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
;
62 EXPORT_SYMBOL_GPL(ovs_net_id
);
64 static struct genl_family dp_packet_genl_family
;
65 static struct genl_family dp_flow_genl_family
;
66 static struct genl_family dp_datapath_genl_family
;
68 static const struct genl_multicast_group ovs_dp_flow_multicast_group
= {
69 .name
= OVS_FLOW_MCGROUP
,
72 static const struct genl_multicast_group ovs_dp_datapath_multicast_group
= {
73 .name
= OVS_DATAPATH_MCGROUP
,
76 static const struct genl_multicast_group ovs_dp_vport_multicast_group
= {
77 .name
= OVS_VPORT_MCGROUP
,
80 /* Check if need to build a reply message.
81 * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
82 static bool ovs_must_notify(struct genl_family
*family
, struct genl_info
*info
,
85 return info
->nlhdr
->nlmsg_flags
& NLM_F_ECHO
||
86 genl_has_listeners(family
, genl_info_net(info
)->genl_sock
,
90 static void ovs_notify(struct genl_family
*family
,
91 struct sk_buff
*skb
, struct genl_info
*info
)
93 genl_notify(family
, skb
, genl_info_net(info
), info
->snd_portid
,
94 0, info
->nlhdr
, GFP_KERNEL
);
100 * All writes e.g. Writes to device state (add/remove datapath, port, set
101 * operations on vports, etc.), Writes to other state (flow table
102 * modifications, set miscellaneous datapath parameters, etc.) are protected
105 * Reads are protected by RCU.
107 * There are a few special cases (mostly stats) that have their own
108 * synchronization but they nest under all of above and don't interact with
111 * The RTNL lock nests inside ovs_mutex.
114 static DEFINE_MUTEX(ovs_mutex
);
118 mutex_lock(&ovs_mutex
);
121 void ovs_unlock(void)
123 mutex_unlock(&ovs_mutex
);
126 #ifdef CONFIG_LOCKDEP
127 int lockdep_ovsl_is_held(void)
130 return lockdep_is_held(&ovs_mutex
);
134 EXPORT_SYMBOL_GPL(lockdep_ovsl_is_held
);
137 static struct vport
*new_vport(const struct vport_parms
*);
138 static int queue_gso_packets(struct datapath
*dp
, struct sk_buff
*,
139 const struct sw_flow_key
*,
140 const struct dp_upcall_info
*);
141 static int queue_userspace_packet(struct datapath
*dp
, struct sk_buff
*,
142 const struct sw_flow_key
*,
143 const struct dp_upcall_info
*);
145 /* Must be called with rcu_read_lock. */
146 static struct datapath
*get_dp_rcu(struct net
*net
, int dp_ifindex
)
148 struct net_device
*dev
= dev_get_by_index_rcu(net
, dp_ifindex
);
151 struct vport
*vport
= ovs_internal_dev_get_vport(dev
);
159 /* The caller must hold either ovs_mutex or rcu_read_lock to keep the
160 * returned dp pointer valid.
162 static inline struct datapath
*get_dp(struct net
*net
, int dp_ifindex
)
166 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_ovsl_is_held());
168 dp
= get_dp_rcu(net
, dp_ifindex
);
174 /* Must be called with rcu_read_lock or ovs_mutex. */
175 const char *ovs_dp_name(const struct datapath
*dp
)
177 struct vport
*vport
= ovs_vport_ovsl_rcu(dp
, OVSP_LOCAL
);
178 return vport
->ops
->get_name(vport
);
181 static int get_dpifindex(const struct datapath
*dp
)
188 local
= ovs_vport_rcu(dp
, OVSP_LOCAL
);
190 ifindex
= netdev_vport_priv(local
)->dev
->ifindex
;
199 static void destroy_dp_rcu(struct rcu_head
*rcu
)
201 struct datapath
*dp
= container_of(rcu
, struct datapath
, rcu
);
203 ovs_flow_tbl_destroy(&dp
->table
);
204 free_percpu(dp
->stats_percpu
);
205 release_net(ovs_dp_get_net(dp
));
210 static struct hlist_head
*vport_hash_bucket(const struct datapath
*dp
,
213 return &dp
->ports
[port_no
& (DP_VPORT_HASH_BUCKETS
- 1)];
216 /* Called with ovs_mutex or RCU read lock. */
217 struct vport
*ovs_lookup_vport(const struct datapath
*dp
, u16 port_no
)
220 struct hlist_head
*head
;
222 head
= vport_hash_bucket(dp
, port_no
);
223 hlist_for_each_entry_rcu(vport
, head
, dp_hash_node
) {
224 if (vport
->port_no
== port_no
)
230 /* Called with ovs_mutex. */
231 static struct vport
*new_vport(const struct vport_parms
*parms
)
235 vport
= ovs_vport_add(parms
);
236 if (!IS_ERR(vport
)) {
237 struct datapath
*dp
= parms
->dp
;
238 struct hlist_head
*head
= vport_hash_bucket(dp
, vport
->port_no
);
240 hlist_add_head_rcu(&vport
->dp_hash_node
, head
);
245 void ovs_dp_detach_port(struct vport
*p
)
249 /* First drop references to device. */
250 hlist_del_rcu(&p
->dp_hash_node
);
252 /* Then destroy it. */
256 /* Must be called with rcu_read_lock. */
257 void ovs_dp_process_packet(struct sk_buff
*skb
, struct sw_flow_key
*key
)
259 const struct vport
*p
= OVS_CB(skb
)->input_vport
;
260 struct datapath
*dp
= p
->dp
;
261 struct sw_flow
*flow
;
262 struct sw_flow_actions
*sf_acts
;
263 struct dp_stats_percpu
*stats
;
267 stats
= this_cpu_ptr(dp
->stats_percpu
);
270 flow
= ovs_flow_tbl_lookup_stats(&dp
->table
, key
, &n_mask_hit
);
271 if (unlikely(!flow
)) {
272 struct dp_upcall_info upcall
;
275 upcall
.cmd
= OVS_PACKET_CMD_MISS
;
276 upcall
.userdata
= NULL
;
277 upcall
.portid
= ovs_vport_find_upcall_portid(p
, skb
);
278 upcall
.egress_tun_info
= NULL
;
279 error
= ovs_dp_upcall(dp
, skb
, key
, &upcall
);
284 stats_counter
= &stats
->n_missed
;
288 ovs_flow_stats_update(flow
, key
->tp
.flags
, skb
);
289 sf_acts
= rcu_dereference(flow
->sf_acts
);
290 ovs_execute_actions(dp
, skb
, sf_acts
, key
);
292 stats_counter
= &stats
->n_hit
;
295 /* Update datapath statistics. */
296 u64_stats_update_begin(&stats
->syncp
);
298 stats
->n_mask_hit
+= n_mask_hit
;
299 u64_stats_update_end(&stats
->syncp
);
302 int ovs_dp_upcall(struct datapath
*dp
, struct sk_buff
*skb
,
303 const struct sw_flow_key
*key
,
304 const struct dp_upcall_info
*upcall_info
)
306 struct dp_stats_percpu
*stats
;
309 if (upcall_info
->portid
== 0) {
314 if (!skb_is_gso(skb
))
315 err
= queue_userspace_packet(dp
, skb
, key
, upcall_info
);
317 err
= queue_gso_packets(dp
, skb
, key
, upcall_info
);
324 stats
= this_cpu_ptr(dp
->stats_percpu
);
326 u64_stats_update_begin(&stats
->syncp
);
328 u64_stats_update_end(&stats
->syncp
);
333 static int queue_gso_packets(struct datapath
*dp
, struct sk_buff
*skb
,
334 const struct sw_flow_key
*key
,
335 const struct dp_upcall_info
*upcall_info
)
337 unsigned short gso_type
= skb_shinfo(skb
)->gso_type
;
338 struct sw_flow_key later_key
;
339 struct sk_buff
*segs
, *nskb
;
340 struct ovs_skb_cb ovs_cb
;
343 ovs_cb
= *OVS_CB(skb
);
344 segs
= __skb_gso_segment(skb
, NETIF_F_SG
, false);
345 *OVS_CB(skb
) = ovs_cb
;
347 return PTR_ERR(segs
);
351 if (gso_type
& SKB_GSO_UDP
) {
352 /* The initial flow key extracted by ovs_flow_key_extract()
353 * in this case is for a first fragment, so we need to
354 * properly mark later fragments.
357 later_key
.ip
.frag
= OVS_FRAG_TYPE_LATER
;
360 /* Queue all of the segments. */
363 *OVS_CB(skb
) = ovs_cb
;
364 if (gso_type
& SKB_GSO_UDP
&& skb
!= segs
)
367 err
= queue_userspace_packet(dp
, skb
, key
, upcall_info
);
371 } while ((skb
= skb
->next
));
373 /* Free all of the segments. */
381 } while ((skb
= nskb
));
385 static size_t upcall_msg_size(const struct dp_upcall_info
*upcall_info
,
388 size_t size
= NLMSG_ALIGN(sizeof(struct ovs_header
))
389 + nla_total_size(hdrlen
) /* OVS_PACKET_ATTR_PACKET */
390 + nla_total_size(ovs_key_attr_size()); /* OVS_PACKET_ATTR_KEY */
392 /* OVS_PACKET_ATTR_USERDATA */
393 if (upcall_info
->userdata
)
394 size
+= NLA_ALIGN(upcall_info
->userdata
->nla_len
);
396 /* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
397 if (upcall_info
->egress_tun_info
)
398 size
+= nla_total_size(ovs_tun_key_attr_size());
403 static int queue_userspace_packet(struct datapath
*dp
, struct sk_buff
*skb
,
404 const struct sw_flow_key
*key
,
405 const struct dp_upcall_info
*upcall_info
)
407 struct ovs_header
*upcall
;
408 struct sk_buff
*nskb
= NULL
;
409 struct sk_buff
*user_skb
= NULL
; /* to be queued to userspace */
411 struct genl_info info
= {
412 .dst_sk
= ovs_dp_get_net(dp
)->genl_sock
,
413 .snd_portid
= upcall_info
->portid
,
419 dp_ifindex
= get_dpifindex(dp
);
423 if (vlan_tx_tag_present(skb
)) {
424 nskb
= skb_clone(skb
, GFP_ATOMIC
);
428 nskb
= __vlan_put_tag(nskb
, nskb
->vlan_proto
, vlan_tx_tag_get(nskb
));
436 if (nla_attr_size(skb
->len
) > USHRT_MAX
) {
441 /* Complete checksum if needed */
442 if (skb
->ip_summed
== CHECKSUM_PARTIAL
&&
443 (err
= skb_checksum_help(skb
)))
446 /* Older versions of OVS user space enforce alignment of the last
447 * Netlink attribute to NLA_ALIGNTO which would require extensive
448 * padding logic. Only perform zerocopy if padding is not required.
450 if (dp
->user_features
& OVS_DP_F_UNALIGNED
)
451 hlen
= skb_zerocopy_headlen(skb
);
455 len
= upcall_msg_size(upcall_info
, hlen
);
456 user_skb
= genlmsg_new_unicast(len
, &info
, GFP_ATOMIC
);
462 upcall
= genlmsg_put(user_skb
, 0, 0, &dp_packet_genl_family
,
463 0, upcall_info
->cmd
);
464 upcall
->dp_ifindex
= dp_ifindex
;
466 nla
= nla_nest_start(user_skb
, OVS_PACKET_ATTR_KEY
);
467 err
= ovs_nla_put_flow(key
, key
, user_skb
);
469 nla_nest_end(user_skb
, nla
);
471 if (upcall_info
->userdata
)
472 __nla_put(user_skb
, OVS_PACKET_ATTR_USERDATA
,
473 nla_len(upcall_info
->userdata
),
474 nla_data(upcall_info
->userdata
));
476 if (upcall_info
->egress_tun_info
) {
477 nla
= nla_nest_start(user_skb
, OVS_PACKET_ATTR_EGRESS_TUN_KEY
);
478 err
= ovs_nla_put_egress_tunnel_key(user_skb
,
479 upcall_info
->egress_tun_info
);
481 nla_nest_end(user_skb
, nla
);
484 /* Only reserve room for attribute header, packet data is added
485 * in skb_zerocopy() */
486 if (!(nla
= nla_reserve(user_skb
, OVS_PACKET_ATTR_PACKET
, 0))) {
490 nla
->nla_len
= nla_attr_size(skb
->len
);
492 err
= skb_zerocopy(user_skb
, skb
, skb
->len
, hlen
);
496 /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
497 if (!(dp
->user_features
& OVS_DP_F_UNALIGNED
)) {
498 size_t plen
= NLA_ALIGN(user_skb
->len
) - user_skb
->len
;
501 memset(skb_put(user_skb
, plen
), 0, plen
);
504 ((struct nlmsghdr
*) user_skb
->data
)->nlmsg_len
= user_skb
->len
;
506 err
= genlmsg_unicast(ovs_dp_get_net(dp
), user_skb
, upcall_info
->portid
);
516 static int ovs_packet_cmd_execute(struct sk_buff
*skb
, struct genl_info
*info
)
518 struct ovs_header
*ovs_header
= info
->userhdr
;
519 struct nlattr
**a
= info
->attrs
;
520 struct sw_flow_actions
*acts
;
521 struct sk_buff
*packet
;
522 struct sw_flow
*flow
;
523 struct sw_flow_actions
*sf_acts
;
526 struct vport
*input_vport
;
531 if (!a
[OVS_PACKET_ATTR_PACKET
] || !a
[OVS_PACKET_ATTR_KEY
] ||
532 !a
[OVS_PACKET_ATTR_ACTIONS
])
535 len
= nla_len(a
[OVS_PACKET_ATTR_PACKET
]);
536 packet
= __dev_alloc_skb(NET_IP_ALIGN
+ len
, GFP_KERNEL
);
540 skb_reserve(packet
, NET_IP_ALIGN
);
542 nla_memcpy(__skb_put(packet
, len
), a
[OVS_PACKET_ATTR_PACKET
], len
);
544 skb_reset_mac_header(packet
);
545 eth
= eth_hdr(packet
);
547 /* Normally, setting the skb 'protocol' field would be handled by a
548 * call to eth_type_trans(), but it assumes there's a sending
549 * device, which we may not have. */
550 if (ntohs(eth
->h_proto
) >= ETH_P_802_3_MIN
)
551 packet
->protocol
= eth
->h_proto
;
553 packet
->protocol
= htons(ETH_P_802_2
);
555 /* Build an sw_flow for sending this packet. */
556 flow
= ovs_flow_alloc();
561 err
= ovs_flow_key_extract_userspace(a
[OVS_PACKET_ATTR_KEY
], packet
,
566 err
= ovs_nla_copy_actions(a
[OVS_PACKET_ATTR_ACTIONS
],
571 rcu_assign_pointer(flow
->sf_acts
, acts
);
572 OVS_CB(packet
)->egress_tun_info
= NULL
;
573 packet
->priority
= flow
->key
.phy
.priority
;
574 packet
->mark
= flow
->key
.phy
.skb_mark
;
577 dp
= get_dp_rcu(sock_net(skb
->sk
), ovs_header
->dp_ifindex
);
582 input_vport
= ovs_vport_rcu(dp
, flow
->key
.phy
.in_port
);
584 input_vport
= ovs_vport_rcu(dp
, OVSP_LOCAL
);
589 OVS_CB(packet
)->input_vport
= input_vport
;
590 sf_acts
= rcu_dereference(flow
->sf_acts
);
593 err
= ovs_execute_actions(dp
, packet
, sf_acts
, &flow
->key
);
597 ovs_flow_free(flow
, false);
603 ovs_flow_free(flow
, false);
610 static const struct nla_policy packet_policy
[OVS_PACKET_ATTR_MAX
+ 1] = {
611 [OVS_PACKET_ATTR_PACKET
] = { .len
= ETH_HLEN
},
612 [OVS_PACKET_ATTR_KEY
] = { .type
= NLA_NESTED
},
613 [OVS_PACKET_ATTR_ACTIONS
] = { .type
= NLA_NESTED
},
616 static const struct genl_ops dp_packet_genl_ops
[] = {
617 { .cmd
= OVS_PACKET_CMD_EXECUTE
,
618 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
619 .policy
= packet_policy
,
620 .doit
= ovs_packet_cmd_execute
624 static struct genl_family dp_packet_genl_family
= {
625 .id
= GENL_ID_GENERATE
,
626 .hdrsize
= sizeof(struct ovs_header
),
627 .name
= OVS_PACKET_FAMILY
,
628 .version
= OVS_PACKET_VERSION
,
629 .maxattr
= OVS_PACKET_ATTR_MAX
,
631 .parallel_ops
= true,
632 .ops
= dp_packet_genl_ops
,
633 .n_ops
= ARRAY_SIZE(dp_packet_genl_ops
),
636 static void get_dp_stats(const struct datapath
*dp
, struct ovs_dp_stats
*stats
,
637 struct ovs_dp_megaflow_stats
*mega_stats
)
641 memset(mega_stats
, 0, sizeof(*mega_stats
));
643 stats
->n_flows
= ovs_flow_tbl_count(&dp
->table
);
644 mega_stats
->n_masks
= ovs_flow_tbl_num_masks(&dp
->table
);
646 stats
->n_hit
= stats
->n_missed
= stats
->n_lost
= 0;
648 for_each_possible_cpu(i
) {
649 const struct dp_stats_percpu
*percpu_stats
;
650 struct dp_stats_percpu local_stats
;
653 percpu_stats
= per_cpu_ptr(dp
->stats_percpu
, i
);
656 start
= u64_stats_fetch_begin_irq(&percpu_stats
->syncp
);
657 local_stats
= *percpu_stats
;
658 } while (u64_stats_fetch_retry_irq(&percpu_stats
->syncp
, start
));
660 stats
->n_hit
+= local_stats
.n_hit
;
661 stats
->n_missed
+= local_stats
.n_missed
;
662 stats
->n_lost
+= local_stats
.n_lost
;
663 mega_stats
->n_mask_hit
+= local_stats
.n_mask_hit
;
667 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions
*acts
)
669 return NLMSG_ALIGN(sizeof(struct ovs_header
))
670 + nla_total_size(ovs_key_attr_size()) /* OVS_FLOW_ATTR_KEY */
671 + nla_total_size(ovs_key_attr_size()) /* OVS_FLOW_ATTR_MASK */
672 + nla_total_size(sizeof(struct ovs_flow_stats
)) /* OVS_FLOW_ATTR_STATS */
673 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
674 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
675 + nla_total_size(acts
->actions_len
); /* OVS_FLOW_ATTR_ACTIONS */
678 /* Called with ovs_mutex or RCU read lock. */
679 static int ovs_flow_cmd_fill_match(const struct sw_flow
*flow
,
686 nla
= nla_nest_start(skb
, OVS_FLOW_ATTR_KEY
);
690 err
= ovs_nla_put_flow(&flow
->unmasked_key
, &flow
->unmasked_key
, skb
);
694 nla_nest_end(skb
, nla
);
696 /* Fill flow mask. */
697 nla
= nla_nest_start(skb
, OVS_FLOW_ATTR_MASK
);
701 err
= ovs_nla_put_flow(&flow
->key
, &flow
->mask
->key
, skb
);
705 nla_nest_end(skb
, nla
);
709 /* Called with ovs_mutex or RCU read lock. */
710 static int ovs_flow_cmd_fill_stats(const struct sw_flow
*flow
,
713 struct ovs_flow_stats stats
;
717 ovs_flow_stats_get(flow
, &stats
, &used
, &tcp_flags
);
720 nla_put_u64(skb
, OVS_FLOW_ATTR_USED
, ovs_flow_used_time(used
)))
723 if (stats
.n_packets
&&
724 nla_put(skb
, OVS_FLOW_ATTR_STATS
, sizeof(struct ovs_flow_stats
), &stats
))
727 if ((u8
)ntohs(tcp_flags
) &&
728 nla_put_u8(skb
, OVS_FLOW_ATTR_TCP_FLAGS
, (u8
)ntohs(tcp_flags
)))
734 /* Called with ovs_mutex or RCU read lock. */
735 static int ovs_flow_cmd_fill_actions(const struct sw_flow
*flow
,
736 struct sk_buff
*skb
, int skb_orig_len
)
738 struct nlattr
*start
;
741 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
742 * this is the first flow to be dumped into 'skb'. This is unusual for
743 * Netlink but individual action lists can be longer than
744 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
745 * The userspace caller can always fetch the actions separately if it
746 * really wants them. (Most userspace callers in fact don't care.)
748 * This can only fail for dump operations because the skb is always
749 * properly sized for single flows.
751 start
= nla_nest_start(skb
, OVS_FLOW_ATTR_ACTIONS
);
753 const struct sw_flow_actions
*sf_acts
;
755 sf_acts
= rcu_dereference_ovsl(flow
->sf_acts
);
756 err
= ovs_nla_put_actions(sf_acts
->actions
,
757 sf_acts
->actions_len
, skb
);
760 nla_nest_end(skb
, start
);
765 nla_nest_cancel(skb
, start
);
767 } else if (skb_orig_len
) {
774 /* Called with ovs_mutex or RCU read lock. */
775 static int ovs_flow_cmd_fill_info(const struct sw_flow
*flow
, int dp_ifindex
,
776 struct sk_buff
*skb
, u32 portid
,
777 u32 seq
, u32 flags
, u8 cmd
)
779 const int skb_orig_len
= skb
->len
;
780 struct ovs_header
*ovs_header
;
783 ovs_header
= genlmsg_put(skb
, portid
, seq
, &dp_flow_genl_family
,
788 ovs_header
->dp_ifindex
= dp_ifindex
;
790 err
= ovs_flow_cmd_fill_match(flow
, skb
);
794 err
= ovs_flow_cmd_fill_stats(flow
, skb
);
798 err
= ovs_flow_cmd_fill_actions(flow
, skb
, skb_orig_len
);
802 return genlmsg_end(skb
, ovs_header
);
805 genlmsg_cancel(skb
, ovs_header
);
809 /* May not be called with RCU read lock. */
810 static struct sk_buff
*ovs_flow_cmd_alloc_info(const struct sw_flow_actions
*acts
,
811 struct genl_info
*info
,
816 if (!always
&& !ovs_must_notify(&dp_flow_genl_family
, info
, 0))
819 skb
= genlmsg_new_unicast(ovs_flow_cmd_msg_size(acts
), info
, GFP_KERNEL
);
821 return ERR_PTR(-ENOMEM
);
826 /* Called with ovs_mutex. */
827 static struct sk_buff
*ovs_flow_cmd_build_info(const struct sw_flow
*flow
,
829 struct genl_info
*info
, u8 cmd
,
835 skb
= ovs_flow_cmd_alloc_info(ovsl_dereference(flow
->sf_acts
), info
,
837 if (IS_ERR_OR_NULL(skb
))
840 retval
= ovs_flow_cmd_fill_info(flow
, dp_ifindex
, skb
,
841 info
->snd_portid
, info
->snd_seq
, 0,
847 static int ovs_flow_cmd_new(struct sk_buff
*skb
, struct genl_info
*info
)
849 struct nlattr
**a
= info
->attrs
;
850 struct ovs_header
*ovs_header
= info
->userhdr
;
851 struct sw_flow
*flow
, *new_flow
;
852 struct sw_flow_mask mask
;
853 struct sk_buff
*reply
;
855 struct sw_flow_actions
*acts
;
856 struct sw_flow_match match
;
859 /* Must have key and actions. */
861 if (!a
[OVS_FLOW_ATTR_KEY
]) {
862 OVS_NLERR("Flow key attribute not present in new flow.\n");
865 if (!a
[OVS_FLOW_ATTR_ACTIONS
]) {
866 OVS_NLERR("Flow actions attribute not present in new flow.\n");
870 /* Most of the time we need to allocate a new flow, do it before
873 new_flow
= ovs_flow_alloc();
874 if (IS_ERR(new_flow
)) {
875 error
= PTR_ERR(new_flow
);
880 ovs_match_init(&match
, &new_flow
->unmasked_key
, &mask
);
881 error
= ovs_nla_get_match(&match
,
882 a
[OVS_FLOW_ATTR_KEY
], a
[OVS_FLOW_ATTR_MASK
]);
886 ovs_flow_mask_key(&new_flow
->key
, &new_flow
->unmasked_key
, &mask
);
888 /* Validate actions. */
889 error
= ovs_nla_copy_actions(a
[OVS_FLOW_ATTR_ACTIONS
], &new_flow
->key
,
892 OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
896 reply
= ovs_flow_cmd_alloc_info(acts
, info
, false);
898 error
= PTR_ERR(reply
);
903 dp
= get_dp(sock_net(skb
->sk
), ovs_header
->dp_ifindex
);
908 /* Check if this is a duplicate flow */
909 flow
= ovs_flow_tbl_lookup(&dp
->table
, &new_flow
->unmasked_key
);
911 rcu_assign_pointer(new_flow
->sf_acts
, acts
);
913 /* Put flow in bucket. */
914 error
= ovs_flow_tbl_insert(&dp
->table
, new_flow
, &mask
);
915 if (unlikely(error
)) {
920 if (unlikely(reply
)) {
921 error
= ovs_flow_cmd_fill_info(new_flow
,
922 ovs_header
->dp_ifindex
,
923 reply
, info
->snd_portid
,
930 struct sw_flow_actions
*old_acts
;
932 /* Bail out if we're not allowed to modify an existing flow.
933 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
934 * because Generic Netlink treats the latter as a dump
935 * request. We also accept NLM_F_EXCL in case that bug ever
938 if (unlikely(info
->nlhdr
->nlmsg_flags
& (NLM_F_CREATE
943 /* The unmasked key has to be the same for flow updates. */
944 if (unlikely(!ovs_flow_cmp_unmasked_key(flow
, &match
))) {
945 flow
= ovs_flow_tbl_lookup_exact(&dp
->table
, &match
);
951 /* Update actions. */
952 old_acts
= ovsl_dereference(flow
->sf_acts
);
953 rcu_assign_pointer(flow
->sf_acts
, acts
);
955 if (unlikely(reply
)) {
956 error
= ovs_flow_cmd_fill_info(flow
,
957 ovs_header
->dp_ifindex
,
958 reply
, info
->snd_portid
,
965 ovs_nla_free_flow_actions(old_acts
);
966 ovs_flow_free(new_flow
, false);
970 ovs_notify(&dp_flow_genl_family
, reply
, info
);
979 ovs_flow_free(new_flow
, false);
984 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
985 static struct sw_flow_actions
*get_flow_actions(const struct nlattr
*a
,
986 const struct sw_flow_key
*key
,
987 const struct sw_flow_mask
*mask
)
989 struct sw_flow_actions
*acts
;
990 struct sw_flow_key masked_key
;
993 ovs_flow_mask_key(&masked_key
, key
, mask
);
994 error
= ovs_nla_copy_actions(a
, &masked_key
, &acts
);
996 OVS_NLERR("Actions may not be safe on all matching packets.\n");
997 return ERR_PTR(error
);
1003 static int ovs_flow_cmd_set(struct sk_buff
*skb
, struct genl_info
*info
)
1005 struct nlattr
**a
= info
->attrs
;
1006 struct ovs_header
*ovs_header
= info
->userhdr
;
1007 struct sw_flow_key key
;
1008 struct sw_flow
*flow
;
1009 struct sw_flow_mask mask
;
1010 struct sk_buff
*reply
= NULL
;
1011 struct datapath
*dp
;
1012 struct sw_flow_actions
*old_acts
= NULL
, *acts
= NULL
;
1013 struct sw_flow_match match
;
1018 if (!a
[OVS_FLOW_ATTR_KEY
]) {
1019 OVS_NLERR("Flow key attribute not present in set flow.\n");
1023 ovs_match_init(&match
, &key
, &mask
);
1024 error
= ovs_nla_get_match(&match
,
1025 a
[OVS_FLOW_ATTR_KEY
], a
[OVS_FLOW_ATTR_MASK
]);
1029 /* Validate actions. */
1030 if (a
[OVS_FLOW_ATTR_ACTIONS
]) {
1031 acts
= get_flow_actions(a
[OVS_FLOW_ATTR_ACTIONS
], &key
, &mask
);
1033 error
= PTR_ERR(acts
);
1037 /* Can allocate before locking if have acts. */
1038 reply
= ovs_flow_cmd_alloc_info(acts
, info
, false);
1039 if (IS_ERR(reply
)) {
1040 error
= PTR_ERR(reply
);
1041 goto err_kfree_acts
;
1046 dp
= get_dp(sock_net(skb
->sk
), ovs_header
->dp_ifindex
);
1047 if (unlikely(!dp
)) {
1049 goto err_unlock_ovs
;
1051 /* Check that the flow exists. */
1052 flow
= ovs_flow_tbl_lookup_exact(&dp
->table
, &match
);
1053 if (unlikely(!flow
)) {
1055 goto err_unlock_ovs
;
1058 /* Update actions, if present. */
1060 old_acts
= ovsl_dereference(flow
->sf_acts
);
1061 rcu_assign_pointer(flow
->sf_acts
, acts
);
1063 if (unlikely(reply
)) {
1064 error
= ovs_flow_cmd_fill_info(flow
,
1065 ovs_header
->dp_ifindex
,
1066 reply
, info
->snd_portid
,
1072 /* Could not alloc without acts before locking. */
1073 reply
= ovs_flow_cmd_build_info(flow
, ovs_header
->dp_ifindex
,
1074 info
, OVS_FLOW_CMD_NEW
, false);
1075 if (unlikely(IS_ERR(reply
))) {
1076 error
= PTR_ERR(reply
);
1077 goto err_unlock_ovs
;
1082 if (a
[OVS_FLOW_ATTR_CLEAR
])
1083 ovs_flow_stats_clear(flow
);
1087 ovs_notify(&dp_flow_genl_family
, reply
, info
);
1089 ovs_nla_free_flow_actions(old_acts
);
1102 static int ovs_flow_cmd_get(struct sk_buff
*skb
, struct genl_info
*info
)
1104 struct nlattr
**a
= info
->attrs
;
1105 struct ovs_header
*ovs_header
= info
->userhdr
;
1106 struct sw_flow_key key
;
1107 struct sk_buff
*reply
;
1108 struct sw_flow
*flow
;
1109 struct datapath
*dp
;
1110 struct sw_flow_match match
;
1113 if (!a
[OVS_FLOW_ATTR_KEY
]) {
1114 OVS_NLERR("Flow get message rejected, Key attribute missing.\n");
1118 ovs_match_init(&match
, &key
, NULL
);
1119 err
= ovs_nla_get_match(&match
, a
[OVS_FLOW_ATTR_KEY
], NULL
);
1124 dp
= get_dp(sock_net(skb
->sk
), ovs_header
->dp_ifindex
);
1130 flow
= ovs_flow_tbl_lookup_exact(&dp
->table
, &match
);
1136 reply
= ovs_flow_cmd_build_info(flow
, ovs_header
->dp_ifindex
, info
,
1137 OVS_FLOW_CMD_NEW
, true);
1138 if (IS_ERR(reply
)) {
1139 err
= PTR_ERR(reply
);
1144 return genlmsg_reply(reply
, info
);
1150 static int ovs_flow_cmd_del(struct sk_buff
*skb
, struct genl_info
*info
)
1152 struct nlattr
**a
= info
->attrs
;
1153 struct ovs_header
*ovs_header
= info
->userhdr
;
1154 struct sw_flow_key key
;
1155 struct sk_buff
*reply
;
1156 struct sw_flow
*flow
;
1157 struct datapath
*dp
;
1158 struct sw_flow_match match
;
1161 if (likely(a
[OVS_FLOW_ATTR_KEY
])) {
1162 ovs_match_init(&match
, &key
, NULL
);
1163 err
= ovs_nla_get_match(&match
, a
[OVS_FLOW_ATTR_KEY
], NULL
);
1169 dp
= get_dp(sock_net(skb
->sk
), ovs_header
->dp_ifindex
);
1170 if (unlikely(!dp
)) {
1175 if (unlikely(!a
[OVS_FLOW_ATTR_KEY
])) {
1176 err
= ovs_flow_tbl_flush(&dp
->table
);
1180 flow
= ovs_flow_tbl_lookup_exact(&dp
->table
, &match
);
1181 if (unlikely(!flow
)) {
1186 ovs_flow_tbl_remove(&dp
->table
, flow
);
1189 reply
= ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force
*) flow
->sf_acts
,
1191 if (likely(reply
)) {
1192 if (likely(!IS_ERR(reply
))) {
1193 rcu_read_lock(); /*To keep RCU checker happy. */
1194 err
= ovs_flow_cmd_fill_info(flow
, ovs_header
->dp_ifindex
,
1195 reply
, info
->snd_portid
,
1201 ovs_notify(&dp_flow_genl_family
, reply
, info
);
1203 netlink_set_err(sock_net(skb
->sk
)->genl_sock
, 0, 0, PTR_ERR(reply
));
1207 ovs_flow_free(flow
, true);
1214 static int ovs_flow_cmd_dump(struct sk_buff
*skb
, struct netlink_callback
*cb
)
1216 struct ovs_header
*ovs_header
= genlmsg_data(nlmsg_data(cb
->nlh
));
1217 struct table_instance
*ti
;
1218 struct datapath
*dp
;
1221 dp
= get_dp_rcu(sock_net(skb
->sk
), ovs_header
->dp_ifindex
);
1227 ti
= rcu_dereference(dp
->table
.ti
);
1229 struct sw_flow
*flow
;
1232 bucket
= cb
->args
[0];
1234 flow
= ovs_flow_tbl_dump_next(ti
, &bucket
, &obj
);
1238 if (ovs_flow_cmd_fill_info(flow
, ovs_header
->dp_ifindex
, skb
,
1239 NETLINK_CB(cb
->skb
).portid
,
1240 cb
->nlh
->nlmsg_seq
, NLM_F_MULTI
,
1241 OVS_FLOW_CMD_NEW
) < 0)
1244 cb
->args
[0] = bucket
;
1251 static const struct nla_policy flow_policy
[OVS_FLOW_ATTR_MAX
+ 1] = {
1252 [OVS_FLOW_ATTR_KEY
] = { .type
= NLA_NESTED
},
1253 [OVS_FLOW_ATTR_ACTIONS
] = { .type
= NLA_NESTED
},
1254 [OVS_FLOW_ATTR_CLEAR
] = { .type
= NLA_FLAG
},
1257 static const struct genl_ops dp_flow_genl_ops
[] = {
1258 { .cmd
= OVS_FLOW_CMD_NEW
,
1259 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1260 .policy
= flow_policy
,
1261 .doit
= ovs_flow_cmd_new
1263 { .cmd
= OVS_FLOW_CMD_DEL
,
1264 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1265 .policy
= flow_policy
,
1266 .doit
= ovs_flow_cmd_del
1268 { .cmd
= OVS_FLOW_CMD_GET
,
1269 .flags
= 0, /* OK for unprivileged users. */
1270 .policy
= flow_policy
,
1271 .doit
= ovs_flow_cmd_get
,
1272 .dumpit
= ovs_flow_cmd_dump
1274 { .cmd
= OVS_FLOW_CMD_SET
,
1275 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1276 .policy
= flow_policy
,
1277 .doit
= ovs_flow_cmd_set
,
1281 static struct genl_family dp_flow_genl_family
= {
1282 .id
= GENL_ID_GENERATE
,
1283 .hdrsize
= sizeof(struct ovs_header
),
1284 .name
= OVS_FLOW_FAMILY
,
1285 .version
= OVS_FLOW_VERSION
,
1286 .maxattr
= OVS_FLOW_ATTR_MAX
,
1288 .parallel_ops
= true,
1289 .ops
= dp_flow_genl_ops
,
1290 .n_ops
= ARRAY_SIZE(dp_flow_genl_ops
),
1291 .mcgrps
= &ovs_dp_flow_multicast_group
,
1295 static size_t ovs_dp_cmd_msg_size(void)
1297 size_t msgsize
= NLMSG_ALIGN(sizeof(struct ovs_header
));
1299 msgsize
+= nla_total_size(IFNAMSIZ
);
1300 msgsize
+= nla_total_size(sizeof(struct ovs_dp_stats
));
1301 msgsize
+= nla_total_size(sizeof(struct ovs_dp_megaflow_stats
));
1302 msgsize
+= nla_total_size(sizeof(u32
)); /* OVS_DP_ATTR_USER_FEATURES */
1307 /* Called with ovs_mutex or RCU read lock. */
1308 static int ovs_dp_cmd_fill_info(struct datapath
*dp
, struct sk_buff
*skb
,
1309 u32 portid
, u32 seq
, u32 flags
, u8 cmd
)
1311 struct ovs_header
*ovs_header
;
1312 struct ovs_dp_stats dp_stats
;
1313 struct ovs_dp_megaflow_stats dp_megaflow_stats
;
1316 ovs_header
= genlmsg_put(skb
, portid
, seq
, &dp_datapath_genl_family
,
1321 ovs_header
->dp_ifindex
= get_dpifindex(dp
);
1323 err
= nla_put_string(skb
, OVS_DP_ATTR_NAME
, ovs_dp_name(dp
));
1325 goto nla_put_failure
;
1327 get_dp_stats(dp
, &dp_stats
, &dp_megaflow_stats
);
1328 if (nla_put(skb
, OVS_DP_ATTR_STATS
, sizeof(struct ovs_dp_stats
),
1330 goto nla_put_failure
;
1332 if (nla_put(skb
, OVS_DP_ATTR_MEGAFLOW_STATS
,
1333 sizeof(struct ovs_dp_megaflow_stats
),
1334 &dp_megaflow_stats
))
1335 goto nla_put_failure
;
1337 if (nla_put_u32(skb
, OVS_DP_ATTR_USER_FEATURES
, dp
->user_features
))
1338 goto nla_put_failure
;
1340 return genlmsg_end(skb
, ovs_header
);
1343 genlmsg_cancel(skb
, ovs_header
);
1348 static struct sk_buff
*ovs_dp_cmd_alloc_info(struct genl_info
*info
)
1350 return genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info
, GFP_KERNEL
);
1353 /* Called with rcu_read_lock or ovs_mutex. */
1354 static struct datapath
*lookup_datapath(struct net
*net
,
1355 const struct ovs_header
*ovs_header
,
1356 struct nlattr
*a
[OVS_DP_ATTR_MAX
+ 1])
1358 struct datapath
*dp
;
1360 if (!a
[OVS_DP_ATTR_NAME
])
1361 dp
= get_dp(net
, ovs_header
->dp_ifindex
);
1363 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
;
1368 return dp
? dp
: ERR_PTR(-ENODEV
);
1371 static void ovs_dp_reset_user_features(struct sk_buff
*skb
, struct genl_info
*info
)
1373 struct datapath
*dp
;
1375 dp
= lookup_datapath(sock_net(skb
->sk
), info
->userhdr
, info
->attrs
);
1379 WARN(dp
->user_features
, "Dropping previously announced user features\n");
1380 dp
->user_features
= 0;
1383 static void ovs_dp_change(struct datapath
*dp
, struct nlattr
*a
[])
1385 if (a
[OVS_DP_ATTR_USER_FEATURES
])
1386 dp
->user_features
= nla_get_u32(a
[OVS_DP_ATTR_USER_FEATURES
]);
1389 static int ovs_dp_cmd_new(struct sk_buff
*skb
, struct genl_info
*info
)
1391 struct nlattr
**a
= info
->attrs
;
1392 struct vport_parms parms
;
1393 struct sk_buff
*reply
;
1394 struct datapath
*dp
;
1395 struct vport
*vport
;
1396 struct ovs_net
*ovs_net
;
1400 if (!a
[OVS_DP_ATTR_NAME
] || !a
[OVS_DP_ATTR_UPCALL_PID
])
1403 reply
= ovs_dp_cmd_alloc_info(info
);
1408 dp
= kzalloc(sizeof(*dp
), GFP_KERNEL
);
1410 goto err_free_reply
;
1412 ovs_dp_set_net(dp
, hold_net(sock_net(skb
->sk
)));
1414 /* Allocate table. */
1415 err
= ovs_flow_tbl_init(&dp
->table
);
1419 dp
->stats_percpu
= netdev_alloc_pcpu_stats(struct dp_stats_percpu
);
1420 if (!dp
->stats_percpu
) {
1422 goto err_destroy_table
;
1425 dp
->ports
= kmalloc(DP_VPORT_HASH_BUCKETS
* sizeof(struct hlist_head
),
1429 goto err_destroy_percpu
;
1432 for (i
= 0; i
< DP_VPORT_HASH_BUCKETS
; i
++)
1433 INIT_HLIST_HEAD(&dp
->ports
[i
]);
1435 /* Set up our datapath device. */
1436 parms
.name
= nla_data(a
[OVS_DP_ATTR_NAME
]);
1437 parms
.type
= OVS_VPORT_TYPE_INTERNAL
;
1438 parms
.options
= NULL
;
1440 parms
.port_no
= OVSP_LOCAL
;
1441 parms
.upcall_portids
= a
[OVS_DP_ATTR_UPCALL_PID
];
1443 ovs_dp_change(dp
, a
);
1445 /* So far only local changes have been made, now need the lock. */
1448 vport
= new_vport(&parms
);
1449 if (IS_ERR(vport
)) {
1450 err
= PTR_ERR(vport
);
1454 if (err
== -EEXIST
) {
1455 /* An outdated user space instance that does not understand
1456 * the concept of user_features has attempted to create a new
1457 * datapath and is likely to reuse it. Drop all user features.
1459 if (info
->genlhdr
->version
< OVS_DP_VER_FEATURES
)
1460 ovs_dp_reset_user_features(skb
, info
);
1463 goto err_destroy_ports_array
;
1466 err
= ovs_dp_cmd_fill_info(dp
, reply
, info
->snd_portid
,
1467 info
->snd_seq
, 0, OVS_DP_CMD_NEW
);
1470 ovs_net
= net_generic(ovs_dp_get_net(dp
), ovs_net_id
);
1471 list_add_tail_rcu(&dp
->list_node
, &ovs_net
->dps
);
1475 ovs_notify(&dp_datapath_genl_family
, reply
, info
);
1478 err_destroy_ports_array
:
1482 free_percpu(dp
->stats_percpu
);
1484 ovs_flow_tbl_destroy(&dp
->table
);
1486 release_net(ovs_dp_get_net(dp
));
1494 /* Called with ovs_mutex. */
1495 static void __dp_destroy(struct datapath
*dp
)
1499 for (i
= 0; i
< DP_VPORT_HASH_BUCKETS
; i
++) {
1500 struct vport
*vport
;
1501 struct hlist_node
*n
;
1503 hlist_for_each_entry_safe(vport
, n
, &dp
->ports
[i
], dp_hash_node
)
1504 if (vport
->port_no
!= OVSP_LOCAL
)
1505 ovs_dp_detach_port(vport
);
1508 list_del_rcu(&dp
->list_node
);
1510 /* OVSP_LOCAL is datapath internal port. We need to make sure that
1511 * all ports in datapath are destroyed first before freeing datapath.
1513 ovs_dp_detach_port(ovs_vport_ovsl(dp
, OVSP_LOCAL
));
1515 /* RCU destroy the flow table */
1516 call_rcu(&dp
->rcu
, destroy_dp_rcu
);
1519 static int ovs_dp_cmd_del(struct sk_buff
*skb
, struct genl_info
*info
)
1521 struct sk_buff
*reply
;
1522 struct datapath
*dp
;
1525 reply
= ovs_dp_cmd_alloc_info(info
);
1530 dp
= lookup_datapath(sock_net(skb
->sk
), info
->userhdr
, info
->attrs
);
1533 goto err_unlock_free
;
1535 err
= ovs_dp_cmd_fill_info(dp
, reply
, info
->snd_portid
,
1536 info
->snd_seq
, 0, OVS_DP_CMD_DEL
);
1542 ovs_notify(&dp_datapath_genl_family
, reply
, info
);
1552 static int ovs_dp_cmd_set(struct sk_buff
*skb
, struct genl_info
*info
)
1554 struct sk_buff
*reply
;
1555 struct datapath
*dp
;
1558 reply
= ovs_dp_cmd_alloc_info(info
);
1563 dp
= lookup_datapath(sock_net(skb
->sk
), info
->userhdr
, info
->attrs
);
1566 goto err_unlock_free
;
1568 ovs_dp_change(dp
, info
->attrs
);
1570 err
= ovs_dp_cmd_fill_info(dp
, reply
, info
->snd_portid
,
1571 info
->snd_seq
, 0, OVS_DP_CMD_NEW
);
1575 ovs_notify(&dp_datapath_genl_family
, reply
, info
);
1585 static int ovs_dp_cmd_get(struct sk_buff
*skb
, struct genl_info
*info
)
1587 struct sk_buff
*reply
;
1588 struct datapath
*dp
;
1591 reply
= ovs_dp_cmd_alloc_info(info
);
1596 dp
= lookup_datapath(sock_net(skb
->sk
), info
->userhdr
, info
->attrs
);
1599 goto err_unlock_free
;
1601 err
= ovs_dp_cmd_fill_info(dp
, reply
, info
->snd_portid
,
1602 info
->snd_seq
, 0, OVS_DP_CMD_NEW
);
1606 return genlmsg_reply(reply
, info
);
1614 static int ovs_dp_cmd_dump(struct sk_buff
*skb
, struct netlink_callback
*cb
)
1616 struct ovs_net
*ovs_net
= net_generic(sock_net(skb
->sk
), ovs_net_id
);
1617 struct datapath
*dp
;
1618 int skip
= cb
->args
[0];
1622 list_for_each_entry_rcu(dp
, &ovs_net
->dps
, list_node
) {
1624 ovs_dp_cmd_fill_info(dp
, skb
, NETLINK_CB(cb
->skb
).portid
,
1625 cb
->nlh
->nlmsg_seq
, NLM_F_MULTI
,
1626 OVS_DP_CMD_NEW
) < 0)
1637 static const struct nla_policy datapath_policy
[OVS_DP_ATTR_MAX
+ 1] = {
1638 [OVS_DP_ATTR_NAME
] = { .type
= NLA_NUL_STRING
, .len
= IFNAMSIZ
- 1 },
1639 [OVS_DP_ATTR_UPCALL_PID
] = { .type
= NLA_U32
},
1640 [OVS_DP_ATTR_USER_FEATURES
] = { .type
= NLA_U32
},
1643 static const struct genl_ops dp_datapath_genl_ops
[] = {
1644 { .cmd
= OVS_DP_CMD_NEW
,
1645 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1646 .policy
= datapath_policy
,
1647 .doit
= ovs_dp_cmd_new
1649 { .cmd
= OVS_DP_CMD_DEL
,
1650 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1651 .policy
= datapath_policy
,
1652 .doit
= ovs_dp_cmd_del
1654 { .cmd
= OVS_DP_CMD_GET
,
1655 .flags
= 0, /* OK for unprivileged users. */
1656 .policy
= datapath_policy
,
1657 .doit
= ovs_dp_cmd_get
,
1658 .dumpit
= ovs_dp_cmd_dump
1660 { .cmd
= OVS_DP_CMD_SET
,
1661 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1662 .policy
= datapath_policy
,
1663 .doit
= ovs_dp_cmd_set
,
1667 static struct genl_family dp_datapath_genl_family
= {
1668 .id
= GENL_ID_GENERATE
,
1669 .hdrsize
= sizeof(struct ovs_header
),
1670 .name
= OVS_DATAPATH_FAMILY
,
1671 .version
= OVS_DATAPATH_VERSION
,
1672 .maxattr
= OVS_DP_ATTR_MAX
,
1674 .parallel_ops
= true,
1675 .ops
= dp_datapath_genl_ops
,
1676 .n_ops
= ARRAY_SIZE(dp_datapath_genl_ops
),
1677 .mcgrps
= &ovs_dp_datapath_multicast_group
,
1681 /* Called with ovs_mutex or RCU read lock. */
1682 static int ovs_vport_cmd_fill_info(struct vport
*vport
, struct sk_buff
*skb
,
1683 u32 portid
, u32 seq
, u32 flags
, u8 cmd
)
1685 struct ovs_header
*ovs_header
;
1686 struct ovs_vport_stats vport_stats
;
1689 ovs_header
= genlmsg_put(skb
, portid
, seq
, &dp_vport_genl_family
,
1694 ovs_header
->dp_ifindex
= get_dpifindex(vport
->dp
);
1696 if (nla_put_u32(skb
, OVS_VPORT_ATTR_PORT_NO
, vport
->port_no
) ||
1697 nla_put_u32(skb
, OVS_VPORT_ATTR_TYPE
, vport
->ops
->type
) ||
1698 nla_put_string(skb
, OVS_VPORT_ATTR_NAME
,
1699 vport
->ops
->get_name(vport
)))
1700 goto nla_put_failure
;
1702 ovs_vport_get_stats(vport
, &vport_stats
);
1703 if (nla_put(skb
, OVS_VPORT_ATTR_STATS
, sizeof(struct ovs_vport_stats
),
1705 goto nla_put_failure
;
1707 if (ovs_vport_get_upcall_portids(vport
, skb
))
1708 goto nla_put_failure
;
1710 err
= ovs_vport_get_options(vport
, skb
);
1711 if (err
== -EMSGSIZE
)
1714 return genlmsg_end(skb
, ovs_header
);
1719 genlmsg_cancel(skb
, ovs_header
);
1723 static struct sk_buff
*ovs_vport_cmd_alloc_info(void)
1725 return nlmsg_new(NLMSG_DEFAULT_SIZE
, GFP_KERNEL
);
1728 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1729 struct sk_buff
*ovs_vport_cmd_build_info(struct vport
*vport
, u32 portid
,
1732 struct sk_buff
*skb
;
1735 skb
= nlmsg_new(NLMSG_DEFAULT_SIZE
, GFP_ATOMIC
);
1737 return ERR_PTR(-ENOMEM
);
1739 retval
= ovs_vport_cmd_fill_info(vport
, skb
, portid
, seq
, 0, cmd
);
1745 /* Called with ovs_mutex or RCU read lock. */
1746 static struct vport
*lookup_vport(struct net
*net
,
1747 const struct ovs_header
*ovs_header
,
1748 struct nlattr
*a
[OVS_VPORT_ATTR_MAX
+ 1])
1750 struct datapath
*dp
;
1751 struct vport
*vport
;
1753 if (a
[OVS_VPORT_ATTR_NAME
]) {
1754 vport
= ovs_vport_locate(net
, nla_data(a
[OVS_VPORT_ATTR_NAME
]));
1756 return ERR_PTR(-ENODEV
);
1757 if (ovs_header
->dp_ifindex
&&
1758 ovs_header
->dp_ifindex
!= get_dpifindex(vport
->dp
))
1759 return ERR_PTR(-ENODEV
);
1761 } else if (a
[OVS_VPORT_ATTR_PORT_NO
]) {
1762 u32 port_no
= nla_get_u32(a
[OVS_VPORT_ATTR_PORT_NO
]);
1764 if (port_no
>= DP_MAX_PORTS
)
1765 return ERR_PTR(-EFBIG
);
1767 dp
= get_dp(net
, ovs_header
->dp_ifindex
);
1769 return ERR_PTR(-ENODEV
);
1771 vport
= ovs_vport_ovsl_rcu(dp
, port_no
);
1773 return ERR_PTR(-ENODEV
);
1776 return ERR_PTR(-EINVAL
);
1779 static int ovs_vport_cmd_new(struct sk_buff
*skb
, struct genl_info
*info
)
1781 struct nlattr
**a
= info
->attrs
;
1782 struct ovs_header
*ovs_header
= info
->userhdr
;
1783 struct vport_parms parms
;
1784 struct sk_buff
*reply
;
1785 struct vport
*vport
;
1786 struct datapath
*dp
;
1790 if (!a
[OVS_VPORT_ATTR_NAME
] || !a
[OVS_VPORT_ATTR_TYPE
] ||
1791 !a
[OVS_VPORT_ATTR_UPCALL_PID
])
1794 port_no
= a
[OVS_VPORT_ATTR_PORT_NO
]
1795 ? nla_get_u32(a
[OVS_VPORT_ATTR_PORT_NO
]) : 0;
1796 if (port_no
>= DP_MAX_PORTS
)
1799 reply
= ovs_vport_cmd_alloc_info();
1805 dp
= get_dp(sock_net(skb
->sk
), ovs_header
->dp_ifindex
);
1808 goto exit_unlock_free
;
1811 vport
= ovs_vport_ovsl(dp
, port_no
);
1814 goto exit_unlock_free
;
1816 for (port_no
= 1; ; port_no
++) {
1817 if (port_no
>= DP_MAX_PORTS
) {
1819 goto exit_unlock_free
;
1821 vport
= ovs_vport_ovsl(dp
, port_no
);
1827 parms
.name
= nla_data(a
[OVS_VPORT_ATTR_NAME
]);
1828 parms
.type
= nla_get_u32(a
[OVS_VPORT_ATTR_TYPE
]);
1829 parms
.options
= a
[OVS_VPORT_ATTR_OPTIONS
];
1831 parms
.port_no
= port_no
;
1832 parms
.upcall_portids
= a
[OVS_VPORT_ATTR_UPCALL_PID
];
1834 vport
= new_vport(&parms
);
1835 err
= PTR_ERR(vport
);
1836 if (IS_ERR(vport
)) {
1839 goto exit_unlock_free
;
1842 err
= ovs_vport_cmd_fill_info(vport
, reply
, info
->snd_portid
,
1843 info
->snd_seq
, 0, OVS_VPORT_CMD_NEW
);
1847 ovs_notify(&dp_vport_genl_family
, reply
, info
);
1856 static int ovs_vport_cmd_set(struct sk_buff
*skb
, struct genl_info
*info
)
1858 struct nlattr
**a
= info
->attrs
;
1859 struct sk_buff
*reply
;
1860 struct vport
*vport
;
1863 reply
= ovs_vport_cmd_alloc_info();
1868 vport
= lookup_vport(sock_net(skb
->sk
), info
->userhdr
, a
);
1869 err
= PTR_ERR(vport
);
1871 goto exit_unlock_free
;
1873 if (a
[OVS_VPORT_ATTR_TYPE
] &&
1874 nla_get_u32(a
[OVS_VPORT_ATTR_TYPE
]) != vport
->ops
->type
) {
1876 goto exit_unlock_free
;
1879 if (a
[OVS_VPORT_ATTR_OPTIONS
]) {
1880 err
= ovs_vport_set_options(vport
, a
[OVS_VPORT_ATTR_OPTIONS
]);
1882 goto exit_unlock_free
;
1886 if (a
[OVS_VPORT_ATTR_UPCALL_PID
]) {
1887 struct nlattr
*ids
= a
[OVS_VPORT_ATTR_UPCALL_PID
];
1889 err
= ovs_vport_set_upcall_portids(vport
, ids
);
1891 goto exit_unlock_free
;
1894 err
= ovs_vport_cmd_fill_info(vport
, reply
, info
->snd_portid
,
1895 info
->snd_seq
, 0, OVS_VPORT_CMD_NEW
);
1899 ovs_notify(&dp_vport_genl_family
, reply
, info
);
1908 static int ovs_vport_cmd_del(struct sk_buff
*skb
, struct genl_info
*info
)
1910 struct nlattr
**a
= info
->attrs
;
1911 struct sk_buff
*reply
;
1912 struct vport
*vport
;
1915 reply
= ovs_vport_cmd_alloc_info();
1920 vport
= lookup_vport(sock_net(skb
->sk
), info
->userhdr
, a
);
1921 err
= PTR_ERR(vport
);
1923 goto exit_unlock_free
;
1925 if (vport
->port_no
== OVSP_LOCAL
) {
1927 goto exit_unlock_free
;
1930 err
= ovs_vport_cmd_fill_info(vport
, reply
, info
->snd_portid
,
1931 info
->snd_seq
, 0, OVS_VPORT_CMD_DEL
);
1933 ovs_dp_detach_port(vport
);
1936 ovs_notify(&dp_vport_genl_family
, reply
, info
);
1945 static int ovs_vport_cmd_get(struct sk_buff
*skb
, struct genl_info
*info
)
1947 struct nlattr
**a
= info
->attrs
;
1948 struct ovs_header
*ovs_header
= info
->userhdr
;
1949 struct sk_buff
*reply
;
1950 struct vport
*vport
;
1953 reply
= ovs_vport_cmd_alloc_info();
1958 vport
= lookup_vport(sock_net(skb
->sk
), ovs_header
, a
);
1959 err
= PTR_ERR(vport
);
1961 goto exit_unlock_free
;
1962 err
= ovs_vport_cmd_fill_info(vport
, reply
, info
->snd_portid
,
1963 info
->snd_seq
, 0, OVS_VPORT_CMD_NEW
);
1967 return genlmsg_reply(reply
, info
);
1975 static int ovs_vport_cmd_dump(struct sk_buff
*skb
, struct netlink_callback
*cb
)
1977 struct ovs_header
*ovs_header
= genlmsg_data(nlmsg_data(cb
->nlh
));
1978 struct datapath
*dp
;
1979 int bucket
= cb
->args
[0], skip
= cb
->args
[1];
1983 dp
= get_dp_rcu(sock_net(skb
->sk
), ovs_header
->dp_ifindex
);
1988 for (i
= bucket
; i
< DP_VPORT_HASH_BUCKETS
; i
++) {
1989 struct vport
*vport
;
1992 hlist_for_each_entry_rcu(vport
, &dp
->ports
[i
], dp_hash_node
) {
1994 ovs_vport_cmd_fill_info(vport
, skb
,
1995 NETLINK_CB(cb
->skb
).portid
,
1998 OVS_VPORT_CMD_NEW
) < 0)
2014 static const struct nla_policy vport_policy
[OVS_VPORT_ATTR_MAX
+ 1] = {
2015 [OVS_VPORT_ATTR_NAME
] = { .type
= NLA_NUL_STRING
, .len
= IFNAMSIZ
- 1 },
2016 [OVS_VPORT_ATTR_STATS
] = { .len
= sizeof(struct ovs_vport_stats
) },
2017 [OVS_VPORT_ATTR_PORT_NO
] = { .type
= NLA_U32
},
2018 [OVS_VPORT_ATTR_TYPE
] = { .type
= NLA_U32
},
2019 [OVS_VPORT_ATTR_UPCALL_PID
] = { .type
= NLA_U32
},
2020 [OVS_VPORT_ATTR_OPTIONS
] = { .type
= NLA_NESTED
},
2023 static const struct genl_ops dp_vport_genl_ops
[] = {
2024 { .cmd
= OVS_VPORT_CMD_NEW
,
2025 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
2026 .policy
= vport_policy
,
2027 .doit
= ovs_vport_cmd_new
2029 { .cmd
= OVS_VPORT_CMD_DEL
,
2030 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
2031 .policy
= vport_policy
,
2032 .doit
= ovs_vport_cmd_del
2034 { .cmd
= OVS_VPORT_CMD_GET
,
2035 .flags
= 0, /* OK for unprivileged users. */
2036 .policy
= vport_policy
,
2037 .doit
= ovs_vport_cmd_get
,
2038 .dumpit
= ovs_vport_cmd_dump
2040 { .cmd
= OVS_VPORT_CMD_SET
,
2041 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
2042 .policy
= vport_policy
,
2043 .doit
= ovs_vport_cmd_set
,
2047 struct genl_family dp_vport_genl_family
= {
2048 .id
= GENL_ID_GENERATE
,
2049 .hdrsize
= sizeof(struct ovs_header
),
2050 .name
= OVS_VPORT_FAMILY
,
2051 .version
= OVS_VPORT_VERSION
,
2052 .maxattr
= OVS_VPORT_ATTR_MAX
,
2054 .parallel_ops
= true,
2055 .ops
= dp_vport_genl_ops
,
2056 .n_ops
= ARRAY_SIZE(dp_vport_genl_ops
),
2057 .mcgrps
= &ovs_dp_vport_multicast_group
,
2061 static struct genl_family
* const dp_genl_families
[] = {
2062 &dp_datapath_genl_family
,
2063 &dp_vport_genl_family
,
2064 &dp_flow_genl_family
,
2065 &dp_packet_genl_family
,
2068 static void dp_unregister_genl(int n_families
)
2072 for (i
= 0; i
< n_families
; i
++)
2073 genl_unregister_family(dp_genl_families
[i
]);
2076 static int dp_register_genl(void)
2081 for (i
= 0; i
< ARRAY_SIZE(dp_genl_families
); i
++) {
2083 err
= genl_register_family(dp_genl_families
[i
]);
2091 dp_unregister_genl(i
);
2095 static int __net_init
ovs_init_net(struct net
*net
)
2097 struct ovs_net
*ovs_net
= net_generic(net
, ovs_net_id
);
2099 INIT_LIST_HEAD(&ovs_net
->dps
);
2100 INIT_WORK(&ovs_net
->dp_notify_work
, ovs_dp_notify_wq
);
2104 static void __net_exit
ovs_exit_net(struct net
*net
)
2106 struct datapath
*dp
, *dp_next
;
2107 struct ovs_net
*ovs_net
= net_generic(net
, ovs_net_id
);
2110 list_for_each_entry_safe(dp
, dp_next
, &ovs_net
->dps
, list_node
)
2114 cancel_work_sync(&ovs_net
->dp_notify_work
);
2117 static struct pernet_operations ovs_net_ops
= {
2118 .init
= ovs_init_net
,
2119 .exit
= ovs_exit_net
,
2121 .size
= sizeof(struct ovs_net
),
2124 static int __init
dp_init(void)
2128 BUILD_BUG_ON(sizeof(struct ovs_skb_cb
) > FIELD_SIZEOF(struct sk_buff
, cb
));
2130 pr_info("Open vSwitch switching datapath\n");
2132 err
= action_fifos_init();
2136 err
= ovs_internal_dev_rtnl_link_register();
2138 goto error_action_fifos_exit
;
2140 err
= ovs_flow_init();
2142 goto error_unreg_rtnl_link
;
2144 err
= ovs_vport_init();
2146 goto error_flow_exit
;
2148 err
= register_pernet_device(&ovs_net_ops
);
2150 goto error_vport_exit
;
2152 err
= register_netdevice_notifier(&ovs_dp_device_notifier
);
2154 goto error_netns_exit
;
2156 err
= ovs_netdev_init();
2158 goto error_unreg_notifier
;
2160 err
= dp_register_genl();
2162 goto error_unreg_netdev
;
2168 error_unreg_notifier
:
2169 unregister_netdevice_notifier(&ovs_dp_device_notifier
);
2171 unregister_pernet_device(&ovs_net_ops
);
2176 error_unreg_rtnl_link
:
2177 ovs_internal_dev_rtnl_link_unregister();
2178 error_action_fifos_exit
:
2179 action_fifos_exit();
2184 static void dp_cleanup(void)
2186 dp_unregister_genl(ARRAY_SIZE(dp_genl_families
));
2188 unregister_netdevice_notifier(&ovs_dp_device_notifier
);
2189 unregister_pernet_device(&ovs_net_ops
);
2193 ovs_internal_dev_rtnl_link_unregister();
2194 action_fifos_exit();
2197 module_init(dp_init
);
2198 module_exit(dp_cleanup
);
2200 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2201 MODULE_LICENSE("GPL");