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 *, u32 pid, 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 flow_table *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 int get_dpifindex(struct datapath *dp)
134 local = get_vport_protected(dp, OVSP_LOCAL);
136 ifindex = vport_get_ifindex(local);
145 static inline size_t br_nlmsg_size(void)
147 return NLMSG_ALIGN(sizeof(struct ifinfomsg))
148 + nla_total_size(IFNAMSIZ) /* IFLA_IFNAME */
149 + nla_total_size(MAX_ADDR_LEN) /* IFLA_ADDRESS */
150 + nla_total_size(4) /* IFLA_MASTER */
151 + nla_total_size(4) /* IFLA_MTU */
152 + nla_total_size(1); /* IFLA_OPERSTATE */
155 /* Caller must hold RTNL lock. */
156 static int dp_fill_ifinfo(struct sk_buff *skb,
157 const struct vport *port,
158 int event, unsigned int flags)
160 struct datapath *dp = port->dp;
161 int ifindex = vport_get_ifindex(port);
162 struct ifinfomsg *hdr;
163 struct nlmsghdr *nlh;
168 nlh = nlmsg_put(skb, 0, 0, event, sizeof(*hdr), flags);
172 hdr = nlmsg_data(nlh);
173 hdr->ifi_family = AF_BRIDGE;
175 hdr->ifi_type = ARPHRD_ETHER;
176 hdr->ifi_index = ifindex;
177 hdr->ifi_flags = vport_get_flags(port);
180 NLA_PUT_STRING(skb, IFLA_IFNAME, vport_get_name(port));
181 NLA_PUT_U32(skb, IFLA_MASTER, get_dpifindex(dp));
182 NLA_PUT_U32(skb, IFLA_MTU, vport_get_mtu(port));
183 #ifdef IFLA_OPERSTATE
184 NLA_PUT_U8(skb, IFLA_OPERSTATE,
185 vport_is_running(port)
186 ? vport_get_operstate(port)
190 NLA_PUT(skb, IFLA_ADDRESS, ETH_ALEN, vport_get_addr(port));
192 return nlmsg_end(skb, nlh);
195 nlmsg_cancel(skb, nlh);
199 /* Caller must hold RTNL lock. */
200 static void dp_ifinfo_notify(int event, struct vport *port)
205 skb = nlmsg_new(br_nlmsg_size(), GFP_KERNEL);
209 err = dp_fill_ifinfo(skb, port, event, 0);
211 /* -EMSGSIZE implies BUG in br_nlmsg_size() */
212 WARN_ON(err == -EMSGSIZE);
216 rtnl_notify(skb, &init_net, 0, RTNLGRP_LINK, NULL, GFP_KERNEL);
220 rtnl_set_sk_err(&init_net, RTNLGRP_LINK, err);
223 static void release_dp(struct kobject *kobj)
225 struct datapath *dp = container_of(kobj, struct datapath, ifobj);
229 static struct kobj_type dp_ktype = {
230 .release = release_dp
233 static void destroy_dp_rcu(struct rcu_head *rcu)
235 struct datapath *dp = container_of(rcu, struct datapath, rcu);
237 flow_tbl_destroy(dp->table);
238 free_percpu(dp->stats_percpu);
239 kobject_put(&dp->ifobj);
242 /* Called with RTNL lock and genl_lock. */
243 static struct vport *new_vport(const struct vport_parms *parms)
247 vport = vport_add(parms);
248 if (!IS_ERR(vport)) {
249 struct datapath *dp = parms->dp;
251 rcu_assign_pointer(dp->ports[parms->port_no], vport);
252 list_add(&vport->node, &dp->port_list);
254 dp_ifinfo_notify(RTM_NEWLINK, vport);
260 /* Called with RTNL lock. */
261 void dp_detach_port(struct vport *p)
265 if (p->port_no != OVSP_LOCAL)
267 dp_ifinfo_notify(RTM_DELLINK, p);
269 /* First drop references to device. */
271 rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
273 /* Then destroy it. */
277 /* Must be called with rcu_read_lock. */
278 void dp_process_received_packet(struct vport *p, struct sk_buff *skb)
280 struct datapath *dp = p->dp;
281 struct sw_flow *flow;
282 struct dp_stats_percpu *stats;
283 int stats_counter_off;
286 OVS_CB(skb)->vport = p;
288 if (!OVS_CB(skb)->flow) {
289 struct sw_flow_key key;
293 /* Extract flow from 'skb' into 'key'. */
294 error = flow_extract(skb, p->port_no, &key, &key_len, &is_frag);
295 if (unlikely(error)) {
300 if (is_frag && dp->drop_frags) {
302 stats_counter_off = offsetof(struct dp_stats_percpu, n_frags);
307 flow = flow_tbl_lookup(rcu_dereference(dp->table), &key, key_len);
308 if (unlikely(!flow)) {
309 struct dp_upcall_info upcall;
311 upcall.cmd = OVS_PACKET_CMD_MISS;
314 upcall.sample_pool = 0;
315 upcall.actions = NULL;
316 upcall.actions_len = 0;
317 dp_upcall(dp, skb, &upcall);
318 stats_counter_off = offsetof(struct dp_stats_percpu, n_missed);
322 OVS_CB(skb)->flow = flow;
325 stats_counter_off = offsetof(struct dp_stats_percpu, n_hit);
326 flow_used(OVS_CB(skb)->flow, skb);
327 execute_actions(dp, skb);
330 /* Update datapath statistics. */
332 stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
334 write_seqcount_begin(&stats->seqlock);
335 (*(u64 *)((u8 *)stats + stats_counter_off))++;
336 write_seqcount_end(&stats->seqlock);
341 static void copy_and_csum_skb(struct sk_buff *skb, void *to)
343 u16 csum_start, csum_offset;
346 get_skb_csum_pointers(skb, &csum_start, &csum_offset);
347 csum_start -= skb_headroom(skb);
349 skb_copy_bits(skb, 0, to, csum_start);
351 csum = skb_copy_and_csum_bits(skb, csum_start, to + csum_start,
352 skb->len - csum_start, 0);
353 *(__sum16 *)(to + csum_start + csum_offset) = csum_fold(csum);
356 static struct genl_family dp_packet_genl_family = {
357 .id = GENL_ID_GENERATE,
358 .hdrsize = sizeof(struct ovs_header),
359 .name = OVS_PACKET_FAMILY,
361 .maxattr = OVS_PACKET_ATTR_MAX
364 int dp_upcall(struct datapath *dp, struct sk_buff *skb, const struct dp_upcall_info *upcall_info)
366 struct dp_stats_percpu *stats;
370 if (OVS_CB(skb)->flow)
371 pid = OVS_CB(skb)->flow->upcall_pid;
373 pid = OVS_CB(skb)->vport->upcall_pid;
381 forward_ip_summed(skb, true);
383 /* Break apart GSO packets into their component pieces. Otherwise
384 * userspace may try to stuff a 64kB packet into a 1500-byte MTU. */
385 if (skb_is_gso(skb)) {
386 struct sk_buff *nskb = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
397 err = queue_userspace_packets(dp, pid, skb, upcall_info);
405 stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
407 write_seqcount_begin(&stats->seqlock);
409 write_seqcount_end(&stats->seqlock);
416 /* Send each packet in the 'skb' list to userspace for 'dp' as directed by
417 * 'upcall_info'. There will be only one packet unless we broke up a GSO
420 static int queue_userspace_packets(struct datapath *dp, u32 pid,
422 const struct dp_upcall_info *upcall_info)
425 struct sk_buff *nskb;
428 dp_ifindex = get_dpifindex(dp);
436 struct ovs_header *upcall;
437 struct sk_buff *user_skb; /* to be queued to userspace */
444 err = vlan_deaccel_tag(skb);
448 if (nla_attr_size(skb->len) > USHRT_MAX) {
453 len = sizeof(struct ovs_header);
454 len += nla_total_size(skb->len);
455 len += nla_total_size(FLOW_BUFSIZE);
456 if (upcall_info->userdata)
457 len += nla_total_size(8);
458 if (upcall_info->sample_pool)
459 len += nla_total_size(4);
460 if (upcall_info->actions_len)
461 len += nla_total_size(upcall_info->actions_len);
463 user_skb = genlmsg_new(len, GFP_ATOMIC);
469 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family, 0, upcall_info->cmd);
470 upcall->dp_ifindex = dp_ifindex;
472 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
473 flow_to_nlattrs(upcall_info->key, user_skb);
474 nla_nest_end(user_skb, nla);
476 if (upcall_info->userdata)
477 nla_put_u64(user_skb, OVS_PACKET_ATTR_USERDATA, upcall_info->userdata);
478 if (upcall_info->sample_pool)
479 nla_put_u32(user_skb, OVS_PACKET_ATTR_SAMPLE_POOL, upcall_info->sample_pool);
480 if (upcall_info->actions_len) {
481 const struct nlattr *actions = upcall_info->actions;
482 u32 actions_len = upcall_info->actions_len;
484 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_ACTIONS);
485 memcpy(__skb_put(user_skb, actions_len), actions, actions_len);
486 nla_nest_end(user_skb, nla);
489 nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
490 if (skb->ip_summed == CHECKSUM_PARTIAL)
491 copy_and_csum_skb(skb, nla_data(nla));
493 skb_copy_bits(skb, 0, nla_data(nla), skb->len);
495 err = genlmsg_unicast(&init_net, user_skb, pid);
506 while ((skb = nskb) != NULL) {
513 /* Called with genl_mutex. */
514 static int flush_flows(int dp_ifindex)
516 struct flow_table *old_table;
517 struct flow_table *new_table;
520 dp = get_dp(dp_ifindex);
524 old_table = get_table_protected(dp);
525 new_table = flow_tbl_alloc(TBL_MIN_BUCKETS);
529 rcu_assign_pointer(dp->table, new_table);
531 flow_tbl_deferred_destroy(old_table);
535 static int validate_actions(const struct nlattr *attr)
537 const struct nlattr *a;
540 nla_for_each_nested(a, attr, rem) {
541 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
542 [OVS_ACTION_ATTR_OUTPUT] = 4,
543 [OVS_ACTION_ATTR_USERSPACE] = 8,
544 [OVS_ACTION_ATTR_PUSH_VLAN] = 2,
545 [OVS_ACTION_ATTR_POP_VLAN] = 0,
546 [OVS_ACTION_ATTR_SET_DL_SRC] = ETH_ALEN,
547 [OVS_ACTION_ATTR_SET_DL_DST] = ETH_ALEN,
548 [OVS_ACTION_ATTR_SET_NW_SRC] = 4,
549 [OVS_ACTION_ATTR_SET_NW_DST] = 4,
550 [OVS_ACTION_ATTR_SET_NW_TOS] = 1,
551 [OVS_ACTION_ATTR_SET_TP_SRC] = 2,
552 [OVS_ACTION_ATTR_SET_TP_DST] = 2,
553 [OVS_ACTION_ATTR_SET_TUNNEL] = 8,
554 [OVS_ACTION_ATTR_SET_PRIORITY] = 4,
555 [OVS_ACTION_ATTR_POP_PRIORITY] = 0,
557 int type = nla_type(a);
559 if (type > OVS_ACTION_ATTR_MAX || nla_len(a) != action_lens[type])
563 case OVS_ACTION_ATTR_UNSPEC:
566 case OVS_ACTION_ATTR_USERSPACE:
567 case OVS_ACTION_ATTR_POP_VLAN:
568 case OVS_ACTION_ATTR_SET_DL_SRC:
569 case OVS_ACTION_ATTR_SET_DL_DST:
570 case OVS_ACTION_ATTR_SET_NW_SRC:
571 case OVS_ACTION_ATTR_SET_NW_DST:
572 case OVS_ACTION_ATTR_SET_TP_SRC:
573 case OVS_ACTION_ATTR_SET_TP_DST:
574 case OVS_ACTION_ATTR_SET_TUNNEL:
575 case OVS_ACTION_ATTR_SET_PRIORITY:
576 case OVS_ACTION_ATTR_POP_PRIORITY:
577 /* No validation needed. */
580 case OVS_ACTION_ATTR_OUTPUT:
581 if (nla_get_u32(a) >= DP_MAX_PORTS)
585 case OVS_ACTION_ATTR_PUSH_VLAN:
586 if (nla_get_be16(a) & htons(VLAN_CFI_MASK))
590 case OVS_ACTION_ATTR_SET_NW_TOS:
591 if (nla_get_u8(a) & INET_ECN_MASK)
605 static void clear_stats(struct sw_flow *flow)
609 flow->packet_count = 0;
610 flow->byte_count = 0;
613 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
615 struct ovs_header *ovs_header = info->userhdr;
616 struct nlattr **a = info->attrs;
617 struct sw_flow_actions *acts;
618 struct sk_buff *packet;
619 struct sw_flow *flow;
628 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
629 !a[OVS_PACKET_ATTR_ACTIONS] ||
630 nla_len(a[OVS_PACKET_ATTR_PACKET]) < ETH_HLEN)
633 err = validate_actions(a[OVS_PACKET_ATTR_ACTIONS]);
637 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
638 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
642 skb_reserve(packet, NET_IP_ALIGN);
644 memcpy(__skb_put(packet, len), nla_data(a[OVS_PACKET_ATTR_PACKET]), len);
646 skb_reset_mac_header(packet);
647 eth = eth_hdr(packet);
649 /* Normally, setting the skb 'protocol' field would be handled by a
650 * call to eth_type_trans(), but it assumes there's a sending
651 * device, which we may not have. */
652 if (ntohs(eth->h_proto) >= 1536)
653 packet->protocol = eth->h_proto;
655 packet->protocol = htons(ETH_P_802_2);
657 /* Build an sw_flow for sending this packet. */
663 err = flow_extract(packet, -1, &flow->key, &key_len, &is_frag);
667 err = flow_metadata_from_nlattrs(&flow->key.eth.in_port,
668 &flow->key.eth.tun_id,
669 a[OVS_PACKET_ATTR_KEY]);
673 flow->hash = flow_hash(&flow->key, key_len);
675 if (a[OVS_PACKET_ATTR_UPCALL_PID])
676 flow->upcall_pid = nla_get_u32(a[OVS_PACKET_ATTR_UPCALL_PID]);
678 flow->upcall_pid = NETLINK_CB(skb).pid;
680 acts = flow_actions_alloc(a[OVS_PACKET_ATTR_ACTIONS]);
684 rcu_assign_pointer(flow->sf_acts, acts);
686 OVS_CB(packet)->flow = flow;
689 dp = get_dp(ovs_header->dp_ifindex);
694 if (flow->key.eth.in_port < DP_MAX_PORTS)
695 OVS_CB(packet)->vport = get_vport_protected(dp,
696 flow->key.eth.in_port);
698 err = execute_actions(dp, packet);
714 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
715 [OVS_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
716 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
717 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
718 [OVS_PACKET_ATTR_UPCALL_PID] = { .type = NLA_U32 },
721 static struct genl_ops dp_packet_genl_ops[] = {
722 { .cmd = OVS_PACKET_CMD_EXECUTE,
723 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
724 .policy = packet_policy,
725 .doit = ovs_packet_cmd_execute
729 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
732 struct flow_table *table = get_table_protected(dp);
734 stats->n_flows = flow_tbl_count(table);
736 stats->n_frags = stats->n_hit = stats->n_missed = stats->n_lost = 0;
737 for_each_possible_cpu(i) {
738 const struct dp_stats_percpu *percpu_stats;
739 struct dp_stats_percpu local_stats;
742 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
745 seqcount = read_seqcount_begin(&percpu_stats->seqlock);
746 local_stats = *percpu_stats;
747 } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount));
749 stats->n_frags += local_stats.n_frags;
750 stats->n_hit += local_stats.n_hit;
751 stats->n_missed += local_stats.n_missed;
752 stats->n_lost += local_stats.n_lost;
756 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
757 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
758 [OVS_FLOW_ATTR_UPCALL_PID] = { .type = NLA_U32 },
759 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
760 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
763 static struct genl_family dp_flow_genl_family = {
764 .id = GENL_ID_GENERATE,
765 .hdrsize = sizeof(struct ovs_header),
766 .name = OVS_FLOW_FAMILY,
768 .maxattr = OVS_FLOW_ATTR_MAX
771 static struct genl_multicast_group dp_flow_multicast_group = {
772 .name = OVS_FLOW_MCGROUP
775 /* Called with genl_lock. */
776 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
777 struct sk_buff *skb, u32 pid, u32 seq, u32 flags, u8 cmd)
779 const int skb_orig_len = skb->len;
780 const struct sw_flow_actions *sf_acts;
781 struct ovs_flow_stats stats;
782 struct ovs_header *ovs_header;
788 sf_acts = rcu_dereference_protected(flow->sf_acts,
789 lockdep_genl_is_held());
791 ovs_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags, cmd);
795 ovs_header->dp_ifindex = get_dpifindex(dp);
797 nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
799 goto nla_put_failure;
800 err = flow_to_nlattrs(&flow->key, skb);
803 nla_nest_end(skb, nla);
805 NLA_PUT_U32(skb, OVS_FLOW_ATTR_UPCALL_PID, flow->upcall_pid);
807 spin_lock_bh(&flow->lock);
809 stats.n_packets = flow->packet_count;
810 stats.n_bytes = flow->byte_count;
811 tcp_flags = flow->tcp_flags;
812 spin_unlock_bh(&flow->lock);
815 NLA_PUT_U64(skb, OVS_FLOW_ATTR_USED, flow_used_time(used));
818 NLA_PUT(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats);
821 NLA_PUT_U8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags);
823 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
824 * this is the first flow to be dumped into 'skb'. This is unusual for
825 * Netlink but individual action lists can be longer than
826 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
827 * The userspace caller can always fetch the actions separately if it
828 * really wants them. (Most userspace callers in fact don't care.)
830 * This can only fail for dump operations because the skb is always
831 * properly sized for single flows.
833 err = nla_put(skb, OVS_FLOW_ATTR_ACTIONS, sf_acts->actions_len,
835 if (err < 0 && skb_orig_len)
838 return genlmsg_end(skb, ovs_header);
843 genlmsg_cancel(skb, ovs_header);
847 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
849 const struct sw_flow_actions *sf_acts;
852 sf_acts = rcu_dereference_protected(flow->sf_acts,
853 lockdep_genl_is_held());
855 len = nla_total_size(FLOW_BUFSIZE); /* OVS_FLOW_ATTR_KEY */
856 len += nla_total_size(sf_acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
857 len += nla_total_size(sizeof(struct ovs_flow_stats)); /* OVS_FLOW_ATTR_STATS */
858 len += nla_total_size(1); /* OVS_FLOW_ATTR_TCP_FLAGS */
859 len += nla_total_size(8); /* OVS_FLOW_ATTR_USED */
860 return genlmsg_new(NLMSG_ALIGN(sizeof(struct ovs_header)) + len, GFP_KERNEL);
863 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow, struct datapath *dp,
864 u32 pid, u32 seq, u8 cmd)
869 skb = ovs_flow_cmd_alloc_info(flow);
871 return ERR_PTR(-ENOMEM);
873 retval = ovs_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd);
878 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
880 struct nlattr **a = info->attrs;
881 struct ovs_header *ovs_header = info->userhdr;
882 struct sw_flow_key key;
883 struct sw_flow *flow;
884 struct sk_buff *reply;
886 struct flow_table *table;
892 if (!a[OVS_FLOW_ATTR_KEY])
894 error = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
898 /* Validate actions. */
899 if (a[OVS_FLOW_ATTR_ACTIONS]) {
900 error = validate_actions(a[OVS_FLOW_ATTR_ACTIONS]);
903 } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
908 dp = get_dp(ovs_header->dp_ifindex);
913 table = get_table_protected(dp);
914 flow = flow_tbl_lookup(table, &key, key_len);
916 struct sw_flow_actions *acts;
918 /* Bail out if we're not allowed to create a new flow. */
920 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
923 /* Expand table, if necessary, to make room. */
924 if (flow_tbl_need_to_expand(table)) {
925 struct flow_table *new_table;
927 new_table = flow_tbl_expand(table);
928 if (!IS_ERR(new_table)) {
929 rcu_assign_pointer(dp->table, new_table);
930 flow_tbl_deferred_destroy(table);
931 table = get_table_protected(dp);
938 error = PTR_ERR(flow);
944 if (a[OVS_FLOW_ATTR_UPCALL_PID])
945 flow->upcall_pid = nla_get_u32(a[OVS_FLOW_ATTR_UPCALL_PID]);
947 flow->upcall_pid = NETLINK_CB(skb).pid;
949 /* Obtain actions. */
950 acts = flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]);
951 error = PTR_ERR(acts);
953 goto error_free_flow;
954 rcu_assign_pointer(flow->sf_acts, acts);
956 /* Put flow in bucket. */
957 flow->hash = flow_hash(&key, key_len);
958 flow_tbl_insert(table, flow);
960 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
961 info->snd_seq, OVS_FLOW_CMD_NEW);
963 /* We found a matching flow. */
964 struct sw_flow_actions *old_acts;
966 /* Bail out if we're not allowed to modify an existing flow.
967 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
968 * because Generic Netlink treats the latter as a dump
969 * request. We also accept NLM_F_EXCL in case that bug ever
973 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
974 info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
977 /* Update actions. */
978 old_acts = rcu_dereference_protected(flow->sf_acts,
979 lockdep_genl_is_held());
980 if (a[OVS_FLOW_ATTR_ACTIONS] &&
981 (old_acts->actions_len != nla_len(a[OVS_FLOW_ATTR_ACTIONS]) ||
982 memcmp(old_acts->actions, nla_data(a[OVS_FLOW_ATTR_ACTIONS]),
983 old_acts->actions_len))) {
984 struct sw_flow_actions *new_acts;
986 new_acts = flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]);
987 error = PTR_ERR(new_acts);
988 if (IS_ERR(new_acts))
991 rcu_assign_pointer(flow->sf_acts, new_acts);
992 flow_deferred_free_acts(old_acts);
995 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
996 info->snd_seq, OVS_FLOW_CMD_NEW);
998 if (a[OVS_FLOW_ATTR_UPCALL_PID])
999 flow->upcall_pid = nla_get_u32(a[OVS_FLOW_ATTR_UPCALL_PID]);
1002 if (a[OVS_FLOW_ATTR_CLEAR]) {
1003 spin_lock_bh(&flow->lock);
1005 spin_unlock_bh(&flow->lock);
1010 genl_notify(reply, genl_info_net(info), info->snd_pid,
1011 dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1013 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1014 dp_flow_multicast_group.id, PTR_ERR(reply));
1023 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1025 struct nlattr **a = info->attrs;
1026 struct ovs_header *ovs_header = info->userhdr;
1027 struct sw_flow_key key;
1028 struct sk_buff *reply;
1029 struct sw_flow *flow;
1030 struct datapath *dp;
1031 struct flow_table *table;
1035 if (!a[OVS_FLOW_ATTR_KEY])
1037 err = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1041 dp = get_dp(ovs_header->dp_ifindex);
1045 table = get_table_protected(dp);
1046 flow = flow_tbl_lookup(table, &key, key_len);
1050 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid, info->snd_seq, OVS_FLOW_CMD_NEW);
1052 return PTR_ERR(reply);
1054 return genlmsg_reply(reply, info);
1057 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1059 struct nlattr **a = info->attrs;
1060 struct ovs_header *ovs_header = info->userhdr;
1061 struct sw_flow_key key;
1062 struct sk_buff *reply;
1063 struct sw_flow *flow;
1064 struct datapath *dp;
1065 struct flow_table *table;
1069 if (!a[OVS_FLOW_ATTR_KEY])
1070 return flush_flows(ovs_header->dp_ifindex);
1071 err = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1075 dp = get_dp(ovs_header->dp_ifindex);
1079 table = get_table_protected(dp);
1080 flow = flow_tbl_lookup(table, &key, key_len);
1084 reply = ovs_flow_cmd_alloc_info(flow);
1088 flow_tbl_remove(table, flow);
1090 err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_pid,
1091 info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1094 flow_deferred_free(flow);
1096 genl_notify(reply, genl_info_net(info), info->snd_pid,
1097 dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1101 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1103 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1104 struct datapath *dp;
1106 dp = get_dp(ovs_header->dp_ifindex);
1111 struct sw_flow *flow;
1114 bucket = cb->args[0];
1116 flow = flow_tbl_next(get_table_protected(dp), &bucket, &obj);
1120 if (ovs_flow_cmd_fill_info(flow, dp, skb, NETLINK_CB(cb->skb).pid,
1121 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1122 OVS_FLOW_CMD_NEW) < 0)
1125 cb->args[0] = bucket;
1131 static struct genl_ops dp_flow_genl_ops[] = {
1132 { .cmd = OVS_FLOW_CMD_NEW,
1133 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1134 .policy = flow_policy,
1135 .doit = ovs_flow_cmd_new_or_set
1137 { .cmd = OVS_FLOW_CMD_DEL,
1138 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1139 .policy = flow_policy,
1140 .doit = ovs_flow_cmd_del
1142 { .cmd = OVS_FLOW_CMD_GET,
1143 .flags = 0, /* OK for unprivileged users. */
1144 .policy = flow_policy,
1145 .doit = ovs_flow_cmd_get,
1146 .dumpit = ovs_flow_cmd_dump
1148 { .cmd = OVS_FLOW_CMD_SET,
1149 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1150 .policy = flow_policy,
1151 .doit = ovs_flow_cmd_new_or_set,
1155 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1156 #ifdef HAVE_NLA_NUL_STRING
1157 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1159 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1160 [OVS_DP_ATTR_IPV4_FRAGS] = { .type = NLA_U32 },
1161 [OVS_DP_ATTR_SAMPLING] = { .type = NLA_U32 },
1164 static struct genl_family dp_datapath_genl_family = {
1165 .id = GENL_ID_GENERATE,
1166 .hdrsize = sizeof(struct ovs_header),
1167 .name = OVS_DATAPATH_FAMILY,
1169 .maxattr = OVS_DP_ATTR_MAX
1172 static struct genl_multicast_group dp_datapath_multicast_group = {
1173 .name = OVS_DATAPATH_MCGROUP
1176 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1177 u32 pid, u32 seq, u32 flags, u8 cmd)
1179 struct ovs_header *ovs_header;
1183 ovs_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family,
1188 ovs_header->dp_ifindex = get_dpifindex(dp);
1191 err = nla_put_string(skb, OVS_DP_ATTR_NAME, dp_name(dp));
1194 goto nla_put_failure;
1196 nla = nla_reserve(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats));
1198 goto nla_put_failure;
1199 get_dp_stats(dp, nla_data(nla));
1201 NLA_PUT_U32(skb, OVS_DP_ATTR_IPV4_FRAGS,
1202 dp->drop_frags ? OVS_DP_FRAG_DROP : OVS_DP_FRAG_ZERO);
1204 if (dp->sflow_probability)
1205 NLA_PUT_U32(skb, OVS_DP_ATTR_SAMPLING, dp->sflow_probability);
1207 return genlmsg_end(skb, ovs_header);
1210 genlmsg_cancel(skb, ovs_header);
1215 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 pid,
1218 struct sk_buff *skb;
1221 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1223 return ERR_PTR(-ENOMEM);
1225 retval = ovs_dp_cmd_fill_info(dp, skb, pid, seq, 0, cmd);
1228 return ERR_PTR(retval);
1233 static int ovs_dp_cmd_validate(struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1235 if (a[OVS_DP_ATTR_IPV4_FRAGS]) {
1236 u32 frags = nla_get_u32(a[OVS_DP_ATTR_IPV4_FRAGS]);
1238 if (frags != OVS_DP_FRAG_ZERO && frags != OVS_DP_FRAG_DROP)
1242 return CHECK_NUL_STRING(a[OVS_DP_ATTR_NAME], IFNAMSIZ - 1);
1245 /* Called with genl_mutex and optionally with RTNL lock also. */
1246 static struct datapath *lookup_datapath(struct ovs_header *ovs_header, struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1248 struct datapath *dp;
1250 if (!a[OVS_DP_ATTR_NAME])
1251 dp = get_dp(ovs_header->dp_ifindex);
1253 struct vport *vport;
1256 vport = vport_locate(nla_data(a[OVS_DP_ATTR_NAME]));
1257 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1260 return dp ? dp : ERR_PTR(-ENODEV);
1263 /* Called with genl_mutex. */
1264 static void change_datapath(struct datapath *dp, struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1266 if (a[OVS_DP_ATTR_IPV4_FRAGS])
1267 dp->drop_frags = nla_get_u32(a[OVS_DP_ATTR_IPV4_FRAGS]) == OVS_DP_FRAG_DROP;
1268 if (a[OVS_DP_ATTR_SAMPLING])
1269 dp->sflow_probability = nla_get_u32(a[OVS_DP_ATTR_SAMPLING]);
1272 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1274 struct nlattr **a = info->attrs;
1275 struct vport_parms parms;
1276 struct sk_buff *reply;
1277 struct datapath *dp;
1278 struct vport *vport;
1282 if (!a[OVS_DP_ATTR_NAME])
1285 err = ovs_dp_cmd_validate(a);
1291 if (!try_module_get(THIS_MODULE))
1292 goto err_unlock_rtnl;
1295 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1297 goto err_put_module;
1298 INIT_LIST_HEAD(&dp->port_list);
1300 /* Initialize kobject for bridge. This will be added as
1301 * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
1302 dp->ifobj.kset = NULL;
1303 kobject_init(&dp->ifobj, &dp_ktype);
1305 /* Allocate table. */
1307 rcu_assign_pointer(dp->table, flow_tbl_alloc(TBL_MIN_BUCKETS));
1312 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1313 if (!dp->stats_percpu) {
1315 goto err_destroy_table;
1318 change_datapath(dp, a);
1320 /* Set up our datapath device. */
1321 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1322 parms.type = OVS_VPORT_TYPE_INTERNAL;
1323 parms.options = NULL;
1325 parms.port_no = OVSP_LOCAL;
1326 if (a[OVS_DP_ATTR_UPCALL_PID])
1327 parms.upcall_pid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1329 parms.upcall_pid = NETLINK_CB(skb).pid;
1331 vport = new_vport(&parms);
1332 if (IS_ERR(vport)) {
1333 err = PTR_ERR(vport);
1337 goto err_destroy_percpu;
1340 reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_NEW);
1341 err = PTR_ERR(reply);
1343 goto err_destroy_local_port;
1345 list_add_tail(&dp->list_node, &dps);
1346 dp_sysfs_add_dp(dp);
1350 genl_notify(reply, genl_info_net(info), info->snd_pid,
1351 dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1354 err_destroy_local_port:
1355 dp_detach_port(get_vport_protected(dp, OVSP_LOCAL));
1357 free_percpu(dp->stats_percpu);
1359 flow_tbl_destroy(get_table_protected(dp));
1363 module_put(THIS_MODULE);
1370 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1372 struct vport *vport, *next_vport;
1373 struct sk_buff *reply;
1374 struct datapath *dp;
1377 err = ovs_dp_cmd_validate(info->attrs);
1382 dp = lookup_datapath(info->userhdr, info->attrs);
1387 reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_DEL);
1388 err = PTR_ERR(reply);
1392 list_for_each_entry_safe (vport, next_vport, &dp->port_list, node)
1393 if (vport->port_no != OVSP_LOCAL)
1394 dp_detach_port(vport);
1396 dp_sysfs_del_dp(dp);
1397 list_del(&dp->list_node);
1398 dp_detach_port(get_vport_protected(dp, OVSP_LOCAL));
1400 /* rtnl_unlock() will wait until all the references to devices that
1401 * are pending unregistration have been dropped. We do it here to
1402 * ensure that any internal devices (which contain DP pointers) are
1403 * fully destroyed before freeing the datapath.
1407 call_rcu(&dp->rcu, destroy_dp_rcu);
1408 module_put(THIS_MODULE);
1410 genl_notify(reply, genl_info_net(info), info->snd_pid,
1411 dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1421 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1423 struct sk_buff *reply;
1424 struct datapath *dp;
1427 err = ovs_dp_cmd_validate(info->attrs);
1431 dp = lookup_datapath(info->userhdr, info->attrs);
1435 change_datapath(dp, info->attrs);
1437 reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_NEW);
1438 if (IS_ERR(reply)) {
1439 err = PTR_ERR(reply);
1440 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1441 dp_datapath_multicast_group.id, err);
1445 genl_notify(reply, genl_info_net(info), info->snd_pid,
1446 dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1450 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1452 struct sk_buff *reply;
1453 struct datapath *dp;
1456 err = ovs_dp_cmd_validate(info->attrs);
1460 dp = lookup_datapath(info->userhdr, info->attrs);
1464 reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_NEW);
1466 return PTR_ERR(reply);
1468 return genlmsg_reply(reply, info);
1471 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1473 struct datapath *dp;
1474 int skip = cb->args[0];
1477 list_for_each_entry (dp, &dps, list_node) {
1480 if (ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).pid,
1481 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1482 OVS_DP_CMD_NEW) < 0)
1492 static struct genl_ops dp_datapath_genl_ops[] = {
1493 { .cmd = OVS_DP_CMD_NEW,
1494 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1495 .policy = datapath_policy,
1496 .doit = ovs_dp_cmd_new
1498 { .cmd = OVS_DP_CMD_DEL,
1499 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1500 .policy = datapath_policy,
1501 .doit = ovs_dp_cmd_del
1503 { .cmd = OVS_DP_CMD_GET,
1504 .flags = 0, /* OK for unprivileged users. */
1505 .policy = datapath_policy,
1506 .doit = ovs_dp_cmd_get,
1507 .dumpit = ovs_dp_cmd_dump
1509 { .cmd = OVS_DP_CMD_SET,
1510 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1511 .policy = datapath_policy,
1512 .doit = ovs_dp_cmd_set,
1516 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1517 #ifdef HAVE_NLA_NUL_STRING
1518 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1519 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1520 [OVS_VPORT_ATTR_ADDRESS] = { .len = ETH_ALEN },
1522 [OVS_VPORT_ATTR_STATS] = { .minlen = sizeof(struct ovs_vport_stats) },
1523 [OVS_VPORT_ATTR_ADDRESS] = { .minlen = ETH_ALEN },
1525 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1526 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1527 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1528 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1531 static struct genl_family dp_vport_genl_family = {
1532 .id = GENL_ID_GENERATE,
1533 .hdrsize = sizeof(struct ovs_header),
1534 .name = OVS_VPORT_FAMILY,
1536 .maxattr = OVS_VPORT_ATTR_MAX
1539 struct genl_multicast_group dp_vport_multicast_group = {
1540 .name = OVS_VPORT_MCGROUP
1543 /* Called with RTNL lock or RCU read lock. */
1544 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1545 u32 pid, u32 seq, u32 flags, u8 cmd)
1547 struct ovs_header *ovs_header;
1552 ovs_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family,
1557 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1559 NLA_PUT_U32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no);
1560 NLA_PUT_U32(skb, OVS_VPORT_ATTR_TYPE, vport_get_type(vport));
1561 NLA_PUT_STRING(skb, OVS_VPORT_ATTR_NAME, vport_get_name(vport));
1562 NLA_PUT_U32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_pid);
1564 nla = nla_reserve(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats));
1566 goto nla_put_failure;
1568 vport_get_stats(vport, nla_data(nla));
1570 NLA_PUT(skb, OVS_VPORT_ATTR_ADDRESS, ETH_ALEN, vport_get_addr(vport));
1572 err = vport_get_options(vport, skb);
1573 if (err == -EMSGSIZE)
1576 ifindex = vport_get_ifindex(vport);
1578 NLA_PUT_U32(skb, OVS_VPORT_ATTR_IFINDEX, ifindex);
1580 return genlmsg_end(skb, ovs_header);
1585 genlmsg_cancel(skb, ovs_header);
1589 /* Called with RTNL lock or RCU read lock. */
1590 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 pid,
1593 struct sk_buff *skb;
1596 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1598 return ERR_PTR(-ENOMEM);
1600 retval = ovs_vport_cmd_fill_info(vport, skb, pid, seq, 0, cmd);
1603 return ERR_PTR(retval);
1608 static int ovs_vport_cmd_validate(struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1610 return CHECK_NUL_STRING(a[OVS_VPORT_ATTR_NAME], IFNAMSIZ - 1);
1613 /* Called with RTNL lock or RCU read lock. */
1614 static struct vport *lookup_vport(struct ovs_header *ovs_header,
1615 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1617 struct datapath *dp;
1618 struct vport *vport;
1620 if (a[OVS_VPORT_ATTR_NAME]) {
1621 vport = vport_locate(nla_data(a[OVS_VPORT_ATTR_NAME]));
1623 return ERR_PTR(-ENODEV);
1625 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1626 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1628 if (port_no >= DP_MAX_PORTS)
1629 return ERR_PTR(-EFBIG);
1631 dp = get_dp(ovs_header->dp_ifindex);
1633 return ERR_PTR(-ENODEV);
1635 vport = get_vport_protected(dp, port_no);
1637 return ERR_PTR(-ENOENT);
1640 return ERR_PTR(-EINVAL);
1643 /* Called with RTNL lock. */
1644 static int change_vport(struct vport *vport, struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1648 if (a[OVS_VPORT_ATTR_STATS])
1649 vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1651 if (a[OVS_VPORT_ATTR_ADDRESS])
1652 err = vport_set_addr(vport, nla_data(a[OVS_VPORT_ATTR_ADDRESS]));
1657 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1659 struct nlattr **a = info->attrs;
1660 struct ovs_header *ovs_header = info->userhdr;
1661 struct vport_parms parms;
1662 struct sk_buff *reply;
1663 struct vport *vport;
1664 struct datapath *dp;
1669 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE])
1672 err = ovs_vport_cmd_validate(a);
1677 dp = get_dp(ovs_header->dp_ifindex);
1682 if (a[OVS_VPORT_ATTR_PORT_NO]) {
1683 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1686 if (port_no >= DP_MAX_PORTS)
1689 vport = get_vport_protected(dp, port_no);
1694 for (port_no = 1; ; port_no++) {
1695 if (port_no >= DP_MAX_PORTS) {
1699 vport = get_vport_protected(dp, port_no);
1705 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1706 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1707 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1709 parms.port_no = port_no;
1710 if (a[OVS_VPORT_ATTR_UPCALL_PID])
1711 parms.upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1713 parms.upcall_pid = NETLINK_CB(skb).pid;
1715 vport = new_vport(&parms);
1716 err = PTR_ERR(vport);
1720 dp_sysfs_add_if(vport);
1722 err = change_vport(vport, a);
1724 reply = ovs_vport_cmd_build_info(vport, info->snd_pid,
1725 info->snd_seq, OVS_VPORT_CMD_NEW);
1727 err = PTR_ERR(reply);
1730 dp_detach_port(vport);
1733 genl_notify(reply, genl_info_net(info), info->snd_pid,
1734 dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1743 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1745 struct nlattr **a = info->attrs;
1746 struct sk_buff *reply;
1747 struct vport *vport;
1750 err = ovs_vport_cmd_validate(a);
1755 vport = lookup_vport(info->userhdr, a);
1756 err = PTR_ERR(vport);
1761 if (a[OVS_VPORT_ATTR_OPTIONS])
1762 err = vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1764 err = change_vport(vport, a);
1765 if (!err && a[OVS_VPORT_ATTR_UPCALL_PID])
1766 vport->upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1768 reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1770 if (IS_ERR(reply)) {
1771 err = PTR_ERR(reply);
1772 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1773 dp_vport_multicast_group.id, err);
1777 genl_notify(reply, genl_info_net(info), info->snd_pid,
1778 dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1786 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1788 struct nlattr **a = info->attrs;
1789 struct sk_buff *reply;
1790 struct vport *vport;
1793 err = ovs_vport_cmd_validate(a);
1798 vport = lookup_vport(info->userhdr, a);
1799 err = PTR_ERR(vport);
1803 if (vport->port_no == OVSP_LOCAL) {
1808 reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1810 err = PTR_ERR(reply);
1814 dp_detach_port(vport);
1816 genl_notify(reply, genl_info_net(info), info->snd_pid,
1817 dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1825 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1827 struct nlattr **a = info->attrs;
1828 struct ovs_header *ovs_header = info->userhdr;
1829 struct sk_buff *reply;
1830 struct vport *vport;
1833 err = ovs_vport_cmd_validate(a);
1838 vport = lookup_vport(ovs_header, a);
1839 err = PTR_ERR(vport);
1843 reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1845 err = PTR_ERR(reply);
1851 return genlmsg_reply(reply, info);
1859 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1861 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1862 struct datapath *dp;
1866 dp = get_dp(ovs_header->dp_ifindex);
1871 for (port_no = cb->args[0]; port_no < DP_MAX_PORTS; port_no++) {
1872 struct vport *vport;
1874 vport = get_vport_protected(dp, port_no);
1878 if (ovs_vport_cmd_fill_info(vport, skb, NETLINK_CB(cb->skb).pid,
1879 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1880 OVS_VPORT_CMD_NEW) < 0)
1885 cb->args[0] = port_no;
1891 static struct genl_ops dp_vport_genl_ops[] = {
1892 { .cmd = OVS_VPORT_CMD_NEW,
1893 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1894 .policy = vport_policy,
1895 .doit = ovs_vport_cmd_new
1897 { .cmd = OVS_VPORT_CMD_DEL,
1898 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1899 .policy = vport_policy,
1900 .doit = ovs_vport_cmd_del
1902 { .cmd = OVS_VPORT_CMD_GET,
1903 .flags = 0, /* OK for unprivileged users. */
1904 .policy = vport_policy,
1905 .doit = ovs_vport_cmd_get,
1906 .dumpit = ovs_vport_cmd_dump
1908 { .cmd = OVS_VPORT_CMD_SET,
1909 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1910 .policy = vport_policy,
1911 .doit = ovs_vport_cmd_set,
1915 struct genl_family_and_ops {
1916 struct genl_family *family;
1917 struct genl_ops *ops;
1919 struct genl_multicast_group *group;
1922 static const struct genl_family_and_ops dp_genl_families[] = {
1923 { &dp_datapath_genl_family,
1924 dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1925 &dp_datapath_multicast_group },
1926 { &dp_vport_genl_family,
1927 dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1928 &dp_vport_multicast_group },
1929 { &dp_flow_genl_family,
1930 dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1931 &dp_flow_multicast_group },
1932 { &dp_packet_genl_family,
1933 dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1937 static void dp_unregister_genl(int n_families)
1941 for (i = 0; i < n_families; i++)
1942 genl_unregister_family(dp_genl_families[i].family);
1945 static int dp_register_genl(void)
1952 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
1953 const struct genl_family_and_ops *f = &dp_genl_families[i];
1955 err = genl_register_family_with_ops(f->family, f->ops,
1962 err = genl_register_mc_group(f->family, f->group);
1971 dp_unregister_genl(n_registered);
1975 static int __init dp_init(void)
1977 struct sk_buff *dummy_skb;
1980 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
1982 printk("Open vSwitch %s, built "__DATE__" "__TIME__"\n", VERSION BUILDNR);
1990 goto error_tnl_exit;
1994 goto error_flow_exit;
1996 err = register_netdevice_notifier(&dp_device_notifier);
1998 goto error_vport_exit;
2000 err = dp_register_genl();
2002 goto error_unreg_notifier;
2006 error_unreg_notifier:
2007 unregister_netdevice_notifier(&dp_device_notifier);
2018 static void dp_cleanup(void)
2021 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2022 unregister_netdevice_notifier(&dp_device_notifier);
2028 module_init(dp_init);
2029 module_exit(dp_cleanup);
2031 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2032 MODULE_LICENSE("GPL");