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
), group
);
89 static void ovs_notify(struct genl_family
*family
,
90 struct sk_buff
*skb
, struct genl_info
*info
)
92 genl_notify(family
, skb
, genl_info_net(info
), info
->snd_portid
,
93 0, info
->nlhdr
, GFP_KERNEL
);
99 * All writes e.g. Writes to device state (add/remove datapath, port, set
100 * operations on vports, etc.), Writes to other state (flow table
101 * modifications, set miscellaneous datapath parameters, etc.) are protected
104 * Reads are protected by RCU.
106 * There are a few special cases (mostly stats) that have their own
107 * synchronization but they nest under all of above and don't interact with
110 * The RTNL lock nests inside ovs_mutex.
113 static DEFINE_MUTEX(ovs_mutex
);
117 mutex_lock(&ovs_mutex
);
120 void ovs_unlock(void)
122 mutex_unlock(&ovs_mutex
);
125 #ifdef CONFIG_LOCKDEP
126 int lockdep_ovsl_is_held(void)
129 return lockdep_is_held(&ovs_mutex
);
133 EXPORT_SYMBOL_GPL(lockdep_ovsl_is_held
);
136 static struct vport
*new_vport(const struct vport_parms
*);
137 static int queue_gso_packets(struct datapath
*dp
, struct sk_buff
*,
138 const struct sw_flow_key
*,
139 const struct dp_upcall_info
*);
140 static int queue_userspace_packet(struct datapath
*dp
, struct sk_buff
*,
141 const struct sw_flow_key
*,
142 const struct dp_upcall_info
*);
144 /* Must be called with rcu_read_lock. */
145 static struct datapath
*get_dp_rcu(struct net
*net
, int dp_ifindex
)
147 struct net_device
*dev
= dev_get_by_index_rcu(net
, dp_ifindex
);
150 struct vport
*vport
= ovs_internal_dev_get_vport(dev
);
158 /* The caller must hold either ovs_mutex or rcu_read_lock to keep the
159 * returned dp pointer valid.
161 static inline struct datapath
*get_dp(struct net
*net
, int dp_ifindex
)
165 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_ovsl_is_held());
167 dp
= get_dp_rcu(net
, dp_ifindex
);
173 /* Must be called with rcu_read_lock or ovs_mutex. */
174 const char *ovs_dp_name(const struct datapath
*dp
)
176 struct vport
*vport
= ovs_vport_ovsl_rcu(dp
, OVSP_LOCAL
);
177 return vport
->ops
->get_name(vport
);
180 static int get_dpifindex(const struct datapath
*dp
)
187 local
= ovs_vport_rcu(dp
, OVSP_LOCAL
);
189 ifindex
= netdev_vport_priv(local
)->dev
->ifindex
;
198 static void destroy_dp_rcu(struct rcu_head
*rcu
)
200 struct datapath
*dp
= container_of(rcu
, struct datapath
, rcu
);
202 ovs_flow_tbl_destroy(&dp
->table
);
203 free_percpu(dp
->stats_percpu
);
204 release_net(ovs_dp_get_net(dp
));
209 static struct hlist_head
*vport_hash_bucket(const struct datapath
*dp
,
212 return &dp
->ports
[port_no
& (DP_VPORT_HASH_BUCKETS
- 1)];
215 /* Called with ovs_mutex or RCU read lock. */
216 struct vport
*ovs_lookup_vport(const struct datapath
*dp
, u16 port_no
)
219 struct hlist_head
*head
;
221 head
= vport_hash_bucket(dp
, port_no
);
222 hlist_for_each_entry_rcu(vport
, head
, dp_hash_node
) {
223 if (vport
->port_no
== port_no
)
229 /* Called with ovs_mutex. */
230 static struct vport
*new_vport(const struct vport_parms
*parms
)
234 vport
= ovs_vport_add(parms
);
235 if (!IS_ERR(vport
)) {
236 struct datapath
*dp
= parms
->dp
;
237 struct hlist_head
*head
= vport_hash_bucket(dp
, vport
->port_no
);
239 hlist_add_head_rcu(&vport
->dp_hash_node
, head
);
244 void ovs_dp_detach_port(struct vport
*p
)
248 /* First drop references to device. */
249 hlist_del_rcu(&p
->dp_hash_node
);
251 /* Then destroy it. */
255 /* Must be called with rcu_read_lock. */
256 void ovs_dp_process_packet(struct sk_buff
*skb
, struct sw_flow_key
*key
)
258 const struct vport
*p
= OVS_CB(skb
)->input_vport
;
259 struct datapath
*dp
= p
->dp
;
260 struct sw_flow
*flow
;
261 struct sw_flow_actions
*sf_acts
;
262 struct dp_stats_percpu
*stats
;
266 stats
= this_cpu_ptr(dp
->stats_percpu
);
269 flow
= ovs_flow_tbl_lookup_stats(&dp
->table
, key
, &n_mask_hit
);
270 if (unlikely(!flow
)) {
271 struct dp_upcall_info upcall
;
274 upcall
.cmd
= OVS_PACKET_CMD_MISS
;
275 upcall
.userdata
= NULL
;
276 upcall
.portid
= ovs_vport_find_upcall_portid(p
, skb
);
277 upcall
.egress_tun_info
= NULL
;
278 error
= ovs_dp_upcall(dp
, skb
, key
, &upcall
);
283 stats_counter
= &stats
->n_missed
;
287 ovs_flow_stats_update(flow
, key
->tp
.flags
, skb
);
288 sf_acts
= rcu_dereference(flow
->sf_acts
);
289 ovs_execute_actions(dp
, skb
, sf_acts
, key
);
291 stats_counter
= &stats
->n_hit
;
294 /* Update datapath statistics. */
295 u64_stats_update_begin(&stats
->syncp
);
297 stats
->n_mask_hit
+= n_mask_hit
;
298 u64_stats_update_end(&stats
->syncp
);
301 int ovs_dp_upcall(struct datapath
*dp
, struct sk_buff
*skb
,
302 const struct sw_flow_key
*key
,
303 const struct dp_upcall_info
*upcall_info
)
305 struct dp_stats_percpu
*stats
;
308 if (upcall_info
->portid
== 0) {
313 if (!skb_is_gso(skb
))
314 err
= queue_userspace_packet(dp
, skb
, key
, upcall_info
);
316 err
= queue_gso_packets(dp
, skb
, key
, upcall_info
);
323 stats
= this_cpu_ptr(dp
->stats_percpu
);
325 u64_stats_update_begin(&stats
->syncp
);
327 u64_stats_update_end(&stats
->syncp
);
332 static int queue_gso_packets(struct datapath
*dp
, struct sk_buff
*skb
,
333 const struct sw_flow_key
*key
,
334 const struct dp_upcall_info
*upcall_info
)
336 unsigned short gso_type
= skb_shinfo(skb
)->gso_type
;
337 struct sw_flow_key later_key
;
338 struct sk_buff
*segs
, *nskb
;
339 struct ovs_skb_cb ovs_cb
;
342 ovs_cb
= *OVS_CB(skb
);
343 segs
= __skb_gso_segment(skb
, NETIF_F_SG
, false);
344 *OVS_CB(skb
) = ovs_cb
;
346 return PTR_ERR(segs
);
350 if (gso_type
& SKB_GSO_UDP
) {
351 /* The initial flow key extracted by ovs_flow_key_extract()
352 * in this case is for a first fragment, so we need to
353 * properly mark later fragments.
356 later_key
.ip
.frag
= OVS_FRAG_TYPE_LATER
;
359 /* Queue all of the segments. */
362 *OVS_CB(skb
) = ovs_cb
;
363 if (gso_type
& SKB_GSO_UDP
&& skb
!= segs
)
366 err
= queue_userspace_packet(dp
, skb
, key
, upcall_info
);
370 } while ((skb
= skb
->next
));
372 /* Free all of the segments. */
380 } while ((skb
= nskb
));
384 static size_t upcall_msg_size(const struct dp_upcall_info
*upcall_info
,
387 size_t size
= NLMSG_ALIGN(sizeof(struct ovs_header
))
388 + nla_total_size(hdrlen
) /* OVS_PACKET_ATTR_PACKET */
389 + nla_total_size(ovs_key_attr_size()); /* OVS_PACKET_ATTR_KEY */
391 /* OVS_PACKET_ATTR_USERDATA */
392 if (upcall_info
->userdata
)
393 size
+= NLA_ALIGN(upcall_info
->userdata
->nla_len
);
395 /* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
396 if (upcall_info
->egress_tun_info
)
397 size
+= nla_total_size(ovs_tun_key_attr_size());
402 static int queue_userspace_packet(struct datapath
*dp
, struct sk_buff
*skb
,
403 const struct sw_flow_key
*key
,
404 const struct dp_upcall_info
*upcall_info
)
406 struct ovs_header
*upcall
;
407 struct sk_buff
*nskb
= NULL
;
408 struct sk_buff
*user_skb
= NULL
; /* to be queued to userspace */
410 struct genl_info info
= {
411 .dst_sk
= ovs_dp_get_net(dp
)->genl_sock
,
412 .snd_portid
= upcall_info
->portid
,
418 dp_ifindex
= get_dpifindex(dp
);
422 if (skb_vlan_tag_present(skb
)) {
423 nskb
= skb_clone(skb
, GFP_ATOMIC
);
427 nskb
= __vlan_hwaccel_push_inside(nskb
);
434 if (nla_attr_size(skb
->len
) > USHRT_MAX
) {
439 /* Complete checksum if needed */
440 if (skb
->ip_summed
== CHECKSUM_PARTIAL
&&
441 (err
= skb_checksum_help(skb
)))
444 /* Older versions of OVS user space enforce alignment of the last
445 * Netlink attribute to NLA_ALIGNTO which would require extensive
446 * padding logic. Only perform zerocopy if padding is not required.
448 if (dp
->user_features
& OVS_DP_F_UNALIGNED
)
449 hlen
= skb_zerocopy_headlen(skb
);
453 len
= upcall_msg_size(upcall_info
, hlen
);
454 user_skb
= genlmsg_new_unicast(len
, &info
, GFP_ATOMIC
);
460 upcall
= genlmsg_put(user_skb
, 0, 0, &dp_packet_genl_family
,
461 0, upcall_info
->cmd
);
462 upcall
->dp_ifindex
= dp_ifindex
;
464 nla
= nla_nest_start(user_skb
, OVS_PACKET_ATTR_KEY
);
465 err
= ovs_nla_put_flow(key
, key
, user_skb
);
467 nla_nest_end(user_skb
, nla
);
469 if (upcall_info
->userdata
)
470 __nla_put(user_skb
, OVS_PACKET_ATTR_USERDATA
,
471 nla_len(upcall_info
->userdata
),
472 nla_data(upcall_info
->userdata
));
474 if (upcall_info
->egress_tun_info
) {
475 nla
= nla_nest_start(user_skb
, OVS_PACKET_ATTR_EGRESS_TUN_KEY
);
476 err
= ovs_nla_put_egress_tunnel_key(user_skb
,
477 upcall_info
->egress_tun_info
);
479 nla_nest_end(user_skb
, nla
);
482 /* Only reserve room for attribute header, packet data is added
483 * in skb_zerocopy() */
484 if (!(nla
= nla_reserve(user_skb
, OVS_PACKET_ATTR_PACKET
, 0))) {
488 nla
->nla_len
= nla_attr_size(skb
->len
);
490 err
= skb_zerocopy(user_skb
, skb
, skb
->len
, hlen
);
494 /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
495 if (!(dp
->user_features
& OVS_DP_F_UNALIGNED
)) {
496 size_t plen
= NLA_ALIGN(user_skb
->len
) - user_skb
->len
;
499 memset(skb_put(user_skb
, plen
), 0, plen
);
502 ((struct nlmsghdr
*) user_skb
->data
)->nlmsg_len
= user_skb
->len
;
504 err
= genlmsg_unicast(ovs_dp_get_net(dp
), user_skb
, upcall_info
->portid
);
514 static int ovs_packet_cmd_execute(struct sk_buff
*skb
, struct genl_info
*info
)
516 struct ovs_header
*ovs_header
= info
->userhdr
;
517 struct nlattr
**a
= info
->attrs
;
518 struct sw_flow_actions
*acts
;
519 struct sk_buff
*packet
;
520 struct sw_flow
*flow
;
521 struct sw_flow_actions
*sf_acts
;
524 struct vport
*input_vport
;
527 bool log
= !a
[OVS_PACKET_ATTR_PROBE
];
530 if (!a
[OVS_PACKET_ATTR_PACKET
] || !a
[OVS_PACKET_ATTR_KEY
] ||
531 !a
[OVS_PACKET_ATTR_ACTIONS
])
534 len
= nla_len(a
[OVS_PACKET_ATTR_PACKET
]);
535 packet
= __dev_alloc_skb(NET_IP_ALIGN
+ len
, GFP_KERNEL
);
539 skb_reserve(packet
, NET_IP_ALIGN
);
541 nla_memcpy(__skb_put(packet
, len
), a
[OVS_PACKET_ATTR_PACKET
], len
);
543 skb_reset_mac_header(packet
);
544 eth
= eth_hdr(packet
);
546 /* Normally, setting the skb 'protocol' field would be handled by a
547 * call to eth_type_trans(), but it assumes there's a sending
548 * device, which we may not have. */
549 if (ntohs(eth
->h_proto
) >= ETH_P_802_3_MIN
)
550 packet
->protocol
= eth
->h_proto
;
552 packet
->protocol
= htons(ETH_P_802_2
);
554 /* Build an sw_flow for sending this packet. */
555 flow
= ovs_flow_alloc();
560 err
= ovs_flow_key_extract_userspace(a
[OVS_PACKET_ATTR_KEY
], packet
,
565 err
= ovs_nla_copy_actions(a
[OVS_PACKET_ATTR_ACTIONS
],
566 &flow
->key
, &acts
, log
);
570 rcu_assign_pointer(flow
->sf_acts
, acts
);
571 OVS_CB(packet
)->egress_tun_info
= NULL
;
572 packet
->priority
= flow
->key
.phy
.priority
;
573 packet
->mark
= flow
->key
.phy
.skb_mark
;
576 dp
= get_dp_rcu(sock_net(skb
->sk
), ovs_header
->dp_ifindex
);
581 input_vport
= ovs_vport_rcu(dp
, flow
->key
.phy
.in_port
);
583 input_vport
= ovs_vport_rcu(dp
, OVSP_LOCAL
);
588 OVS_CB(packet
)->input_vport
= input_vport
;
589 sf_acts
= rcu_dereference(flow
->sf_acts
);
592 err
= ovs_execute_actions(dp
, packet
, sf_acts
, &flow
->key
);
596 ovs_flow_free(flow
, false);
602 ovs_flow_free(flow
, false);
609 static const struct nla_policy packet_policy
[OVS_PACKET_ATTR_MAX
+ 1] = {
610 [OVS_PACKET_ATTR_PACKET
] = { .len
= ETH_HLEN
},
611 [OVS_PACKET_ATTR_KEY
] = { .type
= NLA_NESTED
},
612 [OVS_PACKET_ATTR_ACTIONS
] = { .type
= NLA_NESTED
},
613 [OVS_PACKET_ATTR_PROBE
] = { .type
= NLA_FLAG
},
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 genlmsg_end(skb
, ovs_header
);
806 genlmsg_cancel(skb
, ovs_header
);
810 /* May not be called with RCU read lock. */
811 static struct sk_buff
*ovs_flow_cmd_alloc_info(const struct sw_flow_actions
*acts
,
812 struct genl_info
*info
,
817 if (!always
&& !ovs_must_notify(&dp_flow_genl_family
, info
, 0))
820 skb
= genlmsg_new_unicast(ovs_flow_cmd_msg_size(acts
), info
, GFP_KERNEL
);
822 return ERR_PTR(-ENOMEM
);
827 /* Called with ovs_mutex. */
828 static struct sk_buff
*ovs_flow_cmd_build_info(const struct sw_flow
*flow
,
830 struct genl_info
*info
, u8 cmd
,
836 skb
= ovs_flow_cmd_alloc_info(ovsl_dereference(flow
->sf_acts
), info
,
838 if (IS_ERR_OR_NULL(skb
))
841 retval
= ovs_flow_cmd_fill_info(flow
, dp_ifindex
, skb
,
842 info
->snd_portid
, info
->snd_seq
, 0,
848 static int ovs_flow_cmd_new(struct sk_buff
*skb
, struct genl_info
*info
)
850 struct nlattr
**a
= info
->attrs
;
851 struct ovs_header
*ovs_header
= info
->userhdr
;
852 struct sw_flow
*flow
, *new_flow
;
853 struct sw_flow_mask mask
;
854 struct sk_buff
*reply
;
856 struct sw_flow_actions
*acts
;
857 struct sw_flow_match match
;
859 bool log
= !a
[OVS_FLOW_ATTR_PROBE
];
861 /* Must have key and actions. */
863 if (!a
[OVS_FLOW_ATTR_KEY
]) {
864 OVS_NLERR(log
, "Flow key attr not present in new flow.");
867 if (!a
[OVS_FLOW_ATTR_ACTIONS
]) {
868 OVS_NLERR(log
, "Flow actions attr not present in new flow.");
872 /* Most of the time we need to allocate a new flow, do it before
875 new_flow
= ovs_flow_alloc();
876 if (IS_ERR(new_flow
)) {
877 error
= PTR_ERR(new_flow
);
882 ovs_match_init(&match
, &new_flow
->unmasked_key
, &mask
);
883 error
= ovs_nla_get_match(&match
, a
[OVS_FLOW_ATTR_KEY
],
884 a
[OVS_FLOW_ATTR_MASK
], log
);
888 ovs_flow_mask_key(&new_flow
->key
, &new_flow
->unmasked_key
, &mask
);
890 /* Validate actions. */
891 error
= ovs_nla_copy_actions(a
[OVS_FLOW_ATTR_ACTIONS
], &new_flow
->key
,
894 OVS_NLERR(log
, "Flow actions may not be safe on all matching packets.");
898 reply
= ovs_flow_cmd_alloc_info(acts
, info
, false);
900 error
= PTR_ERR(reply
);
905 dp
= get_dp(sock_net(skb
->sk
), ovs_header
->dp_ifindex
);
910 /* Check if this is a duplicate flow */
911 flow
= ovs_flow_tbl_lookup(&dp
->table
, &new_flow
->unmasked_key
);
913 rcu_assign_pointer(new_flow
->sf_acts
, acts
);
915 /* Put flow in bucket. */
916 error
= ovs_flow_tbl_insert(&dp
->table
, new_flow
, &mask
);
917 if (unlikely(error
)) {
922 if (unlikely(reply
)) {
923 error
= ovs_flow_cmd_fill_info(new_flow
,
924 ovs_header
->dp_ifindex
,
925 reply
, info
->snd_portid
,
932 struct sw_flow_actions
*old_acts
;
934 /* Bail out if we're not allowed to modify an existing flow.
935 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
936 * because Generic Netlink treats the latter as a dump
937 * request. We also accept NLM_F_EXCL in case that bug ever
940 if (unlikely(info
->nlhdr
->nlmsg_flags
& (NLM_F_CREATE
945 /* The unmasked key has to be the same for flow updates. */
946 if (unlikely(!ovs_flow_cmp_unmasked_key(flow
, &match
))) {
947 /* Look for any overlapping flow. */
948 flow
= ovs_flow_tbl_lookup_exact(&dp
->table
, &match
);
954 /* Update actions. */
955 old_acts
= ovsl_dereference(flow
->sf_acts
);
956 rcu_assign_pointer(flow
->sf_acts
, acts
);
958 if (unlikely(reply
)) {
959 error
= ovs_flow_cmd_fill_info(flow
,
960 ovs_header
->dp_ifindex
,
961 reply
, info
->snd_portid
,
968 ovs_nla_free_flow_actions(old_acts
);
969 ovs_flow_free(new_flow
, false);
973 ovs_notify(&dp_flow_genl_family
, reply
, info
);
982 ovs_flow_free(new_flow
, false);
987 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
988 static struct sw_flow_actions
*get_flow_actions(const struct nlattr
*a
,
989 const struct sw_flow_key
*key
,
990 const struct sw_flow_mask
*mask
,
993 struct sw_flow_actions
*acts
;
994 struct sw_flow_key masked_key
;
997 ovs_flow_mask_key(&masked_key
, key
, mask
);
998 error
= ovs_nla_copy_actions(a
, &masked_key
, &acts
, log
);
1001 "Actions may not be safe on all matching packets");
1002 return ERR_PTR(error
);
1008 static int ovs_flow_cmd_set(struct sk_buff
*skb
, struct genl_info
*info
)
1010 struct nlattr
**a
= info
->attrs
;
1011 struct ovs_header
*ovs_header
= info
->userhdr
;
1012 struct sw_flow_key key
;
1013 struct sw_flow
*flow
;
1014 struct sw_flow_mask mask
;
1015 struct sk_buff
*reply
= NULL
;
1016 struct datapath
*dp
;
1017 struct sw_flow_actions
*old_acts
= NULL
, *acts
= NULL
;
1018 struct sw_flow_match match
;
1020 bool log
= !a
[OVS_FLOW_ATTR_PROBE
];
1024 if (!a
[OVS_FLOW_ATTR_KEY
]) {
1025 OVS_NLERR(log
, "Flow key attribute not present in set flow.");
1029 ovs_match_init(&match
, &key
, &mask
);
1030 error
= ovs_nla_get_match(&match
, a
[OVS_FLOW_ATTR_KEY
],
1031 a
[OVS_FLOW_ATTR_MASK
], log
);
1035 /* Validate actions. */
1036 if (a
[OVS_FLOW_ATTR_ACTIONS
]) {
1037 acts
= get_flow_actions(a
[OVS_FLOW_ATTR_ACTIONS
], &key
, &mask
,
1040 error
= PTR_ERR(acts
);
1044 /* Can allocate before locking if have acts. */
1045 reply
= ovs_flow_cmd_alloc_info(acts
, info
, false);
1046 if (IS_ERR(reply
)) {
1047 error
= PTR_ERR(reply
);
1048 goto err_kfree_acts
;
1053 dp
= get_dp(sock_net(skb
->sk
), ovs_header
->dp_ifindex
);
1054 if (unlikely(!dp
)) {
1056 goto err_unlock_ovs
;
1058 /* Check that the flow exists. */
1059 flow
= ovs_flow_tbl_lookup_exact(&dp
->table
, &match
);
1060 if (unlikely(!flow
)) {
1062 goto err_unlock_ovs
;
1065 /* Update actions, if present. */
1067 old_acts
= ovsl_dereference(flow
->sf_acts
);
1068 rcu_assign_pointer(flow
->sf_acts
, acts
);
1070 if (unlikely(reply
)) {
1071 error
= ovs_flow_cmd_fill_info(flow
,
1072 ovs_header
->dp_ifindex
,
1073 reply
, info
->snd_portid
,
1079 /* Could not alloc without acts before locking. */
1080 reply
= ovs_flow_cmd_build_info(flow
, ovs_header
->dp_ifindex
,
1081 info
, OVS_FLOW_CMD_NEW
, false);
1082 if (unlikely(IS_ERR(reply
))) {
1083 error
= PTR_ERR(reply
);
1084 goto err_unlock_ovs
;
1089 if (a
[OVS_FLOW_ATTR_CLEAR
])
1090 ovs_flow_stats_clear(flow
);
1094 ovs_notify(&dp_flow_genl_family
, reply
, info
);
1096 ovs_nla_free_flow_actions(old_acts
);
1109 static int ovs_flow_cmd_get(struct sk_buff
*skb
, struct genl_info
*info
)
1111 struct nlattr
**a
= info
->attrs
;
1112 struct ovs_header
*ovs_header
= info
->userhdr
;
1113 struct sw_flow_key key
;
1114 struct sk_buff
*reply
;
1115 struct sw_flow
*flow
;
1116 struct datapath
*dp
;
1117 struct sw_flow_match match
;
1119 bool log
= !a
[OVS_FLOW_ATTR_PROBE
];
1121 if (!a
[OVS_FLOW_ATTR_KEY
]) {
1123 "Flow get message rejected, Key attribute missing.");
1127 ovs_match_init(&match
, &key
, NULL
);
1128 err
= ovs_nla_get_match(&match
, a
[OVS_FLOW_ATTR_KEY
], NULL
, log
);
1133 dp
= get_dp(sock_net(skb
->sk
), ovs_header
->dp_ifindex
);
1139 flow
= ovs_flow_tbl_lookup_exact(&dp
->table
, &match
);
1145 reply
= ovs_flow_cmd_build_info(flow
, ovs_header
->dp_ifindex
, info
,
1146 OVS_FLOW_CMD_NEW
, true);
1147 if (IS_ERR(reply
)) {
1148 err
= PTR_ERR(reply
);
1153 return genlmsg_reply(reply
, info
);
1159 static int ovs_flow_cmd_del(struct sk_buff
*skb
, struct genl_info
*info
)
1161 struct nlattr
**a
= info
->attrs
;
1162 struct ovs_header
*ovs_header
= info
->userhdr
;
1163 struct sw_flow_key key
;
1164 struct sk_buff
*reply
;
1165 struct sw_flow
*flow
;
1166 struct datapath
*dp
;
1167 struct sw_flow_match match
;
1169 bool log
= !a
[OVS_FLOW_ATTR_PROBE
];
1171 if (likely(a
[OVS_FLOW_ATTR_KEY
])) {
1172 ovs_match_init(&match
, &key
, NULL
);
1173 err
= ovs_nla_get_match(&match
, a
[OVS_FLOW_ATTR_KEY
], NULL
,
1180 dp
= get_dp(sock_net(skb
->sk
), ovs_header
->dp_ifindex
);
1181 if (unlikely(!dp
)) {
1186 if (unlikely(!a
[OVS_FLOW_ATTR_KEY
])) {
1187 err
= ovs_flow_tbl_flush(&dp
->table
);
1191 flow
= ovs_flow_tbl_lookup_exact(&dp
->table
, &match
);
1192 if (unlikely(!flow
)) {
1197 ovs_flow_tbl_remove(&dp
->table
, flow
);
1200 reply
= ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force
*) flow
->sf_acts
,
1202 if (likely(reply
)) {
1203 if (likely(!IS_ERR(reply
))) {
1204 rcu_read_lock(); /*To keep RCU checker happy. */
1205 err
= ovs_flow_cmd_fill_info(flow
, ovs_header
->dp_ifindex
,
1206 reply
, info
->snd_portid
,
1212 ovs_notify(&dp_flow_genl_family
, reply
, info
);
1214 netlink_set_err(sock_net(skb
->sk
)->genl_sock
, 0, 0, PTR_ERR(reply
));
1218 ovs_flow_free(flow
, true);
1225 static int ovs_flow_cmd_dump(struct sk_buff
*skb
, struct netlink_callback
*cb
)
1227 struct ovs_header
*ovs_header
= genlmsg_data(nlmsg_data(cb
->nlh
));
1228 struct table_instance
*ti
;
1229 struct datapath
*dp
;
1232 dp
= get_dp_rcu(sock_net(skb
->sk
), ovs_header
->dp_ifindex
);
1238 ti
= rcu_dereference(dp
->table
.ti
);
1240 struct sw_flow
*flow
;
1243 bucket
= cb
->args
[0];
1245 flow
= ovs_flow_tbl_dump_next(ti
, &bucket
, &obj
);
1249 if (ovs_flow_cmd_fill_info(flow
, ovs_header
->dp_ifindex
, skb
,
1250 NETLINK_CB(cb
->skb
).portid
,
1251 cb
->nlh
->nlmsg_seq
, NLM_F_MULTI
,
1252 OVS_FLOW_CMD_NEW
) < 0)
1255 cb
->args
[0] = bucket
;
1262 static const struct nla_policy flow_policy
[OVS_FLOW_ATTR_MAX
+ 1] = {
1263 [OVS_FLOW_ATTR_KEY
] = { .type
= NLA_NESTED
},
1264 [OVS_FLOW_ATTR_MASK
] = { .type
= NLA_NESTED
},
1265 [OVS_FLOW_ATTR_ACTIONS
] = { .type
= NLA_NESTED
},
1266 [OVS_FLOW_ATTR_CLEAR
] = { .type
= NLA_FLAG
},
1267 [OVS_FLOW_ATTR_PROBE
] = { .type
= NLA_FLAG
},
1270 static const struct genl_ops dp_flow_genl_ops
[] = {
1271 { .cmd
= OVS_FLOW_CMD_NEW
,
1272 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1273 .policy
= flow_policy
,
1274 .doit
= ovs_flow_cmd_new
1276 { .cmd
= OVS_FLOW_CMD_DEL
,
1277 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1278 .policy
= flow_policy
,
1279 .doit
= ovs_flow_cmd_del
1281 { .cmd
= OVS_FLOW_CMD_GET
,
1282 .flags
= 0, /* OK for unprivileged users. */
1283 .policy
= flow_policy
,
1284 .doit
= ovs_flow_cmd_get
,
1285 .dumpit
= ovs_flow_cmd_dump
1287 { .cmd
= OVS_FLOW_CMD_SET
,
1288 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1289 .policy
= flow_policy
,
1290 .doit
= ovs_flow_cmd_set
,
1294 static struct genl_family dp_flow_genl_family
= {
1295 .id
= GENL_ID_GENERATE
,
1296 .hdrsize
= sizeof(struct ovs_header
),
1297 .name
= OVS_FLOW_FAMILY
,
1298 .version
= OVS_FLOW_VERSION
,
1299 .maxattr
= OVS_FLOW_ATTR_MAX
,
1301 .parallel_ops
= true,
1302 .ops
= dp_flow_genl_ops
,
1303 .n_ops
= ARRAY_SIZE(dp_flow_genl_ops
),
1304 .mcgrps
= &ovs_dp_flow_multicast_group
,
1308 static size_t ovs_dp_cmd_msg_size(void)
1310 size_t msgsize
= NLMSG_ALIGN(sizeof(struct ovs_header
));
1312 msgsize
+= nla_total_size(IFNAMSIZ
);
1313 msgsize
+= nla_total_size(sizeof(struct ovs_dp_stats
));
1314 msgsize
+= nla_total_size(sizeof(struct ovs_dp_megaflow_stats
));
1315 msgsize
+= nla_total_size(sizeof(u32
)); /* OVS_DP_ATTR_USER_FEATURES */
1320 /* Called with ovs_mutex. */
1321 static int ovs_dp_cmd_fill_info(struct datapath
*dp
, struct sk_buff
*skb
,
1322 u32 portid
, u32 seq
, u32 flags
, u8 cmd
)
1324 struct ovs_header
*ovs_header
;
1325 struct ovs_dp_stats dp_stats
;
1326 struct ovs_dp_megaflow_stats dp_megaflow_stats
;
1329 ovs_header
= genlmsg_put(skb
, portid
, seq
, &dp_datapath_genl_family
,
1334 ovs_header
->dp_ifindex
= get_dpifindex(dp
);
1336 err
= nla_put_string(skb
, OVS_DP_ATTR_NAME
, ovs_dp_name(dp
));
1338 goto nla_put_failure
;
1340 get_dp_stats(dp
, &dp_stats
, &dp_megaflow_stats
);
1341 if (nla_put(skb
, OVS_DP_ATTR_STATS
, sizeof(struct ovs_dp_stats
),
1343 goto nla_put_failure
;
1345 if (nla_put(skb
, OVS_DP_ATTR_MEGAFLOW_STATS
,
1346 sizeof(struct ovs_dp_megaflow_stats
),
1347 &dp_megaflow_stats
))
1348 goto nla_put_failure
;
1350 if (nla_put_u32(skb
, OVS_DP_ATTR_USER_FEATURES
, dp
->user_features
))
1351 goto nla_put_failure
;
1353 genlmsg_end(skb
, ovs_header
);
1357 genlmsg_cancel(skb
, ovs_header
);
1362 static struct sk_buff
*ovs_dp_cmd_alloc_info(struct genl_info
*info
)
1364 return genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info
, GFP_KERNEL
);
1367 /* Called with rcu_read_lock or ovs_mutex. */
1368 static struct datapath
*lookup_datapath(struct net
*net
,
1369 const struct ovs_header
*ovs_header
,
1370 struct nlattr
*a
[OVS_DP_ATTR_MAX
+ 1])
1372 struct datapath
*dp
;
1374 if (!a
[OVS_DP_ATTR_NAME
])
1375 dp
= get_dp(net
, ovs_header
->dp_ifindex
);
1377 struct vport
*vport
;
1379 vport
= ovs_vport_locate(net
, nla_data(a
[OVS_DP_ATTR_NAME
]));
1380 dp
= vport
&& vport
->port_no
== OVSP_LOCAL
? vport
->dp
: NULL
;
1382 return dp
? dp
: ERR_PTR(-ENODEV
);
1385 static void ovs_dp_reset_user_features(struct sk_buff
*skb
, struct genl_info
*info
)
1387 struct datapath
*dp
;
1389 dp
= lookup_datapath(sock_net(skb
->sk
), info
->userhdr
, info
->attrs
);
1393 WARN(dp
->user_features
, "Dropping previously announced user features\n");
1394 dp
->user_features
= 0;
1397 static void ovs_dp_change(struct datapath
*dp
, struct nlattr
*a
[])
1399 if (a
[OVS_DP_ATTR_USER_FEATURES
])
1400 dp
->user_features
= nla_get_u32(a
[OVS_DP_ATTR_USER_FEATURES
]);
1403 static int ovs_dp_cmd_new(struct sk_buff
*skb
, struct genl_info
*info
)
1405 struct nlattr
**a
= info
->attrs
;
1406 struct vport_parms parms
;
1407 struct sk_buff
*reply
;
1408 struct datapath
*dp
;
1409 struct vport
*vport
;
1410 struct ovs_net
*ovs_net
;
1414 if (!a
[OVS_DP_ATTR_NAME
] || !a
[OVS_DP_ATTR_UPCALL_PID
])
1417 reply
= ovs_dp_cmd_alloc_info(info
);
1422 dp
= kzalloc(sizeof(*dp
), GFP_KERNEL
);
1424 goto err_free_reply
;
1426 ovs_dp_set_net(dp
, hold_net(sock_net(skb
->sk
)));
1428 /* Allocate table. */
1429 err
= ovs_flow_tbl_init(&dp
->table
);
1433 dp
->stats_percpu
= netdev_alloc_pcpu_stats(struct dp_stats_percpu
);
1434 if (!dp
->stats_percpu
) {
1436 goto err_destroy_table
;
1439 dp
->ports
= kmalloc(DP_VPORT_HASH_BUCKETS
* sizeof(struct hlist_head
),
1443 goto err_destroy_percpu
;
1446 for (i
= 0; i
< DP_VPORT_HASH_BUCKETS
; i
++)
1447 INIT_HLIST_HEAD(&dp
->ports
[i
]);
1449 /* Set up our datapath device. */
1450 parms
.name
= nla_data(a
[OVS_DP_ATTR_NAME
]);
1451 parms
.type
= OVS_VPORT_TYPE_INTERNAL
;
1452 parms
.options
= NULL
;
1454 parms
.port_no
= OVSP_LOCAL
;
1455 parms
.upcall_portids
= a
[OVS_DP_ATTR_UPCALL_PID
];
1457 ovs_dp_change(dp
, a
);
1459 /* So far only local changes have been made, now need the lock. */
1462 vport
= new_vport(&parms
);
1463 if (IS_ERR(vport
)) {
1464 err
= PTR_ERR(vport
);
1468 if (err
== -EEXIST
) {
1469 /* An outdated user space instance that does not understand
1470 * the concept of user_features has attempted to create a new
1471 * datapath and is likely to reuse it. Drop all user features.
1473 if (info
->genlhdr
->version
< OVS_DP_VER_FEATURES
)
1474 ovs_dp_reset_user_features(skb
, info
);
1477 goto err_destroy_ports_array
;
1480 err
= ovs_dp_cmd_fill_info(dp
, reply
, info
->snd_portid
,
1481 info
->snd_seq
, 0, OVS_DP_CMD_NEW
);
1484 ovs_net
= net_generic(ovs_dp_get_net(dp
), ovs_net_id
);
1485 list_add_tail_rcu(&dp
->list_node
, &ovs_net
->dps
);
1489 ovs_notify(&dp_datapath_genl_family
, reply
, info
);
1492 err_destroy_ports_array
:
1496 free_percpu(dp
->stats_percpu
);
1498 ovs_flow_tbl_destroy(&dp
->table
);
1500 release_net(ovs_dp_get_net(dp
));
1508 /* Called with ovs_mutex. */
1509 static void __dp_destroy(struct datapath
*dp
)
1513 for (i
= 0; i
< DP_VPORT_HASH_BUCKETS
; i
++) {
1514 struct vport
*vport
;
1515 struct hlist_node
*n
;
1517 hlist_for_each_entry_safe(vport
, n
, &dp
->ports
[i
], dp_hash_node
)
1518 if (vport
->port_no
!= OVSP_LOCAL
)
1519 ovs_dp_detach_port(vport
);
1522 list_del_rcu(&dp
->list_node
);
1524 /* OVSP_LOCAL is datapath internal port. We need to make sure that
1525 * all ports in datapath are destroyed first before freeing datapath.
1527 ovs_dp_detach_port(ovs_vport_ovsl(dp
, OVSP_LOCAL
));
1529 /* RCU destroy the flow table */
1530 call_rcu(&dp
->rcu
, destroy_dp_rcu
);
1533 static int ovs_dp_cmd_del(struct sk_buff
*skb
, struct genl_info
*info
)
1535 struct sk_buff
*reply
;
1536 struct datapath
*dp
;
1539 reply
= ovs_dp_cmd_alloc_info(info
);
1544 dp
= lookup_datapath(sock_net(skb
->sk
), info
->userhdr
, info
->attrs
);
1547 goto err_unlock_free
;
1549 err
= ovs_dp_cmd_fill_info(dp
, reply
, info
->snd_portid
,
1550 info
->snd_seq
, 0, OVS_DP_CMD_DEL
);
1556 ovs_notify(&dp_datapath_genl_family
, reply
, info
);
1566 static int ovs_dp_cmd_set(struct sk_buff
*skb
, struct genl_info
*info
)
1568 struct sk_buff
*reply
;
1569 struct datapath
*dp
;
1572 reply
= ovs_dp_cmd_alloc_info(info
);
1577 dp
= lookup_datapath(sock_net(skb
->sk
), info
->userhdr
, info
->attrs
);
1580 goto err_unlock_free
;
1582 ovs_dp_change(dp
, info
->attrs
);
1584 err
= ovs_dp_cmd_fill_info(dp
, reply
, info
->snd_portid
,
1585 info
->snd_seq
, 0, OVS_DP_CMD_NEW
);
1589 ovs_notify(&dp_datapath_genl_family
, reply
, info
);
1599 static int ovs_dp_cmd_get(struct sk_buff
*skb
, struct genl_info
*info
)
1601 struct sk_buff
*reply
;
1602 struct datapath
*dp
;
1605 reply
= ovs_dp_cmd_alloc_info(info
);
1610 dp
= lookup_datapath(sock_net(skb
->sk
), info
->userhdr
, info
->attrs
);
1613 goto err_unlock_free
;
1615 err
= ovs_dp_cmd_fill_info(dp
, reply
, info
->snd_portid
,
1616 info
->snd_seq
, 0, OVS_DP_CMD_NEW
);
1620 return genlmsg_reply(reply
, info
);
1628 static int ovs_dp_cmd_dump(struct sk_buff
*skb
, struct netlink_callback
*cb
)
1630 struct ovs_net
*ovs_net
= net_generic(sock_net(skb
->sk
), ovs_net_id
);
1631 struct datapath
*dp
;
1632 int skip
= cb
->args
[0];
1636 list_for_each_entry(dp
, &ovs_net
->dps
, list_node
) {
1638 ovs_dp_cmd_fill_info(dp
, skb
, NETLINK_CB(cb
->skb
).portid
,
1639 cb
->nlh
->nlmsg_seq
, NLM_F_MULTI
,
1640 OVS_DP_CMD_NEW
) < 0)
1651 static const struct nla_policy datapath_policy
[OVS_DP_ATTR_MAX
+ 1] = {
1652 [OVS_DP_ATTR_NAME
] = { .type
= NLA_NUL_STRING
, .len
= IFNAMSIZ
- 1 },
1653 [OVS_DP_ATTR_UPCALL_PID
] = { .type
= NLA_U32
},
1654 [OVS_DP_ATTR_USER_FEATURES
] = { .type
= NLA_U32
},
1657 static const struct genl_ops dp_datapath_genl_ops
[] = {
1658 { .cmd
= OVS_DP_CMD_NEW
,
1659 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1660 .policy
= datapath_policy
,
1661 .doit
= ovs_dp_cmd_new
1663 { .cmd
= OVS_DP_CMD_DEL
,
1664 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1665 .policy
= datapath_policy
,
1666 .doit
= ovs_dp_cmd_del
1668 { .cmd
= OVS_DP_CMD_GET
,
1669 .flags
= 0, /* OK for unprivileged users. */
1670 .policy
= datapath_policy
,
1671 .doit
= ovs_dp_cmd_get
,
1672 .dumpit
= ovs_dp_cmd_dump
1674 { .cmd
= OVS_DP_CMD_SET
,
1675 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
1676 .policy
= datapath_policy
,
1677 .doit
= ovs_dp_cmd_set
,
1681 static struct genl_family dp_datapath_genl_family
= {
1682 .id
= GENL_ID_GENERATE
,
1683 .hdrsize
= sizeof(struct ovs_header
),
1684 .name
= OVS_DATAPATH_FAMILY
,
1685 .version
= OVS_DATAPATH_VERSION
,
1686 .maxattr
= OVS_DP_ATTR_MAX
,
1688 .parallel_ops
= true,
1689 .ops
= dp_datapath_genl_ops
,
1690 .n_ops
= ARRAY_SIZE(dp_datapath_genl_ops
),
1691 .mcgrps
= &ovs_dp_datapath_multicast_group
,
1695 /* Called with ovs_mutex or RCU read lock. */
1696 static int ovs_vport_cmd_fill_info(struct vport
*vport
, struct sk_buff
*skb
,
1697 u32 portid
, u32 seq
, u32 flags
, u8 cmd
)
1699 struct ovs_header
*ovs_header
;
1700 struct ovs_vport_stats vport_stats
;
1703 ovs_header
= genlmsg_put(skb
, portid
, seq
, &dp_vport_genl_family
,
1708 ovs_header
->dp_ifindex
= get_dpifindex(vport
->dp
);
1710 if (nla_put_u32(skb
, OVS_VPORT_ATTR_PORT_NO
, vport
->port_no
) ||
1711 nla_put_u32(skb
, OVS_VPORT_ATTR_TYPE
, vport
->ops
->type
) ||
1712 nla_put_string(skb
, OVS_VPORT_ATTR_NAME
,
1713 vport
->ops
->get_name(vport
)))
1714 goto nla_put_failure
;
1716 ovs_vport_get_stats(vport
, &vport_stats
);
1717 if (nla_put(skb
, OVS_VPORT_ATTR_STATS
, sizeof(struct ovs_vport_stats
),
1719 goto nla_put_failure
;
1721 if (ovs_vport_get_upcall_portids(vport
, skb
))
1722 goto nla_put_failure
;
1724 err
= ovs_vport_get_options(vport
, skb
);
1725 if (err
== -EMSGSIZE
)
1728 genlmsg_end(skb
, ovs_header
);
1734 genlmsg_cancel(skb
, ovs_header
);
1738 static struct sk_buff
*ovs_vport_cmd_alloc_info(void)
1740 return nlmsg_new(NLMSG_DEFAULT_SIZE
, GFP_KERNEL
);
1743 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1744 struct sk_buff
*ovs_vport_cmd_build_info(struct vport
*vport
, u32 portid
,
1747 struct sk_buff
*skb
;
1750 skb
= nlmsg_new(NLMSG_DEFAULT_SIZE
, GFP_ATOMIC
);
1752 return ERR_PTR(-ENOMEM
);
1754 retval
= ovs_vport_cmd_fill_info(vport
, skb
, portid
, seq
, 0, cmd
);
1760 /* Called with ovs_mutex or RCU read lock. */
1761 static struct vport
*lookup_vport(struct net
*net
,
1762 const struct ovs_header
*ovs_header
,
1763 struct nlattr
*a
[OVS_VPORT_ATTR_MAX
+ 1])
1765 struct datapath
*dp
;
1766 struct vport
*vport
;
1768 if (a
[OVS_VPORT_ATTR_NAME
]) {
1769 vport
= ovs_vport_locate(net
, nla_data(a
[OVS_VPORT_ATTR_NAME
]));
1771 return ERR_PTR(-ENODEV
);
1772 if (ovs_header
->dp_ifindex
&&
1773 ovs_header
->dp_ifindex
!= get_dpifindex(vport
->dp
))
1774 return ERR_PTR(-ENODEV
);
1776 } else if (a
[OVS_VPORT_ATTR_PORT_NO
]) {
1777 u32 port_no
= nla_get_u32(a
[OVS_VPORT_ATTR_PORT_NO
]);
1779 if (port_no
>= DP_MAX_PORTS
)
1780 return ERR_PTR(-EFBIG
);
1782 dp
= get_dp(net
, ovs_header
->dp_ifindex
);
1784 return ERR_PTR(-ENODEV
);
1786 vport
= ovs_vport_ovsl_rcu(dp
, port_no
);
1788 return ERR_PTR(-ENODEV
);
1791 return ERR_PTR(-EINVAL
);
1794 static int ovs_vport_cmd_new(struct sk_buff
*skb
, struct genl_info
*info
)
1796 struct nlattr
**a
= info
->attrs
;
1797 struct ovs_header
*ovs_header
= info
->userhdr
;
1798 struct vport_parms parms
;
1799 struct sk_buff
*reply
;
1800 struct vport
*vport
;
1801 struct datapath
*dp
;
1805 if (!a
[OVS_VPORT_ATTR_NAME
] || !a
[OVS_VPORT_ATTR_TYPE
] ||
1806 !a
[OVS_VPORT_ATTR_UPCALL_PID
])
1809 port_no
= a
[OVS_VPORT_ATTR_PORT_NO
]
1810 ? nla_get_u32(a
[OVS_VPORT_ATTR_PORT_NO
]) : 0;
1811 if (port_no
>= DP_MAX_PORTS
)
1814 reply
= ovs_vport_cmd_alloc_info();
1820 dp
= get_dp(sock_net(skb
->sk
), ovs_header
->dp_ifindex
);
1823 goto exit_unlock_free
;
1826 vport
= ovs_vport_ovsl(dp
, port_no
);
1829 goto exit_unlock_free
;
1831 for (port_no
= 1; ; port_no
++) {
1832 if (port_no
>= DP_MAX_PORTS
) {
1834 goto exit_unlock_free
;
1836 vport
= ovs_vport_ovsl(dp
, port_no
);
1842 parms
.name
= nla_data(a
[OVS_VPORT_ATTR_NAME
]);
1843 parms
.type
= nla_get_u32(a
[OVS_VPORT_ATTR_TYPE
]);
1844 parms
.options
= a
[OVS_VPORT_ATTR_OPTIONS
];
1846 parms
.port_no
= port_no
;
1847 parms
.upcall_portids
= a
[OVS_VPORT_ATTR_UPCALL_PID
];
1849 vport
= new_vport(&parms
);
1850 err
= PTR_ERR(vport
);
1851 if (IS_ERR(vport
)) {
1854 goto exit_unlock_free
;
1857 err
= ovs_vport_cmd_fill_info(vport
, reply
, info
->snd_portid
,
1858 info
->snd_seq
, 0, OVS_VPORT_CMD_NEW
);
1862 ovs_notify(&dp_vport_genl_family
, reply
, info
);
1871 static int ovs_vport_cmd_set(struct sk_buff
*skb
, struct genl_info
*info
)
1873 struct nlattr
**a
= info
->attrs
;
1874 struct sk_buff
*reply
;
1875 struct vport
*vport
;
1878 reply
= ovs_vport_cmd_alloc_info();
1883 vport
= lookup_vport(sock_net(skb
->sk
), info
->userhdr
, a
);
1884 err
= PTR_ERR(vport
);
1886 goto exit_unlock_free
;
1888 if (a
[OVS_VPORT_ATTR_TYPE
] &&
1889 nla_get_u32(a
[OVS_VPORT_ATTR_TYPE
]) != vport
->ops
->type
) {
1891 goto exit_unlock_free
;
1894 if (a
[OVS_VPORT_ATTR_OPTIONS
]) {
1895 err
= ovs_vport_set_options(vport
, a
[OVS_VPORT_ATTR_OPTIONS
]);
1897 goto exit_unlock_free
;
1901 if (a
[OVS_VPORT_ATTR_UPCALL_PID
]) {
1902 struct nlattr
*ids
= a
[OVS_VPORT_ATTR_UPCALL_PID
];
1904 err
= ovs_vport_set_upcall_portids(vport
, ids
);
1906 goto exit_unlock_free
;
1909 err
= ovs_vport_cmd_fill_info(vport
, reply
, info
->snd_portid
,
1910 info
->snd_seq
, 0, OVS_VPORT_CMD_NEW
);
1914 ovs_notify(&dp_vport_genl_family
, reply
, info
);
1923 static int ovs_vport_cmd_del(struct sk_buff
*skb
, struct genl_info
*info
)
1925 struct nlattr
**a
= info
->attrs
;
1926 struct sk_buff
*reply
;
1927 struct vport
*vport
;
1930 reply
= ovs_vport_cmd_alloc_info();
1935 vport
= lookup_vport(sock_net(skb
->sk
), info
->userhdr
, a
);
1936 err
= PTR_ERR(vport
);
1938 goto exit_unlock_free
;
1940 if (vport
->port_no
== OVSP_LOCAL
) {
1942 goto exit_unlock_free
;
1945 err
= ovs_vport_cmd_fill_info(vport
, reply
, info
->snd_portid
,
1946 info
->snd_seq
, 0, OVS_VPORT_CMD_DEL
);
1948 ovs_dp_detach_port(vport
);
1951 ovs_notify(&dp_vport_genl_family
, reply
, info
);
1960 static int ovs_vport_cmd_get(struct sk_buff
*skb
, struct genl_info
*info
)
1962 struct nlattr
**a
= info
->attrs
;
1963 struct ovs_header
*ovs_header
= info
->userhdr
;
1964 struct sk_buff
*reply
;
1965 struct vport
*vport
;
1968 reply
= ovs_vport_cmd_alloc_info();
1973 vport
= lookup_vport(sock_net(skb
->sk
), ovs_header
, a
);
1974 err
= PTR_ERR(vport
);
1976 goto exit_unlock_free
;
1977 err
= ovs_vport_cmd_fill_info(vport
, reply
, info
->snd_portid
,
1978 info
->snd_seq
, 0, OVS_VPORT_CMD_NEW
);
1982 return genlmsg_reply(reply
, info
);
1990 static int ovs_vport_cmd_dump(struct sk_buff
*skb
, struct netlink_callback
*cb
)
1992 struct ovs_header
*ovs_header
= genlmsg_data(nlmsg_data(cb
->nlh
));
1993 struct datapath
*dp
;
1994 int bucket
= cb
->args
[0], skip
= cb
->args
[1];
1998 dp
= get_dp_rcu(sock_net(skb
->sk
), ovs_header
->dp_ifindex
);
2003 for (i
= bucket
; i
< DP_VPORT_HASH_BUCKETS
; i
++) {
2004 struct vport
*vport
;
2007 hlist_for_each_entry_rcu(vport
, &dp
->ports
[i
], dp_hash_node
) {
2009 ovs_vport_cmd_fill_info(vport
, skb
,
2010 NETLINK_CB(cb
->skb
).portid
,
2013 OVS_VPORT_CMD_NEW
) < 0)
2029 static const struct nla_policy vport_policy
[OVS_VPORT_ATTR_MAX
+ 1] = {
2030 [OVS_VPORT_ATTR_NAME
] = { .type
= NLA_NUL_STRING
, .len
= IFNAMSIZ
- 1 },
2031 [OVS_VPORT_ATTR_STATS
] = { .len
= sizeof(struct ovs_vport_stats
) },
2032 [OVS_VPORT_ATTR_PORT_NO
] = { .type
= NLA_U32
},
2033 [OVS_VPORT_ATTR_TYPE
] = { .type
= NLA_U32
},
2034 [OVS_VPORT_ATTR_UPCALL_PID
] = { .type
= NLA_U32
},
2035 [OVS_VPORT_ATTR_OPTIONS
] = { .type
= NLA_NESTED
},
2038 static const struct genl_ops dp_vport_genl_ops
[] = {
2039 { .cmd
= OVS_VPORT_CMD_NEW
,
2040 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
2041 .policy
= vport_policy
,
2042 .doit
= ovs_vport_cmd_new
2044 { .cmd
= OVS_VPORT_CMD_DEL
,
2045 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
2046 .policy
= vport_policy
,
2047 .doit
= ovs_vport_cmd_del
2049 { .cmd
= OVS_VPORT_CMD_GET
,
2050 .flags
= 0, /* OK for unprivileged users. */
2051 .policy
= vport_policy
,
2052 .doit
= ovs_vport_cmd_get
,
2053 .dumpit
= ovs_vport_cmd_dump
2055 { .cmd
= OVS_VPORT_CMD_SET
,
2056 .flags
= GENL_ADMIN_PERM
, /* Requires CAP_NET_ADMIN privilege. */
2057 .policy
= vport_policy
,
2058 .doit
= ovs_vport_cmd_set
,
2062 struct genl_family dp_vport_genl_family
= {
2063 .id
= GENL_ID_GENERATE
,
2064 .hdrsize
= sizeof(struct ovs_header
),
2065 .name
= OVS_VPORT_FAMILY
,
2066 .version
= OVS_VPORT_VERSION
,
2067 .maxattr
= OVS_VPORT_ATTR_MAX
,
2069 .parallel_ops
= true,
2070 .ops
= dp_vport_genl_ops
,
2071 .n_ops
= ARRAY_SIZE(dp_vport_genl_ops
),
2072 .mcgrps
= &ovs_dp_vport_multicast_group
,
2076 static struct genl_family
* const dp_genl_families
[] = {
2077 &dp_datapath_genl_family
,
2078 &dp_vport_genl_family
,
2079 &dp_flow_genl_family
,
2080 &dp_packet_genl_family
,
2083 static void dp_unregister_genl(int n_families
)
2087 for (i
= 0; i
< n_families
; i
++)
2088 genl_unregister_family(dp_genl_families
[i
]);
2091 static int dp_register_genl(void)
2096 for (i
= 0; i
< ARRAY_SIZE(dp_genl_families
); i
++) {
2098 err
= genl_register_family(dp_genl_families
[i
]);
2106 dp_unregister_genl(i
);
2110 static int __net_init
ovs_init_net(struct net
*net
)
2112 struct ovs_net
*ovs_net
= net_generic(net
, ovs_net_id
);
2114 INIT_LIST_HEAD(&ovs_net
->dps
);
2115 INIT_WORK(&ovs_net
->dp_notify_work
, ovs_dp_notify_wq
);
2119 static void __net_exit
ovs_exit_net(struct net
*net
)
2121 struct datapath
*dp
, *dp_next
;
2122 struct ovs_net
*ovs_net
= net_generic(net
, ovs_net_id
);
2125 list_for_each_entry_safe(dp
, dp_next
, &ovs_net
->dps
, list_node
)
2129 cancel_work_sync(&ovs_net
->dp_notify_work
);
2132 static struct pernet_operations ovs_net_ops
= {
2133 .init
= ovs_init_net
,
2134 .exit
= ovs_exit_net
,
2136 .size
= sizeof(struct ovs_net
),
2139 static int __init
dp_init(void)
2143 BUILD_BUG_ON(sizeof(struct ovs_skb_cb
) > FIELD_SIZEOF(struct sk_buff
, cb
));
2145 pr_info("Open vSwitch switching datapath\n");
2147 err
= action_fifos_init();
2151 err
= ovs_internal_dev_rtnl_link_register();
2153 goto error_action_fifos_exit
;
2155 err
= ovs_flow_init();
2157 goto error_unreg_rtnl_link
;
2159 err
= ovs_vport_init();
2161 goto error_flow_exit
;
2163 err
= register_pernet_device(&ovs_net_ops
);
2165 goto error_vport_exit
;
2167 err
= register_netdevice_notifier(&ovs_dp_device_notifier
);
2169 goto error_netns_exit
;
2171 err
= ovs_netdev_init();
2173 goto error_unreg_notifier
;
2175 err
= dp_register_genl();
2177 goto error_unreg_netdev
;
2183 error_unreg_notifier
:
2184 unregister_netdevice_notifier(&ovs_dp_device_notifier
);
2186 unregister_pernet_device(&ovs_net_ops
);
2191 error_unreg_rtnl_link
:
2192 ovs_internal_dev_rtnl_link_unregister();
2193 error_action_fifos_exit
:
2194 action_fifos_exit();
2199 static void dp_cleanup(void)
2201 dp_unregister_genl(ARRAY_SIZE(dp_genl_families
));
2203 unregister_netdevice_notifier(&ovs_dp_device_notifier
);
2204 unregister_pernet_device(&ovs_net_ops
);
2208 ovs_internal_dev_rtnl_link_unregister();
2209 action_fifos_exit();
2212 module_init(dp_init
);
2213 module_exit(dp_cleanup
);
2215 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2216 MODULE_LICENSE("GPL");