2 * Copyright (c) 2007-2014 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/openvswitch.h>
48 #include <linux/rculist.h>
49 #include <linux/dmi.h>
50 #include <net/genetlink.h>
51 #include <net/net_namespace.h>
52 #include <net/netns/generic.h>
56 #include "flow_table.h"
57 #include "flow_netlink.h"
58 #include "vport-internal_dev.h"
59 #include "vport-netdev.h"
61 int ovs_net_id __read_mostly
;
63 static struct genl_family dp_packet_genl_family
;
64 static struct genl_family dp_flow_genl_family
;
65 static struct genl_family dp_datapath_genl_family
;
67 static const struct genl_multicast_group ovs_dp_flow_multicast_group
= {
68 .name
= OVS_FLOW_MCGROUP
,
71 static const struct genl_multicast_group ovs_dp_datapath_multicast_group
= {
72 .name
= OVS_DATAPATH_MCGROUP
,
75 static const struct genl_multicast_group ovs_dp_vport_multicast_group
= {
76 .name
= OVS_VPORT_MCGROUP
,
79 /* Check if need to build a reply message.
80 * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
81 static bool ovs_must_notify(struct genl_family
*family
, struct genl_info
*info
,
84 return info
->nlhdr
->nlmsg_flags
& NLM_F_ECHO
||
85 genl_has_listeners(family
, genl_info_net(info
)->genl_sock
,
89 static void ovs_notify(struct genl_family
*family
,
90 struct sk_buff
*skb
, struct genl_info
*info
)
92 genl_notify(family
, skb
, genl_info_net(info
), info
->snd_portid
,
93 0, info
->nlhdr
, GFP_KERNEL
);
99 * All writes e.g. Writes to device state (add/remove datapath, port, set
100 * operations on vports, etc.), Writes to other state (flow table
101 * modifications, set miscellaneous datapath parameters, etc.) are protected
104 * Reads are protected by RCU.
106 * There are a few special cases (mostly stats) that have their own
107 * synchronization but they nest under all of above and don't interact with
110 * The RTNL lock nests inside ovs_mutex.
113 static DEFINE_MUTEX(ovs_mutex
);
117 mutex_lock(&ovs_mutex
);
120 void ovs_unlock(void)
122 mutex_unlock(&ovs_mutex
);
125 #ifdef CONFIG_LOCKDEP
126 int lockdep_ovsl_is_held(void)
129 return lockdep_is_held(&ovs_mutex
);
135 static struct vport
*new_vport(const struct vport_parms
*);
136 static int queue_gso_packets(struct datapath
*dp
, struct sk_buff
*,
137 const struct dp_upcall_info
*);
138 static int queue_userspace_packet(struct datapath
*dp
, struct sk_buff
*,
139 const struct dp_upcall_info
*);
141 /* Must be called with rcu_read_lock or ovs_mutex. */
142 static struct datapath
*get_dp(struct net
*net
, int dp_ifindex
)
144 struct datapath
*dp
= NULL
;
145 struct net_device
*dev
;
148 dev
= dev_get_by_index_rcu(net
, dp_ifindex
);
150 struct vport
*vport
= ovs_internal_dev_get_vport(dev
);
159 /* Must be called with rcu_read_lock or ovs_mutex. */
160 static const char *ovs_dp_name(const struct datapath
*dp
)
162 struct vport
*vport
= ovs_vport_ovsl_rcu(dp
, OVSP_LOCAL
);
163 return vport
->ops
->get_name(vport
);
166 static int get_dpifindex(struct datapath
*dp
)
173 local
= ovs_vport_rcu(dp
, OVSP_LOCAL
);
175 ifindex
= netdev_vport_priv(local
)->dev
->ifindex
;
184 static void destroy_dp_rcu(struct rcu_head
*rcu
)
186 struct datapath
*dp
= container_of(rcu
, struct datapath
, rcu
);
188 free_percpu(dp
->stats_percpu
);
189 release_net(ovs_dp_get_net(dp
));
194 static struct hlist_head
*vport_hash_bucket(const struct datapath
*dp
,
197 return &dp
->ports
[port_no
& (DP_VPORT_HASH_BUCKETS
- 1)];
200 /* Called with ovs_mutex or RCU read lock. */
201 struct vport
*ovs_lookup_vport(const struct datapath
*dp
, u16 port_no
)
204 struct hlist_head
*head
;
206 head
= vport_hash_bucket(dp
, port_no
);
207 hlist_for_each_entry_rcu(vport
, head
, dp_hash_node
) {
208 if (vport
->port_no
== port_no
)
214 /* Called with ovs_mutex. */
215 static struct vport
*new_vport(const struct vport_parms
*parms
)
219 vport
= ovs_vport_add(parms
);
220 if (!IS_ERR(vport
)) {
221 struct datapath
*dp
= parms
->dp
;
222 struct hlist_head
*head
= vport_hash_bucket(dp
, vport
->port_no
);
224 hlist_add_head_rcu(&vport
->dp_hash_node
, head
);
229 void ovs_dp_detach_port(struct vport
*p
)
233 /* First drop references to device. */
234 hlist_del_rcu(&p
->dp_hash_node
);
236 /* Then destroy it. */
240 /* Must be called with rcu_read_lock. */
241 void ovs_dp_process_received_packet(struct vport
*p
, struct sk_buff
*skb
)
243 struct datapath
*dp
= p
->dp
;
244 struct sw_flow
*flow
;
245 struct dp_stats_percpu
*stats
;
246 struct sw_flow_key key
;
251 stats
= this_cpu_ptr(dp
->stats_percpu
);
253 /* Extract flow from 'skb' into 'key'. */
254 error
= ovs_flow_extract(skb
, p
->port_no
, &key
);
255 if (unlikely(error
)) {
261 flow
= ovs_flow_tbl_lookup_stats(&dp
->table
, &key
, &n_mask_hit
);
262 if (unlikely(!flow
)) {
263 struct dp_upcall_info upcall
;
265 upcall
.cmd
= OVS_PACKET_CMD_MISS
;
267 upcall
.userdata
= NULL
;
268 upcall
.portid
= ovs_vport_find_upcall_portid(p
, skb
);
269 error
= ovs_dp_upcall(dp
, skb
, &upcall
);
274 stats_counter
= &stats
->n_missed
;
278 OVS_CB(skb
)->flow
= flow
;
279 OVS_CB(skb
)->pkt_key
= &key
;
281 ovs_flow_stats_update(OVS_CB(skb
)->flow
, key
.tp
.flags
, skb
);
282 ovs_execute_actions(dp
, skb
);
283 stats_counter
= &stats
->n_hit
;
286 /* Update datapath statistics. */
287 u64_stats_update_begin(&stats
->syncp
);
289 stats
->n_mask_hit
+= n_mask_hit
;
290 u64_stats_update_end(&stats
->syncp
);
293 int ovs_dp_upcall(struct datapath
*dp
, struct sk_buff
*skb
,
294 const struct dp_upcall_info
*upcall_info
)
296 struct dp_stats_percpu
*stats
;
299 if (upcall_info
->portid
== 0) {
304 if (!skb_is_gso(skb
))
305 err
= queue_userspace_packet(dp
, skb
, upcall_info
);
307 err
= queue_gso_packets(dp
, skb
, upcall_info
);
314 stats
= this_cpu_ptr(dp
->stats_percpu
);
316 u64_stats_update_begin(&stats
->syncp
);
318 u64_stats_update_end(&stats
->syncp
);
323 static int queue_gso_packets(struct datapath
*dp
, struct sk_buff
*skb
,
324 const struct dp_upcall_info
*upcall_info
)
326 unsigned short gso_type
= skb_shinfo(skb
)->gso_type
;
327 struct dp_upcall_info later_info
;
328 struct sw_flow_key later_key
;
329 struct sk_buff
*segs
, *nskb
;
332 segs
= __skb_gso_segment(skb
, NETIF_F_SG
, false);
334 return PTR_ERR(segs
);
336 /* Queue all of the segments. */
339 err
= queue_userspace_packet(dp
, skb
, upcall_info
);
343 if (skb
== segs
&& gso_type
& SKB_GSO_UDP
) {
344 /* The initial flow key extracted by ovs_flow_extract()
345 * in this case is for a first fragment, so we need to
346 * properly mark later fragments.
348 later_key
= *upcall_info
->key
;
349 later_key
.ip
.frag
= OVS_FRAG_TYPE_LATER
;
351 later_info
= *upcall_info
;
352 later_info
.key
= &later_key
;
353 upcall_info
= &later_info
;
355 } while ((skb
= skb
->next
));
357 /* Free all of the segments. */
365 } while ((skb
= nskb
));
369 static size_t key_attr_size(void)
371 return nla_total_size(4) /* OVS_KEY_ATTR_PRIORITY */
372 + nla_total_size(0) /* OVS_KEY_ATTR_TUNNEL */
373 + nla_total_size(8) /* OVS_TUNNEL_KEY_ATTR_ID */
374 + nla_total_size(4) /* OVS_TUNNEL_KEY_ATTR_IPV4_SRC */
375 + nla_total_size(4) /* OVS_TUNNEL_KEY_ATTR_IPV4_DST */
376 + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TOS */
377 + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TTL */
378 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
379 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_CSUM */
380 + nla_total_size(4) /* OVS_KEY_ATTR_IN_PORT */
381 + nla_total_size(4) /* OVS_KEY_ATTR_SKB_MARK */
382 + nla_total_size(12) /* OVS_KEY_ATTR_ETHERNET */
383 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
384 + nla_total_size(4) /* OVS_KEY_ATTR_8021Q */
385 + nla_total_size(0) /* OVS_KEY_ATTR_ENCAP */
386 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
387 + nla_total_size(40) /* OVS_KEY_ATTR_IPV6 */
388 + nla_total_size(2) /* OVS_KEY_ATTR_ICMPV6 */
389 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
392 static size_t upcall_msg_size(const struct nlattr
*userdata
,
395 size_t size
= NLMSG_ALIGN(sizeof(struct ovs_header
))
396 + nla_total_size(hdrlen
) /* OVS_PACKET_ATTR_PACKET */
397 + nla_total_size(key_attr_size()); /* OVS_PACKET_ATTR_KEY */
399 /* OVS_PACKET_ATTR_USERDATA */
401 size
+= NLA_ALIGN(userdata
->nla_len
);
406 static int queue_userspace_packet(struct datapath
*dp
, struct sk_buff
*skb
,
407 const struct dp_upcall_info
*upcall_info
)
409 struct ovs_header
*upcall
;
410 struct sk_buff
*nskb
= NULL
;
411 struct sk_buff
*user_skb
= NULL
; /* to be queued to userspace */
413 struct genl_info info
= {
414 .dst_sk
= ovs_dp_get_net(dp
)->genl_sock
,
415 .snd_portid
= upcall_info
->portid
,
421 dp_ifindex
= get_dpifindex(dp
);
425 if (vlan_tx_tag_present(skb
)) {
426 nskb
= skb_clone(skb
, GFP_ATOMIC
);
430 nskb
= __vlan_put_tag(nskb
, nskb
->vlan_proto
, vlan_tx_tag_get(nskb
));
438 if (nla_attr_size(skb
->len
) > USHRT_MAX
) {
443 /* Complete checksum if needed */
444 if (skb
->ip_summed
== CHECKSUM_PARTIAL
&&
445 (err
= skb_checksum_help(skb
)))
448 /* Older versions of OVS user space enforce alignment of the last
449 * Netlink attribute to NLA_ALIGNTO which would require extensive
450 * padding logic. Only perform zerocopy if padding is not required.
452 if (dp
->user_features
& OVS_DP_F_UNALIGNED
)
453 hlen
= skb_zerocopy_headlen(skb
);
457 len
= upcall_msg_size(upcall_info
->userdata
, hlen
);
458 user_skb
= genlmsg_new_unicast(len
, &info
, GFP_ATOMIC
);
464 upcall
= genlmsg_put(user_skb
, 0, 0, &dp_packet_genl_family
,
465 0, upcall_info
->cmd
);
466 upcall
->dp_ifindex
= dp_ifindex
;
468 nla
= nla_nest_start(user_skb
, OVS_PACKET_ATTR_KEY
);
469 err
= ovs_nla_put_flow(upcall_info
->key
, upcall_info
->key
, user_skb
);
471 nla_nest_end(user_skb
, nla
);
473 if (upcall_info
->userdata
)
474 __nla_put(user_skb
, OVS_PACKET_ATTR_USERDATA
,
475 nla_len(upcall_info
->userdata
),
476 nla_data(upcall_info
->userdata
));
478 /* Only reserve room for attribute header, packet data is added
479 * in skb_zerocopy() */
480 if (!(nla
= nla_reserve(user_skb
, OVS_PACKET_ATTR_PACKET
, 0))) {
484 nla
->nla_len
= nla_attr_size(skb
->len
);
486 err
= skb_zerocopy(user_skb
, skb
, skb
->len
, hlen
);
490 /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
491 if (!(dp
->user_features
& OVS_DP_F_UNALIGNED
)) {
492 size_t plen
= NLA_ALIGN(user_skb
->len
) - user_skb
->len
;
495 memset(skb_put(user_skb
, plen
), 0, plen
);
498 ((struct nlmsghdr
*) user_skb
->data
)->nlmsg_len
= user_skb
->len
;
500 err
= genlmsg_unicast(ovs_dp_get_net(dp
), user_skb
, upcall_info
->portid
);
510 static int ovs_packet_cmd_execute(struct sk_buff
*skb
, struct genl_info
*info
)
512 struct ovs_header
*ovs_header
= info
->userhdr
;
513 struct nlattr
**a
= info
->attrs
;
514 struct sw_flow_actions
*acts
;
515 struct sk_buff
*packet
;
516 struct sw_flow
*flow
;
523 if (!a
[OVS_PACKET_ATTR_PACKET
] || !a
[OVS_PACKET_ATTR_KEY
] ||
524 !a
[OVS_PACKET_ATTR_ACTIONS
])
527 len
= nla_len(a
[OVS_PACKET_ATTR_PACKET
]);
528 packet
= __dev_alloc_skb(NET_IP_ALIGN
+ len
, GFP_KERNEL
);
532 skb_reserve(packet
, NET_IP_ALIGN
);
534 nla_memcpy(__skb_put(packet
, len
), a
[OVS_PACKET_ATTR_PACKET
], len
);
536 skb_reset_mac_header(packet
);
537 eth
= eth_hdr(packet
);
539 /* Normally, setting the skb 'protocol' field would be handled by a
540 * call to eth_type_trans(), but it assumes there's a sending
541 * device, which we may not have. */
542 if (ntohs(eth
->h_proto
) >= ETH_P_802_3_MIN
)
543 packet
->protocol
= eth
->h_proto
;
545 packet
->protocol
= htons(ETH_P_802_2
);
547 /* Build an sw_flow for sending this packet. */
548 flow
= ovs_flow_alloc();
553 err
= ovs_flow_extract(packet
, -1, &flow
->key
);
557 err
= ovs_nla_get_flow_metadata(flow
, a
[OVS_PACKET_ATTR_KEY
]);
560 acts
= ovs_nla_alloc_flow_actions(nla_len(a
[OVS_PACKET_ATTR_ACTIONS
]));
565 err
= ovs_nla_copy_actions(a
[OVS_PACKET_ATTR_ACTIONS
],
566 &flow
->key
, 0, &acts
);
567 rcu_assign_pointer(flow
->sf_acts
, acts
);
571 OVS_CB(packet
)->flow
= flow
;
572 OVS_CB(packet
)->pkt_key
= &flow
->key
;
573 packet
->priority
= flow
->key
.phy
.priority
;
574 packet
->mark
= flow
->key
.phy
.skb_mark
;
577 dp
= get_dp(sock_net(skb
->sk
), ovs_header
->dp_ifindex
);
583 err
= ovs_execute_actions(dp
, packet
);
587 ovs_flow_free(flow
, false);
593 ovs_flow_free(flow
, false);
600 static const struct nla_policy packet_policy
[OVS_PACKET_ATTR_MAX
+ 1] = {
601 [OVS_PACKET_ATTR_PACKET
] = { .len
= ETH_HLEN
},
602 [OVS_PACKET_ATTR_KEY
] = { .type
= NLA_NESTED
},
603 [OVS_PACKET_ATTR_ACTIONS
] = { .type
= NLA_NESTED
},
606 static const struct genl_ops dp_packet_genl_ops
[] = {
607 { .cmd
= OVS_PACKET_CMD_EXECUTE
,
608 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
609 .policy
= packet_policy
,
610 .doit
= ovs_packet_cmd_execute
614 static struct genl_family dp_packet_genl_family
= {
615 .id
= GENL_ID_GENERATE
,
616 .hdrsize
= sizeof(struct ovs_header
),
617 .name
= OVS_PACKET_FAMILY
,
618 .version
= OVS_PACKET_VERSION
,
619 .maxattr
= OVS_PACKET_ATTR_MAX
,
621 .parallel_ops
= true,
622 .ops
= dp_packet_genl_ops
,
623 .n_ops
= ARRAY_SIZE(dp_packet_genl_ops
),
626 static void get_dp_stats(struct datapath
*dp
, struct ovs_dp_stats
*stats
,
627 struct ovs_dp_megaflow_stats
*mega_stats
)
631 memset(mega_stats
, 0, sizeof(*mega_stats
));
633 stats
->n_flows
= ovs_flow_tbl_count(&dp
->table
);
634 mega_stats
->n_masks
= ovs_flow_tbl_num_masks(&dp
->table
);
636 stats
->n_hit
= stats
->n_missed
= stats
->n_lost
= 0;
638 for_each_possible_cpu(i
) {
639 const struct dp_stats_percpu
*percpu_stats
;
640 struct dp_stats_percpu local_stats
;
643 percpu_stats
= per_cpu_ptr(dp
->stats_percpu
, i
);
646 start
= u64_stats_fetch_begin_irq(&percpu_stats
->syncp
);
647 local_stats
= *percpu_stats
;
648 } while (u64_stats_fetch_retry_irq(&percpu_stats
->syncp
, start
));
650 stats
->n_hit
+= local_stats
.n_hit
;
651 stats
->n_missed
+= local_stats
.n_missed
;
652 stats
->n_lost
+= local_stats
.n_lost
;
653 mega_stats
->n_mask_hit
+= local_stats
.n_mask_hit
;
657 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions
*acts
)
659 return NLMSG_ALIGN(sizeof(struct ovs_header
))
660 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
661 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_MASK */
662 + nla_total_size(sizeof(struct ovs_flow_stats
)) /* OVS_FLOW_ATTR_STATS */
663 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
664 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
665 + nla_total_size(acts
->actions_len
); /* OVS_FLOW_ATTR_ACTIONS */
668 /* Called with ovs_mutex or RCU read lock. */
669 static int ovs_flow_cmd_fill_info(const struct sw_flow
*flow
, int dp_ifindex
,
670 struct sk_buff
*skb
, u32 portid
,
671 u32 seq
, u32 flags
, u8 cmd
)
673 const int skb_orig_len
= skb
->len
;
674 struct nlattr
*start
;
675 struct ovs_flow_stats stats
;
678 struct ovs_header
*ovs_header
;
682 ovs_header
= genlmsg_put(skb
, portid
, seq
, &dp_flow_genl_family
, flags
, cmd
);
686 ovs_header
->dp_ifindex
= dp_ifindex
;
689 nla
= nla_nest_start(skb
, OVS_FLOW_ATTR_KEY
);
691 goto nla_put_failure
;
693 err
= ovs_nla_put_flow(&flow
->unmasked_key
, &flow
->unmasked_key
, skb
);
696 nla_nest_end(skb
, nla
);
698 nla
= nla_nest_start(skb
, OVS_FLOW_ATTR_MASK
);
700 goto nla_put_failure
;
702 err
= ovs_nla_put_flow(&flow
->key
, &flow
->mask
->key
, skb
);
706 nla_nest_end(skb
, nla
);
708 ovs_flow_stats_get(flow
, &stats
, &used
, &tcp_flags
);
711 nla_put_u64(skb
, OVS_FLOW_ATTR_USED
, ovs_flow_used_time(used
)))
712 goto nla_put_failure
;
714 if (stats
.n_packets
&&
715 nla_put(skb
, OVS_FLOW_ATTR_STATS
, sizeof(struct ovs_flow_stats
), &stats
))
716 goto nla_put_failure
;
718 if ((u8
)ntohs(tcp_flags
) &&
719 nla_put_u8(skb
, OVS_FLOW_ATTR_TCP_FLAGS
, (u8
)ntohs(tcp_flags
)))
720 goto nla_put_failure
;
722 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
723 * this is the first flow to be dumped into 'skb'. This is unusual for
724 * Netlink but individual action lists can be longer than
725 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
726 * The userspace caller can always fetch the actions separately if it
727 * really wants them. (Most userspace callers in fact don't care.)
729 * This can only fail for dump operations because the skb is always
730 * properly sized for single flows.
732 start
= nla_nest_start(skb
, OVS_FLOW_ATTR_ACTIONS
);
734 const struct sw_flow_actions
*sf_acts
;
736 sf_acts
= rcu_dereference_ovsl(flow
->sf_acts
);
737 err
= ovs_nla_put_actions(sf_acts
->actions
,
738 sf_acts
->actions_len
, skb
);
741 nla_nest_end(skb
, start
);
746 nla_nest_cancel(skb
, start
);
748 } else if (skb_orig_len
)
749 goto nla_put_failure
;
751 return genlmsg_end(skb
, ovs_header
);
756 genlmsg_cancel(skb
, ovs_header
);
760 /* May not be called with RCU read lock. */
761 static struct sk_buff
*ovs_flow_cmd_alloc_info(const struct sw_flow_actions
*acts
,
762 struct genl_info
*info
,
767 if (!always
&& !ovs_must_notify(&dp_flow_genl_family
, info
, 0))
770 skb
= genlmsg_new_unicast(ovs_flow_cmd_msg_size(acts
), info
, GFP_KERNEL
);
772 return ERR_PTR(-ENOMEM
);
777 /* Called with ovs_mutex. */
778 static struct sk_buff
*ovs_flow_cmd_build_info(const struct sw_flow
*flow
,
780 struct genl_info
*info
, u8 cmd
,
786 skb
= ovs_flow_cmd_alloc_info(ovsl_dereference(flow
->sf_acts
), info
,
788 if (IS_ERR_OR_NULL(skb
))
791 retval
= ovs_flow_cmd_fill_info(flow
, dp_ifindex
, skb
,
792 info
->snd_portid
, info
->snd_seq
, 0,
798 static int ovs_flow_cmd_new(struct sk_buff
*skb
, struct genl_info
*info
)
800 struct nlattr
**a
= info
->attrs
;
801 struct ovs_header
*ovs_header
= info
->userhdr
;
802 struct sw_flow
*flow
, *new_flow
;
803 struct sw_flow_mask mask
;
804 struct sk_buff
*reply
;
806 struct sw_flow_actions
*acts
;
807 struct sw_flow_match match
;
810 /* Must have key and actions. */
812 if (!a
[OVS_FLOW_ATTR_KEY
])
814 if (!a
[OVS_FLOW_ATTR_ACTIONS
])
817 /* Most of the time we need to allocate a new flow, do it before
820 new_flow
= ovs_flow_alloc();
821 if (IS_ERR(new_flow
)) {
822 error
= PTR_ERR(new_flow
);
827 ovs_match_init(&match
, &new_flow
->unmasked_key
, &mask
);
828 error
= ovs_nla_get_match(&match
,
829 a
[OVS_FLOW_ATTR_KEY
], a
[OVS_FLOW_ATTR_MASK
]);
833 ovs_flow_mask_key(&new_flow
->key
, &new_flow
->unmasked_key
, &mask
);
835 /* Validate actions. */
836 acts
= ovs_nla_alloc_flow_actions(nla_len(a
[OVS_FLOW_ATTR_ACTIONS
]));
837 error
= PTR_ERR(acts
);
841 error
= ovs_nla_copy_actions(a
[OVS_FLOW_ATTR_ACTIONS
], &new_flow
->key
,
844 OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
848 reply
= ovs_flow_cmd_alloc_info(acts
, info
, false);
850 error
= PTR_ERR(reply
);
855 dp
= get_dp(sock_net(skb
->sk
), ovs_header
->dp_ifindex
);
860 /* Check if this is a duplicate flow */
861 flow
= ovs_flow_tbl_lookup(&dp
->table
, &new_flow
->unmasked_key
);
863 rcu_assign_pointer(new_flow
->sf_acts
, acts
);
865 /* Put flow in bucket. */
866 error
= ovs_flow_tbl_insert(&dp
->table
, new_flow
, &mask
);
867 if (unlikely(error
)) {
872 if (unlikely(reply
)) {
873 error
= ovs_flow_cmd_fill_info(new_flow
,
874 ovs_header
->dp_ifindex
,
875 reply
, info
->snd_portid
,
882 struct sw_flow_actions
*old_acts
;
884 /* Bail out if we're not allowed to modify an existing flow.
885 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
886 * because Generic Netlink treats the latter as a dump
887 * request. We also accept NLM_F_EXCL in case that bug ever
890 if (unlikely(info
->nlhdr
->nlmsg_flags
& (NLM_F_CREATE
895 /* The unmasked key has to be the same for flow updates. */
896 if (unlikely(!ovs_flow_cmp_unmasked_key(flow
, &match
))) {
897 flow
= ovs_flow_tbl_lookup_exact(&dp
->table
, &match
);
903 /* Update actions. */
904 old_acts
= ovsl_dereference(flow
->sf_acts
);
905 rcu_assign_pointer(flow
->sf_acts
, acts
);
907 if (unlikely(reply
)) {
908 error
= ovs_flow_cmd_fill_info(flow
,
909 ovs_header
->dp_ifindex
,
910 reply
, info
->snd_portid
,
917 ovs_nla_free_flow_actions(old_acts
);
918 ovs_flow_free(new_flow
, false);
922 ovs_notify(&dp_flow_genl_family
, reply
, info
);
931 ovs_flow_free(new_flow
, false);
936 static int ovs_flow_cmd_set(struct sk_buff
*skb
, struct genl_info
*info
)
938 struct nlattr
**a
= info
->attrs
;
939 struct ovs_header
*ovs_header
= info
->userhdr
;
940 struct sw_flow_key key
, masked_key
;
941 struct sw_flow
*flow
;
942 struct sw_flow_mask mask
;
943 struct sk_buff
*reply
= NULL
;
945 struct sw_flow_actions
*old_acts
= NULL
, *acts
= NULL
;
946 struct sw_flow_match match
;
951 if (!a
[OVS_FLOW_ATTR_KEY
])
954 ovs_match_init(&match
, &key
, &mask
);
955 error
= ovs_nla_get_match(&match
,
956 a
[OVS_FLOW_ATTR_KEY
], a
[OVS_FLOW_ATTR_MASK
]);
960 /* Validate actions. */
961 if (a
[OVS_FLOW_ATTR_ACTIONS
]) {
962 acts
= ovs_nla_alloc_flow_actions(nla_len(a
[OVS_FLOW_ATTR_ACTIONS
]));
963 error
= PTR_ERR(acts
);
967 ovs_flow_mask_key(&masked_key
, &key
, &mask
);
968 error
= ovs_nla_copy_actions(a
[OVS_FLOW_ATTR_ACTIONS
],
969 &masked_key
, 0, &acts
);
971 OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
976 /* Can allocate before locking if have acts. */
978 reply
= ovs_flow_cmd_alloc_info(acts
, info
, false);
980 error
= PTR_ERR(reply
);
986 dp
= get_dp(sock_net(skb
->sk
), ovs_header
->dp_ifindex
);
991 /* Check that the flow exists. */
992 flow
= ovs_flow_tbl_lookup_exact(&dp
->table
, &match
);
993 if (unlikely(!flow
)) {
998 /* Update actions, if present. */
1000 old_acts
= ovsl_dereference(flow
->sf_acts
);
1001 rcu_assign_pointer(flow
->sf_acts
, acts
);
1003 if (unlikely(reply
)) {
1004 error
= ovs_flow_cmd_fill_info(flow
,
1005 ovs_header
->dp_ifindex
,
1006 reply
, info
->snd_portid
,
1012 /* Could not alloc without acts before locking. */
1013 reply
= ovs_flow_cmd_build_info(flow
, ovs_header
->dp_ifindex
,
1014 info
, OVS_FLOW_CMD_NEW
, false);
1015 if (unlikely(IS_ERR(reply
))) {
1016 error
= PTR_ERR(reply
);
1017 goto err_unlock_ovs
;
1022 if (a
[OVS_FLOW_ATTR_CLEAR
])
1023 ovs_flow_stats_clear(flow
);
1027 ovs_notify(&dp_flow_genl_family
, reply
, info
);
1029 ovs_nla_free_flow_actions(old_acts
);
1042 static int ovs_flow_cmd_get(struct sk_buff
*skb
, struct genl_info
*info
)
1044 struct nlattr
**a
= info
->attrs
;
1045 struct ovs_header
*ovs_header
= info
->userhdr
;
1046 struct sw_flow_key key
;
1047 struct sk_buff
*reply
;
1048 struct sw_flow
*flow
;
1049 struct datapath
*dp
;
1050 struct sw_flow_match match
;
1053 if (!a
[OVS_FLOW_ATTR_KEY
]) {
1054 OVS_NLERR("Flow get message rejected, Key attribute missing.\n");
1058 ovs_match_init(&match
, &key
, NULL
);
1059 err
= ovs_nla_get_match(&match
, a
[OVS_FLOW_ATTR_KEY
], NULL
);
1064 dp
= get_dp(sock_net(skb
->sk
), ovs_header
->dp_ifindex
);
1070 flow
= ovs_flow_tbl_lookup_exact(&dp
->table
, &match
);
1076 reply
= ovs_flow_cmd_build_info(flow
, ovs_header
->dp_ifindex
, info
,
1077 OVS_FLOW_CMD_NEW
, true);
1078 if (IS_ERR(reply
)) {
1079 err
= PTR_ERR(reply
);
1084 return genlmsg_reply(reply
, info
);
1090 static int ovs_flow_cmd_del(struct sk_buff
*skb
, struct genl_info
*info
)
1092 struct nlattr
**a
= info
->attrs
;
1093 struct ovs_header
*ovs_header
= info
->userhdr
;
1094 struct sw_flow_key key
;
1095 struct sk_buff
*reply
;
1096 struct sw_flow
*flow
;
1097 struct datapath
*dp
;
1098 struct sw_flow_match match
;
1101 if (likely(a
[OVS_FLOW_ATTR_KEY
])) {
1102 ovs_match_init(&match
, &key
, NULL
);
1103 err
= ovs_nla_get_match(&match
, a
[OVS_FLOW_ATTR_KEY
], NULL
);
1109 dp
= get_dp(sock_net(skb
->sk
), ovs_header
->dp_ifindex
);
1110 if (unlikely(!dp
)) {
1115 if (unlikely(!a
[OVS_FLOW_ATTR_KEY
])) {
1116 err
= ovs_flow_tbl_flush(&dp
->table
);
1120 flow
= ovs_flow_tbl_lookup_exact(&dp
->table
, &match
);
1121 if (unlikely(!flow
)) {
1126 ovs_flow_tbl_remove(&dp
->table
, flow
);
1129 reply
= ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force
*) flow
->sf_acts
,
1131 if (likely(reply
)) {
1132 if (likely(!IS_ERR(reply
))) {
1133 rcu_read_lock(); /*To keep RCU checker happy. */
1134 err
= ovs_flow_cmd_fill_info(flow
, ovs_header
->dp_ifindex
,
1135 reply
, info
->snd_portid
,
1141 ovs_notify(&dp_flow_genl_family
, reply
, info
);
1143 netlink_set_err(sock_net(skb
->sk
)->genl_sock
, 0, 0, PTR_ERR(reply
));
1147 ovs_flow_free(flow
, true);
1154 static int ovs_flow_cmd_dump(struct sk_buff
*skb
, struct netlink_callback
*cb
)
1156 struct ovs_header
*ovs_header
= genlmsg_data(nlmsg_data(cb
->nlh
));
1157 struct table_instance
*ti
;
1158 struct datapath
*dp
;
1161 dp
= get_dp(sock_net(skb
->sk
), ovs_header
->dp_ifindex
);
1167 ti
= rcu_dereference(dp
->table
.ti
);
1169 struct sw_flow
*flow
;
1172 bucket
= cb
->args
[0];
1174 flow
= ovs_flow_tbl_dump_next(ti
, &bucket
, &obj
);
1178 if (ovs_flow_cmd_fill_info(flow
, ovs_header
->dp_ifindex
, skb
,
1179 NETLINK_CB(cb
->skb
).portid
,
1180 cb
->nlh
->nlmsg_seq
, NLM_F_MULTI
,
1181 OVS_FLOW_CMD_NEW
) < 0)
1184 cb
->args
[0] = bucket
;
1191 static const struct nla_policy flow_policy
[OVS_FLOW_ATTR_MAX
+ 1] = {
1192 [OVS_FLOW_ATTR_KEY
] = { .type
= NLA_NESTED
},
1193 [OVS_FLOW_ATTR_ACTIONS
] = { .type
= NLA_NESTED
},
1194 [OVS_FLOW_ATTR_CLEAR
] = { .type
= NLA_FLAG
},
1197 static const struct genl_ops dp_flow_genl_ops
[] = {
1198 { .cmd
= OVS_FLOW_CMD_NEW
,
1199 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1200 .policy
= flow_policy
,
1201 .doit
= ovs_flow_cmd_new
1203 { .cmd
= OVS_FLOW_CMD_DEL
,
1204 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1205 .policy
= flow_policy
,
1206 .doit
= ovs_flow_cmd_del
1208 { .cmd
= OVS_FLOW_CMD_GET
,
1209 .flags
= 0, /* OK for unprivileged users. */
1210 .policy
= flow_policy
,
1211 .doit
= ovs_flow_cmd_get
,
1212 .dumpit
= ovs_flow_cmd_dump
1214 { .cmd
= OVS_FLOW_CMD_SET
,
1215 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1216 .policy
= flow_policy
,
1217 .doit
= ovs_flow_cmd_set
,
1221 static struct genl_family dp_flow_genl_family
= {
1222 .id
= GENL_ID_GENERATE
,
1223 .hdrsize
= sizeof(struct ovs_header
),
1224 .name
= OVS_FLOW_FAMILY
,
1225 .version
= OVS_FLOW_VERSION
,
1226 .maxattr
= OVS_FLOW_ATTR_MAX
,
1228 .parallel_ops
= true,
1229 .ops
= dp_flow_genl_ops
,
1230 .n_ops
= ARRAY_SIZE(dp_flow_genl_ops
),
1231 .mcgrps
= &ovs_dp_flow_multicast_group
,
1235 static size_t ovs_dp_cmd_msg_size(void)
1237 size_t msgsize
= NLMSG_ALIGN(sizeof(struct ovs_header
));
1239 msgsize
+= nla_total_size(IFNAMSIZ
);
1240 msgsize
+= nla_total_size(sizeof(struct ovs_dp_stats
));
1241 msgsize
+= nla_total_size(sizeof(struct ovs_dp_megaflow_stats
));
1242 msgsize
+= nla_total_size(sizeof(u32
)); /* OVS_DP_ATTR_USER_FEATURES */
1247 /* Called with ovs_mutex or RCU read lock. */
1248 static int ovs_dp_cmd_fill_info(struct datapath
*dp
, struct sk_buff
*skb
,
1249 u32 portid
, u32 seq
, u32 flags
, u8 cmd
)
1251 struct ovs_header
*ovs_header
;
1252 struct ovs_dp_stats dp_stats
;
1253 struct ovs_dp_megaflow_stats dp_megaflow_stats
;
1256 ovs_header
= genlmsg_put(skb
, portid
, seq
, &dp_datapath_genl_family
,
1261 ovs_header
->dp_ifindex
= get_dpifindex(dp
);
1263 err
= nla_put_string(skb
, OVS_DP_ATTR_NAME
, ovs_dp_name(dp
));
1265 goto nla_put_failure
;
1267 get_dp_stats(dp
, &dp_stats
, &dp_megaflow_stats
);
1268 if (nla_put(skb
, OVS_DP_ATTR_STATS
, sizeof(struct ovs_dp_stats
),
1270 goto nla_put_failure
;
1272 if (nla_put(skb
, OVS_DP_ATTR_MEGAFLOW_STATS
,
1273 sizeof(struct ovs_dp_megaflow_stats
),
1274 &dp_megaflow_stats
))
1275 goto nla_put_failure
;
1277 if (nla_put_u32(skb
, OVS_DP_ATTR_USER_FEATURES
, dp
->user_features
))
1278 goto nla_put_failure
;
1280 return genlmsg_end(skb
, ovs_header
);
1283 genlmsg_cancel(skb
, ovs_header
);
1288 static struct sk_buff
*ovs_dp_cmd_alloc_info(struct genl_info
*info
)
1290 return genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info
, GFP_KERNEL
);
1293 /* Called with rcu_read_lock or ovs_mutex. */
1294 static struct datapath
*lookup_datapath(struct net
*net
,
1295 struct ovs_header
*ovs_header
,
1296 struct nlattr
*a
[OVS_DP_ATTR_MAX
+ 1])
1298 struct datapath
*dp
;
1300 if (!a
[OVS_DP_ATTR_NAME
])
1301 dp
= get_dp(net
, ovs_header
->dp_ifindex
);
1303 struct vport
*vport
;
1305 vport
= ovs_vport_locate(net
, nla_data(a
[OVS_DP_ATTR_NAME
]));
1306 dp
= vport
&& vport
->port_no
== OVSP_LOCAL
? vport
->dp
: NULL
;
1308 return dp
? dp
: ERR_PTR(-ENODEV
);
1311 static void ovs_dp_reset_user_features(struct sk_buff
*skb
, struct genl_info
*info
)
1313 struct datapath
*dp
;
1315 dp
= lookup_datapath(sock_net(skb
->sk
), info
->userhdr
, info
->attrs
);
1319 WARN(dp
->user_features
, "Dropping previously announced user features\n");
1320 dp
->user_features
= 0;
1323 static void ovs_dp_change(struct datapath
*dp
, struct nlattr
**a
)
1325 if (a
[OVS_DP_ATTR_USER_FEATURES
])
1326 dp
->user_features
= nla_get_u32(a
[OVS_DP_ATTR_USER_FEATURES
]);
1329 static int ovs_dp_cmd_new(struct sk_buff
*skb
, struct genl_info
*info
)
1331 struct nlattr
**a
= info
->attrs
;
1332 struct vport_parms parms
;
1333 struct sk_buff
*reply
;
1334 struct datapath
*dp
;
1335 struct vport
*vport
;
1336 struct ovs_net
*ovs_net
;
1340 if (!a
[OVS_DP_ATTR_NAME
] || !a
[OVS_DP_ATTR_UPCALL_PID
])
1343 reply
= ovs_dp_cmd_alloc_info(info
);
1348 dp
= kzalloc(sizeof(*dp
), GFP_KERNEL
);
1350 goto err_free_reply
;
1352 ovs_dp_set_net(dp
, hold_net(sock_net(skb
->sk
)));
1354 /* Allocate table. */
1355 err
= ovs_flow_tbl_init(&dp
->table
);
1359 dp
->stats_percpu
= netdev_alloc_pcpu_stats(struct dp_stats_percpu
);
1360 if (!dp
->stats_percpu
) {
1362 goto err_destroy_table
;
1365 dp
->ports
= kmalloc(DP_VPORT_HASH_BUCKETS
* sizeof(struct hlist_head
),
1369 goto err_destroy_percpu
;
1372 for (i
= 0; i
< DP_VPORT_HASH_BUCKETS
; i
++)
1373 INIT_HLIST_HEAD(&dp
->ports
[i
]);
1375 /* Set up our datapath device. */
1376 parms
.name
= nla_data(a
[OVS_DP_ATTR_NAME
]);
1377 parms
.type
= OVS_VPORT_TYPE_INTERNAL
;
1378 parms
.options
= NULL
;
1380 parms
.port_no
= OVSP_LOCAL
;
1381 parms
.upcall_portids
= a
[OVS_DP_ATTR_UPCALL_PID
];
1383 ovs_dp_change(dp
, a
);
1385 /* So far only local changes have been made, now need the lock. */
1388 vport
= new_vport(&parms
);
1389 if (IS_ERR(vport
)) {
1390 err
= PTR_ERR(vport
);
1394 if (err
== -EEXIST
) {
1395 /* An outdated user space instance that does not understand
1396 * the concept of user_features has attempted to create a new
1397 * datapath and is likely to reuse it. Drop all user features.
1399 if (info
->genlhdr
->version
< OVS_DP_VER_FEATURES
)
1400 ovs_dp_reset_user_features(skb
, info
);
1403 goto err_destroy_ports_array
;
1406 err
= ovs_dp_cmd_fill_info(dp
, reply
, info
->snd_portid
,
1407 info
->snd_seq
, 0, OVS_DP_CMD_NEW
);
1410 ovs_net
= net_generic(ovs_dp_get_net(dp
), ovs_net_id
);
1411 list_add_tail_rcu(&dp
->list_node
, &ovs_net
->dps
);
1415 ovs_notify(&dp_datapath_genl_family
, reply
, info
);
1418 err_destroy_ports_array
:
1422 free_percpu(dp
->stats_percpu
);
1424 ovs_flow_tbl_destroy(&dp
->table
, false);
1426 release_net(ovs_dp_get_net(dp
));
1434 /* Called with ovs_mutex. */
1435 static void __dp_destroy(struct datapath
*dp
)
1439 for (i
= 0; i
< DP_VPORT_HASH_BUCKETS
; i
++) {
1440 struct vport
*vport
;
1441 struct hlist_node
*n
;
1443 hlist_for_each_entry_safe(vport
, n
, &dp
->ports
[i
], dp_hash_node
)
1444 if (vport
->port_no
!= OVSP_LOCAL
)
1445 ovs_dp_detach_port(vport
);
1448 list_del_rcu(&dp
->list_node
);
1450 /* OVSP_LOCAL is datapath internal port. We need to make sure that
1451 * all ports in datapath are destroyed first before freeing datapath.
1453 ovs_dp_detach_port(ovs_vport_ovsl(dp
, OVSP_LOCAL
));
1455 /* RCU destroy the flow table */
1456 ovs_flow_tbl_destroy(&dp
->table
, true);
1458 call_rcu(&dp
->rcu
, destroy_dp_rcu
);
1461 static int ovs_dp_cmd_del(struct sk_buff
*skb
, struct genl_info
*info
)
1463 struct sk_buff
*reply
;
1464 struct datapath
*dp
;
1467 reply
= ovs_dp_cmd_alloc_info(info
);
1472 dp
= lookup_datapath(sock_net(skb
->sk
), info
->userhdr
, info
->attrs
);
1475 goto err_unlock_free
;
1477 err
= ovs_dp_cmd_fill_info(dp
, reply
, info
->snd_portid
,
1478 info
->snd_seq
, 0, OVS_DP_CMD_DEL
);
1484 ovs_notify(&dp_datapath_genl_family
, reply
, info
);
1494 static int ovs_dp_cmd_set(struct sk_buff
*skb
, struct genl_info
*info
)
1496 struct sk_buff
*reply
;
1497 struct datapath
*dp
;
1500 reply
= ovs_dp_cmd_alloc_info(info
);
1505 dp
= lookup_datapath(sock_net(skb
->sk
), info
->userhdr
, info
->attrs
);
1508 goto err_unlock_free
;
1510 ovs_dp_change(dp
, info
->attrs
);
1512 err
= ovs_dp_cmd_fill_info(dp
, reply
, info
->snd_portid
,
1513 info
->snd_seq
, 0, OVS_DP_CMD_NEW
);
1517 ovs_notify(&dp_datapath_genl_family
, reply
, info
);
1527 static int ovs_dp_cmd_get(struct sk_buff
*skb
, struct genl_info
*info
)
1529 struct sk_buff
*reply
;
1530 struct datapath
*dp
;
1533 reply
= ovs_dp_cmd_alloc_info(info
);
1538 dp
= lookup_datapath(sock_net(skb
->sk
), info
->userhdr
, info
->attrs
);
1541 goto err_unlock_free
;
1543 err
= ovs_dp_cmd_fill_info(dp
, reply
, info
->snd_portid
,
1544 info
->snd_seq
, 0, OVS_DP_CMD_NEW
);
1548 return genlmsg_reply(reply
, info
);
1556 static int ovs_dp_cmd_dump(struct sk_buff
*skb
, struct netlink_callback
*cb
)
1558 struct ovs_net
*ovs_net
= net_generic(sock_net(skb
->sk
), ovs_net_id
);
1559 struct datapath
*dp
;
1560 int skip
= cb
->args
[0];
1564 list_for_each_entry_rcu(dp
, &ovs_net
->dps
, list_node
) {
1566 ovs_dp_cmd_fill_info(dp
, skb
, NETLINK_CB(cb
->skb
).portid
,
1567 cb
->nlh
->nlmsg_seq
, NLM_F_MULTI
,
1568 OVS_DP_CMD_NEW
) < 0)
1579 static const struct nla_policy datapath_policy
[OVS_DP_ATTR_MAX
+ 1] = {
1580 [OVS_DP_ATTR_NAME
] = { .type
= NLA_NUL_STRING
, .len
= IFNAMSIZ
- 1 },
1581 [OVS_DP_ATTR_UPCALL_PID
] = { .type
= NLA_U32
},
1582 [OVS_DP_ATTR_USER_FEATURES
] = { .type
= NLA_U32
},
1585 static const struct genl_ops dp_datapath_genl_ops
[] = {
1586 { .cmd
= OVS_DP_CMD_NEW
,
1587 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1588 .policy
= datapath_policy
,
1589 .doit
= ovs_dp_cmd_new
1591 { .cmd
= OVS_DP_CMD_DEL
,
1592 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1593 .policy
= datapath_policy
,
1594 .doit
= ovs_dp_cmd_del
1596 { .cmd
= OVS_DP_CMD_GET
,
1597 .flags
= 0, /* OK for unprivileged users. */
1598 .policy
= datapath_policy
,
1599 .doit
= ovs_dp_cmd_get
,
1600 .dumpit
= ovs_dp_cmd_dump
1602 { .cmd
= OVS_DP_CMD_SET
,
1603 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1604 .policy
= datapath_policy
,
1605 .doit
= ovs_dp_cmd_set
,
1609 static struct genl_family dp_datapath_genl_family
= {
1610 .id
= GENL_ID_GENERATE
,
1611 .hdrsize
= sizeof(struct ovs_header
),
1612 .name
= OVS_DATAPATH_FAMILY
,
1613 .version
= OVS_DATAPATH_VERSION
,
1614 .maxattr
= OVS_DP_ATTR_MAX
,
1616 .parallel_ops
= true,
1617 .ops
= dp_datapath_genl_ops
,
1618 .n_ops
= ARRAY_SIZE(dp_datapath_genl_ops
),
1619 .mcgrps
= &ovs_dp_datapath_multicast_group
,
1623 /* Called with ovs_mutex or RCU read lock. */
1624 static int ovs_vport_cmd_fill_info(struct vport
*vport
, struct sk_buff
*skb
,
1625 u32 portid
, u32 seq
, u32 flags
, u8 cmd
)
1627 struct ovs_header
*ovs_header
;
1628 struct ovs_vport_stats vport_stats
;
1631 ovs_header
= genlmsg_put(skb
, portid
, seq
, &dp_vport_genl_family
,
1636 ovs_header
->dp_ifindex
= get_dpifindex(vport
->dp
);
1638 if (nla_put_u32(skb
, OVS_VPORT_ATTR_PORT_NO
, vport
->port_no
) ||
1639 nla_put_u32(skb
, OVS_VPORT_ATTR_TYPE
, vport
->ops
->type
) ||
1640 nla_put_string(skb
, OVS_VPORT_ATTR_NAME
,
1641 vport
->ops
->get_name(vport
)))
1642 goto nla_put_failure
;
1644 ovs_vport_get_stats(vport
, &vport_stats
);
1645 if (nla_put(skb
, OVS_VPORT_ATTR_STATS
, sizeof(struct ovs_vport_stats
),
1647 goto nla_put_failure
;
1649 if (ovs_vport_get_upcall_portids(vport
, skb
))
1650 goto nla_put_failure
;
1652 err
= ovs_vport_get_options(vport
, skb
);
1653 if (err
== -EMSGSIZE
)
1656 return genlmsg_end(skb
, ovs_header
);
1661 genlmsg_cancel(skb
, ovs_header
);
1665 static struct sk_buff
*ovs_vport_cmd_alloc_info(void)
1667 return nlmsg_new(NLMSG_DEFAULT_SIZE
, GFP_KERNEL
);
1670 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1671 struct sk_buff
*ovs_vport_cmd_build_info(struct vport
*vport
, u32 portid
,
1674 struct sk_buff
*skb
;
1677 skb
= nlmsg_new(NLMSG_DEFAULT_SIZE
, GFP_ATOMIC
);
1679 return ERR_PTR(-ENOMEM
);
1681 retval
= ovs_vport_cmd_fill_info(vport
, skb
, portid
, seq
, 0, cmd
);
1687 /* Called with ovs_mutex or RCU read lock. */
1688 static struct vport
*lookup_vport(struct net
*net
,
1689 struct ovs_header
*ovs_header
,
1690 struct nlattr
*a
[OVS_VPORT_ATTR_MAX
+ 1])
1692 struct datapath
*dp
;
1693 struct vport
*vport
;
1695 if (a
[OVS_VPORT_ATTR_NAME
]) {
1696 vport
= ovs_vport_locate(net
, nla_data(a
[OVS_VPORT_ATTR_NAME
]));
1698 return ERR_PTR(-ENODEV
);
1699 if (ovs_header
->dp_ifindex
&&
1700 ovs_header
->dp_ifindex
!= get_dpifindex(vport
->dp
))
1701 return ERR_PTR(-ENODEV
);
1703 } else if (a
[OVS_VPORT_ATTR_PORT_NO
]) {
1704 u32 port_no
= nla_get_u32(a
[OVS_VPORT_ATTR_PORT_NO
]);
1706 if (port_no
>= DP_MAX_PORTS
)
1707 return ERR_PTR(-EFBIG
);
1709 dp
= get_dp(net
, ovs_header
->dp_ifindex
);
1711 return ERR_PTR(-ENODEV
);
1713 vport
= ovs_vport_ovsl_rcu(dp
, port_no
);
1715 return ERR_PTR(-ENODEV
);
1718 return ERR_PTR(-EINVAL
);
1721 static int ovs_vport_cmd_new(struct sk_buff
*skb
, struct genl_info
*info
)
1723 struct nlattr
**a
= info
->attrs
;
1724 struct ovs_header
*ovs_header
= info
->userhdr
;
1725 struct vport_parms parms
;
1726 struct sk_buff
*reply
;
1727 struct vport
*vport
;
1728 struct datapath
*dp
;
1732 if (!a
[OVS_VPORT_ATTR_NAME
] || !a
[OVS_VPORT_ATTR_TYPE
] ||
1733 !a
[OVS_VPORT_ATTR_UPCALL_PID
])
1736 port_no
= a
[OVS_VPORT_ATTR_PORT_NO
]
1737 ? nla_get_u32(a
[OVS_VPORT_ATTR_PORT_NO
]) : 0;
1738 if (port_no
>= DP_MAX_PORTS
)
1741 reply
= ovs_vport_cmd_alloc_info();
1746 dp
= get_dp(sock_net(skb
->sk
), ovs_header
->dp_ifindex
);
1749 goto exit_unlock_free
;
1752 vport
= ovs_vport_ovsl(dp
, port_no
);
1755 goto exit_unlock_free
;
1757 for (port_no
= 1; ; port_no
++) {
1758 if (port_no
>= DP_MAX_PORTS
) {
1760 goto exit_unlock_free
;
1762 vport
= ovs_vport_ovsl(dp
, port_no
);
1768 parms
.name
= nla_data(a
[OVS_VPORT_ATTR_NAME
]);
1769 parms
.type
= nla_get_u32(a
[OVS_VPORT_ATTR_TYPE
]);
1770 parms
.options
= a
[OVS_VPORT_ATTR_OPTIONS
];
1772 parms
.port_no
= port_no
;
1773 parms
.upcall_portids
= a
[OVS_VPORT_ATTR_UPCALL_PID
];
1775 vport
= new_vport(&parms
);
1776 err
= PTR_ERR(vport
);
1778 goto exit_unlock_free
;
1780 err
= ovs_vport_cmd_fill_info(vport
, reply
, info
->snd_portid
,
1781 info
->snd_seq
, 0, OVS_VPORT_CMD_NEW
);
1785 ovs_notify(&dp_vport_genl_family
, reply
, info
);
1794 static int ovs_vport_cmd_set(struct sk_buff
*skb
, struct genl_info
*info
)
1796 struct nlattr
**a
= info
->attrs
;
1797 struct sk_buff
*reply
;
1798 struct vport
*vport
;
1801 reply
= ovs_vport_cmd_alloc_info();
1806 vport
= lookup_vport(sock_net(skb
->sk
), info
->userhdr
, a
);
1807 err
= PTR_ERR(vport
);
1809 goto exit_unlock_free
;
1811 if (a
[OVS_VPORT_ATTR_TYPE
] &&
1812 nla_get_u32(a
[OVS_VPORT_ATTR_TYPE
]) != vport
->ops
->type
) {
1814 goto exit_unlock_free
;
1817 if (a
[OVS_VPORT_ATTR_OPTIONS
]) {
1818 err
= ovs_vport_set_options(vport
, a
[OVS_VPORT_ATTR_OPTIONS
]);
1820 goto exit_unlock_free
;
1824 if (a
[OVS_VPORT_ATTR_UPCALL_PID
]) {
1825 struct nlattr
*ids
= a
[OVS_VPORT_ATTR_UPCALL_PID
];
1827 err
= ovs_vport_set_upcall_portids(vport
, ids
);
1829 goto exit_unlock_free
;
1832 err
= ovs_vport_cmd_fill_info(vport
, reply
, info
->snd_portid
,
1833 info
->snd_seq
, 0, OVS_VPORT_CMD_NEW
);
1837 ovs_notify(&dp_vport_genl_family
, reply
, info
);
1846 static int ovs_vport_cmd_del(struct sk_buff
*skb
, struct genl_info
*info
)
1848 struct nlattr
**a
= info
->attrs
;
1849 struct sk_buff
*reply
;
1850 struct vport
*vport
;
1853 reply
= ovs_vport_cmd_alloc_info();
1858 vport
= lookup_vport(sock_net(skb
->sk
), info
->userhdr
, a
);
1859 err
= PTR_ERR(vport
);
1861 goto exit_unlock_free
;
1863 if (vport
->port_no
== OVSP_LOCAL
) {
1865 goto exit_unlock_free
;
1868 err
= ovs_vport_cmd_fill_info(vport
, reply
, info
->snd_portid
,
1869 info
->snd_seq
, 0, OVS_VPORT_CMD_DEL
);
1871 ovs_dp_detach_port(vport
);
1874 ovs_notify(&dp_vport_genl_family
, reply
, info
);
1883 static int ovs_vport_cmd_get(struct sk_buff
*skb
, struct genl_info
*info
)
1885 struct nlattr
**a
= info
->attrs
;
1886 struct ovs_header
*ovs_header
= info
->userhdr
;
1887 struct sk_buff
*reply
;
1888 struct vport
*vport
;
1891 reply
= ovs_vport_cmd_alloc_info();
1896 vport
= lookup_vport(sock_net(skb
->sk
), ovs_header
, a
);
1897 err
= PTR_ERR(vport
);
1899 goto exit_unlock_free
;
1900 err
= ovs_vport_cmd_fill_info(vport
, reply
, info
->snd_portid
,
1901 info
->snd_seq
, 0, OVS_VPORT_CMD_NEW
);
1905 return genlmsg_reply(reply
, info
);
1913 static int ovs_vport_cmd_dump(struct sk_buff
*skb
, struct netlink_callback
*cb
)
1915 struct ovs_header
*ovs_header
= genlmsg_data(nlmsg_data(cb
->nlh
));
1916 struct datapath
*dp
;
1917 int bucket
= cb
->args
[0], skip
= cb
->args
[1];
1921 dp
= get_dp(sock_net(skb
->sk
), ovs_header
->dp_ifindex
);
1926 for (i
= bucket
; i
< DP_VPORT_HASH_BUCKETS
; i
++) {
1927 struct vport
*vport
;
1930 hlist_for_each_entry_rcu(vport
, &dp
->ports
[i
], dp_hash_node
) {
1932 ovs_vport_cmd_fill_info(vport
, skb
,
1933 NETLINK_CB(cb
->skb
).portid
,
1936 OVS_VPORT_CMD_NEW
) < 0)
1952 static const struct nla_policy vport_policy
[OVS_VPORT_ATTR_MAX
+ 1] = {
1953 [OVS_VPORT_ATTR_NAME
] = { .type
= NLA_NUL_STRING
, .len
= IFNAMSIZ
- 1 },
1954 [OVS_VPORT_ATTR_STATS
] = { .len
= sizeof(struct ovs_vport_stats
) },
1955 [OVS_VPORT_ATTR_PORT_NO
] = { .type
= NLA_U32
},
1956 [OVS_VPORT_ATTR_TYPE
] = { .type
= NLA_U32
},
1957 [OVS_VPORT_ATTR_UPCALL_PID
] = { .type
= NLA_U32
},
1958 [OVS_VPORT_ATTR_OPTIONS
] = { .type
= NLA_NESTED
},
1961 static const struct genl_ops dp_vport_genl_ops
[] = {
1962 { .cmd
= OVS_VPORT_CMD_NEW
,
1963 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1964 .policy
= vport_policy
,
1965 .doit
= ovs_vport_cmd_new
1967 { .cmd
= OVS_VPORT_CMD_DEL
,
1968 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1969 .policy
= vport_policy
,
1970 .doit
= ovs_vport_cmd_del
1972 { .cmd
= OVS_VPORT_CMD_GET
,
1973 .flags
= 0, /* OK for unprivileged users. */
1974 .policy
= vport_policy
,
1975 .doit
= ovs_vport_cmd_get
,
1976 .dumpit
= ovs_vport_cmd_dump
1978 { .cmd
= OVS_VPORT_CMD_SET
,
1979 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1980 .policy
= vport_policy
,
1981 .doit
= ovs_vport_cmd_set
,
1985 struct genl_family dp_vport_genl_family
= {
1986 .id
= GENL_ID_GENERATE
,
1987 .hdrsize
= sizeof(struct ovs_header
),
1988 .name
= OVS_VPORT_FAMILY
,
1989 .version
= OVS_VPORT_VERSION
,
1990 .maxattr
= OVS_VPORT_ATTR_MAX
,
1992 .parallel_ops
= true,
1993 .ops
= dp_vport_genl_ops
,
1994 .n_ops
= ARRAY_SIZE(dp_vport_genl_ops
),
1995 .mcgrps
= &ovs_dp_vport_multicast_group
,
1999 static struct genl_family
* const dp_genl_families
[] = {
2000 &dp_datapath_genl_family
,
2001 &dp_vport_genl_family
,
2002 &dp_flow_genl_family
,
2003 &dp_packet_genl_family
,
2006 static void dp_unregister_genl(int n_families
)
2010 for (i
= 0; i
< n_families
; i
++)
2011 genl_unregister_family(dp_genl_families
[i
]);
2014 static int dp_register_genl(void)
2019 for (i
= 0; i
< ARRAY_SIZE(dp_genl_families
); i
++) {
2021 err
= genl_register_family(dp_genl_families
[i
]);
2029 dp_unregister_genl(i
);
2033 static int __net_init
ovs_init_net(struct net
*net
)
2035 struct ovs_net
*ovs_net
= net_generic(net
, ovs_net_id
);
2037 INIT_LIST_HEAD(&ovs_net
->dps
);
2038 INIT_WORK(&ovs_net
->dp_notify_work
, ovs_dp_notify_wq
);
2042 static void __net_exit
ovs_exit_net(struct net
*net
)
2044 struct datapath
*dp
, *dp_next
;
2045 struct ovs_net
*ovs_net
= net_generic(net
, ovs_net_id
);
2048 list_for_each_entry_safe(dp
, dp_next
, &ovs_net
->dps
, list_node
)
2052 cancel_work_sync(&ovs_net
->dp_notify_work
);
2055 static struct pernet_operations ovs_net_ops
= {
2056 .init
= ovs_init_net
,
2057 .exit
= ovs_exit_net
,
2059 .size
= sizeof(struct ovs_net
),
2062 static int __init
dp_init(void)
2066 BUILD_BUG_ON(sizeof(struct ovs_skb_cb
) > FIELD_SIZEOF(struct sk_buff
, cb
));
2068 pr_info("Open vSwitch switching datapath\n");
2070 err
= ovs_internal_dev_rtnl_link_register();
2074 err
= ovs_flow_init();
2076 goto error_unreg_rtnl_link
;
2078 err
= ovs_vport_init();
2080 goto error_flow_exit
;
2082 err
= register_pernet_device(&ovs_net_ops
);
2084 goto error_vport_exit
;
2086 err
= register_netdevice_notifier(&ovs_dp_device_notifier
);
2088 goto error_netns_exit
;
2090 err
= dp_register_genl();
2092 goto error_unreg_notifier
;
2096 error_unreg_notifier
:
2097 unregister_netdevice_notifier(&ovs_dp_device_notifier
);
2099 unregister_pernet_device(&ovs_net_ops
);
2104 error_unreg_rtnl_link
:
2105 ovs_internal_dev_rtnl_link_unregister();
2110 static void dp_cleanup(void)
2112 dp_unregister_genl(ARRAY_SIZE(dp_genl_families
));
2113 unregister_netdevice_notifier(&ovs_dp_device_notifier
);
2114 unregister_pernet_device(&ovs_net_ops
);
2118 ovs_internal_dev_rtnl_link_unregister();
2121 module_init(dp_init
);
2122 module_exit(dp_cleanup
);
2124 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2125 MODULE_LICENSE("GPL");