2 * Copyright (c) 2007-2012 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/version.h>
40 #include <linux/ethtool.h>
41 #include <linux/wait.h>
42 #include <asm/system.h>
43 #include <asm/div64.h>
44 #include <linux/highmem.h>
45 #include <linux/netfilter_bridge.h>
46 #include <linux/netfilter_ipv4.h>
47 #include <linux/inetdevice.h>
48 #include <linux/list.h>
49 #include <linux/openvswitch.h>
50 #include <linux/rculist.h>
51 #include <linux/dmi.h>
52 #include <net/genetlink.h>
53 #include <net/net_namespace.h>
54 #include <net/netns/generic.h>
59 #include "genl_exec.h"
62 #include "vport-internal_dev.h"
64 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,18) || \
65 LINUX_VERSION_CODE >= KERNEL_VERSION(3,4,0)
66 #error Kernels before 2.6.18 or after 3.3 are not supported by this version of Open vSwitch.
69 #define REHASH_FLOW_INTERVAL (10 * 60 * HZ)
70 static void rehash_flow_table(struct work_struct
*work
);
71 static DECLARE_DELAYED_WORK(rehash_flow_wq
, rehash_flow_table
);
73 int ovs_net_id __read_mostly
;
75 int (*ovs_dp_ioctl_hook
)(struct net_device
*dev
, struct ifreq
*rq
, int cmd
);
76 EXPORT_SYMBOL(ovs_dp_ioctl_hook
);
81 * Writes to device state (add/remove datapath, port, set operations on vports,
82 * etc.) are protected by RTNL.
84 * Writes to other state (flow table modifications, set miscellaneous datapath
85 * parameters, etc.) are protected by genl_mutex. The RTNL lock nests inside
88 * Reads are protected by RCU.
90 * There are a few special cases (mostly stats) that have their own
91 * synchronization but they nest under all of above and don't interact with
95 static struct vport
*new_vport(const struct vport_parms
*);
96 static int queue_gso_packets(struct net
*, int dp_ifindex
, struct sk_buff
*,
97 const struct dp_upcall_info
*);
98 static int queue_userspace_packet(struct net
*, int dp_ifindex
,
100 const struct dp_upcall_info
*);
102 /* Must be called with rcu_read_lock, genl_mutex, or RTNL lock. */
103 static struct datapath
*get_dp(struct net
*net
, int dp_ifindex
)
105 struct datapath
*dp
= NULL
;
106 struct net_device
*dev
;
109 dev
= dev_get_by_index_rcu(net
, dp_ifindex
);
111 struct vport
*vport
= ovs_internal_dev_get_vport(dev
);
120 /* Must be called with rcu_read_lock or RTNL lock. */
121 const char *ovs_dp_name(const struct datapath
*dp
)
123 struct vport
*vport
= ovs_vport_rtnl_rcu(dp
, OVSP_LOCAL
);
124 return vport
->ops
->get_name(vport
);
127 static int get_dpifindex(struct datapath
*dp
)
134 local
= ovs_vport_rcu(dp
, OVSP_LOCAL
);
136 ifindex
= local
->ops
->get_ifindex(local
);
145 static size_t br_nlmsg_size(void)
147 return NLMSG_ALIGN(sizeof(struct ifinfomsg
))
148 + nla_total_size(IFNAMSIZ
) /* IFLA_IFNAME */
149 + nla_total_size(MAX_ADDR_LEN
) /* IFLA_ADDRESS */
150 + nla_total_size(4) /* IFLA_MASTER */
151 + nla_total_size(4) /* IFLA_MTU */
152 + nla_total_size(1); /* IFLA_OPERSTATE */
155 /* Caller must hold RTNL lock. */
156 static int dp_fill_ifinfo(struct sk_buff
*skb
,
157 const struct vport
*port
,
158 int event
, unsigned int flags
)
160 struct datapath
*dp
= port
->dp
;
161 struct ifinfomsg
*hdr
;
162 struct nlmsghdr
*nlh
;
164 if (!port
->ops
->get_ifindex
)
167 nlh
= nlmsg_put(skb
, 0, 0, event
, sizeof(*hdr
), flags
);
171 hdr
= nlmsg_data(nlh
);
172 hdr
->ifi_family
= AF_BRIDGE
;
174 hdr
->ifi_type
= ARPHRD_ETHER
;
175 hdr
->ifi_index
= port
->ops
->get_ifindex(port
);
176 hdr
->ifi_flags
= port
->ops
->get_dev_flags(port
);
179 if (nla_put_string(skb
, IFLA_IFNAME
, port
->ops
->get_name(port
)) ||
180 nla_put_u32(skb
, IFLA_MASTER
, get_dpifindex(dp
)) ||
181 nla_put_u32(skb
, IFLA_MTU
, port
->ops
->get_mtu(port
)) ||
182 #ifdef IFLA_OPERSTATE
183 nla_put_u8(skb
, IFLA_OPERSTATE
,
184 port
->ops
->is_running(port
) ?
185 port
->ops
->get_operstate(port
) :
188 nla_put(skb
, IFLA_ADDRESS
, ETH_ALEN
, port
->ops
->get_addr(port
)))
189 goto nla_put_failure
;
191 return nlmsg_end(skb
, nlh
);
194 nlmsg_cancel(skb
, nlh
);
198 /* Caller must hold RTNL lock. */
199 static void dp_ifinfo_notify(int event
, struct vport
*port
)
204 skb
= nlmsg_new(br_nlmsg_size(), GFP_KERNEL
);
210 err
= dp_fill_ifinfo(skb
, port
, event
, 0);
212 if (err
== -ENODEV
) {
215 /* -EMSGSIZE implies BUG in br_nlmsg_size() */
216 WARN_ON(err
== -EMSGSIZE
);
221 rtnl_notify(skb
, ovs_dp_get_net(port
->dp
), 0, RTNLGRP_LINK
, NULL
, GFP_KERNEL
);
225 rtnl_set_sk_err(ovs_dp_get_net(port
->dp
), RTNLGRP_LINK
, err
);
230 static void release_dp(struct kobject
*kobj
)
232 struct datapath
*dp
= container_of(kobj
, struct datapath
, ifobj
);
236 static struct kobj_type dp_ktype
= {
237 .release
= release_dp
240 static void destroy_dp_rcu(struct rcu_head
*rcu
)
242 struct datapath
*dp
= container_of(rcu
, struct datapath
, rcu
);
244 ovs_flow_tbl_destroy((__force
struct flow_table
*)dp
->table
);
245 free_percpu(dp
->stats_percpu
);
246 release_net(ovs_dp_get_net(dp
));
248 kobject_put(&dp
->ifobj
);
251 static struct hlist_head
*vport_hash_bucket(const struct datapath
*dp
,
254 return &dp
->ports
[port_no
& (DP_VPORT_HASH_BUCKETS
- 1)];
257 struct vport
*ovs_lookup_vport(const struct datapath
*dp
, u16 port_no
)
260 struct hlist_node
*n
;
261 struct hlist_head
*head
;
263 head
= vport_hash_bucket(dp
, port_no
);
264 hlist_for_each_entry_rcu(vport
, n
, head
, dp_hash_node
) {
265 if (vport
->port_no
== port_no
)
271 /* Called with RTNL lock and genl_lock. */
272 static struct vport
*new_vport(const struct vport_parms
*parms
)
276 vport
= ovs_vport_add(parms
);
277 if (!IS_ERR(vport
)) {
278 struct datapath
*dp
= parms
->dp
;
279 struct hlist_head
*head
= vport_hash_bucket(dp
, vport
->port_no
);
281 hlist_add_head_rcu(&vport
->dp_hash_node
, head
);
282 dp_ifinfo_notify(RTM_NEWLINK
, vport
);
287 /* Called with RTNL lock. */
288 void ovs_dp_detach_port(struct vport
*p
)
292 if (p
->port_no
!= OVSP_LOCAL
)
293 ovs_dp_sysfs_del_if(p
);
295 dp_ifinfo_notify(RTM_DELLINK
, p
);
297 /* First drop references to device. */
298 hlist_del_rcu(&p
->dp_hash_node
);
300 /* Then destroy it. */
304 /* Must be called with rcu_read_lock. */
305 void ovs_dp_process_received_packet(struct vport
*p
, struct sk_buff
*skb
)
307 struct datapath
*dp
= p
->dp
;
308 struct sw_flow
*flow
;
309 struct dp_stats_percpu
*stats
;
313 stats
= per_cpu_ptr(dp
->stats_percpu
, smp_processor_id());
315 if (!OVS_CB(skb
)->flow
) {
316 struct sw_flow_key key
;
319 /* Extract flow from 'skb' into 'key'. */
320 error
= ovs_flow_extract(skb
, p
->port_no
, &key
, &key_len
);
321 if (unlikely(error
)) {
327 flow
= ovs_flow_tbl_lookup(rcu_dereference(dp
->table
),
329 if (unlikely(!flow
)) {
330 struct dp_upcall_info upcall
;
332 upcall
.cmd
= OVS_PACKET_CMD_MISS
;
334 upcall
.userdata
= NULL
;
335 upcall
.pid
= p
->upcall_pid
;
336 ovs_dp_upcall(dp
, skb
, &upcall
);
338 stats_counter
= &stats
->n_missed
;
342 OVS_CB(skb
)->flow
= flow
;
345 stats_counter
= &stats
->n_hit
;
346 ovs_flow_used(OVS_CB(skb
)->flow
, skb
);
347 ovs_execute_actions(dp
, skb
);
350 /* Update datapath statistics. */
351 u64_stats_update_begin(&stats
->sync
);
353 u64_stats_update_end(&stats
->sync
);
356 static struct genl_family dp_packet_genl_family
= {
357 .id
= GENL_ID_GENERATE
,
358 .hdrsize
= sizeof(struct ovs_header
),
359 .name
= OVS_PACKET_FAMILY
,
360 .version
= OVS_PACKET_VERSION
,
361 .maxattr
= OVS_PACKET_ATTR_MAX
,
365 int ovs_dp_upcall(struct datapath
*dp
, struct sk_buff
*skb
,
366 const struct dp_upcall_info
*upcall_info
)
368 struct dp_stats_percpu
*stats
;
372 if (upcall_info
->pid
== 0) {
377 dp_ifindex
= get_dpifindex(dp
);
383 forward_ip_summed(skb
, true);
385 if (!skb_is_gso(skb
))
386 err
= queue_userspace_packet(ovs_dp_get_net(dp
), dp_ifindex
, skb
, upcall_info
);
388 err
= queue_gso_packets(ovs_dp_get_net(dp
), dp_ifindex
, skb
, upcall_info
);
395 stats
= per_cpu_ptr(dp
->stats_percpu
, smp_processor_id());
397 u64_stats_update_begin(&stats
->sync
);
399 u64_stats_update_end(&stats
->sync
);
404 static int queue_gso_packets(struct net
*net
, int dp_ifindex
,
406 const struct dp_upcall_info
*upcall_info
)
408 struct dp_upcall_info later_info
;
409 struct sw_flow_key later_key
;
410 struct sk_buff
*segs
, *nskb
;
413 segs
= skb_gso_segment(skb
, NETIF_F_SG
| NETIF_F_HW_CSUM
);
417 /* Queue all of the segments. */
420 err
= queue_userspace_packet(net
, dp_ifindex
, skb
, upcall_info
);
424 if (skb
== segs
&& skb_shinfo(skb
)->gso_type
& SKB_GSO_UDP
) {
425 /* The initial flow key extracted by ovs_flow_extract()
426 * in this case is for a first fragment, so we need to
427 * properly mark later fragments.
429 later_key
= *upcall_info
->key
;
430 later_key
.ip
.frag
= OVS_FRAG_TYPE_LATER
;
432 later_info
= *upcall_info
;
433 later_info
.key
= &later_key
;
434 upcall_info
= &later_info
;
436 } while ((skb
= skb
->next
));
438 /* Free all of the segments. */
446 } while ((skb
= nskb
));
450 static int queue_userspace_packet(struct net
*net
, int dp_ifindex
,
452 const struct dp_upcall_info
*upcall_info
)
454 struct ovs_header
*upcall
;
455 struct sk_buff
*nskb
= NULL
;
456 struct sk_buff
*user_skb
; /* to be queued to userspace */
461 if (vlan_tx_tag_present(skb
)) {
462 nskb
= skb_clone(skb
, GFP_ATOMIC
);
466 err
= vlan_deaccel_tag(nskb
);
473 if (nla_attr_size(skb
->len
) > USHRT_MAX
) {
478 len
= sizeof(struct ovs_header
);
479 len
+= nla_total_size(skb
->len
);
480 len
+= nla_total_size(FLOW_BUFSIZE
);
481 if (upcall_info
->cmd
== OVS_PACKET_CMD_ACTION
)
482 len
+= nla_total_size(8);
484 user_skb
= genlmsg_new(len
, GFP_ATOMIC
);
490 upcall
= genlmsg_put(user_skb
, 0, 0, &dp_packet_genl_family
,
491 0, upcall_info
->cmd
);
492 upcall
->dp_ifindex
= dp_ifindex
;
494 nla
= nla_nest_start(user_skb
, OVS_PACKET_ATTR_KEY
);
495 ovs_flow_to_nlattrs(upcall_info
->key
, user_skb
);
496 nla_nest_end(user_skb
, nla
);
498 if (upcall_info
->userdata
)
499 nla_put_u64(user_skb
, OVS_PACKET_ATTR_USERDATA
,
500 nla_get_u64(upcall_info
->userdata
));
502 nla
= __nla_reserve(user_skb
, OVS_PACKET_ATTR_PACKET
, skb
->len
);
504 skb_copy_and_csum_dev(skb
, nla_data(nla
));
506 err
= genlmsg_unicast(net
, user_skb
, upcall_info
->pid
);
513 /* Called with genl_mutex. */
514 static int flush_flows(struct datapath
*dp
)
516 struct flow_table
*old_table
;
517 struct flow_table
*new_table
;
519 old_table
= genl_dereference(dp
->table
);
520 new_table
= ovs_flow_tbl_alloc(TBL_MIN_BUCKETS
);
524 rcu_assign_pointer(dp
->table
, new_table
);
526 ovs_flow_tbl_deferred_destroy(old_table
);
530 static int validate_actions(const struct nlattr
*attr
,
531 const struct sw_flow_key
*key
, int depth
);
533 static int validate_sample(const struct nlattr
*attr
,
534 const struct sw_flow_key
*key
, int depth
)
536 const struct nlattr
*attrs
[OVS_SAMPLE_ATTR_MAX
+ 1];
537 const struct nlattr
*probability
, *actions
;
538 const struct nlattr
*a
;
541 memset(attrs
, 0, sizeof(attrs
));
542 nla_for_each_nested(a
, attr
, rem
) {
543 int type
= nla_type(a
);
544 if (!type
|| type
> OVS_SAMPLE_ATTR_MAX
|| attrs
[type
])
551 probability
= attrs
[OVS_SAMPLE_ATTR_PROBABILITY
];
552 if (!probability
|| nla_len(probability
) != sizeof(u32
))
555 actions
= attrs
[OVS_SAMPLE_ATTR_ACTIONS
];
556 if (!actions
|| (nla_len(actions
) && nla_len(actions
) < NLA_HDRLEN
))
558 return validate_actions(actions
, key
, depth
+ 1);
561 static int validate_set(const struct nlattr
*a
,
562 const struct sw_flow_key
*flow_key
)
564 const struct nlattr
*ovs_key
= nla_data(a
);
565 int key_type
= nla_type(ovs_key
);
567 /* There can be only one key in a action */
568 if (nla_total_size(nla_len(ovs_key
)) != nla_len(a
))
571 if (key_type
> OVS_KEY_ATTR_MAX
||
572 nla_len(ovs_key
) != ovs_key_lens
[key_type
])
576 const struct ovs_key_ipv4
*ipv4_key
;
578 case OVS_KEY_ATTR_PRIORITY
:
579 case OVS_KEY_ATTR_TUN_ID
:
580 case OVS_KEY_ATTR_ETHERNET
:
583 case OVS_KEY_ATTR_IPV4
:
584 if (flow_key
->eth
.type
!= htons(ETH_P_IP
))
587 if (!flow_key
->ipv4
.addr
.src
|| !flow_key
->ipv4
.addr
.dst
)
590 ipv4_key
= nla_data(ovs_key
);
591 if (ipv4_key
->ipv4_proto
!= flow_key
->ip
.proto
)
594 if (ipv4_key
->ipv4_frag
!= flow_key
->ip
.frag
)
599 case OVS_KEY_ATTR_TCP
:
600 if (flow_key
->ip
.proto
!= IPPROTO_TCP
)
603 if (!flow_key
->ipv4
.tp
.src
|| !flow_key
->ipv4
.tp
.dst
)
608 case OVS_KEY_ATTR_UDP
:
609 if (flow_key
->ip
.proto
!= IPPROTO_UDP
)
612 if (!flow_key
->ipv4
.tp
.src
|| !flow_key
->ipv4
.tp
.dst
)
623 static int validate_userspace(const struct nlattr
*attr
)
625 static const struct nla_policy userspace_policy
[OVS_USERSPACE_ATTR_MAX
+ 1] = {
626 [OVS_USERSPACE_ATTR_PID
] = {.type
= NLA_U32
},
627 [OVS_USERSPACE_ATTR_USERDATA
] = {.type
= NLA_U64
},
629 struct nlattr
*a
[OVS_USERSPACE_ATTR_MAX
+ 1];
632 error
= nla_parse_nested(a
, OVS_USERSPACE_ATTR_MAX
,
633 attr
, userspace_policy
);
637 if (!a
[OVS_USERSPACE_ATTR_PID
] ||
638 !nla_get_u32(a
[OVS_USERSPACE_ATTR_PID
]))
644 static int validate_actions(const struct nlattr
*attr
,
645 const struct sw_flow_key
*key
, int depth
)
647 const struct nlattr
*a
;
650 if (depth
>= SAMPLE_ACTION_DEPTH
)
653 nla_for_each_nested(a
, attr
, rem
) {
654 /* Expected argument lengths, (u32)-1 for variable length. */
655 static const u32 action_lens
[OVS_ACTION_ATTR_MAX
+ 1] = {
656 [OVS_ACTION_ATTR_OUTPUT
] = sizeof(u32
),
657 [OVS_ACTION_ATTR_USERSPACE
] = (u32
)-1,
658 [OVS_ACTION_ATTR_PUSH_VLAN
] = sizeof(struct ovs_action_push_vlan
),
659 [OVS_ACTION_ATTR_POP_VLAN
] = 0,
660 [OVS_ACTION_ATTR_SET
] = (u32
)-1,
661 [OVS_ACTION_ATTR_SAMPLE
] = (u32
)-1
663 const struct ovs_action_push_vlan
*vlan
;
664 int type
= nla_type(a
);
666 if (type
> OVS_ACTION_ATTR_MAX
||
667 (action_lens
[type
] != nla_len(a
) &&
668 action_lens
[type
] != (u32
)-1))
672 case OVS_ACTION_ATTR_UNSPEC
:
675 case OVS_ACTION_ATTR_USERSPACE
:
676 err
= validate_userspace(a
);
681 case OVS_ACTION_ATTR_OUTPUT
:
682 if (nla_get_u32(a
) >= DP_MAX_PORTS
)
687 case OVS_ACTION_ATTR_POP_VLAN
:
690 case OVS_ACTION_ATTR_PUSH_VLAN
:
692 if (vlan
->vlan_tpid
!= htons(ETH_P_8021Q
))
694 if (!(vlan
->vlan_tci
& htons(VLAN_TAG_PRESENT
)))
698 case OVS_ACTION_ATTR_SET
:
699 err
= validate_set(a
, key
);
704 case OVS_ACTION_ATTR_SAMPLE
:
705 err
= validate_sample(a
, key
, depth
);
721 static void clear_stats(struct sw_flow
*flow
)
725 flow
->packet_count
= 0;
726 flow
->byte_count
= 0;
729 static int ovs_packet_cmd_execute(struct sk_buff
*skb
, struct genl_info
*info
)
731 struct ovs_header
*ovs_header
= info
->userhdr
;
732 struct nlattr
**a
= info
->attrs
;
733 struct sw_flow_actions
*acts
;
734 struct sk_buff
*packet
;
735 struct sw_flow
*flow
;
743 if (!a
[OVS_PACKET_ATTR_PACKET
] || !a
[OVS_PACKET_ATTR_KEY
] ||
744 !a
[OVS_PACKET_ATTR_ACTIONS
] ||
745 nla_len(a
[OVS_PACKET_ATTR_PACKET
]) < ETH_HLEN
)
748 len
= nla_len(a
[OVS_PACKET_ATTR_PACKET
]);
749 packet
= __dev_alloc_skb(NET_IP_ALIGN
+ len
, GFP_KERNEL
);
753 skb_reserve(packet
, NET_IP_ALIGN
);
755 memcpy(__skb_put(packet
, len
), nla_data(a
[OVS_PACKET_ATTR_PACKET
]), len
);
757 skb_reset_mac_header(packet
);
758 eth
= eth_hdr(packet
);
760 /* Normally, setting the skb 'protocol' field would be handled by a
761 * call to eth_type_trans(), but it assumes there's a sending
762 * device, which we may not have. */
763 if (ntohs(eth
->h_proto
) >= 1536)
764 packet
->protocol
= eth
->h_proto
;
766 packet
->protocol
= htons(ETH_P_802_2
);
768 /* Build an sw_flow for sending this packet. */
769 flow
= ovs_flow_alloc();
774 err
= ovs_flow_extract(packet
, -1, &flow
->key
, &key_len
);
778 err
= ovs_flow_metadata_from_nlattrs(&flow
->key
.phy
.priority
,
779 &flow
->key
.phy
.in_port
,
780 &flow
->key
.phy
.tun_id
,
781 a
[OVS_PACKET_ATTR_KEY
]);
785 err
= validate_actions(a
[OVS_PACKET_ATTR_ACTIONS
], &flow
->key
, 0);
789 flow
->hash
= ovs_flow_hash(&flow
->key
, key_len
);
791 acts
= ovs_flow_actions_alloc(a
[OVS_PACKET_ATTR_ACTIONS
]);
795 rcu_assign_pointer(flow
->sf_acts
, acts
);
797 OVS_CB(packet
)->flow
= flow
;
798 packet
->priority
= flow
->key
.phy
.priority
;
801 dp
= get_dp(sock_net(skb
->sk
), ovs_header
->dp_ifindex
);
807 err
= ovs_execute_actions(dp
, packet
);
824 static const struct nla_policy packet_policy
[OVS_PACKET_ATTR_MAX
+ 1] = {
825 [OVS_PACKET_ATTR_PACKET
] = { .type
= NLA_UNSPEC
},
826 [OVS_PACKET_ATTR_KEY
] = { .type
= NLA_NESTED
},
827 [OVS_PACKET_ATTR_ACTIONS
] = { .type
= NLA_NESTED
},
830 static struct genl_ops dp_packet_genl_ops
[] = {
831 { .cmd
= OVS_PACKET_CMD_EXECUTE
,
832 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
833 .policy
= packet_policy
,
834 .doit
= ovs_packet_cmd_execute
838 static void get_dp_stats(struct datapath
*dp
, struct ovs_dp_stats
*stats
)
841 struct flow_table
*table
= genl_dereference(dp
->table
);
843 stats
->n_flows
= ovs_flow_tbl_count(table
);
845 stats
->n_hit
= stats
->n_missed
= stats
->n_lost
= 0;
846 for_each_possible_cpu(i
) {
847 const struct dp_stats_percpu
*percpu_stats
;
848 struct dp_stats_percpu local_stats
;
851 percpu_stats
= per_cpu_ptr(dp
->stats_percpu
, i
);
854 start
= u64_stats_fetch_begin_bh(&percpu_stats
->sync
);
855 local_stats
= *percpu_stats
;
856 } while (u64_stats_fetch_retry_bh(&percpu_stats
->sync
, start
));
858 stats
->n_hit
+= local_stats
.n_hit
;
859 stats
->n_missed
+= local_stats
.n_missed
;
860 stats
->n_lost
+= local_stats
.n_lost
;
864 static const struct nla_policy flow_policy
[OVS_FLOW_ATTR_MAX
+ 1] = {
865 [OVS_FLOW_ATTR_KEY
] = { .type
= NLA_NESTED
},
866 [OVS_FLOW_ATTR_ACTIONS
] = { .type
= NLA_NESTED
},
867 [OVS_FLOW_ATTR_CLEAR
] = { .type
= NLA_FLAG
},
870 static struct genl_family dp_flow_genl_family
= {
871 .id
= GENL_ID_GENERATE
,
872 .hdrsize
= sizeof(struct ovs_header
),
873 .name
= OVS_FLOW_FAMILY
,
874 .version
= OVS_FLOW_VERSION
,
875 .maxattr
= OVS_FLOW_ATTR_MAX
,
879 static struct genl_multicast_group ovs_dp_flow_multicast_group
= {
880 .name
= OVS_FLOW_MCGROUP
883 /* Called with genl_lock. */
884 static int ovs_flow_cmd_fill_info(struct sw_flow
*flow
, struct datapath
*dp
,
885 struct sk_buff
*skb
, u32 pid
,
886 u32 seq
, u32 flags
, u8 cmd
)
888 const int skb_orig_len
= skb
->len
;
889 const struct sw_flow_actions
*sf_acts
;
890 struct ovs_flow_stats stats
;
891 struct ovs_header
*ovs_header
;
897 sf_acts
= rcu_dereference_protected(flow
->sf_acts
,
898 lockdep_genl_is_held());
900 ovs_header
= genlmsg_put(skb
, pid
, seq
, &dp_flow_genl_family
, flags
, cmd
);
904 ovs_header
->dp_ifindex
= get_dpifindex(dp
);
906 nla
= nla_nest_start(skb
, OVS_FLOW_ATTR_KEY
);
908 goto nla_put_failure
;
909 err
= ovs_flow_to_nlattrs(&flow
->key
, skb
);
912 nla_nest_end(skb
, nla
);
914 spin_lock_bh(&flow
->lock
);
916 stats
.n_packets
= flow
->packet_count
;
917 stats
.n_bytes
= flow
->byte_count
;
918 tcp_flags
= flow
->tcp_flags
;
919 spin_unlock_bh(&flow
->lock
);
922 nla_put_u64(skb
, OVS_FLOW_ATTR_USED
, ovs_flow_used_time(used
)))
923 goto nla_put_failure
;
925 if (stats
.n_packets
&&
926 nla_put(skb
, OVS_FLOW_ATTR_STATS
,
927 sizeof(struct ovs_flow_stats
), &stats
))
928 goto nla_put_failure
;
931 nla_put_u8(skb
, OVS_FLOW_ATTR_TCP_FLAGS
, tcp_flags
))
932 goto nla_put_failure
;
934 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
935 * this is the first flow to be dumped into 'skb'. This is unusual for
936 * Netlink but individual action lists can be longer than
937 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
938 * The userspace caller can always fetch the actions separately if it
939 * really wants them. (Most userspace callers in fact don't care.)
941 * This can only fail for dump operations because the skb is always
942 * properly sized for single flows.
944 err
= nla_put(skb
, OVS_FLOW_ATTR_ACTIONS
, sf_acts
->actions_len
,
946 if (err
< 0 && skb_orig_len
)
949 return genlmsg_end(skb
, ovs_header
);
954 genlmsg_cancel(skb
, ovs_header
);
958 static struct sk_buff
*ovs_flow_cmd_alloc_info(struct sw_flow
*flow
)
960 const struct sw_flow_actions
*sf_acts
;
963 sf_acts
= rcu_dereference_protected(flow
->sf_acts
,
964 lockdep_genl_is_held());
966 /* OVS_FLOW_ATTR_KEY */
967 len
= nla_total_size(FLOW_BUFSIZE
);
968 /* OVS_FLOW_ATTR_ACTIONS */
969 len
+= nla_total_size(sf_acts
->actions_len
);
970 /* OVS_FLOW_ATTR_STATS */
971 len
+= nla_total_size(sizeof(struct ovs_flow_stats
));
972 /* OVS_FLOW_ATTR_TCP_FLAGS */
973 len
+= nla_total_size(1);
974 /* OVS_FLOW_ATTR_USED */
975 len
+= nla_total_size(8);
977 len
+= NLMSG_ALIGN(sizeof(struct ovs_header
));
979 return genlmsg_new(len
, GFP_KERNEL
);
982 static struct sk_buff
*ovs_flow_cmd_build_info(struct sw_flow
*flow
,
984 u32 pid
, u32 seq
, u8 cmd
)
989 skb
= ovs_flow_cmd_alloc_info(flow
);
991 return ERR_PTR(-ENOMEM
);
993 retval
= ovs_flow_cmd_fill_info(flow
, dp
, skb
, pid
, seq
, 0, cmd
);
998 static int ovs_flow_cmd_new_or_set(struct sk_buff
*skb
, struct genl_info
*info
)
1000 struct nlattr
**a
= info
->attrs
;
1001 struct ovs_header
*ovs_header
= info
->userhdr
;
1002 struct sw_flow_key key
;
1003 struct sw_flow
*flow
;
1004 struct sk_buff
*reply
;
1005 struct datapath
*dp
;
1006 struct flow_table
*table
;
1012 if (!a
[OVS_FLOW_ATTR_KEY
])
1014 error
= ovs_flow_from_nlattrs(&key
, &key_len
, a
[OVS_FLOW_ATTR_KEY
]);
1018 /* Validate actions. */
1019 if (a
[OVS_FLOW_ATTR_ACTIONS
]) {
1020 error
= validate_actions(a
[OVS_FLOW_ATTR_ACTIONS
], &key
, 0);
1023 } else if (info
->genlhdr
->cmd
== OVS_FLOW_CMD_NEW
) {
1028 dp
= get_dp(sock_net(skb
->sk
), ovs_header
->dp_ifindex
);
1033 table
= genl_dereference(dp
->table
);
1034 flow
= ovs_flow_tbl_lookup(table
, &key
, key_len
);
1036 struct sw_flow_actions
*acts
;
1038 /* Bail out if we're not allowed to create a new flow. */
1040 if (info
->genlhdr
->cmd
== OVS_FLOW_CMD_SET
)
1043 /* Expand table, if necessary, to make room. */
1044 if (ovs_flow_tbl_need_to_expand(table
)) {
1045 struct flow_table
*new_table
;
1047 new_table
= ovs_flow_tbl_expand(table
);
1048 if (!IS_ERR(new_table
)) {
1049 rcu_assign_pointer(dp
->table
, new_table
);
1050 ovs_flow_tbl_deferred_destroy(table
);
1051 table
= genl_dereference(dp
->table
);
1055 /* Allocate flow. */
1056 flow
= ovs_flow_alloc();
1058 error
= PTR_ERR(flow
);
1064 /* Obtain actions. */
1065 acts
= ovs_flow_actions_alloc(a
[OVS_FLOW_ATTR_ACTIONS
]);
1066 error
= PTR_ERR(acts
);
1068 goto error_free_flow
;
1069 rcu_assign_pointer(flow
->sf_acts
, acts
);
1071 /* Put flow in bucket. */
1072 flow
->hash
= ovs_flow_hash(&key
, key_len
);
1073 ovs_flow_tbl_insert(table
, flow
);
1075 reply
= ovs_flow_cmd_build_info(flow
, dp
, info
->snd_pid
,
1079 /* We found a matching flow. */
1080 struct sw_flow_actions
*old_acts
;
1081 struct nlattr
*acts_attrs
;
1083 /* Bail out if we're not allowed to modify an existing flow.
1084 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1085 * because Generic Netlink treats the latter as a dump
1086 * request. We also accept NLM_F_EXCL in case that bug ever
1090 if (info
->genlhdr
->cmd
== OVS_FLOW_CMD_NEW
&&
1091 info
->nlhdr
->nlmsg_flags
& (NLM_F_CREATE
| NLM_F_EXCL
))
1094 /* Update actions. */
1095 old_acts
= rcu_dereference_protected(flow
->sf_acts
,
1096 lockdep_genl_is_held());
1097 acts_attrs
= a
[OVS_FLOW_ATTR_ACTIONS
];
1099 (old_acts
->actions_len
!= nla_len(acts_attrs
) ||
1100 memcmp(old_acts
->actions
, nla_data(acts_attrs
),
1101 old_acts
->actions_len
))) {
1102 struct sw_flow_actions
*new_acts
;
1104 new_acts
= ovs_flow_actions_alloc(acts_attrs
);
1105 error
= PTR_ERR(new_acts
);
1106 if (IS_ERR(new_acts
))
1109 rcu_assign_pointer(flow
->sf_acts
, new_acts
);
1110 ovs_flow_deferred_free_acts(old_acts
);
1113 reply
= ovs_flow_cmd_build_info(flow
, dp
, info
->snd_pid
,
1114 info
->snd_seq
, OVS_FLOW_CMD_NEW
);
1117 if (a
[OVS_FLOW_ATTR_CLEAR
]) {
1118 spin_lock_bh(&flow
->lock
);
1120 spin_unlock_bh(&flow
->lock
);
1125 genl_notify(reply
, genl_info_net(info
), info
->snd_pid
,
1126 ovs_dp_flow_multicast_group
.id
, info
->nlhdr
,
1129 netlink_set_err(GENL_SOCK(sock_net(skb
->sk
)), 0,
1130 ovs_dp_flow_multicast_group
.id
, PTR_ERR(reply
));
1139 static int ovs_flow_cmd_get(struct sk_buff
*skb
, struct genl_info
*info
)
1141 struct nlattr
**a
= info
->attrs
;
1142 struct ovs_header
*ovs_header
= info
->userhdr
;
1143 struct sw_flow_key key
;
1144 struct sk_buff
*reply
;
1145 struct sw_flow
*flow
;
1146 struct datapath
*dp
;
1147 struct flow_table
*table
;
1151 if (!a
[OVS_FLOW_ATTR_KEY
])
1153 err
= ovs_flow_from_nlattrs(&key
, &key_len
, a
[OVS_FLOW_ATTR_KEY
]);
1157 dp
= get_dp(sock_net(skb
->sk
), ovs_header
->dp_ifindex
);
1161 table
= genl_dereference(dp
->table
);
1162 flow
= ovs_flow_tbl_lookup(table
, &key
, key_len
);
1166 reply
= ovs_flow_cmd_build_info(flow
, dp
, info
->snd_pid
,
1167 info
->snd_seq
, OVS_FLOW_CMD_NEW
);
1169 return PTR_ERR(reply
);
1171 return genlmsg_reply(reply
, info
);
1174 static int ovs_flow_cmd_del(struct sk_buff
*skb
, struct genl_info
*info
)
1176 struct nlattr
**a
= info
->attrs
;
1177 struct ovs_header
*ovs_header
= info
->userhdr
;
1178 struct sw_flow_key key
;
1179 struct sk_buff
*reply
;
1180 struct sw_flow
*flow
;
1181 struct datapath
*dp
;
1182 struct flow_table
*table
;
1186 dp
= get_dp(sock_net(skb
->sk
), ovs_header
->dp_ifindex
);
1190 if (!a
[OVS_FLOW_ATTR_KEY
])
1191 return flush_flows(dp
);
1193 err
= ovs_flow_from_nlattrs(&key
, &key_len
, a
[OVS_FLOW_ATTR_KEY
]);
1197 table
= genl_dereference(dp
->table
);
1198 flow
= ovs_flow_tbl_lookup(table
, &key
, key_len
);
1202 reply
= ovs_flow_cmd_alloc_info(flow
);
1206 ovs_flow_tbl_remove(table
, flow
);
1208 err
= ovs_flow_cmd_fill_info(flow
, dp
, reply
, info
->snd_pid
,
1209 info
->snd_seq
, 0, OVS_FLOW_CMD_DEL
);
1212 ovs_flow_deferred_free(flow
);
1214 genl_notify(reply
, genl_info_net(info
), info
->snd_pid
,
1215 ovs_dp_flow_multicast_group
.id
, info
->nlhdr
, GFP_KERNEL
);
1219 static int ovs_flow_cmd_dump(struct sk_buff
*skb
, struct netlink_callback
*cb
)
1221 struct ovs_header
*ovs_header
= genlmsg_data(nlmsg_data(cb
->nlh
));
1222 struct datapath
*dp
;
1223 struct flow_table
*table
;
1225 dp
= get_dp(sock_net(skb
->sk
), ovs_header
->dp_ifindex
);
1229 table
= genl_dereference(dp
->table
);
1232 struct sw_flow
*flow
;
1235 bucket
= cb
->args
[0];
1237 flow
= ovs_flow_tbl_next(table
, &bucket
, &obj
);
1241 if (ovs_flow_cmd_fill_info(flow
, dp
, skb
,
1242 NETLINK_CB(cb
->skb
).pid
,
1243 cb
->nlh
->nlmsg_seq
, NLM_F_MULTI
,
1244 OVS_FLOW_CMD_NEW
) < 0)
1247 cb
->args
[0] = bucket
;
1253 static struct genl_ops dp_flow_genl_ops
[] = {
1254 { .cmd
= OVS_FLOW_CMD_NEW
,
1255 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1256 .policy
= flow_policy
,
1257 .doit
= ovs_flow_cmd_new_or_set
1259 { .cmd
= OVS_FLOW_CMD_DEL
,
1260 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1261 .policy
= flow_policy
,
1262 .doit
= ovs_flow_cmd_del
1264 { .cmd
= OVS_FLOW_CMD_GET
,
1265 .flags
= 0, /* OK for unprivileged users. */
1266 .policy
= flow_policy
,
1267 .doit
= ovs_flow_cmd_get
,
1268 .dumpit
= ovs_flow_cmd_dump
1270 { .cmd
= OVS_FLOW_CMD_SET
,
1271 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1272 .policy
= flow_policy
,
1273 .doit
= ovs_flow_cmd_new_or_set
,
1277 static const struct nla_policy datapath_policy
[OVS_DP_ATTR_MAX
+ 1] = {
1278 #ifdef HAVE_NLA_NUL_STRING
1279 [OVS_DP_ATTR_NAME
] = { .type
= NLA_NUL_STRING
, .len
= IFNAMSIZ
- 1 },
1281 [OVS_DP_ATTR_UPCALL_PID
] = { .type
= NLA_U32
},
1284 static struct genl_family dp_datapath_genl_family
= {
1285 .id
= GENL_ID_GENERATE
,
1286 .hdrsize
= sizeof(struct ovs_header
),
1287 .name
= OVS_DATAPATH_FAMILY
,
1288 .version
= OVS_DATAPATH_VERSION
,
1289 .maxattr
= OVS_DP_ATTR_MAX
,
1293 static struct genl_multicast_group ovs_dp_datapath_multicast_group
= {
1294 .name
= OVS_DATAPATH_MCGROUP
1297 static int ovs_dp_cmd_fill_info(struct datapath
*dp
, struct sk_buff
*skb
,
1298 u32 pid
, u32 seq
, u32 flags
, u8 cmd
)
1300 struct ovs_header
*ovs_header
;
1301 struct ovs_dp_stats dp_stats
;
1304 ovs_header
= genlmsg_put(skb
, pid
, seq
, &dp_datapath_genl_family
,
1309 ovs_header
->dp_ifindex
= get_dpifindex(dp
);
1312 err
= nla_put_string(skb
, OVS_DP_ATTR_NAME
, ovs_dp_name(dp
));
1315 goto nla_put_failure
;
1317 get_dp_stats(dp
, &dp_stats
);
1318 if (nla_put(skb
, OVS_DP_ATTR_STATS
, sizeof(struct ovs_dp_stats
), &dp_stats
))
1319 goto nla_put_failure
;
1321 return genlmsg_end(skb
, ovs_header
);
1324 genlmsg_cancel(skb
, ovs_header
);
1329 static struct sk_buff
*ovs_dp_cmd_build_info(struct datapath
*dp
, u32 pid
,
1332 struct sk_buff
*skb
;
1335 skb
= nlmsg_new(NLMSG_DEFAULT_SIZE
, GFP_KERNEL
);
1337 return ERR_PTR(-ENOMEM
);
1339 retval
= ovs_dp_cmd_fill_info(dp
, skb
, pid
, seq
, 0, cmd
);
1342 return ERR_PTR(retval
);
1347 static int ovs_dp_cmd_validate(struct nlattr
*a
[OVS_DP_ATTR_MAX
+ 1])
1349 return CHECK_NUL_STRING(a
[OVS_DP_ATTR_NAME
], IFNAMSIZ
- 1);
1352 /* Called with genl_mutex and optionally with RTNL lock also. */
1353 static struct datapath
*lookup_datapath(struct net
*net
,
1354 struct ovs_header
*ovs_header
,
1355 struct nlattr
*a
[OVS_DP_ATTR_MAX
+ 1])
1357 struct datapath
*dp
;
1359 if (!a
[OVS_DP_ATTR_NAME
])
1360 dp
= get_dp(net
, ovs_header
->dp_ifindex
);
1362 struct vport
*vport
;
1365 vport
= ovs_vport_locate(net
, nla_data(a
[OVS_DP_ATTR_NAME
]));
1366 dp
= vport
&& vport
->port_no
== OVSP_LOCAL
? vport
->dp
: NULL
;
1369 return dp
? dp
: ERR_PTR(-ENODEV
);
1372 static int ovs_dp_cmd_new(struct sk_buff
*skb
, struct genl_info
*info
)
1374 struct nlattr
**a
= info
->attrs
;
1375 struct vport_parms parms
;
1376 struct sk_buff
*reply
;
1377 struct datapath
*dp
;
1378 struct vport
*vport
;
1379 struct ovs_net
*ovs_net
;
1383 if (!a
[OVS_DP_ATTR_NAME
] || !a
[OVS_DP_ATTR_UPCALL_PID
])
1386 err
= ovs_dp_cmd_validate(a
);
1393 dp
= kzalloc(sizeof(*dp
), GFP_KERNEL
);
1395 goto err_unlock_rtnl
;
1397 /* Initialize kobject for bridge. This will be added as
1398 * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
1399 dp
->ifobj
.kset
= NULL
;
1400 kobject_init(&dp
->ifobj
, &dp_ktype
);
1402 /* Allocate table. */
1404 rcu_assign_pointer(dp
->table
, ovs_flow_tbl_alloc(TBL_MIN_BUCKETS
));
1408 dp
->stats_percpu
= alloc_percpu(struct dp_stats_percpu
);
1409 if (!dp
->stats_percpu
) {
1411 goto err_destroy_table
;
1413 ovs_dp_set_net(dp
, hold_net(sock_net(skb
->sk
)));
1415 dp
->ports
= kmalloc(DP_VPORT_HASH_BUCKETS
* sizeof(struct hlist_head
),
1419 goto err_destroy_percpu
;
1422 for (i
= 0; i
< DP_VPORT_HASH_BUCKETS
; i
++)
1423 INIT_HLIST_HEAD(&dp
->ports
[i
]);
1425 /* Set up our datapath device. */
1426 parms
.name
= nla_data(a
[OVS_DP_ATTR_NAME
]);
1427 parms
.type
= OVS_VPORT_TYPE_INTERNAL
;
1428 parms
.options
= NULL
;
1430 parms
.port_no
= OVSP_LOCAL
;
1431 parms
.upcall_pid
= nla_get_u32(a
[OVS_DP_ATTR_UPCALL_PID
]);
1433 vport
= new_vport(&parms
);
1434 if (IS_ERR(vport
)) {
1435 err
= PTR_ERR(vport
);
1439 goto err_destroy_ports_array
;
1442 reply
= ovs_dp_cmd_build_info(dp
, info
->snd_pid
,
1443 info
->snd_seq
, OVS_DP_CMD_NEW
);
1444 err
= PTR_ERR(reply
);
1446 goto err_destroy_local_port
;
1448 ovs_net
= net_generic(ovs_dp_get_net(dp
), ovs_net_id
);
1449 list_add_tail(&dp
->list_node
, &ovs_net
->dps
);
1450 ovs_dp_sysfs_add_dp(dp
);
1454 genl_notify(reply
, genl_info_net(info
), info
->snd_pid
,
1455 ovs_dp_datapath_multicast_group
.id
, info
->nlhdr
,
1459 err_destroy_local_port
:
1460 ovs_dp_detach_port(ovs_vport_rtnl(dp
, OVSP_LOCAL
));
1461 err_destroy_ports_array
:
1464 free_percpu(dp
->stats_percpu
);
1466 ovs_flow_tbl_destroy(genl_dereference(dp
->table
));
1475 /* Called with genl_mutex. */
1476 static void __dp_destroy(struct datapath
*dp
)
1482 for (i
= 0; i
< DP_VPORT_HASH_BUCKETS
; i
++) {
1483 struct vport
*vport
;
1484 struct hlist_node
*node
, *n
;
1486 hlist_for_each_entry_safe(vport
, node
, n
, &dp
->ports
[i
], dp_hash_node
)
1487 if (vport
->port_no
!= OVSP_LOCAL
)
1488 ovs_dp_detach_port(vport
);
1491 ovs_dp_sysfs_del_dp(dp
);
1492 list_del(&dp
->list_node
);
1493 ovs_dp_detach_port(ovs_vport_rtnl(dp
, OVSP_LOCAL
));
1495 /* rtnl_unlock() will wait until all the references to devices that
1496 * are pending unregistration have been dropped. We do it here to
1497 * ensure that any internal devices (which contain DP pointers) are
1498 * fully destroyed before freeing the datapath.
1502 call_rcu(&dp
->rcu
, destroy_dp_rcu
);
1505 static int ovs_dp_cmd_del(struct sk_buff
*skb
, struct genl_info
*info
)
1507 struct sk_buff
*reply
;
1508 struct datapath
*dp
;
1511 err
= ovs_dp_cmd_validate(info
->attrs
);
1515 dp
= lookup_datapath(sock_net(skb
->sk
), info
->userhdr
, info
->attrs
);
1520 reply
= ovs_dp_cmd_build_info(dp
, info
->snd_pid
,
1521 info
->snd_seq
, OVS_DP_CMD_DEL
);
1522 err
= PTR_ERR(reply
);
1528 genl_notify(reply
, genl_info_net(info
), info
->snd_pid
,
1529 ovs_dp_datapath_multicast_group
.id
, info
->nlhdr
,
1535 static int ovs_dp_cmd_set(struct sk_buff
*skb
, struct genl_info
*info
)
1537 struct sk_buff
*reply
;
1538 struct datapath
*dp
;
1541 err
= ovs_dp_cmd_validate(info
->attrs
);
1545 dp
= lookup_datapath(sock_net(skb
->sk
), info
->userhdr
, info
->attrs
);
1549 reply
= ovs_dp_cmd_build_info(dp
, info
->snd_pid
,
1550 info
->snd_seq
, OVS_DP_CMD_NEW
);
1551 if (IS_ERR(reply
)) {
1552 err
= PTR_ERR(reply
);
1553 netlink_set_err(GENL_SOCK(sock_net(skb
->sk
)), 0,
1554 ovs_dp_datapath_multicast_group
.id
, err
);
1558 genl_notify(reply
, genl_info_net(info
), info
->snd_pid
,
1559 ovs_dp_datapath_multicast_group
.id
, info
->nlhdr
,
1565 static int ovs_dp_cmd_get(struct sk_buff
*skb
, struct genl_info
*info
)
1567 struct sk_buff
*reply
;
1568 struct datapath
*dp
;
1571 err
= ovs_dp_cmd_validate(info
->attrs
);
1575 dp
= lookup_datapath(sock_net(skb
->sk
), info
->userhdr
, info
->attrs
);
1579 reply
= ovs_dp_cmd_build_info(dp
, info
->snd_pid
,
1580 info
->snd_seq
, OVS_DP_CMD_NEW
);
1582 return PTR_ERR(reply
);
1584 return genlmsg_reply(reply
, info
);
1587 static int ovs_dp_cmd_dump(struct sk_buff
*skb
, struct netlink_callback
*cb
)
1589 struct ovs_net
*ovs_net
= net_generic(sock_net(skb
->sk
), ovs_net_id
);
1590 struct datapath
*dp
;
1591 int skip
= cb
->args
[0];
1594 list_for_each_entry(dp
, &ovs_net
->dps
, list_node
) {
1596 ovs_dp_cmd_fill_info(dp
, skb
, NETLINK_CB(cb
->skb
).pid
,
1597 cb
->nlh
->nlmsg_seq
, NLM_F_MULTI
,
1598 OVS_DP_CMD_NEW
) < 0)
1608 static struct genl_ops dp_datapath_genl_ops
[] = {
1609 { .cmd
= OVS_DP_CMD_NEW
,
1610 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1611 .policy
= datapath_policy
,
1612 .doit
= ovs_dp_cmd_new
1614 { .cmd
= OVS_DP_CMD_DEL
,
1615 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1616 .policy
= datapath_policy
,
1617 .doit
= ovs_dp_cmd_del
1619 { .cmd
= OVS_DP_CMD_GET
,
1620 .flags
= 0, /* OK for unprivileged users. */
1621 .policy
= datapath_policy
,
1622 .doit
= ovs_dp_cmd_get
,
1623 .dumpit
= ovs_dp_cmd_dump
1625 { .cmd
= OVS_DP_CMD_SET
,
1626 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1627 .policy
= datapath_policy
,
1628 .doit
= ovs_dp_cmd_set
,
1632 static const struct nla_policy vport_policy
[OVS_VPORT_ATTR_MAX
+ 1] = {
1633 #ifdef HAVE_NLA_NUL_STRING
1634 [OVS_VPORT_ATTR_NAME
] = { .type
= NLA_NUL_STRING
, .len
= IFNAMSIZ
- 1 },
1635 [OVS_VPORT_ATTR_STATS
] = { .len
= sizeof(struct ovs_vport_stats
) },
1636 [OVS_VPORT_ATTR_ADDRESS
] = { .len
= ETH_ALEN
},
1638 [OVS_VPORT_ATTR_STATS
] = { .minlen
= sizeof(struct ovs_vport_stats
) },
1639 [OVS_VPORT_ATTR_ADDRESS
] = { .minlen
= ETH_ALEN
},
1641 [OVS_VPORT_ATTR_PORT_NO
] = { .type
= NLA_U32
},
1642 [OVS_VPORT_ATTR_TYPE
] = { .type
= NLA_U32
},
1643 [OVS_VPORT_ATTR_UPCALL_PID
] = { .type
= NLA_U32
},
1644 [OVS_VPORT_ATTR_OPTIONS
] = { .type
= NLA_NESTED
},
1647 static struct genl_family dp_vport_genl_family
= {
1648 .id
= GENL_ID_GENERATE
,
1649 .hdrsize
= sizeof(struct ovs_header
),
1650 .name
= OVS_VPORT_FAMILY
,
1651 .version
= OVS_VPORT_VERSION
,
1652 .maxattr
= OVS_VPORT_ATTR_MAX
,
1656 struct genl_multicast_group ovs_dp_vport_multicast_group
= {
1657 .name
= OVS_VPORT_MCGROUP
1660 /* Called with RTNL lock or RCU read lock. */
1661 static int ovs_vport_cmd_fill_info(struct vport
*vport
, struct sk_buff
*skb
,
1662 u32 pid
, u32 seq
, u32 flags
, u8 cmd
)
1664 struct ovs_header
*ovs_header
;
1665 struct ovs_vport_stats vport_stats
;
1668 ovs_header
= genlmsg_put(skb
, pid
, seq
, &dp_vport_genl_family
,
1673 ovs_header
->dp_ifindex
= get_dpifindex(vport
->dp
);
1675 if (nla_put_u32(skb
, OVS_VPORT_ATTR_PORT_NO
, vport
->port_no
) ||
1676 nla_put_u32(skb
, OVS_VPORT_ATTR_TYPE
, vport
->ops
->type
) ||
1677 nla_put_string(skb
, OVS_VPORT_ATTR_NAME
, vport
->ops
->get_name(vport
)) ||
1678 nla_put_u32(skb
, OVS_VPORT_ATTR_UPCALL_PID
, vport
->upcall_pid
))
1679 goto nla_put_failure
;
1681 ovs_vport_get_stats(vport
, &vport_stats
);
1682 if (nla_put(skb
, OVS_VPORT_ATTR_STATS
, sizeof(struct ovs_vport_stats
),
1684 goto nla_put_failure
;
1686 if (nla_put(skb
, OVS_VPORT_ATTR_ADDRESS
, ETH_ALEN
,
1687 vport
->ops
->get_addr(vport
)))
1688 goto nla_put_failure
;
1690 err
= ovs_vport_get_options(vport
, skb
);
1691 if (err
== -EMSGSIZE
)
1694 return genlmsg_end(skb
, ovs_header
);
1699 genlmsg_cancel(skb
, ovs_header
);
1703 /* Called with RTNL lock or RCU read lock. */
1704 struct sk_buff
*ovs_vport_cmd_build_info(struct vport
*vport
, u32 pid
,
1707 struct sk_buff
*skb
;
1710 skb
= nlmsg_new(NLMSG_DEFAULT_SIZE
, GFP_ATOMIC
);
1712 return ERR_PTR(-ENOMEM
);
1714 retval
= ovs_vport_cmd_fill_info(vport
, skb
, pid
, seq
, 0, cmd
);
1717 return ERR_PTR(retval
);
1722 static int ovs_vport_cmd_validate(struct nlattr
*a
[OVS_VPORT_ATTR_MAX
+ 1])
1724 return CHECK_NUL_STRING(a
[OVS_VPORT_ATTR_NAME
], IFNAMSIZ
- 1);
1727 /* Called with RTNL lock or RCU read lock. */
1728 static struct vport
*lookup_vport(struct net
*net
,
1729 struct ovs_header
*ovs_header
,
1730 struct nlattr
*a
[OVS_VPORT_ATTR_MAX
+ 1])
1732 struct datapath
*dp
;
1733 struct vport
*vport
;
1735 if (a
[OVS_VPORT_ATTR_NAME
]) {
1736 vport
= ovs_vport_locate(net
, nla_data(a
[OVS_VPORT_ATTR_NAME
]));
1738 return ERR_PTR(-ENODEV
);
1739 if (ovs_header
->dp_ifindex
&&
1740 ovs_header
->dp_ifindex
!= get_dpifindex(vport
->dp
))
1741 return ERR_PTR(-ENODEV
);
1743 } else if (a
[OVS_VPORT_ATTR_PORT_NO
]) {
1744 u32 port_no
= nla_get_u32(a
[OVS_VPORT_ATTR_PORT_NO
]);
1746 if (port_no
>= DP_MAX_PORTS
)
1747 return ERR_PTR(-EFBIG
);
1749 dp
= get_dp(net
, ovs_header
->dp_ifindex
);
1751 return ERR_PTR(-ENODEV
);
1753 vport
= ovs_vport_rtnl_rcu(dp
, port_no
);
1755 return ERR_PTR(-ENOENT
);
1758 return ERR_PTR(-EINVAL
);
1761 /* Called with RTNL lock. */
1762 static int change_vport(struct vport
*vport
,
1763 struct nlattr
*a
[OVS_VPORT_ATTR_MAX
+ 1])
1767 if (a
[OVS_VPORT_ATTR_STATS
])
1768 ovs_vport_set_stats(vport
, nla_data(a
[OVS_VPORT_ATTR_STATS
]));
1770 if (a
[OVS_VPORT_ATTR_ADDRESS
])
1771 err
= ovs_vport_set_addr(vport
, nla_data(a
[OVS_VPORT_ATTR_ADDRESS
]));
1776 static int ovs_vport_cmd_new(struct sk_buff
*skb
, struct genl_info
*info
)
1778 struct nlattr
**a
= info
->attrs
;
1779 struct ovs_header
*ovs_header
= info
->userhdr
;
1780 struct vport_parms parms
;
1781 struct sk_buff
*reply
;
1782 struct vport
*vport
;
1783 struct datapath
*dp
;
1788 if (!a
[OVS_VPORT_ATTR_NAME
] || !a
[OVS_VPORT_ATTR_TYPE
] ||
1789 !a
[OVS_VPORT_ATTR_UPCALL_PID
])
1792 err
= ovs_vport_cmd_validate(a
);
1797 dp
= get_dp(sock_net(skb
->sk
), ovs_header
->dp_ifindex
);
1802 if (a
[OVS_VPORT_ATTR_PORT_NO
]) {
1803 port_no
= nla_get_u32(a
[OVS_VPORT_ATTR_PORT_NO
]);
1806 if (port_no
>= DP_MAX_PORTS
)
1809 vport
= ovs_vport_rtnl(dp
, port_no
);
1814 for (port_no
= 1; ; port_no
++) {
1815 if (port_no
>= DP_MAX_PORTS
) {
1819 vport
= ovs_vport_rtnl(dp
, port_no
);
1825 parms
.name
= nla_data(a
[OVS_VPORT_ATTR_NAME
]);
1826 parms
.type
= nla_get_u32(a
[OVS_VPORT_ATTR_TYPE
]);
1827 parms
.options
= a
[OVS_VPORT_ATTR_OPTIONS
];
1829 parms
.port_no
= port_no
;
1830 parms
.upcall_pid
= nla_get_u32(a
[OVS_VPORT_ATTR_UPCALL_PID
]);
1832 vport
= new_vport(&parms
);
1833 err
= PTR_ERR(vport
);
1837 ovs_dp_sysfs_add_if(vport
);
1839 err
= change_vport(vport
, a
);
1841 reply
= ovs_vport_cmd_build_info(vport
, info
->snd_pid
,
1845 err
= PTR_ERR(reply
);
1848 ovs_dp_detach_port(vport
);
1851 genl_notify(reply
, genl_info_net(info
), info
->snd_pid
,
1852 ovs_dp_vport_multicast_group
.id
, info
->nlhdr
, GFP_KERNEL
);
1860 static int ovs_vport_cmd_set(struct sk_buff
*skb
, struct genl_info
*info
)
1862 struct nlattr
**a
= info
->attrs
;
1863 struct sk_buff
*reply
;
1864 struct vport
*vport
;
1867 err
= ovs_vport_cmd_validate(a
);
1872 vport
= lookup_vport(sock_net(skb
->sk
), info
->userhdr
, a
);
1873 err
= PTR_ERR(vport
);
1878 if (a
[OVS_VPORT_ATTR_TYPE
] &&
1879 nla_get_u32(a
[OVS_VPORT_ATTR_TYPE
]) != vport
->ops
->type
)
1882 if (!err
&& a
[OVS_VPORT_ATTR_OPTIONS
])
1883 err
= ovs_vport_set_options(vport
, a
[OVS_VPORT_ATTR_OPTIONS
]);
1885 err
= change_vport(vport
, a
);
1888 if (!err
&& a
[OVS_VPORT_ATTR_UPCALL_PID
])
1889 vport
->upcall_pid
= nla_get_u32(a
[OVS_VPORT_ATTR_UPCALL_PID
]);
1891 reply
= ovs_vport_cmd_build_info(vport
, info
->snd_pid
, info
->snd_seq
,
1893 if (IS_ERR(reply
)) {
1894 err
= PTR_ERR(reply
);
1895 netlink_set_err(GENL_SOCK(sock_net(skb
->sk
)), 0,
1896 ovs_dp_vport_multicast_group
.id
, err
);
1900 genl_notify(reply
, genl_info_net(info
), info
->snd_pid
,
1901 ovs_dp_vport_multicast_group
.id
, info
->nlhdr
, GFP_KERNEL
);
1909 static int ovs_vport_cmd_del(struct sk_buff
*skb
, struct genl_info
*info
)
1911 struct nlattr
**a
= info
->attrs
;
1912 struct sk_buff
*reply
;
1913 struct vport
*vport
;
1916 err
= ovs_vport_cmd_validate(a
);
1921 vport
= lookup_vport(sock_net(skb
->sk
), info
->userhdr
, a
);
1922 err
= PTR_ERR(vport
);
1926 if (vport
->port_no
== OVSP_LOCAL
) {
1931 reply
= ovs_vport_cmd_build_info(vport
, info
->snd_pid
, info
->snd_seq
,
1933 err
= PTR_ERR(reply
);
1937 ovs_dp_detach_port(vport
);
1939 genl_notify(reply
, genl_info_net(info
), info
->snd_pid
,
1940 ovs_dp_vport_multicast_group
.id
, info
->nlhdr
, GFP_KERNEL
);
1948 static int ovs_vport_cmd_get(struct sk_buff
*skb
, struct genl_info
*info
)
1950 struct nlattr
**a
= info
->attrs
;
1951 struct ovs_header
*ovs_header
= info
->userhdr
;
1952 struct sk_buff
*reply
;
1953 struct vport
*vport
;
1956 err
= ovs_vport_cmd_validate(a
);
1961 vport
= lookup_vport(sock_net(skb
->sk
), ovs_header
, a
);
1962 err
= PTR_ERR(vport
);
1966 reply
= ovs_vport_cmd_build_info(vport
, info
->snd_pid
, info
->snd_seq
,
1968 err
= PTR_ERR(reply
);
1974 return genlmsg_reply(reply
, info
);
1982 static int ovs_vport_cmd_dump(struct sk_buff
*skb
, struct netlink_callback
*cb
)
1984 struct ovs_header
*ovs_header
= genlmsg_data(nlmsg_data(cb
->nlh
));
1985 struct datapath
*dp
;
1986 int bucket
= cb
->args
[0], skip
= cb
->args
[1];
1989 dp
= get_dp(sock_net(skb
->sk
), ovs_header
->dp_ifindex
);
1994 for (i
= bucket
; i
< DP_VPORT_HASH_BUCKETS
; i
++) {
1995 struct vport
*vport
;
1996 struct hlist_node
*n
;
1999 hlist_for_each_entry_rcu(vport
, n
, &dp
->ports
[i
], dp_hash_node
) {
2001 ovs_vport_cmd_fill_info(vport
, skb
,
2002 NETLINK_CB(cb
->skb
).pid
,
2005 OVS_VPORT_CMD_NEW
) < 0)
2021 static struct genl_ops dp_vport_genl_ops
[] = {
2022 { .cmd
= OVS_VPORT_CMD_NEW
,
2023 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
2024 .policy
= vport_policy
,
2025 .doit
= ovs_vport_cmd_new
2027 { .cmd
= OVS_VPORT_CMD_DEL
,
2028 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
2029 .policy
= vport_policy
,
2030 .doit
= ovs_vport_cmd_del
2032 { .cmd
= OVS_VPORT_CMD_GET
,
2033 .flags
= 0, /* OK for unprivileged users. */
2034 .policy
= vport_policy
,
2035 .doit
= ovs_vport_cmd_get
,
2036 .dumpit
= ovs_vport_cmd_dump
2038 { .cmd
= OVS_VPORT_CMD_SET
,
2039 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
2040 .policy
= vport_policy
,
2041 .doit
= ovs_vport_cmd_set
,
2045 struct genl_family_and_ops
{
2046 struct genl_family
*family
;
2047 struct genl_ops
*ops
;
2049 struct genl_multicast_group
*group
;
2052 static const struct genl_family_and_ops dp_genl_families
[] = {
2053 { &dp_datapath_genl_family
,
2054 dp_datapath_genl_ops
, ARRAY_SIZE(dp_datapath_genl_ops
),
2055 &ovs_dp_datapath_multicast_group
},
2056 { &dp_vport_genl_family
,
2057 dp_vport_genl_ops
, ARRAY_SIZE(dp_vport_genl_ops
),
2058 &ovs_dp_vport_multicast_group
},
2059 { &dp_flow_genl_family
,
2060 dp_flow_genl_ops
, ARRAY_SIZE(dp_flow_genl_ops
),
2061 &ovs_dp_flow_multicast_group
},
2062 { &dp_packet_genl_family
,
2063 dp_packet_genl_ops
, ARRAY_SIZE(dp_packet_genl_ops
),
2067 static void dp_unregister_genl(int n_families
)
2071 for (i
= 0; i
< n_families
; i
++)
2072 genl_unregister_family(dp_genl_families
[i
].family
);
2075 static int dp_register_genl(void)
2082 for (i
= 0; i
< ARRAY_SIZE(dp_genl_families
); i
++) {
2083 const struct genl_family_and_ops
*f
= &dp_genl_families
[i
];
2085 err
= genl_register_family_with_ops(f
->family
, f
->ops
,
2092 err
= genl_register_mc_group(f
->family
, f
->group
);
2101 dp_unregister_genl(n_registered
);
2105 static int __rehash_flow_table(void *dummy
)
2107 struct datapath
*dp
;
2112 struct ovs_net
*ovs_net
= net_generic(net
, ovs_net_id
);
2114 list_for_each_entry(dp
, &ovs_net
->dps
, list_node
) {
2115 struct flow_table
*old_table
= genl_dereference(dp
->table
);
2116 struct flow_table
*new_table
;
2118 new_table
= ovs_flow_tbl_rehash(old_table
);
2119 if (!IS_ERR(new_table
)) {
2120 rcu_assign_pointer(dp
->table
, new_table
);
2121 ovs_flow_tbl_deferred_destroy(old_table
);
2129 static void rehash_flow_table(struct work_struct
*work
)
2131 genl_exec(__rehash_flow_table
, NULL
);
2132 schedule_delayed_work(&rehash_flow_wq
, REHASH_FLOW_INTERVAL
);
2135 static int dp_destroy_all(void *data
)
2137 struct datapath
*dp
, *dp_next
;
2138 struct ovs_net
*ovs_net
= data
;
2140 list_for_each_entry_safe(dp
, dp_next
, &ovs_net
->dps
, list_node
)
2146 static int __net_init
ovs_init_net(struct net
*net
)
2148 struct ovs_net
*ovs_net
= net_generic(net
, ovs_net_id
);
2150 INIT_LIST_HEAD(&ovs_net
->dps
);
2154 static void __net_exit
ovs_exit_net(struct net
*net
)
2156 struct ovs_net
*ovs_net
= net_generic(net
, ovs_net_id
);
2158 genl_exec(dp_destroy_all
, ovs_net
);
2161 static struct pernet_operations ovs_net_ops
= {
2162 .init
= ovs_init_net
,
2163 .exit
= ovs_exit_net
,
2165 .size
= sizeof(struct ovs_net
),
2168 static int __init
dp_init(void)
2170 struct sk_buff
*dummy_skb
;
2173 BUILD_BUG_ON(sizeof(struct ovs_skb_cb
) > sizeof(dummy_skb
->cb
));
2175 pr_info("Open vSwitch switching datapath %s, built "__DATE__
" "__TIME__
"\n",
2178 err
= genl_exec_init();
2182 err
= ovs_workqueues_init();
2184 goto error_genl_exec
;
2186 err
= ovs_tnl_init();
2190 err
= ovs_flow_init();
2192 goto error_tnl_exit
;
2194 err
= ovs_vport_init();
2196 goto error_flow_exit
;
2198 err
= register_pernet_device(&ovs_net_ops
);
2200 goto error_vport_exit
;
2202 err
= register_netdevice_notifier(&ovs_dp_device_notifier
);
2204 goto error_netns_exit
;
2206 err
= dp_register_genl();
2208 goto error_unreg_notifier
;
2210 schedule_delayed_work(&rehash_flow_wq
, REHASH_FLOW_INTERVAL
);
2214 error_unreg_notifier
:
2215 unregister_netdevice_notifier(&ovs_dp_device_notifier
);
2217 unregister_pernet_device(&ovs_net_ops
);
2225 ovs_workqueues_exit();
2232 static void dp_cleanup(void)
2234 cancel_delayed_work_sync(&rehash_flow_wq
);
2235 dp_unregister_genl(ARRAY_SIZE(dp_genl_families
));
2236 unregister_netdevice_notifier(&ovs_dp_device_notifier
);
2237 unregister_pernet_device(&ovs_net_ops
);
2242 ovs_workqueues_exit();
2246 module_init(dp_init
);
2247 module_exit(dp_cleanup
);
2249 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2250 MODULE_LICENSE("GPL");