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_netlink.h"
59 #include "vport-internal_dev.h"
60 #include "vport-netdev.h"
62 int ovs_net_id __read_mostly
;
64 static void ovs_notify(struct genl_family
*family
,
65 struct sk_buff
*skb
, struct genl_info
*info
,
66 struct genl_multicast_group
*grp
)
68 genl_notify(family
, skb
, genl_info_net(info
), info
->snd_portid
,
69 grp
->id
, 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 net
*, int dp_ifindex
, struct sk_buff
*,
113 const struct dp_upcall_info
*);
114 static int queue_userspace_packet(struct net
*, int dp_ifindex
,
116 const struct dp_upcall_info
*);
118 /* Must be called with rcu_read_lock or ovs_mutex. */
119 static struct datapath
*get_dp(struct net
*net
, int dp_ifindex
)
121 struct datapath
*dp
= NULL
;
122 struct net_device
*dev
;
125 dev
= dev_get_by_index_rcu(net
, dp_ifindex
);
127 struct vport
*vport
= ovs_internal_dev_get_vport(dev
);
136 /* Must be called with rcu_read_lock or ovs_mutex. */
137 const char *ovs_dp_name(const struct datapath
*dp
)
139 struct vport
*vport
= ovs_vport_ovsl_rcu(dp
, OVSP_LOCAL
);
140 return vport
->ops
->get_name(vport
);
143 static int get_dpifindex(struct datapath
*dp
)
150 local
= ovs_vport_rcu(dp
, OVSP_LOCAL
);
152 ifindex
= netdev_vport_priv(local
)->dev
->ifindex
;
161 static void destroy_dp_rcu(struct rcu_head
*rcu
)
163 struct datapath
*dp
= container_of(rcu
, struct datapath
, rcu
);
165 ovs_flow_tbl_destroy(&dp
->table
);
166 free_percpu(dp
->stats_percpu
);
167 release_net(ovs_dp_get_net(dp
));
172 static struct hlist_head
*vport_hash_bucket(const struct datapath
*dp
,
175 return &dp
->ports
[port_no
& (DP_VPORT_HASH_BUCKETS
- 1)];
178 struct vport
*ovs_lookup_vport(const struct datapath
*dp
, u16 port_no
)
181 struct hlist_head
*head
;
183 head
= vport_hash_bucket(dp
, port_no
);
184 hlist_for_each_entry_rcu(vport
, head
, dp_hash_node
) {
185 if (vport
->port_no
== port_no
)
191 /* Called with ovs_mutex. */
192 static struct vport
*new_vport(const struct vport_parms
*parms
)
196 vport
= ovs_vport_add(parms
);
197 if (!IS_ERR(vport
)) {
198 struct datapath
*dp
= parms
->dp
;
199 struct hlist_head
*head
= vport_hash_bucket(dp
, vport
->port_no
);
201 hlist_add_head_rcu(&vport
->dp_hash_node
, head
);
206 void ovs_dp_detach_port(struct vport
*p
)
210 /* First drop references to device. */
211 hlist_del_rcu(&p
->dp_hash_node
);
213 /* Then destroy it. */
217 /* Must be called with rcu_read_lock. */
218 void ovs_dp_process_received_packet(struct vport
*p
, struct sk_buff
*skb
)
220 struct datapath
*dp
= p
->dp
;
221 struct sw_flow
*flow
;
222 struct dp_stats_percpu
*stats
;
223 struct sw_flow_key key
;
228 stats
= this_cpu_ptr(dp
->stats_percpu
);
230 /* Extract flow from 'skb' into 'key'. */
231 error
= ovs_flow_extract(skb
, p
->port_no
, &key
);
232 if (unlikely(error
)) {
238 flow
= ovs_flow_tbl_lookup(&dp
->table
, &key
, &n_mask_hit
);
239 if (unlikely(!flow
)) {
240 struct dp_upcall_info upcall
;
242 upcall
.cmd
= OVS_PACKET_CMD_MISS
;
244 upcall
.userdata
= NULL
;
245 upcall
.portid
= p
->upcall_portid
;
246 ovs_dp_upcall(dp
, skb
, &upcall
);
248 stats_counter
= &stats
->n_missed
;
252 OVS_CB(skb
)->flow
= flow
;
253 OVS_CB(skb
)->pkt_key
= &key
;
255 stats_counter
= &stats
->n_hit
;
256 ovs_flow_used(OVS_CB(skb
)->flow
, skb
);
257 ovs_execute_actions(dp
, skb
);
260 /* Update datapath statistics. */
261 u64_stats_update_begin(&stats
->sync
);
263 stats
->n_mask_hit
+= n_mask_hit
;
264 u64_stats_update_end(&stats
->sync
);
267 static struct genl_family dp_packet_genl_family
= {
268 .id
= GENL_ID_GENERATE
,
269 .hdrsize
= sizeof(struct ovs_header
),
270 .name
= OVS_PACKET_FAMILY
,
271 .version
= OVS_PACKET_VERSION
,
272 .maxattr
= OVS_PACKET_ATTR_MAX
,
274 .parallel_ops
= true,
277 int ovs_dp_upcall(struct datapath
*dp
, struct sk_buff
*skb
,
278 const struct dp_upcall_info
*upcall_info
)
280 struct dp_stats_percpu
*stats
;
284 if (upcall_info
->portid
== 0) {
289 dp_ifindex
= get_dpifindex(dp
);
295 if (!skb_is_gso(skb
))
296 err
= queue_userspace_packet(ovs_dp_get_net(dp
), dp_ifindex
, skb
, upcall_info
);
298 err
= queue_gso_packets(ovs_dp_get_net(dp
), dp_ifindex
, skb
, upcall_info
);
305 stats
= this_cpu_ptr(dp
->stats_percpu
);
307 u64_stats_update_begin(&stats
->sync
);
309 u64_stats_update_end(&stats
->sync
);
314 static int queue_gso_packets(struct net
*net
, int dp_ifindex
,
316 const struct dp_upcall_info
*upcall_info
)
318 unsigned short gso_type
= skb_shinfo(skb
)->gso_type
;
319 struct dp_upcall_info later_info
;
320 struct sw_flow_key later_key
;
321 struct sk_buff
*segs
, *nskb
;
324 segs
= __skb_gso_segment(skb
, NETIF_F_SG
| NETIF_F_HW_CSUM
, false);
326 return PTR_ERR(segs
);
328 /* Queue all of the segments. */
331 err
= queue_userspace_packet(net
, dp_ifindex
, skb
, upcall_info
);
335 if (skb
== segs
&& gso_type
& SKB_GSO_UDP
) {
336 /* The initial flow key extracted by ovs_flow_extract()
337 * in this case is for a first fragment, so we need to
338 * properly mark later fragments.
340 later_key
= *upcall_info
->key
;
341 later_key
.ip
.frag
= OVS_FRAG_TYPE_LATER
;
343 later_info
= *upcall_info
;
344 later_info
.key
= &later_key
;
345 upcall_info
= &later_info
;
347 } while ((skb
= skb
->next
));
349 /* Free all of the segments. */
357 } while ((skb
= nskb
));
361 static size_t key_attr_size(void)
363 return nla_total_size(4) /* OVS_KEY_ATTR_PRIORITY */
364 + nla_total_size(0) /* OVS_KEY_ATTR_TUNNEL */
365 + nla_total_size(8) /* OVS_TUNNEL_KEY_ATTR_ID */
366 + nla_total_size(4) /* OVS_TUNNEL_KEY_ATTR_IPV4_SRC */
367 + nla_total_size(4) /* OVS_TUNNEL_KEY_ATTR_IPV4_DST */
368 + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TOS */
369 + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TTL */
370 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
371 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_CSUM */
372 + nla_total_size(4) /* OVS_KEY_ATTR_IN_PORT */
373 + nla_total_size(4) /* OVS_KEY_ATTR_SKB_MARK */
374 + nla_total_size(12) /* OVS_KEY_ATTR_ETHERNET */
375 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
376 + nla_total_size(4) /* OVS_KEY_ATTR_8021Q */
377 + nla_total_size(0) /* OVS_KEY_ATTR_ENCAP */
378 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
379 + nla_total_size(40) /* OVS_KEY_ATTR_IPV6 */
380 + nla_total_size(2) /* OVS_KEY_ATTR_ICMPV6 */
381 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
384 static size_t upcall_msg_size(const struct sk_buff
*skb
,
385 const struct nlattr
*userdata
)
387 size_t size
= NLMSG_ALIGN(sizeof(struct ovs_header
))
388 + nla_total_size(skb
->len
) /* OVS_PACKET_ATTR_PACKET */
389 + nla_total_size(key_attr_size()); /* OVS_PACKET_ATTR_KEY */
391 /* OVS_PACKET_ATTR_USERDATA */
393 size
+= NLA_ALIGN(userdata
->nla_len
);
398 static int queue_userspace_packet(struct net
*net
, int dp_ifindex
,
400 const struct dp_upcall_info
*upcall_info
)
402 struct ovs_header
*upcall
;
403 struct sk_buff
*nskb
= NULL
;
404 struct sk_buff
*user_skb
; /* to be queued to userspace */
408 if (vlan_tx_tag_present(skb
)) {
409 nskb
= skb_clone(skb
, GFP_ATOMIC
);
413 nskb
= __vlan_put_tag(nskb
, nskb
->vlan_proto
, vlan_tx_tag_get(nskb
));
421 if (nla_attr_size(skb
->len
) > USHRT_MAX
) {
426 user_skb
= genlmsg_new(upcall_msg_size(skb
, upcall_info
->userdata
), GFP_ATOMIC
);
432 upcall
= genlmsg_put(user_skb
, 0, 0, &dp_packet_genl_family
,
433 0, upcall_info
->cmd
);
434 upcall
->dp_ifindex
= dp_ifindex
;
436 nla
= nla_nest_start(user_skb
, OVS_PACKET_ATTR_KEY
);
437 ovs_nla_put_flow(upcall_info
->key
, upcall_info
->key
, user_skb
);
438 nla_nest_end(user_skb
, nla
);
440 if (upcall_info
->userdata
)
441 __nla_put(user_skb
, OVS_PACKET_ATTR_USERDATA
,
442 nla_len(upcall_info
->userdata
),
443 nla_data(upcall_info
->userdata
));
445 nla
= __nla_reserve(user_skb
, OVS_PACKET_ATTR_PACKET
, skb
->len
);
447 skb_copy_and_csum_dev(skb
, nla_data(nla
));
449 genlmsg_end(user_skb
, upcall
);
450 err
= genlmsg_unicast(net
, user_skb
, upcall_info
->portid
);
457 static void clear_stats(struct sw_flow
*flow
)
461 flow
->packet_count
= 0;
462 flow
->byte_count
= 0;
465 static int ovs_packet_cmd_execute(struct sk_buff
*skb
, struct genl_info
*info
)
467 struct ovs_header
*ovs_header
= info
->userhdr
;
468 struct nlattr
**a
= info
->attrs
;
469 struct sw_flow_actions
*acts
;
470 struct sk_buff
*packet
;
471 struct sw_flow
*flow
;
478 if (!a
[OVS_PACKET_ATTR_PACKET
] || !a
[OVS_PACKET_ATTR_KEY
] ||
479 !a
[OVS_PACKET_ATTR_ACTIONS
])
482 len
= nla_len(a
[OVS_PACKET_ATTR_PACKET
]);
483 packet
= __dev_alloc_skb(NET_IP_ALIGN
+ len
, GFP_KERNEL
);
487 skb_reserve(packet
, NET_IP_ALIGN
);
489 nla_memcpy(__skb_put(packet
, len
), a
[OVS_PACKET_ATTR_PACKET
], len
);
491 skb_reset_mac_header(packet
);
492 eth
= eth_hdr(packet
);
494 /* Normally, setting the skb 'protocol' field would be handled by a
495 * call to eth_type_trans(), but it assumes there's a sending
496 * device, which we may not have. */
497 if (ntohs(eth
->h_proto
) >= ETH_P_802_3_MIN
)
498 packet
->protocol
= eth
->h_proto
;
500 packet
->protocol
= htons(ETH_P_802_2
);
502 /* Build an sw_flow for sending this packet. */
503 flow
= ovs_flow_alloc();
508 err
= ovs_flow_extract(packet
, -1, &flow
->key
);
512 err
= ovs_nla_get_flow_metadata(flow
, a
[OVS_PACKET_ATTR_KEY
]);
515 acts
= ovs_nla_alloc_flow_actions(nla_len(a
[OVS_PACKET_ATTR_ACTIONS
]));
520 err
= ovs_nla_copy_actions(a
[OVS_PACKET_ATTR_ACTIONS
],
521 &flow
->key
, 0, &acts
);
522 rcu_assign_pointer(flow
->sf_acts
, acts
);
526 OVS_CB(packet
)->flow
= flow
;
527 OVS_CB(packet
)->pkt_key
= &flow
->key
;
528 packet
->priority
= flow
->key
.phy
.priority
;
529 packet
->mark
= flow
->key
.phy
.skb_mark
;
532 dp
= get_dp(sock_net(skb
->sk
), ovs_header
->dp_ifindex
);
538 err
= ovs_execute_actions(dp
, packet
);
542 ovs_flow_free(flow
, false);
548 ovs_flow_free(flow
, false);
555 static const struct nla_policy packet_policy
[OVS_PACKET_ATTR_MAX
+ 1] = {
556 [OVS_PACKET_ATTR_PACKET
] = { .len
= ETH_HLEN
},
557 [OVS_PACKET_ATTR_KEY
] = { .type
= NLA_NESTED
},
558 [OVS_PACKET_ATTR_ACTIONS
] = { .type
= NLA_NESTED
},
561 static const struct genl_ops dp_packet_genl_ops
[] = {
562 { .cmd
= OVS_PACKET_CMD_EXECUTE
,
563 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
564 .policy
= packet_policy
,
565 .doit
= ovs_packet_cmd_execute
569 static void get_dp_stats(struct datapath
*dp
, struct ovs_dp_stats
*stats
,
570 struct ovs_dp_megaflow_stats
*mega_stats
)
574 memset(mega_stats
, 0, sizeof(*mega_stats
));
576 stats
->n_flows
= ovs_flow_tbl_count(&dp
->table
);
577 mega_stats
->n_masks
= ovs_flow_tbl_num_masks(&dp
->table
);
579 stats
->n_hit
= stats
->n_missed
= stats
->n_lost
= 0;
581 for_each_possible_cpu(i
) {
582 const struct dp_stats_percpu
*percpu_stats
;
583 struct dp_stats_percpu local_stats
;
586 percpu_stats
= per_cpu_ptr(dp
->stats_percpu
, i
);
589 start
= u64_stats_fetch_begin_bh(&percpu_stats
->sync
);
590 local_stats
= *percpu_stats
;
591 } while (u64_stats_fetch_retry_bh(&percpu_stats
->sync
, start
));
593 stats
->n_hit
+= local_stats
.n_hit
;
594 stats
->n_missed
+= local_stats
.n_missed
;
595 stats
->n_lost
+= local_stats
.n_lost
;
596 mega_stats
->n_mask_hit
+= local_stats
.n_mask_hit
;
600 static const struct nla_policy flow_policy
[OVS_FLOW_ATTR_MAX
+ 1] = {
601 [OVS_FLOW_ATTR_KEY
] = { .type
= NLA_NESTED
},
602 [OVS_FLOW_ATTR_ACTIONS
] = { .type
= NLA_NESTED
},
603 [OVS_FLOW_ATTR_CLEAR
] = { .type
= NLA_FLAG
},
606 static struct genl_family dp_flow_genl_family
= {
607 .id
= GENL_ID_GENERATE
,
608 .hdrsize
= sizeof(struct ovs_header
),
609 .name
= OVS_FLOW_FAMILY
,
610 .version
= OVS_FLOW_VERSION
,
611 .maxattr
= OVS_FLOW_ATTR_MAX
,
613 .parallel_ops
= true,
616 static struct genl_multicast_group ovs_dp_flow_multicast_group
= {
617 .name
= OVS_FLOW_MCGROUP
620 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions
*acts
)
622 return NLMSG_ALIGN(sizeof(struct ovs_header
))
623 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
624 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_MASK */
625 + nla_total_size(sizeof(struct ovs_flow_stats
)) /* OVS_FLOW_ATTR_STATS */
626 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
627 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
628 + nla_total_size(acts
->actions_len
); /* OVS_FLOW_ATTR_ACTIONS */
631 /* Called with ovs_mutex. */
632 static int ovs_flow_cmd_fill_info(struct sw_flow
*flow
, struct datapath
*dp
,
633 struct sk_buff
*skb
, u32 portid
,
634 u32 seq
, u32 flags
, u8 cmd
)
636 const int skb_orig_len
= skb
->len
;
637 struct nlattr
*start
;
638 struct ovs_flow_stats stats
;
639 struct ovs_header
*ovs_header
;
645 ovs_header
= genlmsg_put(skb
, portid
, seq
, &dp_flow_genl_family
, flags
, cmd
);
649 ovs_header
->dp_ifindex
= get_dpifindex(dp
);
652 nla
= nla_nest_start(skb
, OVS_FLOW_ATTR_KEY
);
654 goto nla_put_failure
;
656 err
= ovs_nla_put_flow(&flow
->unmasked_key
, &flow
->unmasked_key
, skb
);
659 nla_nest_end(skb
, nla
);
661 nla
= nla_nest_start(skb
, OVS_FLOW_ATTR_MASK
);
663 goto nla_put_failure
;
665 err
= ovs_nla_put_flow(&flow
->key
, &flow
->mask
->key
, skb
);
669 nla_nest_end(skb
, nla
);
671 spin_lock_bh(&flow
->lock
);
673 stats
.n_packets
= flow
->packet_count
;
674 stats
.n_bytes
= flow
->byte_count
;
675 tcp_flags
= (u8
)ntohs(flow
->tcp_flags
);
676 spin_unlock_bh(&flow
->lock
);
679 nla_put_u64(skb
, OVS_FLOW_ATTR_USED
, ovs_flow_used_time(used
)))
680 goto nla_put_failure
;
682 if (stats
.n_packets
&&
683 nla_put(skb
, OVS_FLOW_ATTR_STATS
,
684 sizeof(struct ovs_flow_stats
), &stats
))
685 goto nla_put_failure
;
688 nla_put_u8(skb
, OVS_FLOW_ATTR_TCP_FLAGS
, tcp_flags
))
689 goto nla_put_failure
;
691 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
692 * this is the first flow to be dumped into 'skb'. This is unusual for
693 * Netlink but individual action lists can be longer than
694 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
695 * The userspace caller can always fetch the actions separately if it
696 * really wants them. (Most userspace callers in fact don't care.)
698 * This can only fail for dump operations because the skb is always
699 * properly sized for single flows.
701 start
= nla_nest_start(skb
, OVS_FLOW_ATTR_ACTIONS
);
703 const struct sw_flow_actions
*sf_acts
;
705 sf_acts
= rcu_dereference_check(flow
->sf_acts
,
706 lockdep_ovsl_is_held());
708 err
= ovs_nla_put_actions(sf_acts
->actions
,
709 sf_acts
->actions_len
, skb
);
711 nla_nest_end(skb
, start
);
716 nla_nest_cancel(skb
, start
);
718 } else if (skb_orig_len
)
719 goto nla_put_failure
;
721 return genlmsg_end(skb
, ovs_header
);
726 genlmsg_cancel(skb
, ovs_header
);
730 static struct sk_buff
*ovs_flow_cmd_alloc_info(struct sw_flow
*flow
)
732 const struct sw_flow_actions
*sf_acts
;
734 sf_acts
= ovsl_dereference(flow
->sf_acts
);
736 return genlmsg_new(ovs_flow_cmd_msg_size(sf_acts
), GFP_KERNEL
);
739 static struct sk_buff
*ovs_flow_cmd_build_info(struct sw_flow
*flow
,
741 u32 portid
, u32 seq
, u8 cmd
)
746 skb
= ovs_flow_cmd_alloc_info(flow
);
748 return ERR_PTR(-ENOMEM
);
750 retval
= ovs_flow_cmd_fill_info(flow
, dp
, skb
, portid
, seq
, 0, cmd
);
755 static struct sw_flow
*__ovs_flow_tbl_lookup(struct flow_table
*tbl
,
756 const struct sw_flow_key
*key
)
758 u32 __always_unused n_mask_hit
;
760 return ovs_flow_tbl_lookup(tbl
, key
, &n_mask_hit
);
763 static int ovs_flow_cmd_new_or_set(struct sk_buff
*skb
, struct genl_info
*info
)
765 struct nlattr
**a
= info
->attrs
;
766 struct ovs_header
*ovs_header
= info
->userhdr
;
767 struct sw_flow_key key
, masked_key
;
768 struct sw_flow
*flow
= NULL
;
769 struct sw_flow_mask mask
;
770 struct sk_buff
*reply
;
772 struct sw_flow_actions
*acts
= NULL
;
773 struct sw_flow_match match
;
778 if (!a
[OVS_FLOW_ATTR_KEY
])
781 ovs_match_init(&match
, &key
, &mask
);
782 error
= ovs_nla_get_match(&match
,
783 a
[OVS_FLOW_ATTR_KEY
], a
[OVS_FLOW_ATTR_MASK
]);
787 /* Validate actions. */
788 if (a
[OVS_FLOW_ATTR_ACTIONS
]) {
789 acts
= ovs_nla_alloc_flow_actions(nla_len(a
[OVS_FLOW_ATTR_ACTIONS
]));
790 error
= PTR_ERR(acts
);
794 ovs_flow_mask_key(&masked_key
, &key
, &mask
);
795 error
= ovs_nla_copy_actions(a
[OVS_FLOW_ATTR_ACTIONS
],
796 &masked_key
, 0, &acts
);
798 OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
801 } else if (info
->genlhdr
->cmd
== OVS_FLOW_CMD_NEW
) {
807 dp
= get_dp(sock_net(skb
->sk
), ovs_header
->dp_ifindex
);
812 /* Check if this is a duplicate flow */
813 flow
= __ovs_flow_tbl_lookup(&dp
->table
, &key
);
815 /* Bail out if we're not allowed to create a new flow. */
817 if (info
->genlhdr
->cmd
== OVS_FLOW_CMD_SET
)
821 flow
= ovs_flow_alloc();
823 error
= PTR_ERR(flow
);
828 flow
->key
= masked_key
;
829 flow
->unmasked_key
= key
;
830 rcu_assign_pointer(flow
->sf_acts
, acts
);
832 /* Put flow in bucket. */
833 error
= ovs_flow_tbl_insert(&dp
->table
, flow
, &mask
);
839 reply
= ovs_flow_cmd_build_info(flow
, dp
, info
->snd_portid
,
840 info
->snd_seq
, OVS_FLOW_CMD_NEW
);
842 /* We found a matching flow. */
843 struct sw_flow_actions
*old_acts
;
845 /* Bail out if we're not allowed to modify an existing flow.
846 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
847 * because Generic Netlink treats the latter as a dump
848 * request. We also accept NLM_F_EXCL in case that bug ever
852 if (info
->genlhdr
->cmd
== OVS_FLOW_CMD_NEW
&&
853 info
->nlhdr
->nlmsg_flags
& (NLM_F_CREATE
| NLM_F_EXCL
))
856 /* The unmasked key has to be the same for flow updates. */
858 if (!ovs_flow_cmp_unmasked_key(flow
, &match
)) {
859 OVS_NLERR("Flow modification message rejected, unmasked key does not match.\n");
863 /* Update actions. */
864 old_acts
= ovsl_dereference(flow
->sf_acts
);
865 rcu_assign_pointer(flow
->sf_acts
, acts
);
866 ovs_nla_free_flow_actions(old_acts
);
868 reply
= ovs_flow_cmd_build_info(flow
, dp
, info
->snd_portid
,
869 info
->snd_seq
, OVS_FLOW_CMD_NEW
);
872 if (a
[OVS_FLOW_ATTR_CLEAR
]) {
873 spin_lock_bh(&flow
->lock
);
875 spin_unlock_bh(&flow
->lock
);
881 ovs_notify(&dp_flow_genl_family
, reply
, info
,
882 &ovs_dp_flow_multicast_group
);
884 genl_set_err(&dp_flow_genl_family
, sock_net(skb
->sk
), 0,
885 ovs_dp_flow_multicast_group
.id
, PTR_ERR(reply
));
889 ovs_flow_free(flow
, false);
898 static int ovs_flow_cmd_get(struct sk_buff
*skb
, struct genl_info
*info
)
900 struct nlattr
**a
= info
->attrs
;
901 struct ovs_header
*ovs_header
= info
->userhdr
;
902 struct sw_flow_key key
;
903 struct sk_buff
*reply
;
904 struct sw_flow
*flow
;
906 struct sw_flow_match match
;
909 if (!a
[OVS_FLOW_ATTR_KEY
]) {
910 OVS_NLERR("Flow get message rejected, Key attribute missing.\n");
914 ovs_match_init(&match
, &key
, NULL
);
915 err
= ovs_nla_get_match(&match
, a
[OVS_FLOW_ATTR_KEY
], NULL
);
920 dp
= get_dp(sock_net(skb
->sk
), ovs_header
->dp_ifindex
);
926 flow
= __ovs_flow_tbl_lookup(&dp
->table
, &key
);
927 if (!flow
|| !ovs_flow_cmp_unmasked_key(flow
, &match
)) {
932 reply
= ovs_flow_cmd_build_info(flow
, dp
, info
->snd_portid
,
933 info
->snd_seq
, 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
);
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
,
996 &ovs_dp_flow_multicast_group
);
1003 static int ovs_flow_cmd_dump(struct sk_buff
*skb
, struct netlink_callback
*cb
)
1005 struct ovs_header
*ovs_header
= genlmsg_data(nlmsg_data(cb
->nlh
));
1006 struct table_instance
*ti
;
1007 struct datapath
*dp
;
1010 dp
= get_dp(sock_net(skb
->sk
), ovs_header
->dp_ifindex
);
1016 ti
= rcu_dereference(dp
->table
.ti
);
1018 struct sw_flow
*flow
;
1021 bucket
= cb
->args
[0];
1023 flow
= ovs_flow_tbl_dump_next(ti
, &bucket
, &obj
);
1027 if (ovs_flow_cmd_fill_info(flow
, dp
, skb
,
1028 NETLINK_CB(cb
->skb
).portid
,
1029 cb
->nlh
->nlmsg_seq
, NLM_F_MULTI
,
1030 OVS_FLOW_CMD_NEW
) < 0)
1033 cb
->args
[0] = bucket
;
1040 static const struct genl_ops dp_flow_genl_ops
[] = {
1041 { .cmd
= OVS_FLOW_CMD_NEW
,
1042 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1043 .policy
= flow_policy
,
1044 .doit
= ovs_flow_cmd_new_or_set
1046 { .cmd
= OVS_FLOW_CMD_DEL
,
1047 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1048 .policy
= flow_policy
,
1049 .doit
= ovs_flow_cmd_del
1051 { .cmd
= OVS_FLOW_CMD_GET
,
1052 .flags
= 0, /* OK for unprivileged users. */
1053 .policy
= flow_policy
,
1054 .doit
= ovs_flow_cmd_get
,
1055 .dumpit
= ovs_flow_cmd_dump
1057 { .cmd
= OVS_FLOW_CMD_SET
,
1058 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1059 .policy
= flow_policy
,
1060 .doit
= ovs_flow_cmd_new_or_set
,
1064 static const struct nla_policy datapath_policy
[OVS_DP_ATTR_MAX
+ 1] = {
1065 [OVS_DP_ATTR_NAME
] = { .type
= NLA_NUL_STRING
, .len
= IFNAMSIZ
- 1 },
1066 [OVS_DP_ATTR_UPCALL_PID
] = { .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
));
1094 static int ovs_dp_cmd_fill_info(struct datapath
*dp
, struct sk_buff
*skb
,
1095 u32 portid
, u32 seq
, u32 flags
, u8 cmd
)
1097 struct ovs_header
*ovs_header
;
1098 struct ovs_dp_stats dp_stats
;
1099 struct ovs_dp_megaflow_stats dp_megaflow_stats
;
1102 ovs_header
= genlmsg_put(skb
, portid
, seq
, &dp_datapath_genl_family
,
1107 ovs_header
->dp_ifindex
= get_dpifindex(dp
);
1110 err
= nla_put_string(skb
, OVS_DP_ATTR_NAME
, ovs_dp_name(dp
));
1113 goto nla_put_failure
;
1115 get_dp_stats(dp
, &dp_stats
, &dp_megaflow_stats
);
1116 if (nla_put(skb
, OVS_DP_ATTR_STATS
, sizeof(struct ovs_dp_stats
),
1118 goto nla_put_failure
;
1120 if (nla_put(skb
, OVS_DP_ATTR_MEGAFLOW_STATS
,
1121 sizeof(struct ovs_dp_megaflow_stats
),
1122 &dp_megaflow_stats
))
1123 goto nla_put_failure
;
1125 return genlmsg_end(skb
, ovs_header
);
1128 genlmsg_cancel(skb
, ovs_header
);
1133 static struct sk_buff
*ovs_dp_cmd_build_info(struct datapath
*dp
, u32 portid
,
1136 struct sk_buff
*skb
;
1139 skb
= genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL
);
1141 return ERR_PTR(-ENOMEM
);
1143 retval
= ovs_dp_cmd_fill_info(dp
, skb
, portid
, seq
, 0, cmd
);
1146 return ERR_PTR(retval
);
1151 /* Called with ovs_mutex. */
1152 static struct datapath
*lookup_datapath(struct net
*net
,
1153 struct ovs_header
*ovs_header
,
1154 struct nlattr
*a
[OVS_DP_ATTR_MAX
+ 1])
1156 struct datapath
*dp
;
1158 if (!a
[OVS_DP_ATTR_NAME
])
1159 dp
= get_dp(net
, ovs_header
->dp_ifindex
);
1161 struct vport
*vport
;
1164 vport
= ovs_vport_locate(net
, nla_data(a
[OVS_DP_ATTR_NAME
]));
1165 dp
= vport
&& vport
->port_no
== OVSP_LOCAL
? vport
->dp
: NULL
;
1168 return dp
? dp
: ERR_PTR(-ENODEV
);
1171 static int ovs_dp_cmd_new(struct sk_buff
*skb
, struct genl_info
*info
)
1173 struct nlattr
**a
= info
->attrs
;
1174 struct vport_parms parms
;
1175 struct sk_buff
*reply
;
1176 struct datapath
*dp
;
1177 struct vport
*vport
;
1178 struct ovs_net
*ovs_net
;
1182 if (!a
[OVS_DP_ATTR_NAME
] || !a
[OVS_DP_ATTR_UPCALL_PID
])
1188 dp
= kzalloc(sizeof(*dp
), GFP_KERNEL
);
1190 goto err_unlock_ovs
;
1192 ovs_dp_set_net(dp
, hold_net(sock_net(skb
->sk
)));
1194 /* Allocate table. */
1195 err
= ovs_flow_tbl_init(&dp
->table
);
1199 dp
->stats_percpu
= alloc_percpu(struct dp_stats_percpu
);
1200 if (!dp
->stats_percpu
) {
1202 goto err_destroy_table
;
1205 dp
->ports
= kmalloc(DP_VPORT_HASH_BUCKETS
* sizeof(struct hlist_head
),
1209 goto err_destroy_percpu
;
1212 for (i
= 0; i
< DP_VPORT_HASH_BUCKETS
; i
++)
1213 INIT_HLIST_HEAD(&dp
->ports
[i
]);
1215 /* Set up our datapath device. */
1216 parms
.name
= nla_data(a
[OVS_DP_ATTR_NAME
]);
1217 parms
.type
= OVS_VPORT_TYPE_INTERNAL
;
1218 parms
.options
= NULL
;
1220 parms
.port_no
= OVSP_LOCAL
;
1221 parms
.upcall_portid
= nla_get_u32(a
[OVS_DP_ATTR_UPCALL_PID
]);
1223 vport
= new_vport(&parms
);
1224 if (IS_ERR(vport
)) {
1225 err
= PTR_ERR(vport
);
1229 goto err_destroy_ports_array
;
1232 reply
= ovs_dp_cmd_build_info(dp
, info
->snd_portid
,
1233 info
->snd_seq
, OVS_DP_CMD_NEW
);
1234 err
= PTR_ERR(reply
);
1236 goto err_destroy_local_port
;
1238 ovs_net
= net_generic(ovs_dp_get_net(dp
), ovs_net_id
);
1239 list_add_tail_rcu(&dp
->list_node
, &ovs_net
->dps
);
1243 ovs_notify(&dp_datapath_genl_family
, reply
, info
,
1244 &ovs_dp_datapath_multicast_group
);
1247 err_destroy_local_port
:
1248 ovs_dp_detach_port(ovs_vport_ovsl(dp
, OVSP_LOCAL
));
1249 err_destroy_ports_array
:
1252 free_percpu(dp
->stats_percpu
);
1254 ovs_flow_tbl_destroy(&dp
->table
);
1256 release_net(ovs_dp_get_net(dp
));
1264 /* Called with ovs_mutex. */
1265 static void __dp_destroy(struct datapath
*dp
)
1269 for (i
= 0; i
< DP_VPORT_HASH_BUCKETS
; i
++) {
1270 struct vport
*vport
;
1271 struct hlist_node
*n
;
1273 hlist_for_each_entry_safe(vport
, n
, &dp
->ports
[i
], dp_hash_node
)
1274 if (vport
->port_no
!= OVSP_LOCAL
)
1275 ovs_dp_detach_port(vport
);
1278 list_del_rcu(&dp
->list_node
);
1280 /* OVSP_LOCAL is datapath internal port. We need to make sure that
1281 * all port in datapath are destroyed first before freeing datapath.
1283 ovs_dp_detach_port(ovs_vport_ovsl(dp
, OVSP_LOCAL
));
1285 call_rcu(&dp
->rcu
, destroy_dp_rcu
);
1288 static int ovs_dp_cmd_del(struct sk_buff
*skb
, struct genl_info
*info
)
1290 struct sk_buff
*reply
;
1291 struct datapath
*dp
;
1295 dp
= lookup_datapath(sock_net(skb
->sk
), info
->userhdr
, info
->attrs
);
1300 reply
= ovs_dp_cmd_build_info(dp
, info
->snd_portid
,
1301 info
->snd_seq
, OVS_DP_CMD_DEL
);
1302 err
= PTR_ERR(reply
);
1309 ovs_notify(&dp_datapath_genl_family
, reply
, info
,
1310 &ovs_dp_datapath_multicast_group
);
1318 static int ovs_dp_cmd_set(struct sk_buff
*skb
, struct genl_info
*info
)
1320 struct sk_buff
*reply
;
1321 struct datapath
*dp
;
1325 dp
= lookup_datapath(sock_net(skb
->sk
), info
->userhdr
, info
->attrs
);
1330 reply
= ovs_dp_cmd_build_info(dp
, info
->snd_portid
,
1331 info
->snd_seq
, OVS_DP_CMD_NEW
);
1332 if (IS_ERR(reply
)) {
1333 err
= PTR_ERR(reply
);
1334 genl_set_err(&dp_datapath_genl_family
, sock_net(skb
->sk
), 0,
1335 ovs_dp_datapath_multicast_group
.id
, err
);
1341 ovs_notify(&dp_datapath_genl_family
, reply
, info
,
1342 &ovs_dp_datapath_multicast_group
);
1350 static int ovs_dp_cmd_get(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
);
1363 reply
= ovs_dp_cmd_build_info(dp
, info
->snd_portid
,
1364 info
->snd_seq
, OVS_DP_CMD_NEW
);
1365 if (IS_ERR(reply
)) {
1366 err
= PTR_ERR(reply
);
1371 return genlmsg_reply(reply
, info
);
1378 static int ovs_dp_cmd_dump(struct sk_buff
*skb
, struct netlink_callback
*cb
)
1380 struct ovs_net
*ovs_net
= net_generic(sock_net(skb
->sk
), ovs_net_id
);
1381 struct datapath
*dp
;
1382 int skip
= cb
->args
[0];
1386 list_for_each_entry_rcu(dp
, &ovs_net
->dps
, list_node
) {
1388 ovs_dp_cmd_fill_info(dp
, skb
, NETLINK_CB(cb
->skb
).portid
,
1389 cb
->nlh
->nlmsg_seq
, NLM_F_MULTI
,
1390 OVS_DP_CMD_NEW
) < 0)
1401 static const struct genl_ops dp_datapath_genl_ops
[] = {
1402 { .cmd
= OVS_DP_CMD_NEW
,
1403 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1404 .policy
= datapath_policy
,
1405 .doit
= ovs_dp_cmd_new
1407 { .cmd
= OVS_DP_CMD_DEL
,
1408 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1409 .policy
= datapath_policy
,
1410 .doit
= ovs_dp_cmd_del
1412 { .cmd
= OVS_DP_CMD_GET
,
1413 .flags
= 0, /* OK for unprivileged users. */
1414 .policy
= datapath_policy
,
1415 .doit
= ovs_dp_cmd_get
,
1416 .dumpit
= ovs_dp_cmd_dump
1418 { .cmd
= OVS_DP_CMD_SET
,
1419 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1420 .policy
= datapath_policy
,
1421 .doit
= ovs_dp_cmd_set
,
1425 static const struct nla_policy vport_policy
[OVS_VPORT_ATTR_MAX
+ 1] = {
1426 [OVS_VPORT_ATTR_NAME
] = { .type
= NLA_NUL_STRING
, .len
= IFNAMSIZ
- 1 },
1427 [OVS_VPORT_ATTR_STATS
] = { .len
= sizeof(struct ovs_vport_stats
) },
1428 [OVS_VPORT_ATTR_PORT_NO
] = { .type
= NLA_U32
},
1429 [OVS_VPORT_ATTR_TYPE
] = { .type
= NLA_U32
},
1430 [OVS_VPORT_ATTR_UPCALL_PID
] = { .type
= NLA_U32
},
1431 [OVS_VPORT_ATTR_OPTIONS
] = { .type
= NLA_NESTED
},
1434 struct genl_family dp_vport_genl_family
= {
1435 .id
= GENL_ID_GENERATE
,
1436 .hdrsize
= sizeof(struct ovs_header
),
1437 .name
= OVS_VPORT_FAMILY
,
1438 .version
= OVS_VPORT_VERSION
,
1439 .maxattr
= OVS_VPORT_ATTR_MAX
,
1441 .parallel_ops
= true,
1444 struct genl_multicast_group ovs_dp_vport_multicast_group
= {
1445 .name
= OVS_VPORT_MCGROUP
1448 /* Called with ovs_mutex or RCU read lock. */
1449 static int ovs_vport_cmd_fill_info(struct vport
*vport
, struct sk_buff
*skb
,
1450 u32 portid
, u32 seq
, u32 flags
, u8 cmd
)
1452 struct ovs_header
*ovs_header
;
1453 struct ovs_vport_stats vport_stats
;
1456 ovs_header
= genlmsg_put(skb
, portid
, seq
, &dp_vport_genl_family
,
1461 ovs_header
->dp_ifindex
= get_dpifindex(vport
->dp
);
1463 if (nla_put_u32(skb
, OVS_VPORT_ATTR_PORT_NO
, vport
->port_no
) ||
1464 nla_put_u32(skb
, OVS_VPORT_ATTR_TYPE
, vport
->ops
->type
) ||
1465 nla_put_string(skb
, OVS_VPORT_ATTR_NAME
, vport
->ops
->get_name(vport
)) ||
1466 nla_put_u32(skb
, OVS_VPORT_ATTR_UPCALL_PID
, vport
->upcall_portid
))
1467 goto nla_put_failure
;
1469 ovs_vport_get_stats(vport
, &vport_stats
);
1470 if (nla_put(skb
, OVS_VPORT_ATTR_STATS
, sizeof(struct ovs_vport_stats
),
1472 goto nla_put_failure
;
1474 err
= ovs_vport_get_options(vport
, skb
);
1475 if (err
== -EMSGSIZE
)
1478 return genlmsg_end(skb
, ovs_header
);
1483 genlmsg_cancel(skb
, ovs_header
);
1487 /* Called with ovs_mutex or RCU read lock. */
1488 struct sk_buff
*ovs_vport_cmd_build_info(struct vport
*vport
, u32 portid
,
1491 struct sk_buff
*skb
;
1494 skb
= nlmsg_new(NLMSG_DEFAULT_SIZE
, GFP_ATOMIC
);
1496 return ERR_PTR(-ENOMEM
);
1498 retval
= ovs_vport_cmd_fill_info(vport
, skb
, portid
, seq
, 0, cmd
);
1504 /* Called with ovs_mutex or RCU read lock. */
1505 static struct vport
*lookup_vport(struct net
*net
,
1506 struct ovs_header
*ovs_header
,
1507 struct nlattr
*a
[OVS_VPORT_ATTR_MAX
+ 1])
1509 struct datapath
*dp
;
1510 struct vport
*vport
;
1512 if (a
[OVS_VPORT_ATTR_NAME
]) {
1513 vport
= ovs_vport_locate(net
, nla_data(a
[OVS_VPORT_ATTR_NAME
]));
1515 return ERR_PTR(-ENODEV
);
1516 if (ovs_header
->dp_ifindex
&&
1517 ovs_header
->dp_ifindex
!= get_dpifindex(vport
->dp
))
1518 return ERR_PTR(-ENODEV
);
1520 } else if (a
[OVS_VPORT_ATTR_PORT_NO
]) {
1521 u32 port_no
= nla_get_u32(a
[OVS_VPORT_ATTR_PORT_NO
]);
1523 if (port_no
>= DP_MAX_PORTS
)
1524 return ERR_PTR(-EFBIG
);
1526 dp
= get_dp(net
, ovs_header
->dp_ifindex
);
1528 return ERR_PTR(-ENODEV
);
1530 vport
= ovs_vport_ovsl_rcu(dp
, port_no
);
1532 return ERR_PTR(-ENODEV
);
1535 return ERR_PTR(-EINVAL
);
1538 static int ovs_vport_cmd_new(struct sk_buff
*skb
, struct genl_info
*info
)
1540 struct nlattr
**a
= info
->attrs
;
1541 struct ovs_header
*ovs_header
= info
->userhdr
;
1542 struct vport_parms parms
;
1543 struct sk_buff
*reply
;
1544 struct vport
*vport
;
1545 struct datapath
*dp
;
1550 if (!a
[OVS_VPORT_ATTR_NAME
] || !a
[OVS_VPORT_ATTR_TYPE
] ||
1551 !a
[OVS_VPORT_ATTR_UPCALL_PID
])
1555 dp
= get_dp(sock_net(skb
->sk
), ovs_header
->dp_ifindex
);
1560 if (a
[OVS_VPORT_ATTR_PORT_NO
]) {
1561 port_no
= nla_get_u32(a
[OVS_VPORT_ATTR_PORT_NO
]);
1564 if (port_no
>= DP_MAX_PORTS
)
1567 vport
= ovs_vport_ovsl(dp
, port_no
);
1572 for (port_no
= 1; ; port_no
++) {
1573 if (port_no
>= DP_MAX_PORTS
) {
1577 vport
= ovs_vport_ovsl(dp
, port_no
);
1583 parms
.name
= nla_data(a
[OVS_VPORT_ATTR_NAME
]);
1584 parms
.type
= nla_get_u32(a
[OVS_VPORT_ATTR_TYPE
]);
1585 parms
.options
= a
[OVS_VPORT_ATTR_OPTIONS
];
1587 parms
.port_no
= port_no
;
1588 parms
.upcall_portid
= nla_get_u32(a
[OVS_VPORT_ATTR_UPCALL_PID
]);
1590 vport
= new_vport(&parms
);
1591 err
= PTR_ERR(vport
);
1596 reply
= ovs_vport_cmd_build_info(vport
, info
->snd_portid
, info
->snd_seq
,
1598 if (IS_ERR(reply
)) {
1599 err
= PTR_ERR(reply
);
1600 ovs_dp_detach_port(vport
);
1604 ovs_notify(&dp_vport_genl_family
, reply
, info
,
1605 &ovs_dp_vport_multicast_group
);
1613 static int ovs_vport_cmd_set(struct sk_buff
*skb
, struct genl_info
*info
)
1615 struct nlattr
**a
= info
->attrs
;
1616 struct sk_buff
*reply
;
1617 struct vport
*vport
;
1621 vport
= lookup_vport(sock_net(skb
->sk
), info
->userhdr
, a
);
1622 err
= PTR_ERR(vport
);
1626 if (a
[OVS_VPORT_ATTR_TYPE
] &&
1627 nla_get_u32(a
[OVS_VPORT_ATTR_TYPE
]) != vport
->ops
->type
) {
1632 reply
= nlmsg_new(NLMSG_DEFAULT_SIZE
, GFP_KERNEL
);
1638 if (a
[OVS_VPORT_ATTR_OPTIONS
]) {
1639 err
= ovs_vport_set_options(vport
, a
[OVS_VPORT_ATTR_OPTIONS
]);
1644 if (a
[OVS_VPORT_ATTR_UPCALL_PID
])
1645 vport
->upcall_portid
= nla_get_u32(a
[OVS_VPORT_ATTR_UPCALL_PID
]);
1647 err
= ovs_vport_cmd_fill_info(vport
, reply
, info
->snd_portid
,
1648 info
->snd_seq
, 0, OVS_VPORT_CMD_NEW
);
1652 ovs_notify(&dp_vport_genl_family
, reply
, info
,
1653 &ovs_dp_vport_multicast_group
);
1663 static int ovs_vport_cmd_del(struct sk_buff
*skb
, struct genl_info
*info
)
1665 struct nlattr
**a
= info
->attrs
;
1666 struct sk_buff
*reply
;
1667 struct vport
*vport
;
1671 vport
= lookup_vport(sock_net(skb
->sk
), info
->userhdr
, a
);
1672 err
= PTR_ERR(vport
);
1676 if (vport
->port_no
== OVSP_LOCAL
) {
1681 reply
= ovs_vport_cmd_build_info(vport
, info
->snd_portid
,
1682 info
->snd_seq
, OVS_VPORT_CMD_DEL
);
1683 err
= PTR_ERR(reply
);
1688 ovs_dp_detach_port(vport
);
1690 ovs_notify(&dp_vport_genl_family
, reply
, info
,
1691 &ovs_dp_vport_multicast_group
);
1698 static int ovs_vport_cmd_get(struct sk_buff
*skb
, struct genl_info
*info
)
1700 struct nlattr
**a
= info
->attrs
;
1701 struct ovs_header
*ovs_header
= info
->userhdr
;
1702 struct sk_buff
*reply
;
1703 struct vport
*vport
;
1707 vport
= lookup_vport(sock_net(skb
->sk
), ovs_header
, a
);
1708 err
= PTR_ERR(vport
);
1712 reply
= ovs_vport_cmd_build_info(vport
, info
->snd_portid
,
1713 info
->snd_seq
, OVS_VPORT_CMD_NEW
);
1714 err
= PTR_ERR(reply
);
1720 return genlmsg_reply(reply
, info
);
1727 static int ovs_vport_cmd_dump(struct sk_buff
*skb
, struct netlink_callback
*cb
)
1729 struct ovs_header
*ovs_header
= genlmsg_data(nlmsg_data(cb
->nlh
));
1730 struct datapath
*dp
;
1731 int bucket
= cb
->args
[0], skip
= cb
->args
[1];
1734 dp
= get_dp(sock_net(skb
->sk
), ovs_header
->dp_ifindex
);
1739 for (i
= bucket
; i
< DP_VPORT_HASH_BUCKETS
; i
++) {
1740 struct vport
*vport
;
1743 hlist_for_each_entry_rcu(vport
, &dp
->ports
[i
], dp_hash_node
) {
1745 ovs_vport_cmd_fill_info(vport
, skb
,
1746 NETLINK_CB(cb
->skb
).portid
,
1749 OVS_VPORT_CMD_NEW
) < 0)
1765 static const struct genl_ops dp_vport_genl_ops
[] = {
1766 { .cmd
= OVS_VPORT_CMD_NEW
,
1767 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1768 .policy
= vport_policy
,
1769 .doit
= ovs_vport_cmd_new
1771 { .cmd
= OVS_VPORT_CMD_DEL
,
1772 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1773 .policy
= vport_policy
,
1774 .doit
= ovs_vport_cmd_del
1776 { .cmd
= OVS_VPORT_CMD_GET
,
1777 .flags
= 0, /* OK for unprivileged users. */
1778 .policy
= vport_policy
,
1779 .doit
= ovs_vport_cmd_get
,
1780 .dumpit
= ovs_vport_cmd_dump
1782 { .cmd
= OVS_VPORT_CMD_SET
,
1783 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1784 .policy
= vport_policy
,
1785 .doit
= ovs_vport_cmd_set
,
1789 struct genl_family_and_ops
{
1790 struct genl_family
*family
;
1791 const struct genl_ops
*ops
;
1793 struct genl_multicast_group
*group
;
1796 static const struct genl_family_and_ops dp_genl_families
[] = {
1797 { &dp_datapath_genl_family
,
1798 dp_datapath_genl_ops
, ARRAY_SIZE(dp_datapath_genl_ops
),
1799 &ovs_dp_datapath_multicast_group
},
1800 { &dp_vport_genl_family
,
1801 dp_vport_genl_ops
, ARRAY_SIZE(dp_vport_genl_ops
),
1802 &ovs_dp_vport_multicast_group
},
1803 { &dp_flow_genl_family
,
1804 dp_flow_genl_ops
, ARRAY_SIZE(dp_flow_genl_ops
),
1805 &ovs_dp_flow_multicast_group
},
1806 { &dp_packet_genl_family
,
1807 dp_packet_genl_ops
, ARRAY_SIZE(dp_packet_genl_ops
),
1811 static void dp_unregister_genl(int n_families
)
1815 for (i
= 0; i
< n_families
; i
++)
1816 genl_unregister_family(dp_genl_families
[i
].family
);
1819 static int dp_register_genl(void)
1826 for (i
= 0; i
< ARRAY_SIZE(dp_genl_families
); i
++) {
1827 const struct genl_family_and_ops
*f
= &dp_genl_families
[i
];
1829 f
->family
->ops
= f
->ops
;
1830 f
->family
->n_ops
= f
->n_ops
;
1831 err
= genl_register_family(f
->family
);
1837 err
= genl_register_mc_group(f
->family
, f
->group
);
1846 dp_unregister_genl(n_registered
);
1850 static int __net_init
ovs_init_net(struct net
*net
)
1852 struct ovs_net
*ovs_net
= net_generic(net
, ovs_net_id
);
1854 INIT_LIST_HEAD(&ovs_net
->dps
);
1855 INIT_WORK(&ovs_net
->dp_notify_work
, ovs_dp_notify_wq
);
1859 static void __net_exit
ovs_exit_net(struct net
*net
)
1861 struct datapath
*dp
, *dp_next
;
1862 struct ovs_net
*ovs_net
= net_generic(net
, ovs_net_id
);
1865 list_for_each_entry_safe(dp
, dp_next
, &ovs_net
->dps
, list_node
)
1869 cancel_work_sync(&ovs_net
->dp_notify_work
);
1872 static struct pernet_operations ovs_net_ops
= {
1873 .init
= ovs_init_net
,
1874 .exit
= ovs_exit_net
,
1876 .size
= sizeof(struct ovs_net
),
1879 static int __init
dp_init(void)
1883 BUILD_BUG_ON(sizeof(struct ovs_skb_cb
) > FIELD_SIZEOF(struct sk_buff
, cb
));
1885 pr_info("Open vSwitch switching datapath\n");
1887 err
= ovs_flow_init();
1891 err
= ovs_vport_init();
1893 goto error_flow_exit
;
1895 err
= register_pernet_device(&ovs_net_ops
);
1897 goto error_vport_exit
;
1899 err
= register_netdevice_notifier(&ovs_dp_device_notifier
);
1901 goto error_netns_exit
;
1903 err
= dp_register_genl();
1905 goto error_unreg_notifier
;
1909 error_unreg_notifier
:
1910 unregister_netdevice_notifier(&ovs_dp_device_notifier
);
1912 unregister_pernet_device(&ovs_net_ops
);
1921 static void dp_cleanup(void)
1923 dp_unregister_genl(ARRAY_SIZE(dp_genl_families
));
1924 unregister_netdevice_notifier(&ovs_dp_device_notifier
);
1925 unregister_pernet_device(&ovs_net_ops
);
1931 module_init(dp_init
);
1932 module_exit(dp_cleanup
);
1934 MODULE_DESCRIPTION("Open vSwitch switching datapath");
1935 MODULE_LICENSE("GPL");