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
->sync
);
261 stats
->n_mask_hit
+= n_mask_hit
;
262 u64_stats_update_end(&stats
->sync
);
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
->sync
);
300 u64_stats_update_end(&stats
->sync
);
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 skb_zerocopy(user_skb
, skb
, skb
->len
, hlen
);
469 /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
470 if (!(dp
->user_features
& OVS_DP_F_UNALIGNED
)) {
471 size_t plen
= NLA_ALIGN(user_skb
->len
) - user_skb
->len
;
474 memset(skb_put(user_skb
, plen
), 0, plen
);
477 ((struct nlmsghdr
*) user_skb
->data
)->nlmsg_len
= user_skb
->len
;
479 err
= genlmsg_unicast(ovs_dp_get_net(dp
), user_skb
, upcall_info
->portid
);
485 static int ovs_packet_cmd_execute(struct sk_buff
*skb
, struct genl_info
*info
)
487 struct ovs_header
*ovs_header
= info
->userhdr
;
488 struct nlattr
**a
= info
->attrs
;
489 struct sw_flow_actions
*acts
;
490 struct sk_buff
*packet
;
491 struct sw_flow
*flow
;
498 if (!a
[OVS_PACKET_ATTR_PACKET
] || !a
[OVS_PACKET_ATTR_KEY
] ||
499 !a
[OVS_PACKET_ATTR_ACTIONS
])
502 len
= nla_len(a
[OVS_PACKET_ATTR_PACKET
]);
503 packet
= __dev_alloc_skb(NET_IP_ALIGN
+ len
, GFP_KERNEL
);
507 skb_reserve(packet
, NET_IP_ALIGN
);
509 nla_memcpy(__skb_put(packet
, len
), a
[OVS_PACKET_ATTR_PACKET
], len
);
511 skb_reset_mac_header(packet
);
512 eth
= eth_hdr(packet
);
514 /* Normally, setting the skb 'protocol' field would be handled by a
515 * call to eth_type_trans(), but it assumes there's a sending
516 * device, which we may not have. */
517 if (ntohs(eth
->h_proto
) >= ETH_P_802_3_MIN
)
518 packet
->protocol
= eth
->h_proto
;
520 packet
->protocol
= htons(ETH_P_802_2
);
522 /* Build an sw_flow for sending this packet. */
523 flow
= ovs_flow_alloc(false);
528 err
= ovs_flow_extract(packet
, -1, &flow
->key
);
532 err
= ovs_nla_get_flow_metadata(flow
, a
[OVS_PACKET_ATTR_KEY
]);
535 acts
= ovs_nla_alloc_flow_actions(nla_len(a
[OVS_PACKET_ATTR_ACTIONS
]));
540 err
= ovs_nla_copy_actions(a
[OVS_PACKET_ATTR_ACTIONS
],
541 &flow
->key
, 0, &acts
);
542 rcu_assign_pointer(flow
->sf_acts
, acts
);
546 OVS_CB(packet
)->flow
= flow
;
547 OVS_CB(packet
)->pkt_key
= &flow
->key
;
548 packet
->priority
= flow
->key
.phy
.priority
;
549 packet
->mark
= flow
->key
.phy
.skb_mark
;
552 dp
= get_dp(sock_net(skb
->sk
), ovs_header
->dp_ifindex
);
558 err
= ovs_execute_actions(dp
, packet
);
562 ovs_flow_free(flow
, false);
568 ovs_flow_free(flow
, false);
575 static const struct nla_policy packet_policy
[OVS_PACKET_ATTR_MAX
+ 1] = {
576 [OVS_PACKET_ATTR_PACKET
] = { .len
= ETH_HLEN
},
577 [OVS_PACKET_ATTR_KEY
] = { .type
= NLA_NESTED
},
578 [OVS_PACKET_ATTR_ACTIONS
] = { .type
= NLA_NESTED
},
581 static const struct genl_ops dp_packet_genl_ops
[] = {
582 { .cmd
= OVS_PACKET_CMD_EXECUTE
,
583 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
584 .policy
= packet_policy
,
585 .doit
= ovs_packet_cmd_execute
589 static void get_dp_stats(struct datapath
*dp
, struct ovs_dp_stats
*stats
,
590 struct ovs_dp_megaflow_stats
*mega_stats
)
594 memset(mega_stats
, 0, sizeof(*mega_stats
));
596 stats
->n_flows
= ovs_flow_tbl_count(&dp
->table
);
597 mega_stats
->n_masks
= ovs_flow_tbl_num_masks(&dp
->table
);
599 stats
->n_hit
= stats
->n_missed
= stats
->n_lost
= 0;
601 for_each_possible_cpu(i
) {
602 const struct dp_stats_percpu
*percpu_stats
;
603 struct dp_stats_percpu local_stats
;
606 percpu_stats
= per_cpu_ptr(dp
->stats_percpu
, i
);
609 start
= u64_stats_fetch_begin_bh(&percpu_stats
->sync
);
610 local_stats
= *percpu_stats
;
611 } while (u64_stats_fetch_retry_bh(&percpu_stats
->sync
, start
));
613 stats
->n_hit
+= local_stats
.n_hit
;
614 stats
->n_missed
+= local_stats
.n_missed
;
615 stats
->n_lost
+= local_stats
.n_lost
;
616 mega_stats
->n_mask_hit
+= local_stats
.n_mask_hit
;
620 static const struct nla_policy flow_policy
[OVS_FLOW_ATTR_MAX
+ 1] = {
621 [OVS_FLOW_ATTR_KEY
] = { .type
= NLA_NESTED
},
622 [OVS_FLOW_ATTR_ACTIONS
] = { .type
= NLA_NESTED
},
623 [OVS_FLOW_ATTR_CLEAR
] = { .type
= NLA_FLAG
},
626 static struct genl_family dp_flow_genl_family
= {
627 .id
= GENL_ID_GENERATE
,
628 .hdrsize
= sizeof(struct ovs_header
),
629 .name
= OVS_FLOW_FAMILY
,
630 .version
= OVS_FLOW_VERSION
,
631 .maxattr
= OVS_FLOW_ATTR_MAX
,
633 .parallel_ops
= true,
636 static struct genl_multicast_group ovs_dp_flow_multicast_group
= {
637 .name
= OVS_FLOW_MCGROUP
640 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions
*acts
)
642 return NLMSG_ALIGN(sizeof(struct ovs_header
))
643 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
644 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_MASK */
645 + nla_total_size(sizeof(struct ovs_flow_stats
)) /* OVS_FLOW_ATTR_STATS */
646 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
647 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
648 + nla_total_size(acts
->actions_len
); /* OVS_FLOW_ATTR_ACTIONS */
651 /* Called with ovs_mutex. */
652 static int ovs_flow_cmd_fill_info(struct sw_flow
*flow
, struct datapath
*dp
,
653 struct sk_buff
*skb
, u32 portid
,
654 u32 seq
, u32 flags
, u8 cmd
)
656 const int skb_orig_len
= skb
->len
;
657 struct nlattr
*start
;
658 struct ovs_flow_stats stats
;
661 struct ovs_header
*ovs_header
;
665 ovs_header
= genlmsg_put(skb
, portid
, seq
, &dp_flow_genl_family
, flags
, cmd
);
669 ovs_header
->dp_ifindex
= get_dpifindex(dp
);
672 nla
= nla_nest_start(skb
, OVS_FLOW_ATTR_KEY
);
674 goto nla_put_failure
;
676 err
= ovs_nla_put_flow(&flow
->unmasked_key
, &flow
->unmasked_key
, skb
);
679 nla_nest_end(skb
, nla
);
681 nla
= nla_nest_start(skb
, OVS_FLOW_ATTR_MASK
);
683 goto nla_put_failure
;
685 err
= ovs_nla_put_flow(&flow
->key
, &flow
->mask
->key
, skb
);
689 nla_nest_end(skb
, nla
);
691 ovs_flow_stats_get(flow
, &stats
, &used
, &tcp_flags
);
693 nla_put_u64(skb
, OVS_FLOW_ATTR_USED
, ovs_flow_used_time(used
)))
694 goto nla_put_failure
;
696 if (stats
.n_packets
&&
697 nla_put(skb
, OVS_FLOW_ATTR_STATS
, sizeof(struct ovs_flow_stats
), &stats
))
698 goto nla_put_failure
;
700 if ((u8
)ntohs(tcp_flags
) &&
701 nla_put_u8(skb
, OVS_FLOW_ATTR_TCP_FLAGS
, (u8
)ntohs(tcp_flags
)))
702 goto nla_put_failure
;
704 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
705 * this is the first flow to be dumped into 'skb'. This is unusual for
706 * Netlink but individual action lists can be longer than
707 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
708 * The userspace caller can always fetch the actions separately if it
709 * really wants them. (Most userspace callers in fact don't care.)
711 * This can only fail for dump operations because the skb is always
712 * properly sized for single flows.
714 start
= nla_nest_start(skb
, OVS_FLOW_ATTR_ACTIONS
);
716 const struct sw_flow_actions
*sf_acts
;
718 sf_acts
= rcu_dereference_ovsl(flow
->sf_acts
);
720 err
= ovs_nla_put_actions(sf_acts
->actions
,
721 sf_acts
->actions_len
, skb
);
723 nla_nest_end(skb
, start
);
728 nla_nest_cancel(skb
, start
);
730 } else if (skb_orig_len
)
731 goto nla_put_failure
;
733 return genlmsg_end(skb
, ovs_header
);
738 genlmsg_cancel(skb
, ovs_header
);
742 static struct sk_buff
*ovs_flow_cmd_alloc_info(struct sw_flow
*flow
,
743 struct genl_info
*info
)
747 len
= ovs_flow_cmd_msg_size(ovsl_dereference(flow
->sf_acts
));
749 return genlmsg_new_unicast(len
, info
, GFP_KERNEL
);
752 static struct sk_buff
*ovs_flow_cmd_build_info(struct sw_flow
*flow
,
754 struct genl_info
*info
,
760 skb
= ovs_flow_cmd_alloc_info(flow
, info
);
762 return ERR_PTR(-ENOMEM
);
764 retval
= ovs_flow_cmd_fill_info(flow
, dp
, skb
, info
->snd_portid
,
765 info
->snd_seq
, 0, cmd
);
770 static int ovs_flow_cmd_new_or_set(struct sk_buff
*skb
, struct genl_info
*info
)
772 struct nlattr
**a
= info
->attrs
;
773 struct ovs_header
*ovs_header
= info
->userhdr
;
774 struct sw_flow_key key
, masked_key
;
775 struct sw_flow
*flow
= NULL
;
776 struct sw_flow_mask mask
;
777 struct sk_buff
*reply
;
779 struct sw_flow_actions
*acts
= NULL
;
780 struct sw_flow_match match
;
786 if (!a
[OVS_FLOW_ATTR_KEY
])
789 ovs_match_init(&match
, &key
, &mask
);
790 error
= ovs_nla_get_match(&match
, &exact_5tuple
,
791 a
[OVS_FLOW_ATTR_KEY
], a
[OVS_FLOW_ATTR_MASK
]);
795 /* Validate actions. */
796 if (a
[OVS_FLOW_ATTR_ACTIONS
]) {
797 acts
= ovs_nla_alloc_flow_actions(nla_len(a
[OVS_FLOW_ATTR_ACTIONS
]));
798 error
= PTR_ERR(acts
);
802 ovs_flow_mask_key(&masked_key
, &key
, &mask
);
803 error
= ovs_nla_copy_actions(a
[OVS_FLOW_ATTR_ACTIONS
],
804 &masked_key
, 0, &acts
);
806 OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
809 } else if (info
->genlhdr
->cmd
== OVS_FLOW_CMD_NEW
) {
815 dp
= get_dp(sock_net(skb
->sk
), ovs_header
->dp_ifindex
);
820 /* Check if this is a duplicate flow */
821 flow
= ovs_flow_tbl_lookup(&dp
->table
, &key
);
823 /* Bail out if we're not allowed to create a new flow. */
825 if (info
->genlhdr
->cmd
== OVS_FLOW_CMD_SET
)
829 flow
= ovs_flow_alloc(!exact_5tuple
);
831 error
= PTR_ERR(flow
);
835 flow
->key
= masked_key
;
836 flow
->unmasked_key
= key
;
837 rcu_assign_pointer(flow
->sf_acts
, acts
);
839 /* Put flow in bucket. */
840 error
= ovs_flow_tbl_insert(&dp
->table
, flow
, &mask
);
846 reply
= ovs_flow_cmd_build_info(flow
, dp
, info
, OVS_FLOW_CMD_NEW
);
848 /* We found a matching flow. */
849 struct sw_flow_actions
*old_acts
;
851 /* Bail out if we're not allowed to modify an existing flow.
852 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
853 * because Generic Netlink treats the latter as a dump
854 * request. We also accept NLM_F_EXCL in case that bug ever
858 if (info
->genlhdr
->cmd
== OVS_FLOW_CMD_NEW
&&
859 info
->nlhdr
->nlmsg_flags
& (NLM_F_CREATE
| NLM_F_EXCL
))
862 /* The unmasked key has to be the same for flow updates. */
863 if (!ovs_flow_cmp_unmasked_key(flow
, &match
))
866 /* Update actions. */
867 old_acts
= ovsl_dereference(flow
->sf_acts
);
868 rcu_assign_pointer(flow
->sf_acts
, acts
);
869 ovs_nla_free_flow_actions(old_acts
);
871 reply
= ovs_flow_cmd_build_info(flow
, dp
, info
, OVS_FLOW_CMD_NEW
);
874 if (a
[OVS_FLOW_ATTR_CLEAR
])
875 ovs_flow_stats_clear(flow
);
880 ovs_notify(&dp_flow_genl_family
, reply
, info
);
882 genl_set_err(&dp_flow_genl_family
, sock_net(skb
->sk
), 0,
887 ovs_flow_free(flow
, false);
896 static int ovs_flow_cmd_get(struct sk_buff
*skb
, struct genl_info
*info
)
898 struct nlattr
**a
= info
->attrs
;
899 struct ovs_header
*ovs_header
= info
->userhdr
;
900 struct sw_flow_key key
;
901 struct sk_buff
*reply
;
902 struct sw_flow
*flow
;
904 struct sw_flow_match match
;
907 if (!a
[OVS_FLOW_ATTR_KEY
]) {
908 OVS_NLERR("Flow get message rejected, Key attribute missing.\n");
912 ovs_match_init(&match
, &key
, NULL
);
913 err
= ovs_nla_get_match(&match
, NULL
, a
[OVS_FLOW_ATTR_KEY
], NULL
);
918 dp
= get_dp(sock_net(skb
->sk
), ovs_header
->dp_ifindex
);
924 flow
= ovs_flow_tbl_lookup(&dp
->table
, &key
);
925 if (!flow
|| !ovs_flow_cmp_unmasked_key(flow
, &match
)) {
930 reply
= ovs_flow_cmd_build_info(flow
, dp
, info
, OVS_FLOW_CMD_NEW
);
932 err
= PTR_ERR(reply
);
937 return genlmsg_reply(reply
, info
);
943 static int ovs_flow_cmd_del(struct sk_buff
*skb
, struct genl_info
*info
)
945 struct nlattr
**a
= info
->attrs
;
946 struct ovs_header
*ovs_header
= info
->userhdr
;
947 struct sw_flow_key key
;
948 struct sk_buff
*reply
;
949 struct sw_flow
*flow
;
951 struct sw_flow_match match
;
955 dp
= get_dp(sock_net(skb
->sk
), ovs_header
->dp_ifindex
);
961 if (!a
[OVS_FLOW_ATTR_KEY
]) {
962 err
= ovs_flow_tbl_flush(&dp
->table
);
966 ovs_match_init(&match
, &key
, NULL
);
967 err
= ovs_nla_get_match(&match
, NULL
, a
[OVS_FLOW_ATTR_KEY
], NULL
);
971 flow
= ovs_flow_tbl_lookup(&dp
->table
, &key
);
972 if (!flow
|| !ovs_flow_cmp_unmasked_key(flow
, &match
)) {
977 reply
= ovs_flow_cmd_alloc_info(flow
, info
);
983 ovs_flow_tbl_remove(&dp
->table
, flow
);
985 err
= ovs_flow_cmd_fill_info(flow
, dp
, reply
, info
->snd_portid
,
986 info
->snd_seq
, 0, OVS_FLOW_CMD_DEL
);
989 ovs_flow_free(flow
, true);
992 ovs_notify(&dp_flow_genl_family
, reply
, info
);
999 static int ovs_flow_cmd_dump(struct sk_buff
*skb
, struct netlink_callback
*cb
)
1001 struct ovs_header
*ovs_header
= genlmsg_data(nlmsg_data(cb
->nlh
));
1002 struct table_instance
*ti
;
1003 struct datapath
*dp
;
1006 dp
= get_dp(sock_net(skb
->sk
), ovs_header
->dp_ifindex
);
1012 ti
= rcu_dereference(dp
->table
.ti
);
1014 struct sw_flow
*flow
;
1017 bucket
= cb
->args
[0];
1019 flow
= ovs_flow_tbl_dump_next(ti
, &bucket
, &obj
);
1023 if (ovs_flow_cmd_fill_info(flow
, dp
, skb
,
1024 NETLINK_CB(cb
->skb
).portid
,
1025 cb
->nlh
->nlmsg_seq
, NLM_F_MULTI
,
1026 OVS_FLOW_CMD_NEW
) < 0)
1029 cb
->args
[0] = bucket
;
1036 static const struct genl_ops dp_flow_genl_ops
[] = {
1037 { .cmd
= OVS_FLOW_CMD_NEW
,
1038 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1039 .policy
= flow_policy
,
1040 .doit
= ovs_flow_cmd_new_or_set
1042 { .cmd
= OVS_FLOW_CMD_DEL
,
1043 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1044 .policy
= flow_policy
,
1045 .doit
= ovs_flow_cmd_del
1047 { .cmd
= OVS_FLOW_CMD_GET
,
1048 .flags
= 0, /* OK for unprivileged users. */
1049 .policy
= flow_policy
,
1050 .doit
= ovs_flow_cmd_get
,
1051 .dumpit
= ovs_flow_cmd_dump
1053 { .cmd
= OVS_FLOW_CMD_SET
,
1054 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1055 .policy
= flow_policy
,
1056 .doit
= ovs_flow_cmd_new_or_set
,
1060 static const struct nla_policy datapath_policy
[OVS_DP_ATTR_MAX
+ 1] = {
1061 [OVS_DP_ATTR_NAME
] = { .type
= NLA_NUL_STRING
, .len
= IFNAMSIZ
- 1 },
1062 [OVS_DP_ATTR_UPCALL_PID
] = { .type
= NLA_U32
},
1063 [OVS_DP_ATTR_USER_FEATURES
] = { .type
= NLA_U32
},
1066 static struct genl_family dp_datapath_genl_family
= {
1067 .id
= GENL_ID_GENERATE
,
1068 .hdrsize
= sizeof(struct ovs_header
),
1069 .name
= OVS_DATAPATH_FAMILY
,
1070 .version
= OVS_DATAPATH_VERSION
,
1071 .maxattr
= OVS_DP_ATTR_MAX
,
1073 .parallel_ops
= true,
1076 static struct genl_multicast_group ovs_dp_datapath_multicast_group
= {
1077 .name
= OVS_DATAPATH_MCGROUP
1080 static size_t ovs_dp_cmd_msg_size(void)
1082 size_t msgsize
= NLMSG_ALIGN(sizeof(struct ovs_header
));
1084 msgsize
+= nla_total_size(IFNAMSIZ
);
1085 msgsize
+= nla_total_size(sizeof(struct ovs_dp_stats
));
1086 msgsize
+= nla_total_size(sizeof(struct ovs_dp_megaflow_stats
));
1087 msgsize
+= nla_total_size(sizeof(u32
)); /* OVS_DP_ATTR_USER_FEATURES */
1092 static int ovs_dp_cmd_fill_info(struct datapath
*dp
, struct sk_buff
*skb
,
1093 u32 portid
, u32 seq
, u32 flags
, u8 cmd
)
1095 struct ovs_header
*ovs_header
;
1096 struct ovs_dp_stats dp_stats
;
1097 struct ovs_dp_megaflow_stats dp_megaflow_stats
;
1100 ovs_header
= genlmsg_put(skb
, portid
, seq
, &dp_datapath_genl_family
,
1105 ovs_header
->dp_ifindex
= get_dpifindex(dp
);
1108 err
= nla_put_string(skb
, OVS_DP_ATTR_NAME
, ovs_dp_name(dp
));
1111 goto nla_put_failure
;
1113 get_dp_stats(dp
, &dp_stats
, &dp_megaflow_stats
);
1114 if (nla_put(skb
, OVS_DP_ATTR_STATS
, sizeof(struct ovs_dp_stats
),
1116 goto nla_put_failure
;
1118 if (nla_put(skb
, OVS_DP_ATTR_MEGAFLOW_STATS
,
1119 sizeof(struct ovs_dp_megaflow_stats
),
1120 &dp_megaflow_stats
))
1121 goto nla_put_failure
;
1123 if (nla_put_u32(skb
, OVS_DP_ATTR_USER_FEATURES
, dp
->user_features
))
1124 goto nla_put_failure
;
1126 return genlmsg_end(skb
, ovs_header
);
1129 genlmsg_cancel(skb
, ovs_header
);
1134 static struct sk_buff
*ovs_dp_cmd_build_info(struct datapath
*dp
,
1135 struct genl_info
*info
, u8 cmd
)
1137 struct sk_buff
*skb
;
1140 skb
= genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info
, GFP_KERNEL
);
1142 return ERR_PTR(-ENOMEM
);
1144 retval
= ovs_dp_cmd_fill_info(dp
, skb
, info
->snd_portid
, info
->snd_seq
, 0, cmd
);
1147 return ERR_PTR(retval
);
1152 /* Called with ovs_mutex. */
1153 static struct datapath
*lookup_datapath(struct net
*net
,
1154 struct ovs_header
*ovs_header
,
1155 struct nlattr
*a
[OVS_DP_ATTR_MAX
+ 1])
1157 struct datapath
*dp
;
1159 if (!a
[OVS_DP_ATTR_NAME
])
1160 dp
= get_dp(net
, ovs_header
->dp_ifindex
);
1162 struct vport
*vport
;
1165 vport
= ovs_vport_locate(net
, nla_data(a
[OVS_DP_ATTR_NAME
]));
1166 dp
= vport
&& vport
->port_no
== OVSP_LOCAL
? vport
->dp
: NULL
;
1169 return dp
? dp
: ERR_PTR(-ENODEV
);
1172 static void ovs_dp_reset_user_features(struct sk_buff
*skb
, struct genl_info
*info
)
1174 struct datapath
*dp
;
1176 dp
= lookup_datapath(sock_net(skb
->sk
), info
->userhdr
, info
->attrs
);
1180 WARN(dp
->user_features
, "Dropping previously announced user features\n");
1181 dp
->user_features
= 0;
1184 static void ovs_dp_change(struct datapath
*dp
, struct nlattr
**a
)
1186 if (a
[OVS_DP_ATTR_USER_FEATURES
])
1187 dp
->user_features
= nla_get_u32(a
[OVS_DP_ATTR_USER_FEATURES
]);
1190 static int ovs_dp_cmd_new(struct sk_buff
*skb
, struct genl_info
*info
)
1192 struct nlattr
**a
= info
->attrs
;
1193 struct vport_parms parms
;
1194 struct sk_buff
*reply
;
1195 struct datapath
*dp
;
1196 struct vport
*vport
;
1197 struct ovs_net
*ovs_net
;
1201 if (!a
[OVS_DP_ATTR_NAME
] || !a
[OVS_DP_ATTR_UPCALL_PID
])
1207 dp
= kzalloc(sizeof(*dp
), GFP_KERNEL
);
1209 goto err_unlock_ovs
;
1211 ovs_dp_set_net(dp
, hold_net(sock_net(skb
->sk
)));
1213 /* Allocate table. */
1214 err
= ovs_flow_tbl_init(&dp
->table
);
1218 dp
->stats_percpu
= alloc_percpu(struct dp_stats_percpu
);
1219 if (!dp
->stats_percpu
) {
1221 goto err_destroy_table
;
1224 for_each_possible_cpu(i
) {
1225 struct dp_stats_percpu
*dpath_stats
;
1226 dpath_stats
= per_cpu_ptr(dp
->stats_percpu
, i
);
1227 u64_stats_init(&dpath_stats
->sync
);
1230 dp
->ports
= kmalloc(DP_VPORT_HASH_BUCKETS
* sizeof(struct hlist_head
),
1234 goto err_destroy_percpu
;
1237 for (i
= 0; i
< DP_VPORT_HASH_BUCKETS
; i
++)
1238 INIT_HLIST_HEAD(&dp
->ports
[i
]);
1240 /* Set up our datapath device. */
1241 parms
.name
= nla_data(a
[OVS_DP_ATTR_NAME
]);
1242 parms
.type
= OVS_VPORT_TYPE_INTERNAL
;
1243 parms
.options
= NULL
;
1245 parms
.port_no
= OVSP_LOCAL
;
1246 parms
.upcall_portid
= nla_get_u32(a
[OVS_DP_ATTR_UPCALL_PID
]);
1248 ovs_dp_change(dp
, a
);
1250 vport
= new_vport(&parms
);
1251 if (IS_ERR(vport
)) {
1252 err
= PTR_ERR(vport
);
1256 if (err
== -EEXIST
) {
1257 /* An outdated user space instance that does not understand
1258 * the concept of user_features has attempted to create a new
1259 * datapath and is likely to reuse it. Drop all user features.
1261 if (info
->genlhdr
->version
< OVS_DP_VER_FEATURES
)
1262 ovs_dp_reset_user_features(skb
, info
);
1265 goto err_destroy_ports_array
;
1268 reply
= ovs_dp_cmd_build_info(dp
, info
, OVS_DP_CMD_NEW
);
1269 err
= PTR_ERR(reply
);
1271 goto err_destroy_local_port
;
1273 ovs_net
= net_generic(ovs_dp_get_net(dp
), ovs_net_id
);
1274 list_add_tail_rcu(&dp
->list_node
, &ovs_net
->dps
);
1278 ovs_notify(&dp_datapath_genl_family
, reply
, info
);
1281 err_destroy_local_port
:
1282 ovs_dp_detach_port(ovs_vport_ovsl(dp
, OVSP_LOCAL
));
1283 err_destroy_ports_array
:
1286 free_percpu(dp
->stats_percpu
);
1288 ovs_flow_tbl_destroy(&dp
->table
, false);
1290 release_net(ovs_dp_get_net(dp
));
1298 /* Called with ovs_mutex. */
1299 static void __dp_destroy(struct datapath
*dp
)
1303 for (i
= 0; i
< DP_VPORT_HASH_BUCKETS
; i
++) {
1304 struct vport
*vport
;
1305 struct hlist_node
*n
;
1307 hlist_for_each_entry_safe(vport
, n
, &dp
->ports
[i
], dp_hash_node
)
1308 if (vport
->port_no
!= OVSP_LOCAL
)
1309 ovs_dp_detach_port(vport
);
1312 list_del_rcu(&dp
->list_node
);
1314 /* OVSP_LOCAL is datapath internal port. We need to make sure that
1315 * all ports in datapath are destroyed first before freeing datapath.
1317 ovs_dp_detach_port(ovs_vport_ovsl(dp
, OVSP_LOCAL
));
1319 /* RCU destroy the flow table */
1320 ovs_flow_tbl_destroy(&dp
->table
, true);
1322 call_rcu(&dp
->rcu
, destroy_dp_rcu
);
1325 static int ovs_dp_cmd_del(struct sk_buff
*skb
, struct genl_info
*info
)
1327 struct sk_buff
*reply
;
1328 struct datapath
*dp
;
1332 dp
= lookup_datapath(sock_net(skb
->sk
), info
->userhdr
, info
->attrs
);
1337 reply
= ovs_dp_cmd_build_info(dp
, info
, OVS_DP_CMD_DEL
);
1338 err
= PTR_ERR(reply
);
1345 ovs_notify(&dp_datapath_genl_family
, reply
, info
);
1353 static int ovs_dp_cmd_set(struct sk_buff
*skb
, struct genl_info
*info
)
1355 struct sk_buff
*reply
;
1356 struct datapath
*dp
;
1360 dp
= lookup_datapath(sock_net(skb
->sk
), info
->userhdr
, info
->attrs
);
1365 ovs_dp_change(dp
, info
->attrs
);
1367 reply
= ovs_dp_cmd_build_info(dp
, info
, OVS_DP_CMD_NEW
);
1368 if (IS_ERR(reply
)) {
1369 err
= PTR_ERR(reply
);
1370 genl_set_err(&dp_datapath_genl_family
, sock_net(skb
->sk
), 0,
1377 ovs_notify(&dp_datapath_genl_family
, reply
, info
);
1385 static int ovs_dp_cmd_get(struct sk_buff
*skb
, struct genl_info
*info
)
1387 struct sk_buff
*reply
;
1388 struct datapath
*dp
;
1392 dp
= lookup_datapath(sock_net(skb
->sk
), info
->userhdr
, info
->attrs
);
1398 reply
= ovs_dp_cmd_build_info(dp
, info
, OVS_DP_CMD_NEW
);
1399 if (IS_ERR(reply
)) {
1400 err
= PTR_ERR(reply
);
1405 return genlmsg_reply(reply
, info
);
1412 static int ovs_dp_cmd_dump(struct sk_buff
*skb
, struct netlink_callback
*cb
)
1414 struct ovs_net
*ovs_net
= net_generic(sock_net(skb
->sk
), ovs_net_id
);
1415 struct datapath
*dp
;
1416 int skip
= cb
->args
[0];
1420 list_for_each_entry_rcu(dp
, &ovs_net
->dps
, list_node
) {
1422 ovs_dp_cmd_fill_info(dp
, skb
, NETLINK_CB(cb
->skb
).portid
,
1423 cb
->nlh
->nlmsg_seq
, NLM_F_MULTI
,
1424 OVS_DP_CMD_NEW
) < 0)
1435 static const struct genl_ops dp_datapath_genl_ops
[] = {
1436 { .cmd
= OVS_DP_CMD_NEW
,
1437 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1438 .policy
= datapath_policy
,
1439 .doit
= ovs_dp_cmd_new
1441 { .cmd
= OVS_DP_CMD_DEL
,
1442 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1443 .policy
= datapath_policy
,
1444 .doit
= ovs_dp_cmd_del
1446 { .cmd
= OVS_DP_CMD_GET
,
1447 .flags
= 0, /* OK for unprivileged users. */
1448 .policy
= datapath_policy
,
1449 .doit
= ovs_dp_cmd_get
,
1450 .dumpit
= ovs_dp_cmd_dump
1452 { .cmd
= OVS_DP_CMD_SET
,
1453 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1454 .policy
= datapath_policy
,
1455 .doit
= ovs_dp_cmd_set
,
1459 static const struct nla_policy vport_policy
[OVS_VPORT_ATTR_MAX
+ 1] = {
1460 [OVS_VPORT_ATTR_NAME
] = { .type
= NLA_NUL_STRING
, .len
= IFNAMSIZ
- 1 },
1461 [OVS_VPORT_ATTR_STATS
] = { .len
= sizeof(struct ovs_vport_stats
) },
1462 [OVS_VPORT_ATTR_PORT_NO
] = { .type
= NLA_U32
},
1463 [OVS_VPORT_ATTR_TYPE
] = { .type
= NLA_U32
},
1464 [OVS_VPORT_ATTR_UPCALL_PID
] = { .type
= NLA_U32
},
1465 [OVS_VPORT_ATTR_OPTIONS
] = { .type
= NLA_NESTED
},
1468 struct genl_family dp_vport_genl_family
= {
1469 .id
= GENL_ID_GENERATE
,
1470 .hdrsize
= sizeof(struct ovs_header
),
1471 .name
= OVS_VPORT_FAMILY
,
1472 .version
= OVS_VPORT_VERSION
,
1473 .maxattr
= OVS_VPORT_ATTR_MAX
,
1475 .parallel_ops
= true,
1478 static struct genl_multicast_group ovs_dp_vport_multicast_group
= {
1479 .name
= OVS_VPORT_MCGROUP
1482 /* Called with ovs_mutex or RCU read lock. */
1483 static int ovs_vport_cmd_fill_info(struct vport
*vport
, struct sk_buff
*skb
,
1484 u32 portid
, u32 seq
, u32 flags
, u8 cmd
)
1486 struct ovs_header
*ovs_header
;
1487 struct ovs_vport_stats vport_stats
;
1490 ovs_header
= genlmsg_put(skb
, portid
, seq
, &dp_vport_genl_family
,
1495 ovs_header
->dp_ifindex
= get_dpifindex(vport
->dp
);
1497 if (nla_put_u32(skb
, OVS_VPORT_ATTR_PORT_NO
, vport
->port_no
) ||
1498 nla_put_u32(skb
, OVS_VPORT_ATTR_TYPE
, vport
->ops
->type
) ||
1499 nla_put_string(skb
, OVS_VPORT_ATTR_NAME
, vport
->ops
->get_name(vport
)) ||
1500 nla_put_u32(skb
, OVS_VPORT_ATTR_UPCALL_PID
, vport
->upcall_portid
))
1501 goto nla_put_failure
;
1503 ovs_vport_get_stats(vport
, &vport_stats
);
1504 if (nla_put(skb
, OVS_VPORT_ATTR_STATS
, sizeof(struct ovs_vport_stats
),
1506 goto nla_put_failure
;
1508 err
= ovs_vport_get_options(vport
, skb
);
1509 if (err
== -EMSGSIZE
)
1512 return genlmsg_end(skb
, ovs_header
);
1517 genlmsg_cancel(skb
, ovs_header
);
1521 /* Called with ovs_mutex or RCU read lock. */
1522 struct sk_buff
*ovs_vport_cmd_build_info(struct vport
*vport
, u32 portid
,
1525 struct sk_buff
*skb
;
1528 skb
= nlmsg_new(NLMSG_DEFAULT_SIZE
, GFP_ATOMIC
);
1530 return ERR_PTR(-ENOMEM
);
1532 retval
= ovs_vport_cmd_fill_info(vport
, skb
, portid
, seq
, 0, cmd
);
1538 /* Called with ovs_mutex or RCU read lock. */
1539 static struct vport
*lookup_vport(struct net
*net
,
1540 struct ovs_header
*ovs_header
,
1541 struct nlattr
*a
[OVS_VPORT_ATTR_MAX
+ 1])
1543 struct datapath
*dp
;
1544 struct vport
*vport
;
1546 if (a
[OVS_VPORT_ATTR_NAME
]) {
1547 vport
= ovs_vport_locate(net
, nla_data(a
[OVS_VPORT_ATTR_NAME
]));
1549 return ERR_PTR(-ENODEV
);
1550 if (ovs_header
->dp_ifindex
&&
1551 ovs_header
->dp_ifindex
!= get_dpifindex(vport
->dp
))
1552 return ERR_PTR(-ENODEV
);
1554 } else if (a
[OVS_VPORT_ATTR_PORT_NO
]) {
1555 u32 port_no
= nla_get_u32(a
[OVS_VPORT_ATTR_PORT_NO
]);
1557 if (port_no
>= DP_MAX_PORTS
)
1558 return ERR_PTR(-EFBIG
);
1560 dp
= get_dp(net
, ovs_header
->dp_ifindex
);
1562 return ERR_PTR(-ENODEV
);
1564 vport
= ovs_vport_ovsl_rcu(dp
, port_no
);
1566 return ERR_PTR(-ENODEV
);
1569 return ERR_PTR(-EINVAL
);
1572 static int ovs_vport_cmd_new(struct sk_buff
*skb
, struct genl_info
*info
)
1574 struct nlattr
**a
= info
->attrs
;
1575 struct ovs_header
*ovs_header
= info
->userhdr
;
1576 struct vport_parms parms
;
1577 struct sk_buff
*reply
;
1578 struct vport
*vport
;
1579 struct datapath
*dp
;
1584 if (!a
[OVS_VPORT_ATTR_NAME
] || !a
[OVS_VPORT_ATTR_TYPE
] ||
1585 !a
[OVS_VPORT_ATTR_UPCALL_PID
])
1589 dp
= get_dp(sock_net(skb
->sk
), ovs_header
->dp_ifindex
);
1594 if (a
[OVS_VPORT_ATTR_PORT_NO
]) {
1595 port_no
= nla_get_u32(a
[OVS_VPORT_ATTR_PORT_NO
]);
1598 if (port_no
>= DP_MAX_PORTS
)
1601 vport
= ovs_vport_ovsl(dp
, port_no
);
1606 for (port_no
= 1; ; port_no
++) {
1607 if (port_no
>= DP_MAX_PORTS
) {
1611 vport
= ovs_vport_ovsl(dp
, port_no
);
1617 parms
.name
= nla_data(a
[OVS_VPORT_ATTR_NAME
]);
1618 parms
.type
= nla_get_u32(a
[OVS_VPORT_ATTR_TYPE
]);
1619 parms
.options
= a
[OVS_VPORT_ATTR_OPTIONS
];
1621 parms
.port_no
= port_no
;
1622 parms
.upcall_portid
= nla_get_u32(a
[OVS_VPORT_ATTR_UPCALL_PID
]);
1624 vport
= new_vport(&parms
);
1625 err
= PTR_ERR(vport
);
1630 reply
= ovs_vport_cmd_build_info(vport
, info
->snd_portid
, info
->snd_seq
,
1632 if (IS_ERR(reply
)) {
1633 err
= PTR_ERR(reply
);
1634 ovs_dp_detach_port(vport
);
1638 ovs_notify(&dp_vport_genl_family
, reply
, info
);
1646 static int ovs_vport_cmd_set(struct sk_buff
*skb
, struct genl_info
*info
)
1648 struct nlattr
**a
= info
->attrs
;
1649 struct sk_buff
*reply
;
1650 struct vport
*vport
;
1654 vport
= lookup_vport(sock_net(skb
->sk
), info
->userhdr
, a
);
1655 err
= PTR_ERR(vport
);
1659 if (a
[OVS_VPORT_ATTR_TYPE
] &&
1660 nla_get_u32(a
[OVS_VPORT_ATTR_TYPE
]) != vport
->ops
->type
) {
1665 reply
= nlmsg_new(NLMSG_DEFAULT_SIZE
, GFP_KERNEL
);
1671 if (a
[OVS_VPORT_ATTR_OPTIONS
]) {
1672 err
= ovs_vport_set_options(vport
, a
[OVS_VPORT_ATTR_OPTIONS
]);
1677 if (a
[OVS_VPORT_ATTR_UPCALL_PID
])
1678 vport
->upcall_portid
= nla_get_u32(a
[OVS_VPORT_ATTR_UPCALL_PID
]);
1680 err
= ovs_vport_cmd_fill_info(vport
, reply
, info
->snd_portid
,
1681 info
->snd_seq
, 0, OVS_VPORT_CMD_NEW
);
1685 ovs_notify(&dp_vport_genl_family
, reply
, info
);
1695 static int ovs_vport_cmd_del(struct sk_buff
*skb
, struct genl_info
*info
)
1697 struct nlattr
**a
= info
->attrs
;
1698 struct sk_buff
*reply
;
1699 struct vport
*vport
;
1703 vport
= lookup_vport(sock_net(skb
->sk
), info
->userhdr
, a
);
1704 err
= PTR_ERR(vport
);
1708 if (vport
->port_no
== OVSP_LOCAL
) {
1713 reply
= ovs_vport_cmd_build_info(vport
, info
->snd_portid
,
1714 info
->snd_seq
, OVS_VPORT_CMD_DEL
);
1715 err
= PTR_ERR(reply
);
1720 ovs_dp_detach_port(vport
);
1722 ovs_notify(&dp_vport_genl_family
, reply
, info
);
1729 static int ovs_vport_cmd_get(struct sk_buff
*skb
, struct genl_info
*info
)
1731 struct nlattr
**a
= info
->attrs
;
1732 struct ovs_header
*ovs_header
= info
->userhdr
;
1733 struct sk_buff
*reply
;
1734 struct vport
*vport
;
1738 vport
= lookup_vport(sock_net(skb
->sk
), ovs_header
, a
);
1739 err
= PTR_ERR(vport
);
1743 reply
= ovs_vport_cmd_build_info(vport
, info
->snd_portid
,
1744 info
->snd_seq
, OVS_VPORT_CMD_NEW
);
1745 err
= PTR_ERR(reply
);
1751 return genlmsg_reply(reply
, info
);
1758 static int ovs_vport_cmd_dump(struct sk_buff
*skb
, struct netlink_callback
*cb
)
1760 struct ovs_header
*ovs_header
= genlmsg_data(nlmsg_data(cb
->nlh
));
1761 struct datapath
*dp
;
1762 int bucket
= cb
->args
[0], skip
= cb
->args
[1];
1765 dp
= get_dp(sock_net(skb
->sk
), ovs_header
->dp_ifindex
);
1770 for (i
= bucket
; i
< DP_VPORT_HASH_BUCKETS
; i
++) {
1771 struct vport
*vport
;
1774 hlist_for_each_entry_rcu(vport
, &dp
->ports
[i
], dp_hash_node
) {
1776 ovs_vport_cmd_fill_info(vport
, skb
,
1777 NETLINK_CB(cb
->skb
).portid
,
1780 OVS_VPORT_CMD_NEW
) < 0)
1796 static const struct genl_ops dp_vport_genl_ops
[] = {
1797 { .cmd
= OVS_VPORT_CMD_NEW
,
1798 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1799 .policy
= vport_policy
,
1800 .doit
= ovs_vport_cmd_new
1802 { .cmd
= OVS_VPORT_CMD_DEL
,
1803 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1804 .policy
= vport_policy
,
1805 .doit
= ovs_vport_cmd_del
1807 { .cmd
= OVS_VPORT_CMD_GET
,
1808 .flags
= 0, /* OK for unprivileged users. */
1809 .policy
= vport_policy
,
1810 .doit
= ovs_vport_cmd_get
,
1811 .dumpit
= ovs_vport_cmd_dump
1813 { .cmd
= OVS_VPORT_CMD_SET
,
1814 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1815 .policy
= vport_policy
,
1816 .doit
= ovs_vport_cmd_set
,
1820 struct genl_family_and_ops
{
1821 struct genl_family
*family
;
1822 const struct genl_ops
*ops
;
1824 const struct genl_multicast_group
*group
;
1827 static const struct genl_family_and_ops dp_genl_families
[] = {
1828 { &dp_datapath_genl_family
,
1829 dp_datapath_genl_ops
, ARRAY_SIZE(dp_datapath_genl_ops
),
1830 &ovs_dp_datapath_multicast_group
},
1831 { &dp_vport_genl_family
,
1832 dp_vport_genl_ops
, ARRAY_SIZE(dp_vport_genl_ops
),
1833 &ovs_dp_vport_multicast_group
},
1834 { &dp_flow_genl_family
,
1835 dp_flow_genl_ops
, ARRAY_SIZE(dp_flow_genl_ops
),
1836 &ovs_dp_flow_multicast_group
},
1837 { &dp_packet_genl_family
,
1838 dp_packet_genl_ops
, ARRAY_SIZE(dp_packet_genl_ops
),
1842 static void dp_unregister_genl(int n_families
)
1846 for (i
= 0; i
< n_families
; i
++)
1847 genl_unregister_family(dp_genl_families
[i
].family
);
1850 static int dp_register_genl(void)
1857 for (i
= 0; i
< ARRAY_SIZE(dp_genl_families
); i
++) {
1858 const struct genl_family_and_ops
*f
= &dp_genl_families
[i
];
1860 f
->family
->ops
= f
->ops
;
1861 f
->family
->n_ops
= f
->n_ops
;
1862 f
->family
->mcgrps
= f
->group
;
1863 f
->family
->n_mcgrps
= f
->group
? 1 : 0;
1864 err
= genl_register_family(f
->family
);
1873 dp_unregister_genl(n_registered
);
1877 static int __net_init
ovs_init_net(struct net
*net
)
1879 struct ovs_net
*ovs_net
= net_generic(net
, ovs_net_id
);
1881 INIT_LIST_HEAD(&ovs_net
->dps
);
1882 INIT_WORK(&ovs_net
->dp_notify_work
, ovs_dp_notify_wq
);
1886 static void __net_exit
ovs_exit_net(struct net
*net
)
1888 struct datapath
*dp
, *dp_next
;
1889 struct ovs_net
*ovs_net
= net_generic(net
, ovs_net_id
);
1892 list_for_each_entry_safe(dp
, dp_next
, &ovs_net
->dps
, list_node
)
1896 cancel_work_sync(&ovs_net
->dp_notify_work
);
1899 static struct pernet_operations ovs_net_ops
= {
1900 .init
= ovs_init_net
,
1901 .exit
= ovs_exit_net
,
1903 .size
= sizeof(struct ovs_net
),
1906 static int __init
dp_init(void)
1910 BUILD_BUG_ON(sizeof(struct ovs_skb_cb
) > FIELD_SIZEOF(struct sk_buff
, cb
));
1912 pr_info("Open vSwitch switching datapath\n");
1914 err
= ovs_flow_init();
1918 err
= ovs_vport_init();
1920 goto error_flow_exit
;
1922 err
= register_pernet_device(&ovs_net_ops
);
1924 goto error_vport_exit
;
1926 err
= register_netdevice_notifier(&ovs_dp_device_notifier
);
1928 goto error_netns_exit
;
1930 err
= dp_register_genl();
1932 goto error_unreg_notifier
;
1936 error_unreg_notifier
:
1937 unregister_netdevice_notifier(&ovs_dp_device_notifier
);
1939 unregister_pernet_device(&ovs_net_ops
);
1948 static void dp_cleanup(void)
1950 dp_unregister_genl(ARRAY_SIZE(dp_genl_families
));
1951 unregister_netdevice_notifier(&ovs_dp_device_notifier
);
1952 unregister_pernet_device(&ovs_net_ops
);
1958 module_init(dp_init
);
1959 module_exit(dp_cleanup
);
1961 MODULE_DESCRIPTION("Open vSwitch switching datapath");
1962 MODULE_LICENSE("GPL");