update_begin(&stats
->syncp
);
300 u64_stats_update_end(&stats
->syncp
);
305 static int queue_gso_packets(struct datapath
*dp
, struct sk_buff
*skb
,
306 const struct dp_upcall_info
*upcall_info
)
308 unsigned short gso_type
= skb_shinfo(skb
)->gso_type
;
309 struct dp_upcall_info later_info
;
310 struct sw_flow_key later_key
;
311 struct sk_buff
*segs
, *nskb
;
314 segs
= __skb_gso_segment(skb
, NETIF_F_SG
, false);
316 return PTR_ERR(segs
);
318 /* Queue all of the segments. */
321 err
= queue_userspace_packet(dp
, skb
, upcall_info
);
325 if (skb
== segs
&& gso_type
& SKB_GSO_UDP
) {
326 /* The initial flow key extracted by ovs_flow_extract()
327 * in this case is for a first fragment, so we need to
328 * properly mark later fragments.
330 later_key
= *upcall_info
->key
;
331 later_key
.ip
.frag
= OVS_FRAG_TYPE_LATER
;
333 later_info
= *upcall_info
;
334 later_info
.key
= &later_key
;
335 upcall_info
= &later_info
;
337 } while ((skb
= skb
->next
));
339 /* Free all of the segments. */
347 } while ((skb
= nskb
));
351 static size_t key_attr_size(void)
353 return nla_total_size(4) /* OVS_KEY_ATTR_PRIORITY */
354 + nla_total_size(0) /* OVS_KEY_ATTR_TUNNEL */
355 + nla_total_size(8) /* OVS_TUNNEL_KEY_ATTR_ID */
356 + nla_total_size(4) /* OVS_TUNNEL_KEY_ATTR_IPV4_SRC */
357 + nla_total_size(4) /* OVS_TUNNEL_KEY_ATTR_IPV4_DST */
358 + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TOS */
359 + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TTL */
360 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
361 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_CSUM */
362 + nla_total_size(4) /* OVS_KEY_ATTR_IN_PORT */
363 + nla_total_size(4) /* OVS_KEY_ATTR_SKB_MARK */
364 + nla_total_size(12) /* OVS_KEY_ATTR_ETHERNET */
365 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
366 + nla_total_size(4) /* OVS_KEY_ATTR_8021Q */
367 + nla_total_size(0) /* OVS_KEY_ATTR_ENCAP */
368 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
369 + nla_total_size(40) /* OVS_KEY_ATTR_IPV6 */
370 + nla_total_size(2) /* OVS_KEY_ATTR_ICMPV6 */
371 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
374 static size_t upcall_msg_size(const struct nlattr
*userdata
,
377 size_t size
= NLMSG_ALIGN(sizeof(struct ovs_header
))
378 + nla_total_size(hdrlen
) /* OVS_PACKET_ATTR_PACKET */
379 + nla_total_size(key_attr_size()); /* OVS_PACKET_ATTR_KEY */
381 /* OVS_PACKET_ATTR_USERDATA */
383 size
+= NLA_ALIGN(userdata
->nla_len
);
388 static int queue_userspace_packet(struct datapath
*dp
, struct sk_buff
*skb
,
389 const struct dp_upcall_info
*upcall_info
)
391 struct ovs_header
*upcall
;
392 struct sk_buff
*nskb
= NULL
;
393 struct sk_buff
*user_skb
; /* to be queued to userspace */
395 struct genl_info info
= {
396 .dst_sk
= ovs_dp_get_net(dp
)->genl_sock
,
397 .snd_portid
= upcall_info
->portid
,
403 dp_ifindex
= get_dpifindex(dp
);
407 if (vlan_tx_tag_present(skb
)) {
408 nskb
= skb_clone(skb
, GFP_ATOMIC
);
412 nskb
= __vlan_put_tag(nskb
, nskb
->vlan_proto
, vlan_tx_tag_get(nskb
));
420 if (nla_attr_size(skb
->len
) > USHRT_MAX
) {
425 /* Complete checksum if needed */
426 if (skb
->ip_summed
== CHECKSUM_PARTIAL
&&
427 (err
= skb_checksum_help(skb
)))
430 /* Older versions of OVS user space enforce alignment of the last
431 * Netlink attribute to NLA_ALIGNTO which would require extensive
432 * padding logic. Only perform zerocopy if padding is not required.
434 if (dp
->user_features
& OVS_DP_F_UNALIGNED
)
435 hlen
= skb_zerocopy_headlen(skb
);
439 len
= upcall_msg_size(upcall_info
->userdata
, hlen
);
440 user_skb
= genlmsg_new_unicast(len
, &info
, GFP_ATOMIC
);
446 upcall
= genlmsg_put(user_skb
, 0, 0, &dp_packet_genl_family
,
447 0, upcall_info
->cmd
);
448 upcall
->dp_ifindex
= dp_ifindex
;
450 nla
= nla_nest_start(user_skb
, OVS_PACKET_ATTR_KEY
);
451 ovs_nla_put_flow(upcall_info
->key
, upcall_info
->key
, user_skb
);
452 nla_nest_end(user_skb
, nla
);
454 if (upcall_info
->userdata
)
455 __nla_put(user_skb
, OVS_PACKET_ATTR_USERDATA
,
456 nla_len(upcall_info
->userdata
),
457 nla_data(upcall_info
->userdata
));
459 /* Only reserve room for attribute header, packet data is added
460 * in skb_zerocopy() */
461 if (!(nla
= nla_reserve(user_skb
, OVS_PACKET_ATTR_PACKET
, 0))) {
465 nla
->nla_len
= nla_attr_size(skb
->len
);
467 err
= skb_zerocopy(user_skb
, skb
, skb
->len
, hlen
);
471 /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
472 if (!(dp
->user_features
& OVS_DP_F_UNALIGNED
)) {
473 size_t plen
= NLA_ALIGN(user_skb
->len
) - user_skb
->len
;
476 memset(skb_put(user_skb
, plen
), 0, plen
);
479 ((struct nlmsghdr
*) user_skb
->data
)->nlmsg_len
= user_skb
->len
;
481 err
= genlmsg_unicast(ovs_dp_get_net(dp
), user_skb
, upcall_info
->portid
);
489 static int ovs_packet_cmd_execute(struct sk_buff
*skb
, struct genl_info
*info
)
491 struct ovs_header
*ovs_header
= info
->userhdr
;
492 struct nlattr
**a
= info
->attrs
;
493 struct sw_flow_actions
*acts
;
494 struct sk_buff
*packet
;
495 struct sw_flow
*flow
;
502 if (!a
[OVS_PACKET_ATTR_PACKET
] || !a
[OVS_PACKET_ATTR_KEY
] ||
503 !a
[OVS_PACKET_ATTR_ACTIONS
])
506 len
= nla_len(a
[OVS_PACKET_ATTR_PACKET
]);
507 packet
= __dev_alloc_skb(NET_IP_ALIGN
+ len
, GFP_KERNEL
);
511 skb_reserve(packet
, NET_IP_ALIGN
);
513 nla_memcpy(__skb_put(packet
, len
), a
[OVS_PACKET_ATTR_PACKET
], len
);
515 skb_reset_mac_header(packet
);
516 eth
= eth_hdr(packet
);
518 /* Normally, setting the skb 'protocol' field would be handled by a
519 * call to eth_type_trans(), but it assumes there's a sending
520 * device, which we may not have. */
521 if (ntohs(eth
->h_proto
) >= ETH_P_802_3_MIN
)
522 packet
->protocol
= eth
->h_proto
;
524 packet
->protocol
= htons(ETH_P_802_2
);
526 /* Build an sw_flow for sending this packet. */
527 flow
= ovs_flow_alloc();
532 err
= ovs_flow_extract(packet
, -1, &flow
->key
);
536 err
= ovs_nla_get_flow_metadata(flow
, a
[OVS_PACKET_ATTR_KEY
]);
539 acts
= ovs_nla_alloc_flow_actions(nla_len(a
[OVS_PACKET_ATTR_ACTIONS
]));
544 err
= ovs_nla_copy_actions(a
[OVS_PACKET_ATTR_ACTIONS
],
545 &flow
->key
, 0, &acts
);
546 rcu_assign_pointer(flow
->sf_acts
, acts
);
550 OVS_CB(packet
)->flow
= flow
;
551 OVS_CB(packet
)->pkt_key
= &flow
->key
;
552 packet
->priority
= flow
->key
.phy
.priority
;
553 packet
->mark
= flow
->key
.phy
.skb_mark
;
556 dp
= get_dp(sock_net(skb
->sk
), ovs_header
->dp_ifindex
);
562 err
= ovs_execute_actions(dp
, packet
);
566 ovs_flow_free(flow
, false);
572 ovs_flow_free(flow
, false);
579 static const struct nla_policy packet_policy
[OVS_PACKET_ATTR_MAX
+ 1] = {
580 [OVS_PACKET_ATTR_PACKET
] = { .len
= ETH_HLEN
},
581 [OVS_PACKET_ATTR_KEY
] = { .type
= NLA_NESTED
},
582 [OVS_PACKET_ATTR_ACTIONS
] = { .type
= NLA_NESTED
},
585 static const struct genl_ops dp_packet_genl_ops
[] = {
586 { .cmd
= OVS_PACKET_CMD_EXECUTE
,
587 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
588 .policy
= packet_policy
,
589 .doit
= ovs_packet_cmd_execute
593 static void get_dp_stats(struct datapath
*dp
, struct ovs_dp_stats
*stats
,
594 struct ovs_dp_megaflow_stats
*mega_stats
)
598 memset(mega_stats
, 0, sizeof(*mega_stats
));
600 stats
->n_flows
= ovs_flow_tbl_count(&dp
->table
);
601 mega_stats
->n_masks
= ovs_flow_tbl_num_masks(&dp
->table
);
603 stats
->n_hit
= stats
->n_missed
= stats
->n_lost
= 0;
605 for_each_possible_cpu(i
) {
606 const struct dp_stats_percpu
*percpu_stats
;
607 struct dp_stats_percpu local_stats
;
610 percpu_stats
= per_cpu_ptr(dp
->stats_percpu
, i
);
613 start
= u64_stats_fetch_begin_irq(&percpu_stats
->syncp
);
614 local_stats
= *percpu_stats
;
615 } while (u64_stats_fetch_retry_irq(&percpu_stats
->syncp
, start
));
617 stats
->n_hit
+= local_stats
.n_hit
;
618 stats
->n_missed
+= local_stats
.n_missed
;
619 stats
->n_lost
+= local_stats
.n_lost
;
620 mega_stats
->n_mask_hit
+= local_stats
.n_mask_hit
;
624 static const struct nla_policy flow_policy
[OVS_FLOW_ATTR_MAX
+ 1] = {
625 [OVS_FLOW_ATTR_KEY
] = { .type
= NLA_NESTED
},
626 [OVS_FLOW_ATTR_ACTIONS
] = { .type
= NLA_NESTED
},
627 [OVS_FLOW_ATTR_CLEAR
] = { .type
= NLA_FLAG
},
630 static struct genl_family dp_flow_genl_family
= {
631 .id
= GENL_ID_GENERATE
,
632 .hdrsize
= sizeof(struct ovs_header
),
633 .name
= OVS_FLOW_FAMILY
,
634 .version
= OVS_FLOW_VERSION
,
635 .maxattr
= OVS_FLOW_ATTR_MAX
,
637 .parallel_ops
= true,
640 static struct genl_multicast_group ovs_dp_flow_multicast_group
= {
641 .name
= OVS_FLOW_MCGROUP
644 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions
*acts
)
646 return NLMSG_ALIGN(sizeof(struct ovs_header
))
647 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
648 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_MASK */
649 + nla_total_size(sizeof(struct ovs_flow_stats
)) /* OVS_FLOW_ATTR_STATS */
650 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
651 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
652 + nla_total_size(acts
->actions_len
); /* OVS_FLOW_ATTR_ACTIONS */
655 /* Called with ovs_mutex. */
656 static int ovs_flow_cmd_fill_info(struct sw_flow
*flow
, struct datapath
*dp
,
657 struct sk_buff
*skb
, u32 portid
,
658 u32 seq
, u32 flags
, u8 cmd
)
660 const int skb_orig_len
= skb
->len
;
661 struct nlattr
*start
;
662 struct ovs_flow_stats stats
;
665 struct ovs_header
*ovs_header
;
669 ovs_header
= genlmsg_put(skb
, portid
, seq
, &dp_flow_genl_family
, flags
, cmd
);
673 ovs_header
->dp_ifindex
= get_dpifindex(dp
);
676 nla
= nla_nest_start(skb
, OVS_FLOW_ATTR_KEY
);
678 goto nla_put_failure
;
680 err
= ovs_nla_put_flow(&flow
->unmasked_key
, &flow
->unmasked_key
, skb
);
683 nla_nest_end(skb
, nla
);
685 nla
= nla_nest_start(skb
, OVS_FLOW_ATTR_MASK
);
687 goto nla_put_failure
;
689 err
= ovs_nla_put_flow(&flow
->key
, &flow
->mask
->key
, skb
);
693 nla_nest_end(skb
, nla
);
695 ovs_flow_stats_get(flow
, &stats
, &used
, &tcp_flags
);
697 nla_put_u64(skb
, OVS_FLOW_ATTR_USED
, ovs_flow_used_time(used
)))
698 goto nla_put_failure
;
700 if (stats
.n_packets
&&
701 nla_put(skb
, OVS_FLOW_ATTR_STATS
, sizeof(struct ovs_flow_stats
), &stats
))
702 goto nla_put_failure
;
704 if ((u8
)ntohs(tcp_flags
) &&
705 nla_put_u8(skb
, OVS_FLOW_ATTR_TCP_FLAGS
, (u8
)ntohs(tcp_flags
)))
706 goto nla_put_failure
;
708 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
709 * this is the first flow to be dumped into 'skb'. This is unusual for
710 * Netlink but individual action lists can be longer than
711 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
712 * The userspace caller can always fetch the actions separately if it
713 * really wants them. (Most userspace callers in fact don't care.)
715 * This can only fail for dump operations because the skb is always
716 * properly sized for single flows.
718 start
= nla_nest_start(skb
, OVS_FLOW_ATTR_ACTIONS
);
720 const struct sw_flow_actions
*sf_acts
;
722 sf_acts
= rcu_dereference_ovsl(flow
->sf_acts
);
724 err
= ovs_nla_put_actions(sf_acts
->actions
,
725 sf_acts
->actions_len
, skb
);
727 nla_nest_end(skb
, start
);
732 nla_nest_cancel(skb
, start
);
734 } else if (skb_orig_len
)
735 goto nla_put_failure
;
737 return genlmsg_end(skb
, ovs_header
);
742 genlmsg_cancel(skb
, ovs_header
);
746 static struct sk_buff
*ovs_flow_cmd_alloc_info(struct sw_flow
*flow
,
747 struct genl_info
*info
)
751 len
= ovs_flow_cmd_msg_size(ovsl_dereference(flow
->sf_acts
));
753 return genlmsg_new_unicast(len
, info
, GFP_KERNEL
);
756 static struct sk_buff
*ovs_flow_cmd_build_info(struct sw_flow
*flow
,
758 struct genl_info
*info
,
764 skb
= ovs_flow_cmd_alloc_info(flow
, info
);
766 return ERR_PTR(-ENOMEM
);
768 retval
= ovs_flow_cmd_fill_info(flow
, dp
, skb
, info
->snd_portid
,
769 info
->snd_seq
, 0, cmd
);
774 static int ovs_flow_cmd_new_or_set(struct sk_buff
*skb
, struct genl_info
*info
)
776 struct nlattr
**a
= info
->attrs
;
777 struct ovs_header
*ovs_header
= info
->userhdr
;
778 struct sw_flow_key key
, masked_key
;
779 struct sw_flow
*flow
= NULL
;
780 struct sw_flow_mask mask
;
781 struct sk_buff
*reply
;
783 struct sw_flow_actions
*acts
= NULL
;
784 struct sw_flow_match match
;
789 if (!a
[OVS_FLOW_ATTR_KEY
])
792 ovs_match_init(&match
, &key
, &mask
);
793 error
= ovs_nla_get_match(&match
,
794 a
[OVS_FLOW_ATTR_KEY
], a
[OVS_FLOW_ATTR_MASK
]);
798 /* Validate actions. */
799 if (a
[OVS_FLOW_ATTR_ACTIONS
]) {
800 acts
= ovs_nla_alloc_flow_actions(nla_len(a
[OVS_FLOW_ATTR_ACTIONS
]));
801 error
= PTR_ERR(acts
);
805 ovs_flow_mask_key(&masked_key
, &key
, &mask
);
806 error
= ovs_nla_copy_actions(a
[OVS_FLOW_ATTR_ACTIONS
],
807 &masked_key
, 0, &acts
);
809 OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
812 } else if (info
->genlhdr
->cmd
== OVS_FLOW_CMD_NEW
) {
818 dp
= get_dp(sock_net(skb
->sk
), ovs_header
->dp_ifindex
);
823 /* Check if this is a duplicate flow */
824 flow
= ovs_flow_tbl_lookup(&dp
->table
, &key
);
826 /* Bail out if we're not allowed to create a new flow. */
828 if (info
->genlhdr
->cmd
== OVS_FLOW_CMD_SET
)
832 flow
= ovs_flow_alloc();
834 error
= PTR_ERR(flow
);
838 flow
->key
= masked_key
;
839 flow
->unmasked_key
= key
;
840 rcu_assign_pointer(flow
->sf_acts
, acts
);
842 /* Put flow in bucket. */
843 error
= ovs_flow_tbl_insert(&dp
->table
, flow
, &mask
);
849 reply
= ovs_flow_cmd_build_info(flow
, dp
, info
, OVS_FLOW_CMD_NEW
);
851 /* We found a matching flow. */
852 struct sw_flow_actions
*old_acts
;
854 /* Bail out if we're not allowed to modify an existing flow.
855 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
856 * because Generic Netlink treats the latter as a dump
857 * request. We also accept NLM_F_EXCL in case that bug ever
861 if (info
->genlhdr
->cmd
== OVS_FLOW_CMD_NEW
&&
862 info
->nlhdr
->nlmsg_flags
& (NLM_F_CREATE
| NLM_F_EXCL
))
865 /* The unmasked key has to be the same for flow updates. */
866 if (!ovs_flow_cmp_unmasked_key(flow
, &match
))
869 /* Update actions. */
870 old_acts
= ovsl_dereference(flow
->sf_acts
);
871 rcu_assign_pointer(flow
->sf_acts
, acts
);
872 ovs_nla_free_flow_actions(old_acts
);
874 reply
= ovs_flow_cmd_build_info(flow
, dp
, info
, OVS_FLOW_CMD_NEW
);
877 if (a
[OVS_FLOW_ATTR_CLEAR
])
878 ovs_flow_stats_clear(flow
);
883 ovs_notify(&dp_flow_genl_family
, reply
, info
);
885 genl_set_err(&dp_flow_genl_family
, sock_net(skb
->sk
), 0,
890 ovs_flow_free(flow
, false);
899 static int ovs_flow_cmd_get(struct sk_buff
*skb
, struct genl_info
*info
)
901 struct nlattr
**a
= info
->attrs
;
902 struct ovs_header
*ovs_header
= info
->userhdr
;
903 struct sw_flow_key key
;
904 struct sk_buff
*reply
;
905 struct sw_flow
*flow
;
907 struct sw_flow_match match
;
910 if (!a
[OVS_FLOW_ATTR_KEY
]) {
911 OVS_NLERR("Flow get message rejected, Key attribute missing.\n");
915 ovs_match_init(&match
, &key
, NULL
);
916 err
= ovs_nla_get_match(&match
, a
[OVS_FLOW_ATTR_KEY
], NULL
);
921 dp
= get_dp(sock_net(skb
->sk
), ovs_header
->dp_ifindex
);
927 flow
= ovs_flow_tbl_lookup(&dp
->table
, &key
);
928 if (!flow
|| !ovs_flow_cmp_unmasked_key(flow
, &match
)) {
933 reply
= ovs_flow_cmd_build_info(flow
, dp
, info
, OVS_FLOW_CMD_NEW
);
935 err
= PTR_ERR(reply
);
940 return genlmsg_reply(reply
, info
);
946 static int ovs_flow_cmd_del(struct sk_buff
*skb
, struct genl_info
*info
)
948 struct nlattr
**a
= info
->attrs
;
949 struct ovs_header
*ovs_header
= info
->userhdr
;
950 struct sw_flow_key key
;
951 struct sk_buff
*reply
;
952 struct sw_flow
*flow
;
954 struct sw_flow_match match
;
958 dp
= get_dp(sock_net(skb
->sk
), ovs_header
->dp_ifindex
);
964 if (!a
[OVS_FLOW_ATTR_KEY
]) {
965 err
= ovs_flow_tbl_flush(&dp
->table
);
969 ovs_match_init(&match
, &key
, NULL
);
970 err
= ovs_nla_get_match(&match
, a
[OVS_FLOW_ATTR_KEY
], NULL
);
974 flow
= ovs_flow_tbl_lookup(&dp
->table
, &key
);
975 if (!flow
|| !ovs_flow_cmp_unmasked_key(flow
, &match
)) {
980 reply
= ovs_flow_cmd_alloc_info(flow
, info
);
986 ovs_flow_tbl_remove(&dp
->table
, flow
);
988 err
= ovs_flow_cmd_fill_info(flow
, dp
, reply
, info
->snd_portid
,
989 info
->snd_seq
, 0, OVS_FLOW_CMD_DEL
);
992 ovs_flow_free(flow
, true);
995 ovs_notify(&dp_flow_genl_family
, reply
, info
);
1002 static int ovs_flow_cmd_dump(struct sk_buff
*skb
, struct netlink_callback
*cb
)
1004 struct ovs_header
*ovs_header
= genlmsg_data(nlmsg_data(cb
->nlh
));
1005 struct table_instance
*ti
;
1006 struct datapath
*dp
;
1009 dp
= get_dp(sock_net(skb
->sk
), ovs_header
->dp_ifindex
);
1015 ti
= rcu_dereference(dp
->table
.ti
);
1017 struct sw_flow
*flow
;
1020 bucket
= cb
->args
[0];
1022 flow
= ovs_flow_tbl_dump_next(ti
, &bucket
, &obj
);
1026 if (ovs_flow_cmd_fill_info(flow
, dp
, skb
,
1027 NETLINK_CB(cb
->skb
).portid
,
1028 cb
->nlh
->nlmsg_seq
, NLM_F_MULTI
,
1029 OVS_FLOW_CMD_NEW
) < 0)
1032 cb
->args
[0] = bucket
;
1039 static const struct genl_ops dp_flow_genl_ops
[] = {
1040 { .cmd
= OVS_FLOW_CMD_NEW
,
1041 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1042 .policy
= flow_policy
,
1043 .doit
= ovs_flow_cmd_new_or_set
1045 { .cmd
= OVS_FLOW_CMD_DEL
,
1046 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1047 .policy
= flow_policy
,
1048 .doit
= ovs_flow_cmd_del
1050 { .cmd
= OVS_FLOW_CMD_GET
,
1051 .flags
= 0, /* OK for unprivileged users. */
1052 .policy
= flow_policy
,
1053 .doit
= ovs_flow_cmd_get
,
1054 .dumpit
= ovs_flow_cmd_dump
1056 { .cmd
= OVS_FLOW_CMD_SET
,
1057 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1058 .policy
= flow_policy
,
1059 .doit
= ovs_flow_cmd_new_or_set
,
1063 static const struct nla_policy datapath_policy
[OVS_DP_ATTR_MAX
+ 1] = {
1064 [OVS_DP_ATTR_NAME
] = { .type
= NLA_NUL_STRING
, .len
= IFNAMSIZ
- 1 },
1065 [OVS_DP_ATTR_UPCALL_PID
] = { .type
= NLA_U32
},
1066 [OVS_DP_ATTR_USER_FEATURES
] = { .type
= NLA_U32
},
1069 static struct genl_family dp_datapath_genl_family
= {
1070 .id
= GENL_ID_GENERATE
,
1071 .hdrsize
= sizeof(struct ovs_header
),
1072 .name
= OVS_DATAPATH_FAMILY
,
1073 .version
= OVS_DATAPATH_VERSION
,
1074 .maxattr
= OVS_DP_ATTR_MAX
,
1076 .parallel_ops
= true,
1079 static struct genl_multicast_group ovs_dp_datapath_multicast_group
= {
1080 .name
= OVS_DATAPATH_MCGROUP
1083 static size_t ovs_dp_cmd_msg_size(void)
1085 size_t msgsize
= NLMSG_ALIGN(sizeof(struct ovs_header
));
1087 msgsize
+= nla_total_size(IFNAMSIZ
);
1088 msgsize
+= nla_total_size(sizeof(struct ovs_dp_stats
));
1089 msgsize
+= nla_total_size(sizeof(struct ovs_dp_megaflow_stats
));
1090 msgsize
+= nla_total_size(sizeof(u32
)); /* OVS_DP_ATTR_USER_FEATURES */
1095 static int ovs_dp_cmd_fill_info(struct datapath
*dp
, struct sk_buff
*skb
,
1096 u32 portid
, u32 seq
, u32 flags
, u8 cmd
)
1098 struct ovs_header
*ovs_header
;
1099 struct ovs_dp_stats dp_stats
;
1100 struct ovs_dp_megaflow_stats dp_megaflow_stats
;
1103 ovs_header
= genlmsg_put(skb
, portid
, seq
, &dp_datapath_genl_family
,
1108 ovs_header
->dp_ifindex
= get_dpifindex(dp
);
1111 err
= nla_put_string(skb
, OVS_DP_ATTR_NAME
, ovs_dp_name(dp
));
1114 goto nla_put_failure
;
1116 get_dp_stats(dp
, &dp_stats
, &dp_megaflow_stats
);
1117 if (nla_put(skb
, OVS_DP_ATTR_STATS
, sizeof(struct ovs_dp_stats
),
1119 goto nla_put_failure
;
1121 if (nla_put(skb
, OVS_DP_ATTR_MEGAFLOW_STATS
,
1122 sizeof(struct ovs_dp_megaflow_stats
),
1123 &dp_megaflow_stats
))
1124 goto nla_put_failure
;
1126 if (nla_put_u32(skb
, OVS_DP_ATTR_USER_FEATURES
, dp
->user_features
))
1127 goto nla_put_failure
;
1129 return genlmsg_end(skb
, ovs_header
);
1132 genlmsg_cancel(skb
, ovs_header
);
1137 static struct sk_buff
*ovs_dp_cmd_build_info(struct datapath
*dp
,
1138 struct genl_info
*info
, u8 cmd
)
1140 struct sk_buff
*skb
;
1143 skb
= genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info
, GFP_KERNEL
);
1145 return ERR_PTR(-ENOMEM
);
1147 retval
= ovs_dp_cmd_fill_info(dp
, skb
, info
->snd_portid
, info
->snd_seq
, 0, cmd
);
1150 return ERR_PTR(retval
);
1155 /* Called with ovs_mutex. */
1156 static struct datapath
*lookup_datapath(struct net
*net
,
1157 struct ovs_header
*ovs_header
,
1158 struct nlattr
*a
[OVS_DP_ATTR_MAX
+ 1])
1160 struct datapath
*dp
;
1162 if (!a
[OVS_DP_ATTR_NAME
])
1163 dp
= get_dp(net
, ovs_header
->dp_ifindex
);
1165 struct vport
*vport
;
1168 vport
= ovs_vport_locate(net
, nla_data(a
[OVS_DP_ATTR_NAME
]));
1169 dp
= vport
&& vport
->port_no
== OVSP_LOCAL
? vport
->dp
: NULL
;
1172 return dp
? dp
: ERR_PTR(-ENODEV
);
1175 static void ovs_dp_reset_user_features(struct sk_buff
*skb
, struct genl_info
*info
)
1177 struct datapath
*dp
;
1179 dp
= lookup_datapath(sock_net(skb
->sk
), info
->userhdr
, info
->attrs
);
1183 WARN(dp
->user_features
, "Dropping previously announced user features\n");
1184 dp
->user_features
= 0;
1187 static void ovs_dp_change(struct datapath
*dp
, struct nlattr
**a
)
1189 if (a
[OVS_DP_ATTR_USER_FEATURES
])
1190 dp
->user_features
= nla_get_u32(a
[OVS_DP_ATTR_USER_FEATURES
]);
1193 static int ovs_dp_cmd_new(struct sk_buff
*skb
, struct genl_info
*info
)
1195 struct nlattr
**a
= info
->attrs
;
1196 struct vport_parms parms
;
1197 struct sk_buff
*reply
;
1198 struct datapath
*dp
;
1199 struct vport
*vport
;
1200 struct ovs_net
*ovs_net
;
1204 if (!a
[OVS_DP_ATTR_NAME
] || !a
[OVS_DP_ATTR_UPCALL_PID
])
1210 dp
= kzalloc(sizeof(*dp
), GFP_KERNEL
);
1212 goto err_unlock_ovs
;
1214 ovs_dp_set_net(dp
, hold_net(sock_net(skb
->sk
)));
1216 /* Allocate table. */
1217 err
= ovs_flow_tbl_init(&dp
->table
);
1221 dp
->stats_percpu
= netdev_alloc_pcpu_stats(struct dp_stats_percpu
);
1222 if (!dp
->stats_percpu
) {
1224 goto err_destroy_table
;
1227 dp
->ports
= kmalloc(DP_VPORT_HASH_BUCKETS
* sizeof(struct hlist_head
),
1231 goto err_destroy_percpu
;
1234 for (i
= 0; i
< DP_VPORT_HASH_BUCKETS
; i
++)
1235 INIT_HLIST_HEAD(&dp
->ports
[i
]);
1237 /* Set up our datapath device. */
1238 parms
.name
= nla_data(a
[OVS_DP_ATTR_NAME
]);
1239 parms
.type
= OVS_VPORT_TYPE_INTERNAL
;
1240 parms
.options
= NULL
;
1242 parms
.port_no
= OVSP_LOCAL
;
1243 parms
.upcall_portid
= nla_get_u32(a
[OVS_DP_ATTR_UPCALL_PID
]);
1245 ovs_dp_change(dp
, a
);
1247 vport
= new_vport(&parms
);
1248 if (IS_ERR(vport
)) {
1249 err
= PTR_ERR(vport
);
1253 if (err
== -EEXIST
) {
1254 /* An outdated user space instance that does not understand
1255 * the concept of user_features has attempted to create a new
1256 * datapath and is likely to reuse it. Drop all user features.
1258 if (info
->genlhdr
->version
< OVS_DP_VER_FEATURES
)
1259 ovs_dp_reset_user_features(skb
, info
);
1262 goto err_destroy_ports_array
;
1265 reply
= ovs_dp_cmd_build_info(dp
, info
, OVS_DP_CMD_NEW
);
1266 err
= PTR_ERR(reply
);
1268 goto err_destroy_local_port
;
1270 ovs_net
= net_generic(ovs_dp_get_net(dp
), ovs_net_id
);
1271 list_add_tail_rcu(&dp
->list_node
, &ovs_net
->dps
);
1275 ovs_notify(&dp_datapath_genl_family
, reply
, info
);
1278 err_destroy_local_port
:
1279 ovs_dp_detach_port(ovs_vport_ovsl(dp
, OVSP_LOCAL
));
1280 err_destroy_ports_array
:
1283 free_percpu(dp
->stats_percpu
);
1285 ovs_flow_tbl_destroy(&dp
->table
, false);
1287 release_net(ovs_dp_get_net(dp
));
1295 /* Called with ovs_mutex. */
1296 static void __dp_destroy(struct datapath
*dp
)
1300 for (i
= 0; i
< DP_VPORT_HASH_BUCKETS
; i
++) {
1301 struct vport
*vport
;
1302 struct hlist_node
*n
;
1304 hlist_for_each_entry_safe(vport
, n
, &dp
->ports
[i
], dp_hash_node
)
1305 if (vport
->port_no
!= OVSP_LOCAL
)
1306 ovs_dp_detach_port(vport
);
1309 list_del_rcu(&dp
->list_node
);
1311 /* OVSP_LOCAL is datapath internal port. We need to make sure that
1312 * all ports in datapath are destroyed first before freeing datapath.
1314 ovs_dp_detach_port(ovs_vport_ovsl(dp
, OVSP_LOCAL
));
1316 /* RCU destroy the flow table */
1317 ovs_flow_tbl_destroy(&dp
->table
, true);
1319 call_rcu(&dp
->rcu
, destroy_dp_rcu
);
1322 static int ovs_dp_cmd_del(struct sk_buff
*skb
, struct genl_info
*info
)
1324 struct sk_buff
*reply
;
1325 struct datapath
*dp
;
1329 dp
= lookup_datapath(sock_net(skb
->sk
), info
->userhdr
, info
->attrs
);
1334 reply
= ovs_dp_cmd_build_info(dp
, info
, OVS_DP_CMD_DEL
);
1335 err
= PTR_ERR(reply
);
1342 ovs_notify(&dp_datapath_genl_family
, reply
, info
);
1350 static int ovs_dp_cmd_set(struct sk_buff
*skb
, struct genl_info
*info
)
1352 struct sk_buff
*reply
;
1353 struct datapath
*dp
;
1357 dp
= lookup_datapath(sock_net(skb
->sk
), info
->userhdr
, info
->attrs
);
1362 ovs_dp_change(dp
, info
->attrs
);
1364 reply
= ovs_dp_cmd_build_info(dp
, info
, OVS_DP_CMD_NEW
);
1365 if (IS_ERR(reply
)) {
1366 err
= PTR_ERR(reply
);
1367 genl_set_err(&dp_datapath_genl_family
, sock_net(skb
->sk
), 0,
1374 ovs_notify(&dp_datapath_genl_family
, reply
, info
);
1382 static int ovs_dp_cmd_get(struct sk_buff
*skb
, struct genl_info
*info
)
1384 struct sk_buff
*reply
;
1385 struct datapath
*dp
;
1389 dp
= lookup_datapath(sock_net(skb
->sk
), info
->userhdr
, info
->attrs
);
1395 reply
= ovs_dp_cmd_build_info(dp
, info
, OVS_DP_CMD_NEW
);
1396 if (IS_ERR(reply
)) {
1397 err
= PTR_ERR(reply
);
1402 return genlmsg_reply(reply
, info
);
1409 static int ovs_dp_cmd_dump(struct sk_buff
*skb
, struct netlink_callback
*cb
)
1411 struct ovs_net
*ovs_net
= net_generic(sock_net(skb
->sk
), ovs_net_id
);
1412 struct datapath
*dp
;
1413 int skip
= cb
->args
[0];
1417 list_for_each_entry_rcu(dp
, &ovs_net
->dps
, list_node
) {
1419 ovs_dp_cmd_fill_info(dp
, skb
, NETLINK_CB(cb
->skb
).portid
,
1420 cb
->nlh
->nlmsg_seq
, NLM_F_MULTI
,
1421 OVS_DP_CMD_NEW
) < 0)
1432 static const struct genl_ops dp_datapath_genl_ops
[] = {
1433 { .cmd
= OVS_DP_CMD_NEW
,
1434 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1435 .policy
= datapath_policy
,
1436 .doit
= ovs_dp_cmd_new
1438 { .cmd
= OVS_DP_CMD_DEL
,
1439 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1440 .policy
= datapath_policy
,
1441 .doit
= ovs_dp_cmd_del
1443 { .cmd
= OVS_DP_CMD_GET
,
1444 .flags
= 0, /* OK for unprivileged users. */
1445 .policy
= datapath_policy
,
1446 .doit
= ovs_dp_cmd_get
,
1447 .dumpit
= ovs_dp_cmd_dump
1449 { .cmd
= OVS_DP_CMD_SET
,
1450 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1451 .policy
= datapath_policy
,
1452 .doit
= ovs_dp_cmd_set
,
1456 static const struct nla_policy vport_policy
[OVS_VPORT_ATTR_MAX
+ 1] = {
1457 [OVS_VPORT_ATTR_NAME
] = { .type
= NLA_NUL_STRING
, .len
= IFNAMSIZ
- 1 },
1458 [OVS_VPORT_ATTR_STATS
] = { .len
= sizeof(struct ovs_vport_stats
) },
1459 [OVS_VPORT_ATTR_PORT_NO
] = { .type
= NLA_U32
},
1460 [OVS_VPORT_ATTR_TYPE
] = { .type
= NLA_U32
},
1461 [OVS_VPORT_ATTR_UPCALL_PID
] = { .type
= NLA_U32
},
1462 [OVS_VPORT_ATTR_OPTIONS
] = { .type
= NLA_NESTED
},
1465 struct genl_family dp_vport_genl_family
= {
1466 .id
= GENL_ID_GENERATE
,
1467 .hdrsize
= sizeof(struct ovs_header
),
1468 .name
= OVS_VPORT_FAMILY
,
1469 .version
= OVS_VPORT_VERSION
,
1470 .maxattr
= OVS_VPORT_ATTR_MAX
,
1472 .parallel_ops
= true,
1475 static struct genl_multicast_group ovs_dp_vport_multicast_group
= {
1476 .name
= OVS_VPORT_MCGROUP
1479 /* Called with ovs_mutex or RCU read lock. */
1480 static int ovs_vport_cmd_fill_info(struct vport
*vport
, struct sk_buff
*skb
,
1481 u32 portid
, u32 seq
, u32 flags
, u8 cmd
)
1483 struct ovs_header
*ovs_header
;
1484 struct ovs_vport_stats vport_stats
;
1487 ovs_header
= genlmsg_put(skb
, portid
, seq
, &dp_vport_genl_family
,
1492 ovs_header
->dp_ifindex
= get_dpifindex(vport
->dp
);
1494 if (nla_put_u32(skb
, OVS_VPORT_ATTR_PORT_NO
, vport
->port_no
) ||
1495 nla_put_u32(skb
, OVS_VPORT_ATTR_TYPE
, vport
->ops
->type
) ||
1496 nla_put_string(skb
, OVS_VPORT_ATTR_NAME
, vport
->ops
->get_name(vport
)) ||
1497 nla_put_u32(skb
, OVS_VPORT_ATTR_UPCALL_PID
, vport
->upcall_portid
))
1498 goto nla_put_failure
;
1500 ovs_vport_get_stats(vport
, &vport_stats
);
1501 if (nla_put(skb
, OVS_VPORT_ATTR_STATS
, sizeof(struct ovs_vport_stats
),
1503 goto nla_put_failure
;
1505 err
= ovs_vport_get_options(vport
, skb
);
1506 if (err
== -EMSGSIZE
)
1509 return genlmsg_end(skb
, ovs_header
);
1514 genlmsg_cancel(skb
, ovs_header
);
1518 /* Called with ovs_mutex or RCU read lock. */
1519 struct sk_buff
*ovs_vport_cmd_build_info(struct vport
*vport
, u32 portid
,
1522 struct sk_buff
*skb
;
1525 skb
= nlmsg_new(NLMSG_DEFAULT_SIZE
, GFP_ATOMIC
);
1527 return ERR_PTR(-ENOMEM
);
1529 retval
= ovs_vport_cmd_fill_info(vport
, skb
, portid
, seq
, 0, cmd
);
1535 /* Called with ovs_mutex or RCU read lock. */
1536 static struct vport
*lookup_vport(struct net
*net
,
1537 struct ovs_header
*ovs_header
,
1538 struct nlattr
*a
[OVS_VPORT_ATTR_MAX
+ 1])
1540 struct datapath
*dp
;
1541 struct vport
*vport
;
1543 if (a
[OVS_VPORT_ATTR_NAME
]) {
1544 vport
= ovs_vport_locate(net
, nla_data(a
[OVS_VPORT_ATTR_NAME
]));
1546 return ERR_PTR(-ENODEV
);
1547 if (ovs_header
->dp_ifindex
&&
1548 ovs_header
->dp_ifindex
!= get_dpifindex(vport
->dp
))
1549 return ERR_PTR(-ENODEV
);
1551 } else if (a
[OVS_VPORT_ATTR_PORT_NO
]) {
1552 u32 port_no
= nla_get_u32(a
[OVS_VPORT_ATTR_PORT_NO
]);
1554 if (port_no
>= DP_MAX_PORTS
)
1555 return ERR_PTR(-EFBIG
);
1557 dp
= get_dp(net
, ovs_header
->dp_ifindex
);
1559 return ERR_PTR(-ENODEV
);
1561 vport
= ovs_vport_ovsl_rcu(dp
, port_no
);
1563 return ERR_PTR(-ENODEV
);
1566 return ERR_PTR(-EINVAL
);
1569 static int ovs_vport_cmd_new(struct sk_buff
*skb
, struct genl_info
*info
)
1571 struct nlattr
**a
= info
->attrs
;
1572 struct ovs_header
*ovs_header
= info
->userhdr
;
1573 struct vport_parms parms
;
1574 struct sk_buff
*reply
;
1575 struct vport
*vport
;
1576 struct datapath
*dp
;
1581 if (!a
[OVS_VPORT_ATTR_NAME
] || !a
[OVS_VPORT_ATTR_TYPE
] ||
1582 !a
[OVS_VPORT_ATTR_UPCALL_PID
])
1586 dp
= get_dp(sock_net(skb
->sk
), ovs_header
->dp_ifindex
);
1591 if (a
[OVS_VPORT_ATTR_PORT_NO
]) {
1592 port_no
= nla_get_u32(a
[OVS_VPORT_ATTR_PORT_NO
]);
1595 if (port_no
>= DP_MAX_PORTS
)
1598 vport
= ovs_vport_ovsl(dp
, port_no
);
1603 for (port_no
= 1; ; port_no
++) {
1604 if (port_no
>= DP_MAX_PORTS
) {
1608 vport
= ovs_vport_ovsl(dp
, port_no
);
1614 parms
.name
= nla_data(a
[OVS_VPORT_ATTR_NAME
]);
1615 parms
.type
= nla_get_u32(a
[OVS_VPORT_ATTR_TYPE
]);
1616 parms
.options
= a
[OVS_VPORT_ATTR_OPTIONS
];
1618 parms
.port_no
= port_no
;
1619 parms
.upcall_portid
= nla_get_u32(a
[OVS_VPORT_ATTR_UPCALL_PID
]);
1621 vport
= new_vport(&parms
);
1622 err
= PTR_ERR(vport
);
1627 reply
= ovs_vport_cmd_build_info(vport
, info
->snd_portid
, info
->snd_seq
,
1629 if (IS_ERR(reply
)) {
1630 err
= PTR_ERR(reply
);
1631 ovs_dp_detach_port(vport
);
1635 ovs_notify(&dp_vport_genl_family
, reply
, info
);
1643 static int ovs_vport_cmd_set(struct sk_buff
*skb
, struct genl_info
*info
)
1645 struct nlattr
**a
= info
->attrs
;
1646 struct sk_buff
*reply
;
1647 struct vport
*vport
;
1651 vport
= lookup_vport(sock_net(skb
->sk
), info
->userhdr
, a
);
1652 err
= PTR_ERR(vport
);
1656 if (a
[OVS_VPORT_ATTR_TYPE
] &&
1657 nla_get_u32(a
[OVS_VPORT_ATTR_TYPE
]) != vport
->ops
->type
) {
1662 reply
= nlmsg_new(NLMSG_DEFAULT_SIZE
, GFP_KERNEL
);
1668 if (a
[OVS_VPORT_ATTR_OPTIONS
]) {
1669 err
= ovs_vport_set_options(vport
, a
[OVS_VPORT_ATTR_OPTIONS
]);
1674 if (a
[OVS_VPORT_ATTR_UPCALL_PID
])
1675 vport
->upcall_portid
= nla_get_u32(a
[OVS_VPORT_ATTR_UPCALL_PID
]);
1677 err
= ovs_vport_cmd_fill_info(vport
, reply
, info
->snd_portid
,
1678 info
->snd_seq
, 0, OVS_VPORT_CMD_NEW
);
1682 ovs_notify(&dp_vport_genl_family
, reply
, info
);
1692 static int ovs_vport_cmd_del(struct sk_buff
*skb
, struct genl_info
*info
)
1694 struct nlattr
**a
= info
->attrs
;
1695 struct sk_buff
*reply
;
1696 struct vport
*vport
;
1700 vport
= lookup_vport(sock_net(skb
->sk
), info
->userhdr
, a
);
1701 err
= PTR_ERR(vport
);
1705 if (vport
->port_no
== OVSP_LOCAL
) {
1710 reply
= ovs_vport_cmd_build_info(vport
, info
->snd_portid
,
1711 info
->snd_seq
, OVS_VPORT_CMD_DEL
);
1712 err
= PTR_ERR(reply
);
1717 ovs_dp_detach_port(vport
);
1719 ovs_notify(&dp_vport_genl_family
, reply
, info
);
1726 static int ovs_vport_cmd_get(struct sk_buff
*skb
, struct genl_info
*info
)
1728 struct nlattr
**a
= info
->attrs
;
1729 struct ovs_header
*ovs_header
= info
->userhdr
;
1730 struct sk_buff
*reply
;
1731 struct vport
*vport
;
1735 vport
= lookup_vport(sock_net(skb
->sk
), ovs_header
, a
);
1736 err
= PTR_ERR(vport
);
1740 reply
= ovs_vport_cmd_build_info(vport
, info
->snd_portid
,
1741 info
->snd_seq
, OVS_VPORT_CMD_NEW
);
1742 err
= PTR_ERR(reply
);
1748 return genlmsg_reply(reply
, info
);
1755 static int ovs_vport_cmd_dump(struct sk_buff
*skb
, struct netlink_callback
*cb
)
1757 struct ovs_header
*ovs_header
= genlmsg_data(nlmsg_data(cb
->nlh
));
1758 struct datapath
*dp
;
1759 int bucket
= cb
->args
[0], skip
= cb
->args
[1];
1763 dp
= get_dp(sock_net(skb
->sk
), ovs_header
->dp_ifindex
);
1768 for (i
= bucket
; i
< DP_VPORT_HASH_BUCKETS
; i
++) {
1769 struct vport
*vport
;
1772 hlist_for_each_entry_rcu(vport
, &dp
->ports
[i
], dp_hash_node
) {
1774 ovs_vport_cmd_fill_info(vport
, skb
,
1775 NETLINK_CB(cb
->skb
).portid
,
1778 OVS_VPORT_CMD_NEW
) < 0)
1794 static const struct genl_ops dp_vport_genl_ops
[] = {
1795 { .cmd
= OVS_VPORT_CMD_NEW
,
1796 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1797 .policy
= vport_policy
,
1798 .doit
= ovs_vport_cmd_new
1800 { .cmd
= OVS_VPORT_CMD_DEL
,
1801 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1802 .policy
= vport_policy
,
1803 .doit
= ovs_vport_cmd_del
1805 { .cmd
= OVS_VPORT_CMD_GET
,
1806 .flags
= 0, /* OK for unprivileged users. */
1807 .policy
= vport_policy
,
1808 .doit
= ovs_vport_cmd_get
,
1809 .dumpit
= ovs_vport_cmd_dump
1811 { .cmd
= OVS_VPORT_CMD_SET
,
1812 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1813 .policy
= vport_policy
,
1814 .doit
= ovs_vport_cmd_set
,
1818 struct genl_family_and_ops
{
1819 struct genl_family
*family
;
1820 const struct genl_ops
*ops
;
1822 const struct genl_multicast_group
*group
;
1825 static const struct genl_family_and_ops dp_genl_families
[] = {
1826 { &dp_datapath_genl_family
,
1827 dp_datapath_genl_ops
, ARRAY_SIZE(dp_datapath_genl_ops
),
1828 &ovs_dp_datapath_multicast_group
},
1829 { &dp_vport_genl_family
,
1830 dp_vport_genl_ops
, ARRAY_SIZE(dp_vport_genl_ops
),
1831 &ovs_dp_vport_multicast_group
},
1832 { &dp_flow_genl_family
,
1833 dp_flow_genl_ops
, ARRAY_SIZE(dp_flow_genl_ops
),
1834 &ovs_dp_flow_multicast_group
},
1835 { &dp_packet_genl_family
,
1836 dp_packet_genl_ops
, ARRAY_SIZE(dp_packet_genl_ops
),
1840 static void dp_unregister_genl(int n_families
)
1844 for (i
= 0; i
< n_families
; i
++)
1845 genl_unregister_family(dp_genl_families
[i
].family
);
1848 static int dp_register_genl(void)
1855 for (i
= 0; i
< ARRAY_SIZE(dp_genl_families
); i
++) {
1856 const struct genl_family_and_ops
*f
= &dp_genl_families
[i
];
1858 f
->family
->ops
= f
->ops
;
1859 f
->family
->n_ops
= f
->n_ops
;
1860 f
->family
->mcgrps
= f
->group
;
1861 f
->family
->n_mcgrps
= f
->group
? 1 : 0;
1862 err
= genl_register_family(f
->family
);
1871 dp_unregister_genl(n_registered
);
1875 static int __net_init
ovs_init_net(struct net
*net
)
1877 struct ovs_net
*ovs_net
= net_generic(net
, ovs_net_id
);
1879 INIT_LIST_HEAD(&ovs_net
->dps
);
1880 INIT_WORK(&ovs_net
->dp_notify_work
, ovs_dp_notify_wq
);
1884 static void __net_exit
ovs_exit_net(struct net
*net
)
1886 struct datapath
*dp
, *dp_next
;
1887 struct ovs_net
*ovs_net
= net_generic(net
, ovs_net_id
);
1890 list_for_each_entry_safe(dp
, dp_next
, &ovs_net
->dps
, list_node
)
1894 cancel_work_sync(&ovs_net
->dp_notify_work
);
1897 static struct pernet_operations ovs_net_ops
= {
1898 .init
= ovs_init_net
,
1899 .exit
= ovs_exit_net
,
1901 .size
= sizeof(struct ovs_net
),
1904 static int __init
dp_init(void)
1908 BUILD_BUG_ON(sizeof(struct ovs_skb_cb
) > FIELD_SIZEOF(struct sk_buff
, cb
));
1910 pr_info("Open vSwitch switching datapath\n");
1912 err
= ovs_flow_init();
1916 err
= ovs_vport_init();
1918 goto error_flow_exit
;
1920 err
= register_pernet_device(&ovs_net_ops
);
1922 goto error_vport_exit
;
1924 err
= register_netdevice_notifier(&ovs_dp_device_notifier
);
1926 goto error_netns_exit
;
1928 err
= dp_register_genl();
1930 goto error_unreg_notifier
;
1934 error_unreg_notifier
:
1935 unregister_netdevice_notifier(&ovs_dp_device_notifier
);
1937 unregister_pernet_device(&ovs_net_ops
);
1946 static void dp_cleanup(void)
1948 dp_unregister_genl(ARRAY_SIZE(dp_genl_families
));
1949 unregister_netdevice_notifier(&ovs_dp_device_notifier
);
1950 unregister_pernet_device(&ovs_net_ops
);
1956 module_init(dp_init
);
1957 module_exit(dp_cleanup
);
1959 MODULE_DESCRIPTION("Open vSwitch switching datapath");
1960 MODULE_LICENSE("GPL");