2 * Copyright (c) 2007, 2008, 2009, 2010, 2011 Nicira Networks.
3 * Distributed under the terms of the GNU GPL version 2.
5 * Significant portions of this file may be copied from parts of the Linux
6 * kernel, by Linus Torvalds and others.
9 /* Functions for managing the dp interface/device. */
11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
13 #include <linux/init.h>
14 #include <linux/module.h>
15 #include <linux/if_arp.h>
16 #include <linux/if_vlan.h>
19 #include <linux/jhash.h>
20 #include <linux/delay.h>
21 #include <linux/time.h>
22 #include <linux/etherdevice.h>
23 #include <linux/genetlink.h>
24 #include <linux/kernel.h>
25 #include <linux/kthread.h>
26 #include <linux/mutex.h>
27 #include <linux/percpu.h>
28 #include <linux/rcupdate.h>
29 #include <linux/tcp.h>
30 #include <linux/udp.h>
31 #include <linux/version.h>
32 #include <linux/ethtool.h>
33 #include <linux/wait.h>
34 #include <asm/system.h>
35 #include <asm/div64.h>
37 #include <linux/highmem.h>
38 #include <linux/netfilter_bridge.h>
39 #include <linux/netfilter_ipv4.h>
40 #include <linux/inetdevice.h>
41 #include <linux/list.h>
42 #include <linux/rculist.h>
43 #include <linux/dmi.h>
44 #include <net/inet_ecn.h>
45 #include <net/genetlink.h>
47 #include "openvswitch/datapath-protocol.h"
54 #include "vport-internal_dev.h"
56 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,18) || \
57 LINUX_VERSION_CODE >= KERNEL_VERSION(3,1,0)
58 #error Kernels before 2.6.18 or after 3.0 are not supported by this version of Open vSwitch.
61 int (*dp_ioctl_hook)(struct net_device *dev, struct ifreq *rq, int cmd);
62 EXPORT_SYMBOL(dp_ioctl_hook);
67 * Writes to device state (add/remove datapath, port, set operations on vports,
68 * etc.) are protected by RTNL.
70 * Writes to other state (flow table modifications, set miscellaneous datapath
71 * parameters such as drop frags, etc.) are protected by genl_mutex. The RTNL
72 * lock nests inside genl_mutex.
74 * Reads are protected by RCU.
76 * There are a few special cases (mostly stats) that have their own
77 * synchronization but they nest under all of above and don't interact with
81 /* Global list of datapaths to enable dumping them all out.
82 * Protected by genl_mutex.
84 static LIST_HEAD(dps);
86 static struct vport *new_vport(const struct vport_parms *);
87 static int queue_userspace_packets(struct datapath *, struct sk_buff *,
88 const struct dp_upcall_info *);
90 /* Must be called with rcu_read_lock, genl_mutex, or RTNL lock. */
91 struct datapath *get_dp(int dp_ifindex)
93 struct datapath *dp = NULL;
94 struct net_device *dev;
97 dev = dev_get_by_index_rcu(&init_net, dp_ifindex);
99 struct vport *vport = internal_dev_get_vport(dev);
107 EXPORT_SYMBOL_GPL(get_dp);
109 /* Must be called with genl_mutex. */
110 static struct tbl *get_table_protected(struct datapath *dp)
112 return rcu_dereference_protected(dp->table, lockdep_genl_is_held());
115 /* Must be called with rcu_read_lock or RTNL lock. */
116 static struct vport *get_vport_protected(struct datapath *dp, u16 port_no)
118 return rcu_dereference_rtnl(dp->ports[port_no]);
121 /* Must be called with rcu_read_lock or RTNL lock. */
122 const char *dp_name(const struct datapath *dp)
124 return vport_get_name(rcu_dereference_rtnl(dp->ports[OVSP_LOCAL]));
127 static inline size_t br_nlmsg_size(void)
129 return NLMSG_ALIGN(sizeof(struct ifinfomsg))
130 + nla_total_size(IFNAMSIZ) /* IFLA_IFNAME */
131 + nla_total_size(MAX_ADDR_LEN) /* IFLA_ADDRESS */
132 + nla_total_size(4) /* IFLA_MASTER */
133 + nla_total_size(4) /* IFLA_MTU */
134 + nla_total_size(1); /* IFLA_OPERSTATE */
137 /* Caller must hold RTNL lock. */
138 static int dp_fill_ifinfo(struct sk_buff *skb,
139 const struct vport *port,
140 int event, unsigned int flags)
142 struct datapath *dp = port->dp;
143 int ifindex = vport_get_ifindex(port);
144 struct ifinfomsg *hdr;
145 struct nlmsghdr *nlh;
150 nlh = nlmsg_put(skb, 0, 0, event, sizeof(*hdr), flags);
154 hdr = nlmsg_data(nlh);
155 hdr->ifi_family = AF_BRIDGE;
157 hdr->ifi_type = ARPHRD_ETHER;
158 hdr->ifi_index = ifindex;
159 hdr->ifi_flags = vport_get_flags(port);
162 NLA_PUT_STRING(skb, IFLA_IFNAME, vport_get_name(port));
163 NLA_PUT_U32(skb, IFLA_MASTER,
164 vport_get_ifindex(get_vport_protected(dp, OVSP_LOCAL)));
165 NLA_PUT_U32(skb, IFLA_MTU, vport_get_mtu(port));
166 #ifdef IFLA_OPERSTATE
167 NLA_PUT_U8(skb, IFLA_OPERSTATE,
168 vport_is_running(port)
169 ? vport_get_operstate(port)
173 NLA_PUT(skb, IFLA_ADDRESS, ETH_ALEN, vport_get_addr(port));
175 return nlmsg_end(skb, nlh);
178 nlmsg_cancel(skb, nlh);
182 /* Caller must hold RTNL lock. */
183 static void dp_ifinfo_notify(int event, struct vport *port)
188 skb = nlmsg_new(br_nlmsg_size(), GFP_KERNEL);
192 err = dp_fill_ifinfo(skb, port, event, 0);
194 /* -EMSGSIZE implies BUG in br_nlmsg_size() */
195 WARN_ON(err == -EMSGSIZE);
199 rtnl_notify(skb, &init_net, 0, RTNLGRP_LINK, NULL, GFP_KERNEL);
203 rtnl_set_sk_err(&init_net, RTNLGRP_LINK, err);
206 static void release_dp(struct kobject *kobj)
208 struct datapath *dp = container_of(kobj, struct datapath, ifobj);
212 static struct kobj_type dp_ktype = {
213 .release = release_dp
216 static void destroy_dp_rcu(struct rcu_head *rcu)
218 struct datapath *dp = container_of(rcu, struct datapath, rcu);
220 tbl_destroy((struct tbl __force *)dp->table, flow_free_tbl);
221 free_percpu(dp->stats_percpu);
222 kobject_put(&dp->ifobj);
225 /* Called with RTNL lock and genl_lock. */
226 static struct vport *new_vport(const struct vport_parms *parms)
230 vport = vport_add(parms);
231 if (!IS_ERR(vport)) {
232 struct datapath *dp = parms->dp;
234 rcu_assign_pointer(dp->ports[parms->port_no], vport);
235 list_add(&vport->node, &dp->port_list);
237 dp_ifinfo_notify(RTM_NEWLINK, vport);
243 /* Called with RTNL lock. */
244 int dp_detach_port(struct vport *p)
248 if (p->port_no != OVSP_LOCAL)
250 dp_ifinfo_notify(RTM_DELLINK, p);
252 /* First drop references to device. */
254 rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
256 /* Then destroy it. */
260 /* Must be called with rcu_read_lock. */
261 void dp_process_received_packet(struct vport *p, struct sk_buff *skb)
263 struct datapath *dp = p->dp;
264 struct dp_stats_percpu *stats;
265 int stats_counter_off;
268 OVS_CB(skb)->vport = p;
270 if (!OVS_CB(skb)->flow) {
271 struct sw_flow_key key;
272 struct tbl_node *flow_node;
276 /* Extract flow from 'skb' into 'key'. */
277 error = flow_extract(skb, p->port_no, &key, &key_len, &is_frag);
278 if (unlikely(error)) {
283 if (is_frag && dp->drop_frags) {
285 stats_counter_off = offsetof(struct dp_stats_percpu, n_frags);
290 flow_node = tbl_lookup(rcu_dereference(dp->table), &key, key_len,
291 flow_hash(&key, key_len), flow_cmp);
292 if (unlikely(!flow_node)) {
293 struct dp_upcall_info upcall;
295 upcall.cmd = OVS_PACKET_CMD_MISS;
298 upcall.sample_pool = 0;
299 upcall.actions = NULL;
300 upcall.actions_len = 0;
301 dp_upcall(dp, skb, &upcall);
302 stats_counter_off = offsetof(struct dp_stats_percpu, n_missed);
306 OVS_CB(skb)->flow = flow_cast(flow_node);
309 stats_counter_off = offsetof(struct dp_stats_percpu, n_hit);
310 flow_used(OVS_CB(skb)->flow, skb);
311 execute_actions(dp, skb);
314 /* Update datapath statistics. */
316 stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
318 write_seqcount_begin(&stats->seqlock);
319 (*(u64 *)((u8 *)stats + stats_counter_off))++;
320 write_seqcount_end(&stats->seqlock);
325 static void copy_and_csum_skb(struct sk_buff *skb, void *to)
327 u16 csum_start, csum_offset;
330 get_skb_csum_pointers(skb, &csum_start, &csum_offset);
331 csum_start -= skb_headroom(skb);
333 skb_copy_bits(skb, 0, to, csum_start);
335 csum = skb_copy_and_csum_bits(skb, csum_start, to + csum_start,
336 skb->len - csum_start, 0);
337 *(__sum16 *)(to + csum_start + csum_offset) = csum_fold(csum);
340 static struct genl_family dp_packet_genl_family = {
341 .id = GENL_ID_GENERATE,
342 .hdrsize = sizeof(struct ovs_header),
343 .name = OVS_PACKET_FAMILY,
345 .maxattr = OVS_PACKET_ATTR_MAX
348 /* Generic Netlink multicast groups for upcalls.
350 * We really want three unique multicast groups per datapath, but we can't even
351 * get one, because genl_register_mc_group() takes genl_lock, which is also
352 * held during Generic Netlink message processing, so trying to acquire
353 * multicast groups during OVS_DP_NEW processing deadlocks. Instead, we
354 * preallocate a few groups and use them round-robin for datapaths. Collision
355 * isn't fatal--multicast listeners should check that the family is the one
356 * that they want and discard others--but it wastes time and memory to receive
359 #define PACKET_N_MC_GROUPS 16
360 static struct genl_multicast_group packet_mc_groups[PACKET_N_MC_GROUPS];
362 static u32 packet_mc_group(struct datapath *dp, u8 cmd)
365 BUILD_BUG_ON_NOT_POWER_OF_2(PACKET_N_MC_GROUPS);
367 idx = jhash_2words(dp->dp_ifindex, cmd, 0) & (PACKET_N_MC_GROUPS - 1);
368 return packet_mc_groups[idx].id;
371 static int packet_register_mc_groups(void)
375 for (i = 0; i < PACKET_N_MC_GROUPS; i++) {
376 struct genl_multicast_group *group = &packet_mc_groups[i];
379 sprintf(group->name, "packet%d", i);
380 error = genl_register_mc_group(&dp_packet_genl_family, group);
387 int dp_upcall(struct datapath *dp, struct sk_buff *skb, const struct dp_upcall_info *upcall_info)
389 struct dp_stats_percpu *stats;
392 WARN_ON_ONCE(skb_shared(skb));
394 forward_ip_summed(skb, true);
396 /* Break apart GSO packets into their component pieces. Otherwise
397 * userspace may try to stuff a 64kB packet into a 1500-byte MTU. */
398 if (skb_is_gso(skb)) {
399 struct sk_buff *nskb = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
410 err = queue_userspace_packets(dp, skb, upcall_info);
418 stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
420 write_seqcount_begin(&stats->seqlock);
422 write_seqcount_end(&stats->seqlock);
429 /* Send each packet in the 'skb' list to userspace for 'dp' as directed by
430 * 'upcall_info'. There will be only one packet unless we broke up a GSO
433 static int queue_userspace_packets(struct datapath *dp, struct sk_buff *skb,
434 const struct dp_upcall_info *upcall_info)
436 u32 group = packet_mc_group(dp, upcall_info->cmd);
437 struct sk_buff *nskb;
441 struct ovs_header *upcall;
442 struct sk_buff *user_skb; /* to be queued to userspace */
449 err = vlan_deaccel_tag(skb);
453 if (nla_attr_size(skb->len) > USHRT_MAX)
456 len = sizeof(struct ovs_header);
457 len += nla_total_size(skb->len);
458 len += nla_total_size(FLOW_BUFSIZE);
459 if (upcall_info->userdata)
460 len += nla_total_size(8);
461 if (upcall_info->sample_pool)
462 len += nla_total_size(4);
463 if (upcall_info->actions_len)
464 len += nla_total_size(upcall_info->actions_len);
466 user_skb = genlmsg_new(len, GFP_ATOMIC);
468 netlink_set_err(INIT_NET_GENL_SOCK, 0, group, -ENOBUFS);
472 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family, 0, upcall_info->cmd);
473 upcall->dp_ifindex = dp->dp_ifindex;
475 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
476 flow_to_nlattrs(upcall_info->key, user_skb);
477 nla_nest_end(user_skb, nla);
479 if (upcall_info->userdata)
480 nla_put_u64(user_skb, OVS_PACKET_ATTR_USERDATA, upcall_info->userdata);
481 if (upcall_info->sample_pool)
482 nla_put_u32(user_skb, OVS_PACKET_ATTR_SAMPLE_POOL, upcall_info->sample_pool);
483 if (upcall_info->actions_len) {
484 const struct nlattr *actions = upcall_info->actions;
485 u32 actions_len = upcall_info->actions_len;
487 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_ACTIONS);
488 memcpy(__skb_put(user_skb, actions_len), actions, actions_len);
489 nla_nest_end(user_skb, nla);
492 nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
493 if (skb->ip_summed == CHECKSUM_PARTIAL)
494 copy_and_csum_skb(skb, nla_data(nla));
496 skb_copy_bits(skb, 0, nla_data(nla), skb->len);
498 err = genlmsg_multicast(user_skb, 0, group, GFP_ATOMIC);
509 while ((skb = nskb) != NULL) {
516 /* Called with genl_mutex. */
517 static int flush_flows(int dp_ifindex)
519 struct tbl *old_table;
520 struct tbl *new_table;
523 dp = get_dp(dp_ifindex);
527 old_table = get_table_protected(dp);
528 new_table = tbl_create(TBL_MIN_BUCKETS);
532 rcu_assign_pointer(dp->table, new_table);
534 tbl_deferred_destroy(old_table, flow_free_tbl);
539 static int validate_actions(const struct nlattr *attr)
541 const struct nlattr *a;
544 nla_for_each_nested(a, attr, rem) {
545 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
546 [OVS_ACTION_ATTR_OUTPUT] = 4,
547 [OVS_ACTION_ATTR_USERSPACE] = 8,
548 [OVS_ACTION_ATTR_PUSH_VLAN] = 2,
549 [OVS_ACTION_ATTR_POP_VLAN] = 0,
550 [OVS_ACTION_ATTR_SET_DL_SRC] = ETH_ALEN,
551 [OVS_ACTION_ATTR_SET_DL_DST] = ETH_ALEN,
552 [OVS_ACTION_ATTR_SET_NW_SRC] = 4,
553 [OVS_ACTION_ATTR_SET_NW_DST] = 4,
554 [OVS_ACTION_ATTR_SET_NW_TOS] = 1,
555 [OVS_ACTION_ATTR_SET_TP_SRC] = 2,
556 [OVS_ACTION_ATTR_SET_TP_DST] = 2,
557 [OVS_ACTION_ATTR_SET_TUNNEL] = 8,
558 [OVS_ACTION_ATTR_SET_PRIORITY] = 4,
559 [OVS_ACTION_ATTR_POP_PRIORITY] = 0,
561 int type = nla_type(a);
563 if (type > OVS_ACTION_ATTR_MAX || nla_len(a) != action_lens[type])
567 case OVS_ACTION_ATTR_UNSPEC:
570 case OVS_ACTION_ATTR_USERSPACE:
571 case OVS_ACTION_ATTR_POP_VLAN:
572 case OVS_ACTION_ATTR_SET_DL_SRC:
573 case OVS_ACTION_ATTR_SET_DL_DST:
574 case OVS_ACTION_ATTR_SET_NW_SRC:
575 case OVS_ACTION_ATTR_SET_NW_DST:
576 case OVS_ACTION_ATTR_SET_TP_SRC:
577 case OVS_ACTION_ATTR_SET_TP_DST:
578 case OVS_ACTION_ATTR_SET_TUNNEL:
579 case OVS_ACTION_ATTR_SET_PRIORITY:
580 case OVS_ACTION_ATTR_POP_PRIORITY:
581 /* No validation needed. */
584 case OVS_ACTION_ATTR_OUTPUT:
585 if (nla_get_u32(a) >= DP_MAX_PORTS)
589 case OVS_ACTION_ATTR_PUSH_VLAN:
590 if (nla_get_be16(a) & htons(VLAN_CFI_MASK))
594 case OVS_ACTION_ATTR_SET_NW_TOS:
595 if (nla_get_u8(a) & INET_ECN_MASK)
609 static void clear_stats(struct sw_flow *flow)
613 flow->packet_count = 0;
614 flow->byte_count = 0;
617 /* Called with genl_mutex. */
618 static int expand_table(struct datapath *dp)
620 struct tbl *old_table = get_table_protected(dp);
621 struct tbl *new_table;
623 new_table = tbl_expand(old_table);
624 if (IS_ERR(new_table)) {
625 if (PTR_ERR(new_table) != -ENOSPC)
626 return PTR_ERR(new_table);
628 rcu_assign_pointer(dp->table, new_table);
629 tbl_deferred_destroy(old_table, NULL);
635 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
637 struct ovs_header *ovs_header = info->userhdr;
638 struct nlattr **a = info->attrs;
639 struct sw_flow_actions *acts;
640 struct sk_buff *packet;
641 struct sw_flow *flow;
650 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
651 !a[OVS_PACKET_ATTR_ACTIONS] ||
652 nla_len(a[OVS_PACKET_ATTR_PACKET]) < ETH_HLEN)
655 err = validate_actions(a[OVS_PACKET_ATTR_ACTIONS]);
659 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
660 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
664 skb_reserve(packet, NET_IP_ALIGN);
666 memcpy(__skb_put(packet, len), nla_data(a[OVS_PACKET_ATTR_PACKET]), len);
668 skb_reset_mac_header(packet);
669 eth = eth_hdr(packet);
671 /* Normally, setting the skb 'protocol' field would be handled by a
672 * call to eth_type_trans(), but it assumes there's a sending
673 * device, which we may not have. */
674 if (ntohs(eth->h_proto) >= 1536)
675 packet->protocol = eth->h_proto;
677 packet->protocol = htons(ETH_P_802_2);
679 /* Build an sw_flow for sending this packet. */
685 err = flow_extract(packet, -1, &flow->key, &key_len, &is_frag);
689 err = flow_metadata_from_nlattrs(&flow->key.eth.in_port,
690 &flow->key.eth.tun_id,
691 a[OVS_PACKET_ATTR_KEY]);
695 flow->tbl_node.hash = flow_hash(&flow->key, key_len);
697 acts = flow_actions_alloc(a[OVS_PACKET_ATTR_ACTIONS]);
701 rcu_assign_pointer(flow->sf_acts, acts);
703 OVS_CB(packet)->flow = flow;
706 dp = get_dp(ovs_header->dp_ifindex);
710 err = execute_actions(dp, packet);
726 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
727 [OVS_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
728 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
729 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
732 static struct genl_ops dp_packet_genl_ops[] = {
733 { .cmd = OVS_PACKET_CMD_EXECUTE,
734 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
735 .policy = packet_policy,
736 .doit = ovs_packet_cmd_execute
740 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
743 struct tbl *table = get_table_protected(dp);
745 stats->n_flows = tbl_count(table);
747 stats->n_frags = stats->n_hit = stats->n_missed = stats->n_lost = 0;
748 for_each_possible_cpu(i) {
749 const struct dp_stats_percpu *percpu_stats;
750 struct dp_stats_percpu local_stats;
753 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
756 seqcount = read_seqcount_begin(&percpu_stats->seqlock);
757 local_stats = *percpu_stats;
758 } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount));
760 stats->n_frags += local_stats.n_frags;
761 stats->n_hit += local_stats.n_hit;
762 stats->n_missed += local_stats.n_missed;
763 stats->n_lost += local_stats.n_lost;
767 /* MTU of the dp pseudo-device: ETH_DATA_LEN or the minimum of the ports.
768 * Called with RTNL lock.
770 int dp_min_mtu(const struct datapath *dp)
777 list_for_each_entry (p, &dp->port_list, node) {
780 /* Skip any internal ports, since that's what we're trying to
782 if (is_internal_vport(p))
785 dev_mtu = vport_get_mtu(p);
788 if (!mtu || dev_mtu < mtu)
792 return mtu ? mtu : ETH_DATA_LEN;
795 /* Sets the MTU of all datapath devices to the minimum of the ports
796 * Called with RTNL lock.
798 void set_internal_devs_mtu(const struct datapath *dp)
805 mtu = dp_min_mtu(dp);
807 list_for_each_entry (p, &dp->port_list, node) {
808 if (is_internal_vport(p))
809 vport_set_mtu(p, mtu);
813 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
814 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
815 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
816 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
819 static struct genl_family dp_flow_genl_family = {
820 .id = GENL_ID_GENERATE,
821 .hdrsize = sizeof(struct ovs_header),
822 .name = OVS_FLOW_FAMILY,
824 .maxattr = OVS_FLOW_ATTR_MAX
827 static struct genl_multicast_group dp_flow_multicast_group = {
828 .name = OVS_FLOW_MCGROUP
831 /* Called with genl_lock. */
832 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
833 struct sk_buff *skb, u32 pid, u32 seq, u32 flags, u8 cmd)
835 const int skb_orig_len = skb->len;
836 const struct sw_flow_actions *sf_acts;
837 struct ovs_flow_stats stats;
838 struct ovs_header *ovs_header;
844 sf_acts = rcu_dereference_protected(flow->sf_acts,
845 lockdep_genl_is_held());
847 ovs_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags, cmd);
851 ovs_header->dp_ifindex = dp->dp_ifindex;
853 nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
855 goto nla_put_failure;
856 err = flow_to_nlattrs(&flow->key, skb);
859 nla_nest_end(skb, nla);
861 spin_lock_bh(&flow->lock);
863 stats.n_packets = flow->packet_count;
864 stats.n_bytes = flow->byte_count;
865 tcp_flags = flow->tcp_flags;
866 spin_unlock_bh(&flow->lock);
869 NLA_PUT_U64(skb, OVS_FLOW_ATTR_USED, flow_used_time(used));
872 NLA_PUT(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats);
875 NLA_PUT_U8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags);
877 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
878 * this is the first flow to be dumped into 'skb'. This is unusual for
879 * Netlink but individual action lists can be longer than
880 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
881 * The userspace caller can always fetch the actions separately if it
882 * really wants them. (Most userspace callers in fact don't care.)
884 * This can only fail for dump operations because the skb is always
885 * properly sized for single flows.
887 err = nla_put(skb, OVS_FLOW_ATTR_ACTIONS, sf_acts->actions_len,
889 if (err < 0 && skb_orig_len)
892 return genlmsg_end(skb, ovs_header);
897 genlmsg_cancel(skb, ovs_header);
901 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
903 const struct sw_flow_actions *sf_acts;
906 sf_acts = rcu_dereference_protected(flow->sf_acts,
907 lockdep_genl_is_held());
909 len = nla_total_size(FLOW_BUFSIZE); /* OVS_FLOW_ATTR_KEY */
910 len += nla_total_size(sf_acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
911 len += nla_total_size(sizeof(struct ovs_flow_stats)); /* OVS_FLOW_ATTR_STATS */
912 len += nla_total_size(1); /* OVS_FLOW_ATTR_TCP_FLAGS */
913 len += nla_total_size(8); /* OVS_FLOW_ATTR_USED */
914 return genlmsg_new(NLMSG_ALIGN(sizeof(struct ovs_header)) + len, GFP_KERNEL);
917 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow, struct datapath *dp,
918 u32 pid, u32 seq, u8 cmd)
923 skb = ovs_flow_cmd_alloc_info(flow);
925 return ERR_PTR(-ENOMEM);
927 retval = ovs_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd);
932 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
934 struct nlattr **a = info->attrs;
935 struct ovs_header *ovs_header = info->userhdr;
936 struct tbl_node *flow_node;
937 struct sw_flow_key key;
938 struct sw_flow *flow;
939 struct sk_buff *reply;
948 if (!a[OVS_FLOW_ATTR_KEY])
950 error = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
954 /* Validate actions. */
955 if (a[OVS_FLOW_ATTR_ACTIONS]) {
956 error = validate_actions(a[OVS_FLOW_ATTR_ACTIONS]);
959 } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
964 dp = get_dp(ovs_header->dp_ifindex);
969 hash = flow_hash(&key, key_len);
970 table = get_table_protected(dp);
971 flow_node = tbl_lookup(table, &key, key_len, hash, flow_cmp);
973 struct sw_flow_actions *acts;
975 /* Bail out if we're not allowed to create a new flow. */
977 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
980 /* Expand table, if necessary, to make room. */
981 if (tbl_count(table) >= tbl_n_buckets(table)) {
982 error = expand_table(dp);
985 table = get_table_protected(dp);
991 error = PTR_ERR(flow);
997 /* Obtain actions. */
998 acts = flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]);
999 error = PTR_ERR(acts);
1001 goto error_free_flow;
1002 rcu_assign_pointer(flow->sf_acts, acts);
1004 /* Put flow in bucket. */
1005 error = tbl_insert(table, &flow->tbl_node, hash);
1007 goto error_free_flow;
1009 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1010 info->snd_seq, OVS_FLOW_CMD_NEW);
1012 /* We found a matching flow. */
1013 struct sw_flow_actions *old_acts;
1015 /* Bail out if we're not allowed to modify an existing flow.
1016 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1017 * because Generic Netlink treats the latter as a dump
1018 * request. We also accept NLM_F_EXCL in case that bug ever
1022 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
1023 info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1026 /* Update actions. */
1027 flow = flow_cast(flow_node);
1028 old_acts = rcu_dereference_protected(flow->sf_acts,
1029 lockdep_genl_is_held());
1030 if (a[OVS_FLOW_ATTR_ACTIONS] &&
1031 (old_acts->actions_len != nla_len(a[OVS_FLOW_ATTR_ACTIONS]) ||
1032 memcmp(old_acts->actions, nla_data(a[OVS_FLOW_ATTR_ACTIONS]),
1033 old_acts->actions_len))) {
1034 struct sw_flow_actions *new_acts;
1036 new_acts = flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]);
1037 error = PTR_ERR(new_acts);
1038 if (IS_ERR(new_acts))
1041 rcu_assign_pointer(flow->sf_acts, new_acts);
1042 flow_deferred_free_acts(old_acts);
1045 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1046 info->snd_seq, OVS_FLOW_CMD_NEW);
1049 if (a[OVS_FLOW_ATTR_CLEAR]) {
1050 spin_lock_bh(&flow->lock);
1052 spin_unlock_bh(&flow->lock);
1057 genl_notify(reply, genl_info_net(info), info->snd_pid,
1058 dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1060 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1061 dp_flow_multicast_group.id, PTR_ERR(reply));
1070 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1072 struct nlattr **a = info->attrs;
1073 struct ovs_header *ovs_header = info->userhdr;
1074 struct sw_flow_key key;
1075 struct tbl_node *flow_node;
1076 struct sk_buff *reply;
1077 struct sw_flow *flow;
1078 struct datapath *dp;
1083 if (!a[OVS_FLOW_ATTR_KEY])
1085 err = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1089 dp = get_dp(ovs_header->dp_ifindex);
1093 table = get_table_protected(dp);
1094 flow_node = tbl_lookup(table, &key, key_len, flow_hash(&key, key_len),
1099 flow = flow_cast(flow_node);
1100 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid, info->snd_seq, OVS_FLOW_CMD_NEW);
1102 return PTR_ERR(reply);
1104 return genlmsg_reply(reply, info);
1107 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1109 struct nlattr **a = info->attrs;
1110 struct ovs_header *ovs_header = info->userhdr;
1111 struct sw_flow_key key;
1112 struct tbl_node *flow_node;
1113 struct sk_buff *reply;
1114 struct sw_flow *flow;
1115 struct datapath *dp;
1120 if (!a[OVS_FLOW_ATTR_KEY])
1121 return flush_flows(ovs_header->dp_ifindex);
1122 err = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1126 dp = get_dp(ovs_header->dp_ifindex);
1130 table = get_table_protected(dp);
1131 flow_node = tbl_lookup(table, &key, key_len, flow_hash(&key, key_len),
1135 flow = flow_cast(flow_node);
1137 reply = ovs_flow_cmd_alloc_info(flow);
1141 err = tbl_remove(table, flow_node);
1147 err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_pid,
1148 info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1151 flow_deferred_free(flow);
1153 genl_notify(reply, genl_info_net(info), info->snd_pid,
1154 dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1158 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1160 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1161 struct datapath *dp;
1163 dp = get_dp(ovs_header->dp_ifindex);
1168 struct tbl_node *flow_node;
1169 struct sw_flow *flow;
1172 bucket = cb->args[0];
1174 flow_node = tbl_next(get_table_protected(dp), &bucket, &obj);
1178 flow = flow_cast(flow_node);
1179 if (ovs_flow_cmd_fill_info(flow, dp, skb, NETLINK_CB(cb->skb).pid,
1180 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1181 OVS_FLOW_CMD_NEW) < 0)
1184 cb->args[0] = bucket;
1190 static struct genl_ops dp_flow_genl_ops[] = {
1191 { .cmd = OVS_FLOW_CMD_NEW,
1192 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1193 .policy = flow_policy,
1194 .doit = ovs_flow_cmd_new_or_set
1196 { .cmd = OVS_FLOW_CMD_DEL,
1197 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1198 .policy = flow_policy,
1199 .doit = ovs_flow_cmd_del
1201 { .cmd = OVS_FLOW_CMD_GET,
1202 .flags = 0, /* OK for unprivileged users. */
1203 .policy = flow_policy,
1204 .doit = ovs_flow_cmd_get,
1205 .dumpit = ovs_flow_cmd_dump
1207 { .cmd = OVS_FLOW_CMD_SET,
1208 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1209 .policy = flow_policy,
1210 .doit = ovs_flow_cmd_new_or_set,
1214 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1215 #ifdef HAVE_NLA_NUL_STRING
1216 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1218 [OVS_DP_ATTR_IPV4_FRAGS] = { .type = NLA_U32 },
1219 [OVS_DP_ATTR_SAMPLING] = { .type = NLA_U32 },
1222 static struct genl_family dp_datapath_genl_family = {
1223 .id = GENL_ID_GENERATE,
1224 .hdrsize = sizeof(struct ovs_header),
1225 .name = OVS_DATAPATH_FAMILY,
1227 .maxattr = OVS_DP_ATTR_MAX
1230 static struct genl_multicast_group dp_datapath_multicast_group = {
1231 .name = OVS_DATAPATH_MCGROUP
1234 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1235 u32 pid, u32 seq, u32 flags, u8 cmd)
1237 struct ovs_header *ovs_header;
1241 ovs_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family,
1246 ovs_header->dp_ifindex = dp->dp_ifindex;
1249 err = nla_put_string(skb, OVS_DP_ATTR_NAME, dp_name(dp));
1252 goto nla_put_failure;
1254 nla = nla_reserve(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats));
1256 goto nla_put_failure;
1257 get_dp_stats(dp, nla_data(nla));
1259 NLA_PUT_U32(skb, OVS_DP_ATTR_IPV4_FRAGS,
1260 dp->drop_frags ? OVS_DP_FRAG_DROP : OVS_DP_FRAG_ZERO);
1262 if (dp->sflow_probability)
1263 NLA_PUT_U32(skb, OVS_DP_ATTR_SAMPLING, dp->sflow_probability);
1265 nla = nla_nest_start(skb, OVS_DP_ATTR_MCGROUPS);
1267 goto nla_put_failure;
1268 NLA_PUT_U32(skb, OVS_PACKET_CMD_MISS, packet_mc_group(dp, OVS_PACKET_CMD_MISS));
1269 NLA_PUT_U32(skb, OVS_PACKET_CMD_ACTION, packet_mc_group(dp, OVS_PACKET_CMD_ACTION));
1270 NLA_PUT_U32(skb, OVS_PACKET_CMD_SAMPLE, packet_mc_group(dp, OVS_PACKET_CMD_SAMPLE));
1271 nla_nest_end(skb, nla);
1273 return genlmsg_end(skb, ovs_header);
1276 genlmsg_cancel(skb, ovs_header);
1281 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 pid,
1284 struct sk_buff *skb;
1287 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1289 return ERR_PTR(-ENOMEM);
1291 retval = ovs_dp_cmd_fill_info(dp, skb, pid, seq, 0, cmd);
1294 return ERR_PTR(retval);
1299 static int ovs_dp_cmd_validate(struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1301 if (a[OVS_DP_ATTR_IPV4_FRAGS]) {
1302 u32 frags = nla_get_u32(a[OVS_DP_ATTR_IPV4_FRAGS]);
1304 if (frags != OVS_DP_FRAG_ZERO && frags != OVS_DP_FRAG_DROP)
1308 return CHECK_NUL_STRING(a[OVS_DP_ATTR_NAME], IFNAMSIZ - 1);
1311 /* Called with genl_mutex and optionally with RTNL lock also. */
1312 static struct datapath *lookup_datapath(struct ovs_header *ovs_header, struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1314 struct datapath *dp;
1316 if (!a[OVS_DP_ATTR_NAME])
1317 dp = get_dp(ovs_header->dp_ifindex);
1319 struct vport *vport;
1322 vport = vport_locate(nla_data(a[OVS_DP_ATTR_NAME]));
1323 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1326 return dp ? dp : ERR_PTR(-ENODEV);
1329 /* Called with genl_mutex. */
1330 static void change_datapath(struct datapath *dp, struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1332 if (a[OVS_DP_ATTR_IPV4_FRAGS])
1333 dp->drop_frags = nla_get_u32(a[OVS_DP_ATTR_IPV4_FRAGS]) == OVS_DP_FRAG_DROP;
1334 if (a[OVS_DP_ATTR_SAMPLING])
1335 dp->sflow_probability = nla_get_u32(a[OVS_DP_ATTR_SAMPLING]);
1338 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1340 struct nlattr **a = info->attrs;
1341 struct vport_parms parms;
1342 struct sk_buff *reply;
1343 struct datapath *dp;
1344 struct vport *vport;
1348 if (!a[OVS_DP_ATTR_NAME])
1351 err = ovs_dp_cmd_validate(a);
1357 if (!try_module_get(THIS_MODULE))
1358 goto err_unlock_rtnl;
1361 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1363 goto err_put_module;
1364 INIT_LIST_HEAD(&dp->port_list);
1366 /* Initialize kobject for bridge. This will be added as
1367 * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
1368 dp->ifobj.kset = NULL;
1369 kobject_init(&dp->ifobj, &dp_ktype);
1371 /* Allocate table. */
1373 rcu_assign_pointer(dp->table, tbl_create(TBL_MIN_BUCKETS));
1377 /* Set up our datapath device. */
1378 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1379 parms.type = OVS_VPORT_TYPE_INTERNAL;
1380 parms.options = NULL;
1382 parms.port_no = OVSP_LOCAL;
1383 vport = new_vport(&parms);
1384 if (IS_ERR(vport)) {
1385 err = PTR_ERR(vport);
1389 goto err_destroy_table;
1391 dp->dp_ifindex = vport_get_ifindex(vport);
1394 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1395 if (!dp->stats_percpu) {
1397 goto err_destroy_local_port;
1400 change_datapath(dp, a);
1402 reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_NEW);
1403 err = PTR_ERR(reply);
1405 goto err_destroy_local_port;
1407 list_add_tail(&dp->list_node, &dps);
1408 dp_sysfs_add_dp(dp);
1412 genl_notify(reply, genl_info_net(info), info->snd_pid,
1413 dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1416 err_destroy_local_port:
1417 dp_detach_port(get_vport_protected(dp, OVSP_LOCAL));
1419 tbl_destroy(get_table_protected(dp), NULL);
1423 module_put(THIS_MODULE);
1430 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1432 struct vport *vport, *next_vport;
1433 struct sk_buff *reply;
1434 struct datapath *dp;
1437 err = ovs_dp_cmd_validate(info->attrs);
1442 dp = lookup_datapath(info->userhdr, info->attrs);
1447 reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_DEL);
1448 err = PTR_ERR(reply);
1452 list_for_each_entry_safe (vport, next_vport, &dp->port_list, node)
1453 if (vport->port_no != OVSP_LOCAL)
1454 dp_detach_port(vport);
1456 dp_sysfs_del_dp(dp);
1457 list_del(&dp->list_node);
1458 dp_detach_port(get_vport_protected(dp, OVSP_LOCAL));
1460 /* rtnl_unlock() will wait until all the references to devices that
1461 * are pending unregistration have been dropped. We do it here to
1462 * ensure that any internal devices (which contain DP pointers) are
1463 * fully destroyed before freeing the datapath.
1467 call_rcu(&dp->rcu, destroy_dp_rcu);
1468 module_put(THIS_MODULE);
1470 genl_notify(reply, genl_info_net(info), info->snd_pid,
1471 dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1481 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1483 struct sk_buff *reply;
1484 struct datapath *dp;
1487 err = ovs_dp_cmd_validate(info->attrs);
1491 dp = lookup_datapath(info->userhdr, info->attrs);
1495 change_datapath(dp, info->attrs);
1497 reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_NEW);
1498 if (IS_ERR(reply)) {
1499 err = PTR_ERR(reply);
1500 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1501 dp_datapath_multicast_group.id, err);
1505 genl_notify(reply, genl_info_net(info), info->snd_pid,
1506 dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1510 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1512 struct sk_buff *reply;
1513 struct datapath *dp;
1516 err = ovs_dp_cmd_validate(info->attrs);
1520 dp = lookup_datapath(info->userhdr, info->attrs);
1524 reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_NEW);
1526 return PTR_ERR(reply);
1528 return genlmsg_reply(reply, info);
1531 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1533 struct datapath *dp;
1534 int skip = cb->args[0];
1537 list_for_each_entry (dp, &dps, list_node) {
1540 if (ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).pid,
1541 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1542 OVS_DP_CMD_NEW) < 0)
1552 static struct genl_ops dp_datapath_genl_ops[] = {
1553 { .cmd = OVS_DP_CMD_NEW,
1554 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1555 .policy = datapath_policy,
1556 .doit = ovs_dp_cmd_new
1558 { .cmd = OVS_DP_CMD_DEL,
1559 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1560 .policy = datapath_policy,
1561 .doit = ovs_dp_cmd_del
1563 { .cmd = OVS_DP_CMD_GET,
1564 .flags = 0, /* OK for unprivileged users. */
1565 .policy = datapath_policy,
1566 .doit = ovs_dp_cmd_get,
1567 .dumpit = ovs_dp_cmd_dump
1569 { .cmd = OVS_DP_CMD_SET,
1570 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1571 .policy = datapath_policy,
1572 .doit = ovs_dp_cmd_set,
1576 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1577 #ifdef HAVE_NLA_NUL_STRING
1578 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1579 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1580 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1581 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct rtnl_link_stats64) },
1582 [OVS_VPORT_ATTR_ADDRESS] = { .len = ETH_ALEN },
1584 [OVS_VPORT_ATTR_STATS] = { .minlen = sizeof(struct rtnl_link_stats64) },
1585 [OVS_VPORT_ATTR_ADDRESS] = { .minlen = ETH_ALEN },
1587 [OVS_VPORT_ATTR_MTU] = { .type = NLA_U32 },
1588 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1591 static struct genl_family dp_vport_genl_family = {
1592 .id = GENL_ID_GENERATE,
1593 .hdrsize = sizeof(struct ovs_header),
1594 .name = OVS_VPORT_FAMILY,
1596 .maxattr = OVS_VPORT_ATTR_MAX
1599 struct genl_multicast_group dp_vport_multicast_group = {
1600 .name = OVS_VPORT_MCGROUP
1603 /* Called with RTNL lock or RCU read lock. */
1604 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1605 u32 pid, u32 seq, u32 flags, u8 cmd)
1607 struct ovs_header *ovs_header;
1613 ovs_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family,
1618 ovs_header->dp_ifindex = vport->dp->dp_ifindex;
1620 NLA_PUT_U32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no);
1621 NLA_PUT_U32(skb, OVS_VPORT_ATTR_TYPE, vport_get_type(vport));
1622 NLA_PUT_STRING(skb, OVS_VPORT_ATTR_NAME, vport_get_name(vport));
1624 nla = nla_reserve(skb, OVS_VPORT_ATTR_STATS, sizeof(struct rtnl_link_stats64));
1626 goto nla_put_failure;
1627 if (vport_get_stats(vport, nla_data(nla)))
1628 __skb_trim(skb, skb->len - nla->nla_len);
1630 NLA_PUT(skb, OVS_VPORT_ATTR_ADDRESS, ETH_ALEN, vport_get_addr(vport));
1632 mtu = vport_get_mtu(vport);
1634 NLA_PUT_U32(skb, OVS_VPORT_ATTR_MTU, mtu);
1636 err = vport_get_options(vport, skb);
1637 if (err == -EMSGSIZE)
1640 ifindex = vport_get_ifindex(vport);
1642 NLA_PUT_U32(skb, OVS_VPORT_ATTR_IFINDEX, ifindex);
1644 return genlmsg_end(skb, ovs_header);
1649 genlmsg_cancel(skb, ovs_header);
1653 /* Called with RTNL lock or RCU read lock. */
1654 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 pid,
1657 struct sk_buff *skb;
1660 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1662 return ERR_PTR(-ENOMEM);
1664 retval = ovs_vport_cmd_fill_info(vport, skb, pid, seq, 0, cmd);
1667 return ERR_PTR(retval);
1672 static int ovs_vport_cmd_validate(struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1674 return CHECK_NUL_STRING(a[OVS_VPORT_ATTR_NAME], IFNAMSIZ - 1);
1677 /* Called with RTNL lock or RCU read lock. */
1678 static struct vport *lookup_vport(struct ovs_header *ovs_header,
1679 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1681 struct datapath *dp;
1682 struct vport *vport;
1684 if (a[OVS_VPORT_ATTR_NAME]) {
1685 vport = vport_locate(nla_data(a[OVS_VPORT_ATTR_NAME]));
1687 return ERR_PTR(-ENODEV);
1689 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1690 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1692 if (port_no >= DP_MAX_PORTS)
1693 return ERR_PTR(-EFBIG);
1695 dp = get_dp(ovs_header->dp_ifindex);
1697 return ERR_PTR(-ENODEV);
1699 vport = get_vport_protected(dp, port_no);
1701 return ERR_PTR(-ENOENT);
1704 return ERR_PTR(-EINVAL);
1707 /* Called with RTNL lock. */
1708 static int change_vport(struct vport *vport, struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1711 if (a[OVS_VPORT_ATTR_STATS])
1712 err = vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1713 if (!err && a[OVS_VPORT_ATTR_ADDRESS])
1714 err = vport_set_addr(vport, nla_data(a[OVS_VPORT_ATTR_ADDRESS]));
1715 if (!err && a[OVS_VPORT_ATTR_MTU])
1716 err = vport_set_mtu(vport, nla_get_u32(a[OVS_VPORT_ATTR_MTU]));
1720 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1722 struct nlattr **a = info->attrs;
1723 struct ovs_header *ovs_header = info->userhdr;
1724 struct vport_parms parms;
1725 struct sk_buff *reply;
1726 struct vport *vport;
1727 struct datapath *dp;
1732 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE])
1735 err = ovs_vport_cmd_validate(a);
1740 dp = get_dp(ovs_header->dp_ifindex);
1745 if (a[OVS_VPORT_ATTR_PORT_NO]) {
1746 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1749 if (port_no >= DP_MAX_PORTS)
1752 vport = get_vport_protected(dp, port_no);
1757 for (port_no = 1; ; port_no++) {
1758 if (port_no >= DP_MAX_PORTS) {
1762 vport = get_vport_protected(dp, port_no);
1768 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1769 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1770 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1772 parms.port_no = port_no;
1774 vport = new_vport(&parms);
1775 err = PTR_ERR(vport);
1779 set_internal_devs_mtu(dp);
1780 dp_sysfs_add_if(vport);
1782 err = change_vport(vport, a);
1784 reply = ovs_vport_cmd_build_info(vport, info->snd_pid,
1785 info->snd_seq, OVS_VPORT_CMD_NEW);
1787 err = PTR_ERR(reply);
1790 dp_detach_port(vport);
1793 genl_notify(reply, genl_info_net(info), info->snd_pid,
1794 dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1803 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1805 struct nlattr **a = info->attrs;
1806 struct sk_buff *reply;
1807 struct vport *vport;
1810 err = ovs_vport_cmd_validate(a);
1815 vport = lookup_vport(info->userhdr, a);
1816 err = PTR_ERR(vport);
1821 if (a[OVS_VPORT_ATTR_OPTIONS])
1822 err = vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1824 err = change_vport(vport, a);
1826 reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1828 if (IS_ERR(reply)) {
1829 err = PTR_ERR(reply);
1830 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1831 dp_vport_multicast_group.id, err);
1835 genl_notify(reply, genl_info_net(info), info->snd_pid,
1836 dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1844 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1846 struct nlattr **a = info->attrs;
1847 struct sk_buff *reply;
1848 struct vport *vport;
1851 err = ovs_vport_cmd_validate(a);
1856 vport = lookup_vport(info->userhdr, a);
1857 err = PTR_ERR(vport);
1861 if (vport->port_no == OVSP_LOCAL) {
1866 reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1868 err = PTR_ERR(reply);
1872 err = dp_detach_port(vport);
1874 genl_notify(reply, genl_info_net(info), info->snd_pid,
1875 dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1883 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1885 struct nlattr **a = info->attrs;
1886 struct ovs_header *ovs_header = info->userhdr;
1887 struct sk_buff *reply;
1888 struct vport *vport;
1891 err = ovs_vport_cmd_validate(a);
1896 vport = lookup_vport(ovs_header, a);
1897 err = PTR_ERR(vport);
1901 reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1903 err = PTR_ERR(reply);
1909 return genlmsg_reply(reply, info);
1917 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1919 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1920 struct datapath *dp;
1924 dp = get_dp(ovs_header->dp_ifindex);
1929 for (port_no = cb->args[0]; port_no < DP_MAX_PORTS; port_no++) {
1930 struct vport *vport;
1932 vport = get_vport_protected(dp, port_no);
1936 if (ovs_vport_cmd_fill_info(vport, skb, NETLINK_CB(cb->skb).pid,
1937 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1938 OVS_VPORT_CMD_NEW) < 0)
1943 cb->args[0] = port_no;
1949 static struct genl_ops dp_vport_genl_ops[] = {
1950 { .cmd = OVS_VPORT_CMD_NEW,
1951 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1952 .policy = vport_policy,
1953 .doit = ovs_vport_cmd_new
1955 { .cmd = OVS_VPORT_CMD_DEL,
1956 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1957 .policy = vport_policy,
1958 .doit = ovs_vport_cmd_del
1960 { .cmd = OVS_VPORT_CMD_GET,
1961 .flags = 0, /* OK for unprivileged users. */
1962 .policy = vport_policy,
1963 .doit = ovs_vport_cmd_get,
1964 .dumpit = ovs_vport_cmd_dump
1966 { .cmd = OVS_VPORT_CMD_SET,
1967 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1968 .policy = vport_policy,
1969 .doit = ovs_vport_cmd_set,
1973 struct genl_family_and_ops {
1974 struct genl_family *family;
1975 struct genl_ops *ops;
1977 struct genl_multicast_group *group;
1980 static const struct genl_family_and_ops dp_genl_families[] = {
1981 { &dp_datapath_genl_family,
1982 dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1983 &dp_datapath_multicast_group },
1984 { &dp_vport_genl_family,
1985 dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1986 &dp_vport_multicast_group },
1987 { &dp_flow_genl_family,
1988 dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1989 &dp_flow_multicast_group },
1990 { &dp_packet_genl_family,
1991 dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1995 static void dp_unregister_genl(int n_families)
1999 for (i = 0; i < n_families; i++)
2000 genl_unregister_family(dp_genl_families[i].family);
2003 static int dp_register_genl(void)
2010 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2011 const struct genl_family_and_ops *f = &dp_genl_families[i];
2013 err = genl_register_family_with_ops(f->family, f->ops,
2020 err = genl_register_mc_group(f->family, f->group);
2026 err = packet_register_mc_groups();
2032 dp_unregister_genl(n_registered);
2036 static int __init dp_init(void)
2038 struct sk_buff *dummy_skb;
2041 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
2043 printk("Open vSwitch %s, built "__DATE__" "__TIME__"\n", VERSION BUILDNR);
2051 goto error_flow_exit;
2053 err = register_netdevice_notifier(&dp_device_notifier);
2055 goto error_vport_exit;
2057 err = dp_register_genl();
2059 goto error_unreg_notifier;
2063 error_unreg_notifier:
2064 unregister_netdevice_notifier(&dp_device_notifier);
2073 static void dp_cleanup(void)
2076 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2077 unregister_netdevice_notifier(&dp_device_notifier);
2082 module_init(dp_init);
2083 module_exit(dp_cleanup);
2085 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2086 MODULE_LICENSE("GPL");