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 int (*dp_ioctl_hook)(struct net_device *dev, struct ifreq *rq, int cmd);
57 EXPORT_SYMBOL(dp_ioctl_hook);
62 * Writes to device state (add/remove datapath, port, set operations on vports,
63 * etc.) are protected by RTNL.
65 * Writes to other state (flow table modifications, set miscellaneous datapath
66 * parameters such as drop frags, etc.) are protected by genl_mutex. The RTNL
67 * lock nests inside genl_mutex.
69 * Reads are protected by RCU.
71 * There are a few special cases (mostly stats) that have their own
72 * synchronization but they nest under all of above and don't interact with
76 /* Global list of datapaths to enable dumping them all out.
77 * Protected by genl_mutex.
79 static LIST_HEAD(dps);
81 static struct vport *new_vport(const struct vport_parms *);
82 static int queue_control_packets(struct datapath *, struct sk_buff *,
83 const struct dp_upcall_info *);
85 /* Must be called with rcu_read_lock, genl_mutex, or RTNL lock. */
86 struct datapath *get_dp(int dp_ifindex)
88 struct datapath *dp = NULL;
89 struct net_device *dev;
92 dev = dev_get_by_index_rcu(&init_net, dp_ifindex);
94 struct vport *vport = internal_dev_get_vport(dev);
102 EXPORT_SYMBOL_GPL(get_dp);
104 /* Must be called with genl_mutex. */
105 static struct tbl *get_table_protected(struct datapath *dp)
107 return rcu_dereference_protected(dp->table, lockdep_genl_is_held());
110 /* Must be called with rcu_read_lock or RTNL lock. */
111 static struct vport *get_vport_protected(struct datapath *dp, u16 port_no)
113 return rcu_dereference_rtnl(dp->ports[port_no]);
116 /* Must be called with rcu_read_lock or RTNL lock. */
117 const char *dp_name(const struct datapath *dp)
119 return vport_get_name(rcu_dereference_rtnl(dp->ports[ODPP_LOCAL]));
122 static inline size_t br_nlmsg_size(void)
124 return NLMSG_ALIGN(sizeof(struct ifinfomsg))
125 + nla_total_size(IFNAMSIZ) /* IFLA_IFNAME */
126 + nla_total_size(MAX_ADDR_LEN) /* IFLA_ADDRESS */
127 + nla_total_size(4) /* IFLA_MASTER */
128 + nla_total_size(4) /* IFLA_MTU */
129 + nla_total_size(4) /* IFLA_LINK */
130 + nla_total_size(1); /* IFLA_OPERSTATE */
133 /* Caller must hold RTNL lock. */
134 static int dp_fill_ifinfo(struct sk_buff *skb,
135 const struct vport *port,
136 int event, unsigned int flags)
138 struct datapath *dp = port->dp;
139 int ifindex = vport_get_ifindex(port);
140 int iflink = vport_get_iflink(port);
141 struct ifinfomsg *hdr;
142 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, ODPP_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 if (ifindex != iflink)
176 NLA_PUT_U32(skb, IFLA_LINK,iflink);
178 return nlmsg_end(skb, nlh);
181 nlmsg_cancel(skb, nlh);
185 /* Caller must hold RTNL lock. */
186 static void dp_ifinfo_notify(int event, struct vport *port)
191 skb = nlmsg_new(br_nlmsg_size(), GFP_KERNEL);
195 err = dp_fill_ifinfo(skb, port, event, 0);
197 /* -EMSGSIZE implies BUG in br_nlmsg_size() */
198 WARN_ON(err == -EMSGSIZE);
202 rtnl_notify(skb, &init_net, 0, RTNLGRP_LINK, NULL, GFP_KERNEL);
206 rtnl_set_sk_err(&init_net, RTNLGRP_LINK, err);
209 static void release_dp(struct kobject *kobj)
211 struct datapath *dp = container_of(kobj, struct datapath, ifobj);
215 static struct kobj_type dp_ktype = {
216 .release = release_dp
219 static void destroy_dp_rcu(struct rcu_head *rcu)
221 struct datapath *dp = container_of(rcu, struct datapath, rcu);
223 tbl_destroy((struct tbl __force *)dp->table, flow_free_tbl);
224 free_percpu(dp->stats_percpu);
225 kobject_put(&dp->ifobj);
228 /* Called with RTNL lock and genl_lock. */
229 static struct vport *new_vport(const struct vport_parms *parms)
233 vport = vport_add(parms);
234 if (!IS_ERR(vport)) {
235 struct datapath *dp = parms->dp;
237 rcu_assign_pointer(dp->ports[parms->port_no], vport);
238 list_add(&vport->node, &dp->port_list);
240 dp_ifinfo_notify(RTM_NEWLINK, vport);
246 /* Called with RTNL lock. */
247 int dp_detach_port(struct vport *p)
251 if (p->port_no != ODPP_LOCAL)
253 dp_ifinfo_notify(RTM_DELLINK, p);
255 /* First drop references to device. */
257 rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
259 /* Then destroy it. */
263 /* Must be called with rcu_read_lock. */
264 void dp_process_received_packet(struct vport *p, struct sk_buff *skb)
266 struct datapath *dp = p->dp;
267 struct dp_stats_percpu *stats;
268 int stats_counter_off;
271 OVS_CB(skb)->vport = p;
273 if (!OVS_CB(skb)->flow) {
274 struct sw_flow_key key;
275 struct tbl_node *flow_node;
279 /* Extract flow from 'skb' into 'key'. */
280 error = flow_extract(skb, p->port_no, &key, &key_len, &is_frag);
281 if (unlikely(error)) {
286 if (is_frag && dp->drop_frags) {
288 stats_counter_off = offsetof(struct dp_stats_percpu, n_frags);
293 flow_node = tbl_lookup(rcu_dereference(dp->table), &key, key_len,
294 flow_hash(&key, key_len), flow_cmp);
295 if (unlikely(!flow_node)) {
296 struct dp_upcall_info upcall;
298 upcall.cmd = ODP_PACKET_CMD_MISS;
301 upcall.sample_pool = 0;
302 upcall.actions = NULL;
303 upcall.actions_len = 0;
304 dp_upcall(dp, skb, &upcall);
305 stats_counter_off = offsetof(struct dp_stats_percpu, n_missed);
309 OVS_CB(skb)->flow = flow_cast(flow_node);
312 stats_counter_off = offsetof(struct dp_stats_percpu, n_hit);
313 flow_used(OVS_CB(skb)->flow, skb);
314 execute_actions(dp, skb);
317 /* Update datapath statistics. */
319 stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
321 write_seqcount_begin(&stats->seqlock);
322 (*(u64 *)((u8 *)stats + stats_counter_off))++;
323 write_seqcount_end(&stats->seqlock);
328 static void copy_and_csum_skb(struct sk_buff *skb, void *to)
330 u16 csum_start, csum_offset;
333 get_skb_csum_pointers(skb, &csum_start, &csum_offset);
334 csum_start -= skb_headroom(skb);
336 skb_copy_bits(skb, 0, to, csum_start);
338 csum = skb_copy_and_csum_bits(skb, csum_start, to + csum_start,
339 skb->len - csum_start, 0);
340 *(__sum16 *)(to + csum_start + csum_offset) = csum_fold(csum);
343 static struct genl_family dp_packet_genl_family = {
344 .id = GENL_ID_GENERATE,
345 .hdrsize = sizeof(struct odp_header),
346 .name = ODP_PACKET_FAMILY,
348 .maxattr = ODP_PACKET_ATTR_MAX
351 /* Generic Netlink multicast groups for upcalls.
353 * We really want three unique multicast groups per datapath, but we can't even
354 * get one, because genl_register_mc_group() takes genl_lock, which is also
355 * held during Generic Netlink message processing, so trying to acquire
356 * multicast groups during ODP_DP_NEW processing deadlocks. Instead, we
357 * preallocate a few groups and use them round-robin for datapaths. Collision
358 * isn't fatal--multicast listeners should check that the family is the one
359 * that they want and discard others--but it wastes time and memory to receive
362 #define PACKET_N_MC_GROUPS 16
363 static struct genl_multicast_group packet_mc_groups[PACKET_N_MC_GROUPS];
365 static u32 packet_mc_group(struct datapath *dp, u8 cmd)
368 BUILD_BUG_ON_NOT_POWER_OF_2(PACKET_N_MC_GROUPS);
370 idx = jhash_2words(dp->dp_ifindex, cmd, 0) & (PACKET_N_MC_GROUPS - 1);
371 return packet_mc_groups[idx].id;
374 static int packet_register_mc_groups(void)
378 for (i = 0; i < PACKET_N_MC_GROUPS; i++) {
379 struct genl_multicast_group *group = &packet_mc_groups[i];
382 sprintf(group->name, "packet%d", i);
383 error = genl_register_mc_group(&dp_packet_genl_family, group);
390 int dp_upcall(struct datapath *dp, struct sk_buff *skb, const struct dp_upcall_info *upcall_info)
392 struct dp_stats_percpu *stats;
395 WARN_ON_ONCE(skb_shared(skb));
397 forward_ip_summed(skb);
399 err = vswitch_skb_checksum_setup(skb);
403 /* Break apart GSO packets into their component pieces. Otherwise
404 * userspace may try to stuff a 64kB packet into a 1500-byte MTU. */
405 if (skb_is_gso(skb)) {
406 struct sk_buff *nskb = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
416 err = queue_control_packets(dp, skb, upcall_info);
426 stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
428 write_seqcount_begin(&stats->seqlock);
430 write_seqcount_end(&stats->seqlock);
437 /* Send each packet in the 'skb' list to userspace for 'dp' as directed by
438 * 'upcall_info'. There will be only one packet unless we broke up a GSO
441 static int queue_control_packets(struct datapath *dp, struct sk_buff *skb,
442 const struct dp_upcall_info *upcall_info)
444 u32 group = packet_mc_group(dp, upcall_info->cmd);
445 struct sk_buff *nskb;
449 struct odp_header *upcall;
450 struct sk_buff *user_skb; /* to be queued to userspace */
457 err = vlan_deaccel_tag(skb);
461 if (nla_attr_size(skb->len) > USHRT_MAX)
464 len = sizeof(struct odp_header);
465 len += nla_total_size(skb->len);
466 len += nla_total_size(FLOW_BUFSIZE);
467 if (upcall_info->userdata)
468 len += nla_total_size(8);
469 if (upcall_info->sample_pool)
470 len += nla_total_size(4);
471 if (upcall_info->actions_len)
472 len += nla_total_size(upcall_info->actions_len);
474 user_skb = genlmsg_new(len, GFP_ATOMIC);
476 netlink_set_err(INIT_NET_GENL_SOCK, 0, group, -ENOBUFS);
480 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family, 0, upcall_info->cmd);
481 upcall->dp_ifindex = dp->dp_ifindex;
483 nla = nla_nest_start(user_skb, ODP_PACKET_ATTR_KEY);
484 flow_to_nlattrs(upcall_info->key, user_skb);
485 nla_nest_end(user_skb, nla);
487 if (upcall_info->userdata)
488 nla_put_u64(user_skb, ODP_PACKET_ATTR_USERDATA, upcall_info->userdata);
489 if (upcall_info->sample_pool)
490 nla_put_u32(user_skb, ODP_PACKET_ATTR_SAMPLE_POOL, upcall_info->sample_pool);
491 if (upcall_info->actions_len) {
492 const struct nlattr *actions = upcall_info->actions;
493 u32 actions_len = upcall_info->actions_len;
495 nla = nla_nest_start(user_skb, ODP_PACKET_ATTR_ACTIONS);
496 memcpy(__skb_put(user_skb, actions_len), actions, actions_len);
497 nla_nest_end(user_skb, nla);
500 nla = __nla_reserve(user_skb, ODP_PACKET_ATTR_PACKET, skb->len);
501 if (skb->ip_summed == CHECKSUM_PARTIAL)
502 copy_and_csum_skb(skb, nla_data(nla));
504 skb_copy_bits(skb, 0, nla_data(nla), skb->len);
506 err = genlmsg_multicast(user_skb, 0, group, GFP_ATOMIC);
517 while ((skb = nskb) != NULL) {
524 /* Called with genl_mutex. */
525 static int flush_flows(int dp_ifindex)
527 struct tbl *old_table;
528 struct tbl *new_table;
531 dp = get_dp(dp_ifindex);
535 old_table = get_table_protected(dp);
536 new_table = tbl_create(TBL_MIN_BUCKETS);
540 rcu_assign_pointer(dp->table, new_table);
542 tbl_deferred_destroy(old_table, flow_free_tbl);
547 static int validate_actions(const struct nlattr *attr)
549 const struct nlattr *a;
552 nla_for_each_nested(a, attr, rem) {
553 static const u32 action_lens[ODP_ACTION_ATTR_MAX + 1] = {
554 [ODP_ACTION_ATTR_OUTPUT] = 4,
555 [ODP_ACTION_ATTR_CONTROLLER] = 8,
556 [ODP_ACTION_ATTR_SET_DL_TCI] = 2,
557 [ODP_ACTION_ATTR_STRIP_VLAN] = 0,
558 [ODP_ACTION_ATTR_SET_DL_SRC] = ETH_ALEN,
559 [ODP_ACTION_ATTR_SET_DL_DST] = ETH_ALEN,
560 [ODP_ACTION_ATTR_SET_NW_SRC] = 4,
561 [ODP_ACTION_ATTR_SET_NW_DST] = 4,
562 [ODP_ACTION_ATTR_SET_NW_TOS] = 1,
563 [ODP_ACTION_ATTR_SET_TP_SRC] = 2,
564 [ODP_ACTION_ATTR_SET_TP_DST] = 2,
565 [ODP_ACTION_ATTR_SET_TUNNEL] = 8,
566 [ODP_ACTION_ATTR_SET_PRIORITY] = 4,
567 [ODP_ACTION_ATTR_POP_PRIORITY] = 0,
568 [ODP_ACTION_ATTR_DROP_SPOOFED_ARP] = 0,
570 int type = nla_type(a);
572 if (type > ODP_ACTION_ATTR_MAX || nla_len(a) != action_lens[type])
576 case ODP_ACTION_ATTR_UNSPEC:
579 case ODP_ACTION_ATTR_CONTROLLER:
580 case ODP_ACTION_ATTR_STRIP_VLAN:
581 case ODP_ACTION_ATTR_SET_DL_SRC:
582 case ODP_ACTION_ATTR_SET_DL_DST:
583 case ODP_ACTION_ATTR_SET_NW_SRC:
584 case ODP_ACTION_ATTR_SET_NW_DST:
585 case ODP_ACTION_ATTR_SET_TP_SRC:
586 case ODP_ACTION_ATTR_SET_TP_DST:
587 case ODP_ACTION_ATTR_SET_TUNNEL:
588 case ODP_ACTION_ATTR_SET_PRIORITY:
589 case ODP_ACTION_ATTR_POP_PRIORITY:
590 case ODP_ACTION_ATTR_DROP_SPOOFED_ARP:
591 /* No validation needed. */
594 case ODP_ACTION_ATTR_OUTPUT:
595 if (nla_get_u32(a) >= DP_MAX_PORTS)
599 case ODP_ACTION_ATTR_SET_DL_TCI:
600 if (nla_get_be16(a) & htons(VLAN_CFI_MASK))
604 case ODP_ACTION_ATTR_SET_NW_TOS:
605 if (nla_get_u8(a) & INET_ECN_MASK)
619 static void clear_stats(struct sw_flow *flow)
623 flow->packet_count = 0;
624 flow->byte_count = 0;
627 /* Called with genl_mutex. */
628 static int expand_table(struct datapath *dp)
630 struct tbl *old_table = get_table_protected(dp);
631 struct tbl *new_table;
633 new_table = tbl_expand(old_table);
634 if (IS_ERR(new_table))
635 return PTR_ERR(new_table);
637 rcu_assign_pointer(dp->table, new_table);
638 tbl_deferred_destroy(old_table, NULL);
643 static int odp_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
645 struct odp_header *odp_header = info->userhdr;
646 struct nlattr **a = info->attrs;
647 struct sw_flow_actions *acts;
648 struct sk_buff *packet;
649 struct sw_flow *flow;
658 if (!a[ODP_PACKET_ATTR_PACKET] || !a[ODP_PACKET_ATTR_ACTIONS] ||
659 nla_len(a[ODP_PACKET_ATTR_PACKET]) < ETH_HLEN)
662 err = validate_actions(a[ODP_PACKET_ATTR_ACTIONS]);
666 len = nla_len(a[ODP_PACKET_ATTR_PACKET]);
667 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
671 skb_reserve(packet, NET_IP_ALIGN);
673 memcpy(__skb_put(packet, len), nla_data(a[ODP_PACKET_ATTR_PACKET]), len);
675 skb_reset_mac_header(packet);
676 eth = eth_hdr(packet);
678 /* Normally, setting the skb 'protocol' field would be handled by a
679 * call to eth_type_trans(), but it assumes there's a sending
680 * device, which we may not have. */
681 if (ntohs(eth->h_proto) >= 1536)
682 packet->protocol = eth->h_proto;
684 packet->protocol = htons(ETH_P_802_2);
686 /* Build an sw_flow for sending this packet. */
692 err = flow_extract(packet, -1, &flow->key, &key_len, &is_frag);
695 flow->tbl_node.hash = flow_hash(&flow->key, key_len);
697 acts = flow_actions_alloc(a[ODP_PACKET_ATTR_ACTIONS]);
701 rcu_assign_pointer(flow->sf_acts, acts);
703 OVS_CB(packet)->flow = flow;
706 dp = get_dp(odp_header->dp_ifindex);
710 err = execute_actions(dp, packet);
726 static const struct nla_policy packet_policy[ODP_PACKET_ATTR_MAX + 1] = {
727 [ODP_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
728 [ODP_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
731 static struct genl_ops dp_packet_genl_ops[] = {
732 { .cmd = ODP_PACKET_CMD_EXECUTE,
733 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
734 .policy = packet_policy,
735 .doit = odp_packet_cmd_execute
739 static void get_dp_stats(struct datapath *dp, struct odp_stats *stats)
743 stats->n_frags = stats->n_hit = stats->n_missed = stats->n_lost = 0;
744 for_each_possible_cpu(i) {
745 const struct dp_stats_percpu *percpu_stats;
746 struct dp_stats_percpu local_stats;
749 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
752 seqcount = read_seqcount_begin(&percpu_stats->seqlock);
753 local_stats = *percpu_stats;
754 } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount));
756 stats->n_frags += local_stats.n_frags;
757 stats->n_hit += local_stats.n_hit;
758 stats->n_missed += local_stats.n_missed;
759 stats->n_lost += local_stats.n_lost;
763 /* MTU of the dp pseudo-device: ETH_DATA_LEN or the minimum of the ports.
764 * Called with RTNL lock.
766 int dp_min_mtu(const struct datapath *dp)
773 list_for_each_entry (p, &dp->port_list, node) {
776 /* Skip any internal ports, since that's what we're trying to
778 if (is_internal_vport(p))
781 dev_mtu = vport_get_mtu(p);
784 if (!mtu || dev_mtu < mtu)
788 return mtu ? mtu : ETH_DATA_LEN;
791 /* Sets the MTU of all datapath devices to the minimum of the ports
792 * Called with RTNL lock.
794 void set_internal_devs_mtu(const struct datapath *dp)
801 mtu = dp_min_mtu(dp);
803 list_for_each_entry (p, &dp->port_list, node) {
804 if (is_internal_vport(p))
805 vport_set_mtu(p, mtu);
809 static const struct nla_policy flow_policy[ODP_FLOW_ATTR_MAX + 1] = {
810 [ODP_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
811 [ODP_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
812 [ODP_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
815 static struct genl_family dp_flow_genl_family = {
816 .id = GENL_ID_GENERATE,
817 .hdrsize = sizeof(struct odp_header),
818 .name = ODP_FLOW_FAMILY,
820 .maxattr = ODP_FLOW_ATTR_MAX
823 static struct genl_multicast_group dp_flow_multicast_group = {
824 .name = ODP_FLOW_MCGROUP
827 /* Called with genl_lock. */
828 static int odp_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
829 struct sk_buff *skb, u32 pid, u32 seq, u32 flags, u8 cmd)
831 const int skb_orig_len = skb->len;
832 const struct sw_flow_actions *sf_acts;
833 struct odp_flow_stats stats;
834 struct odp_header *odp_header;
840 sf_acts = rcu_dereference_protected(flow->sf_acts,
841 lockdep_genl_is_held());
843 odp_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags, cmd);
847 odp_header->dp_ifindex = dp->dp_ifindex;
849 nla = nla_nest_start(skb, ODP_FLOW_ATTR_KEY);
851 goto nla_put_failure;
852 err = flow_to_nlattrs(&flow->key, skb);
855 nla_nest_end(skb, nla);
857 spin_lock_bh(&flow->lock);
859 stats.n_packets = flow->packet_count;
860 stats.n_bytes = flow->byte_count;
861 tcp_flags = flow->tcp_flags;
862 spin_unlock_bh(&flow->lock);
865 NLA_PUT_U64(skb, ODP_FLOW_ATTR_USED, flow_used_time(used));
868 NLA_PUT(skb, ODP_FLOW_ATTR_STATS, sizeof(struct odp_flow_stats), &stats);
871 NLA_PUT_U8(skb, ODP_FLOW_ATTR_TCP_FLAGS, tcp_flags);
873 /* If ODP_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
874 * this is the first flow to be dumped into 'skb'. This is unusual for
875 * Netlink but individual action lists can be longer than
876 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
877 * The userspace caller can always fetch the actions separately if it
878 * really wants them. (Most userspace callers in fact don't care.)
880 * This can only fail for dump operations because the skb is always
881 * properly sized for single flows.
883 err = nla_put(skb, ODP_FLOW_ATTR_ACTIONS, sf_acts->actions_len,
885 if (err < 0 && skb_orig_len)
888 return genlmsg_end(skb, odp_header);
893 genlmsg_cancel(skb, odp_header);
897 static struct sk_buff *odp_flow_cmd_alloc_info(struct sw_flow *flow)
899 const struct sw_flow_actions *sf_acts;
902 sf_acts = rcu_dereference_protected(flow->sf_acts,
903 lockdep_genl_is_held());
905 len = nla_total_size(FLOW_BUFSIZE); /* ODP_FLOW_ATTR_KEY */
906 len += nla_total_size(sf_acts->actions_len); /* ODP_FLOW_ATTR_ACTIONS */
907 len += nla_total_size(sizeof(struct odp_flow_stats)); /* ODP_FLOW_ATTR_STATS */
908 len += nla_total_size(1); /* ODP_FLOW_ATTR_TCP_FLAGS */
909 len += nla_total_size(8); /* ODP_FLOW_ATTR_USED */
910 return genlmsg_new(NLMSG_ALIGN(sizeof(struct odp_header)) + len, GFP_KERNEL);
913 static struct sk_buff *odp_flow_cmd_build_info(struct sw_flow *flow, struct datapath *dp,
914 u32 pid, u32 seq, u8 cmd)
919 skb = odp_flow_cmd_alloc_info(flow);
921 return ERR_PTR(-ENOMEM);
923 retval = odp_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd);
928 static int odp_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
930 struct nlattr **a = info->attrs;
931 struct odp_header *odp_header = info->userhdr;
932 struct tbl_node *flow_node;
933 struct sw_flow_key key;
934 struct sw_flow *flow;
935 struct sk_buff *reply;
944 if (!a[ODP_FLOW_ATTR_KEY])
946 error = flow_from_nlattrs(&key, &key_len, a[ODP_FLOW_ATTR_KEY]);
950 /* Validate actions. */
951 if (a[ODP_FLOW_ATTR_ACTIONS]) {
952 error = validate_actions(a[ODP_FLOW_ATTR_ACTIONS]);
955 } else if (info->genlhdr->cmd == ODP_FLOW_CMD_NEW) {
960 dp = get_dp(odp_header->dp_ifindex);
965 hash = flow_hash(&key, key_len);
966 table = get_table_protected(dp);
967 flow_node = tbl_lookup(table, &key, key_len, hash, flow_cmp);
969 struct sw_flow_actions *acts;
971 /* Bail out if we're not allowed to create a new flow. */
973 if (info->genlhdr->cmd == ODP_FLOW_CMD_SET)
976 /* Expand table, if necessary, to make room. */
977 if (tbl_count(table) >= tbl_n_buckets(table)) {
978 error = expand_table(dp);
981 table = get_table_protected(dp);
987 error = PTR_ERR(flow);
993 /* Obtain actions. */
994 acts = flow_actions_alloc(a[ODP_FLOW_ATTR_ACTIONS]);
995 error = PTR_ERR(acts);
997 goto error_free_flow;
998 rcu_assign_pointer(flow->sf_acts, acts);
1000 /* Put flow in bucket. */
1001 error = tbl_insert(table, &flow->tbl_node, hash);
1003 goto error_free_flow;
1005 reply = odp_flow_cmd_build_info(flow, dp, info->snd_pid,
1006 info->snd_seq, ODP_FLOW_CMD_NEW);
1008 /* We found a matching flow. */
1009 struct sw_flow_actions *old_acts;
1011 /* Bail out if we're not allowed to modify an existing flow.
1012 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1013 * because Generic Netlink treats the latter as a dump
1014 * request. We also accept NLM_F_EXCL in case that bug ever
1018 if (info->genlhdr->cmd == ODP_FLOW_CMD_NEW &&
1019 info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1022 /* Update actions. */
1023 flow = flow_cast(flow_node);
1024 old_acts = rcu_dereference_protected(flow->sf_acts,
1025 lockdep_genl_is_held());
1026 if (a[ODP_FLOW_ATTR_ACTIONS] &&
1027 (old_acts->actions_len != nla_len(a[ODP_FLOW_ATTR_ACTIONS]) ||
1028 memcmp(old_acts->actions, nla_data(a[ODP_FLOW_ATTR_ACTIONS]),
1029 old_acts->actions_len))) {
1030 struct sw_flow_actions *new_acts;
1032 new_acts = flow_actions_alloc(a[ODP_FLOW_ATTR_ACTIONS]);
1033 error = PTR_ERR(new_acts);
1034 if (IS_ERR(new_acts))
1037 rcu_assign_pointer(flow->sf_acts, new_acts);
1038 flow_deferred_free_acts(old_acts);
1041 reply = odp_flow_cmd_build_info(flow, dp, info->snd_pid,
1042 info->snd_seq, ODP_FLOW_CMD_NEW);
1045 if (a[ODP_FLOW_ATTR_CLEAR]) {
1046 spin_lock_bh(&flow->lock);
1048 spin_unlock_bh(&flow->lock);
1053 genl_notify(reply, genl_info_net(info), info->snd_pid,
1054 dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1056 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1057 dp_flow_multicast_group.id, PTR_ERR(reply));
1066 static int odp_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1068 struct nlattr **a = info->attrs;
1069 struct odp_header *odp_header = info->userhdr;
1070 struct sw_flow_key key;
1071 struct tbl_node *flow_node;
1072 struct sk_buff *reply;
1073 struct sw_flow *flow;
1074 struct datapath *dp;
1079 if (!a[ODP_FLOW_ATTR_KEY])
1081 err = flow_from_nlattrs(&key, &key_len, a[ODP_FLOW_ATTR_KEY]);
1085 dp = get_dp(odp_header->dp_ifindex);
1089 table = get_table_protected(dp);
1090 flow_node = tbl_lookup(table, &key, key_len, flow_hash(&key, key_len),
1095 flow = flow_cast(flow_node);
1096 reply = odp_flow_cmd_build_info(flow, dp, info->snd_pid, info->snd_seq, ODP_FLOW_CMD_NEW);
1098 return PTR_ERR(reply);
1100 return genlmsg_reply(reply, info);
1103 static int odp_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1105 struct nlattr **a = info->attrs;
1106 struct odp_header *odp_header = info->userhdr;
1107 struct sw_flow_key key;
1108 struct tbl_node *flow_node;
1109 struct sk_buff *reply;
1110 struct sw_flow *flow;
1111 struct datapath *dp;
1116 if (!a[ODP_FLOW_ATTR_KEY])
1117 return flush_flows(odp_header->dp_ifindex);
1118 err = flow_from_nlattrs(&key, &key_len, a[ODP_FLOW_ATTR_KEY]);
1122 dp = get_dp(odp_header->dp_ifindex);
1126 table = get_table_protected(dp);
1127 flow_node = tbl_lookup(table, &key, key_len, flow_hash(&key, key_len),
1131 flow = flow_cast(flow_node);
1133 reply = odp_flow_cmd_alloc_info(flow);
1137 err = tbl_remove(table, flow_node);
1143 err = odp_flow_cmd_fill_info(flow, dp, reply, info->snd_pid,
1144 info->snd_seq, 0, ODP_FLOW_CMD_DEL);
1147 flow_deferred_free(flow);
1149 genl_notify(reply, genl_info_net(info), info->snd_pid,
1150 dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1154 static int odp_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1156 struct odp_header *odp_header = genlmsg_data(nlmsg_data(cb->nlh));
1157 struct datapath *dp;
1159 dp = get_dp(odp_header->dp_ifindex);
1164 struct tbl_node *flow_node;
1165 struct sw_flow *flow;
1168 bucket = cb->args[0];
1170 flow_node = tbl_next(get_table_protected(dp), &bucket, &obj);
1174 flow = flow_cast(flow_node);
1175 if (odp_flow_cmd_fill_info(flow, dp, skb, NETLINK_CB(cb->skb).pid,
1176 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1177 ODP_FLOW_CMD_NEW) < 0)
1180 cb->args[0] = bucket;
1186 static struct genl_ops dp_flow_genl_ops[] = {
1187 { .cmd = ODP_FLOW_CMD_NEW,
1188 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1189 .policy = flow_policy,
1190 .doit = odp_flow_cmd_new_or_set
1192 { .cmd = ODP_FLOW_CMD_DEL,
1193 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1194 .policy = flow_policy,
1195 .doit = odp_flow_cmd_del
1197 { .cmd = ODP_FLOW_CMD_GET,
1198 .flags = 0, /* OK for unprivileged users. */
1199 .policy = flow_policy,
1200 .doit = odp_flow_cmd_get,
1201 .dumpit = odp_flow_cmd_dump
1203 { .cmd = ODP_FLOW_CMD_SET,
1204 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1205 .policy = flow_policy,
1206 .doit = odp_flow_cmd_new_or_set,
1210 static const struct nla_policy datapath_policy[ODP_DP_ATTR_MAX + 1] = {
1211 #ifdef HAVE_NLA_NUL_STRING
1212 [ODP_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1214 [ODP_DP_ATTR_IPV4_FRAGS] = { .type = NLA_U32 },
1215 [ODP_DP_ATTR_SAMPLING] = { .type = NLA_U32 },
1218 static struct genl_family dp_datapath_genl_family = {
1219 .id = GENL_ID_GENERATE,
1220 .hdrsize = sizeof(struct odp_header),
1221 .name = ODP_DATAPATH_FAMILY,
1223 .maxattr = ODP_DP_ATTR_MAX
1226 static struct genl_multicast_group dp_datapath_multicast_group = {
1227 .name = ODP_DATAPATH_MCGROUP
1230 static int odp_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1231 u32 pid, u32 seq, u32 flags, u8 cmd)
1233 struct odp_header *odp_header;
1237 odp_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family,
1242 odp_header->dp_ifindex = dp->dp_ifindex;
1245 err = nla_put_string(skb, ODP_DP_ATTR_NAME, dp_name(dp));
1248 goto nla_put_failure;
1250 nla = nla_reserve(skb, ODP_DP_ATTR_STATS, sizeof(struct odp_stats));
1252 goto nla_put_failure;
1253 get_dp_stats(dp, nla_data(nla));
1255 NLA_PUT_U32(skb, ODP_DP_ATTR_IPV4_FRAGS,
1256 dp->drop_frags ? ODP_DP_FRAG_DROP : ODP_DP_FRAG_ZERO);
1258 if (dp->sflow_probability)
1259 NLA_PUT_U32(skb, ODP_DP_ATTR_SAMPLING, dp->sflow_probability);
1261 nla = nla_nest_start(skb, ODP_DP_ATTR_MCGROUPS);
1263 goto nla_put_failure;
1264 NLA_PUT_U32(skb, ODP_PACKET_CMD_MISS, packet_mc_group(dp, ODP_PACKET_CMD_MISS));
1265 NLA_PUT_U32(skb, ODP_PACKET_CMD_ACTION, packet_mc_group(dp, ODP_PACKET_CMD_ACTION));
1266 NLA_PUT_U32(skb, ODP_PACKET_CMD_SAMPLE, packet_mc_group(dp, ODP_PACKET_CMD_SAMPLE));
1267 nla_nest_end(skb, nla);
1269 return genlmsg_end(skb, odp_header);
1272 genlmsg_cancel(skb, odp_header);
1277 static struct sk_buff *odp_dp_cmd_build_info(struct datapath *dp, u32 pid,
1280 struct sk_buff *skb;
1283 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1285 return ERR_PTR(-ENOMEM);
1287 retval = odp_dp_cmd_fill_info(dp, skb, pid, seq, 0, cmd);
1290 return ERR_PTR(retval);
1295 static int odp_dp_cmd_validate(struct nlattr *a[ODP_DP_ATTR_MAX + 1])
1297 if (a[ODP_DP_ATTR_IPV4_FRAGS]) {
1298 u32 frags = nla_get_u32(a[ODP_DP_ATTR_IPV4_FRAGS]);
1300 if (frags != ODP_DP_FRAG_ZERO && frags != ODP_DP_FRAG_DROP)
1304 return CHECK_NUL_STRING(a[ODP_DP_ATTR_NAME], IFNAMSIZ - 1);
1307 /* Called with genl_mutex and optionally with RTNL lock also. */
1308 static struct datapath *lookup_datapath(struct odp_header *odp_header, struct nlattr *a[ODP_DP_ATTR_MAX + 1])
1310 struct datapath *dp;
1312 if (!a[ODP_DP_ATTR_NAME])
1313 dp = get_dp(odp_header->dp_ifindex);
1315 struct vport *vport;
1318 vport = vport_locate(nla_data(a[ODP_DP_ATTR_NAME]));
1319 dp = vport && vport->port_no == ODPP_LOCAL ? vport->dp : NULL;
1322 return dp ? dp : ERR_PTR(-ENODEV);
1325 /* Called with genl_mutex. */
1326 static void change_datapath(struct datapath *dp, struct nlattr *a[ODP_DP_ATTR_MAX + 1])
1328 if (a[ODP_DP_ATTR_IPV4_FRAGS])
1329 dp->drop_frags = nla_get_u32(a[ODP_DP_ATTR_IPV4_FRAGS]) == ODP_DP_FRAG_DROP;
1330 if (a[ODP_DP_ATTR_SAMPLING])
1331 dp->sflow_probability = nla_get_u32(a[ODP_DP_ATTR_SAMPLING]);
1334 static int odp_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1336 struct nlattr **a = info->attrs;
1337 struct vport_parms parms;
1338 struct sk_buff *reply;
1339 struct datapath *dp;
1340 struct vport *vport;
1344 if (!a[ODP_DP_ATTR_NAME])
1347 err = odp_dp_cmd_validate(a);
1353 if (!try_module_get(THIS_MODULE))
1354 goto err_unlock_rtnl;
1357 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1359 goto err_put_module;
1360 INIT_LIST_HEAD(&dp->port_list);
1362 /* Initialize kobject for bridge. This will be added as
1363 * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
1364 dp->ifobj.kset = NULL;
1365 kobject_init(&dp->ifobj, &dp_ktype);
1367 /* Allocate table. */
1369 rcu_assign_pointer(dp->table, tbl_create(TBL_MIN_BUCKETS));
1373 /* Set up our datapath device. */
1374 parms.name = nla_data(a[ODP_DP_ATTR_NAME]);
1375 parms.type = ODP_VPORT_TYPE_INTERNAL;
1376 parms.options = NULL;
1378 parms.port_no = ODPP_LOCAL;
1379 vport = new_vport(&parms);
1380 if (IS_ERR(vport)) {
1381 err = PTR_ERR(vport);
1385 goto err_destroy_table;
1387 dp->dp_ifindex = vport_get_ifindex(vport);
1390 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1391 if (!dp->stats_percpu) {
1393 goto err_destroy_local_port;
1396 change_datapath(dp, a);
1398 reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_NEW);
1399 err = PTR_ERR(reply);
1401 goto err_destroy_local_port;
1403 list_add_tail(&dp->list_node, &dps);
1404 dp_sysfs_add_dp(dp);
1408 genl_notify(reply, genl_info_net(info), info->snd_pid,
1409 dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1412 err_destroy_local_port:
1413 dp_detach_port(get_vport_protected(dp, ODPP_LOCAL));
1415 tbl_destroy(get_table_protected(dp), NULL);
1419 module_put(THIS_MODULE);
1426 static int odp_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1428 struct vport *vport, *next_vport;
1429 struct sk_buff *reply;
1430 struct datapath *dp;
1433 err = odp_dp_cmd_validate(info->attrs);
1438 dp = lookup_datapath(info->userhdr, info->attrs);
1443 reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_DEL);
1444 err = PTR_ERR(reply);
1448 list_for_each_entry_safe (vport, next_vport, &dp->port_list, node)
1449 if (vport->port_no != ODPP_LOCAL)
1450 dp_detach_port(vport);
1452 dp_sysfs_del_dp(dp);
1453 list_del(&dp->list_node);
1454 dp_detach_port(get_vport_protected(dp, ODPP_LOCAL));
1456 /* rtnl_unlock() will wait until all the references to devices that
1457 * are pending unregistration have been dropped. We do it here to
1458 * ensure that any internal devices (which contain DP pointers) are
1459 * fully destroyed before freeing the datapath.
1463 call_rcu(&dp->rcu, destroy_dp_rcu);
1464 module_put(THIS_MODULE);
1466 genl_notify(reply, genl_info_net(info), info->snd_pid,
1467 dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1477 static int odp_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1479 struct sk_buff *reply;
1480 struct datapath *dp;
1483 err = odp_dp_cmd_validate(info->attrs);
1487 dp = lookup_datapath(info->userhdr, info->attrs);
1491 change_datapath(dp, info->attrs);
1493 reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_NEW);
1494 if (IS_ERR(reply)) {
1495 err = PTR_ERR(reply);
1496 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1497 dp_datapath_multicast_group.id, err);
1501 genl_notify(reply, genl_info_net(info), info->snd_pid,
1502 dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1506 static int odp_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1508 struct sk_buff *reply;
1509 struct datapath *dp;
1512 err = odp_dp_cmd_validate(info->attrs);
1516 dp = lookup_datapath(info->userhdr, info->attrs);
1520 reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_NEW);
1522 return PTR_ERR(reply);
1524 return genlmsg_reply(reply, info);
1527 static int odp_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1529 struct datapath *dp;
1530 int skip = cb->args[0];
1533 list_for_each_entry (dp, &dps, list_node) {
1536 if (odp_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).pid,
1537 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1538 ODP_DP_CMD_NEW) < 0)
1548 static struct genl_ops dp_datapath_genl_ops[] = {
1549 { .cmd = ODP_DP_CMD_NEW,
1550 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1551 .policy = datapath_policy,
1552 .doit = odp_dp_cmd_new
1554 { .cmd = ODP_DP_CMD_DEL,
1555 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1556 .policy = datapath_policy,
1557 .doit = odp_dp_cmd_del
1559 { .cmd = ODP_DP_CMD_GET,
1560 .flags = 0, /* OK for unprivileged users. */
1561 .policy = datapath_policy,
1562 .doit = odp_dp_cmd_get,
1563 .dumpit = odp_dp_cmd_dump
1565 { .cmd = ODP_DP_CMD_SET,
1566 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1567 .policy = datapath_policy,
1568 .doit = odp_dp_cmd_set,
1572 static const struct nla_policy vport_policy[ODP_VPORT_ATTR_MAX + 1] = {
1573 #ifdef HAVE_NLA_NUL_STRING
1574 [ODP_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1575 [ODP_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1576 [ODP_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1577 [ODP_VPORT_ATTR_STATS] = { .len = sizeof(struct rtnl_link_stats64) },
1578 [ODP_VPORT_ATTR_ADDRESS] = { .len = ETH_ALEN },
1580 [ODP_VPORT_ATTR_STATS] = { .minlen = sizeof(struct rtnl_link_stats64) },
1581 [ODP_VPORT_ATTR_ADDRESS] = { .minlen = ETH_ALEN },
1583 [ODP_VPORT_ATTR_MTU] = { .type = NLA_U32 },
1584 [ODP_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1587 static struct genl_family dp_vport_genl_family = {
1588 .id = GENL_ID_GENERATE,
1589 .hdrsize = sizeof(struct odp_header),
1590 .name = ODP_VPORT_FAMILY,
1592 .maxattr = ODP_VPORT_ATTR_MAX
1595 static struct genl_multicast_group dp_vport_multicast_group = {
1596 .name = ODP_VPORT_MCGROUP
1599 /* Called with RTNL lock or RCU read lock. */
1600 static int odp_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1601 u32 pid, u32 seq, u32 flags, u8 cmd)
1603 struct odp_header *odp_header;
1605 int ifindex, iflink;
1609 odp_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family,
1614 odp_header->dp_ifindex = vport->dp->dp_ifindex;
1616 NLA_PUT_U32(skb, ODP_VPORT_ATTR_PORT_NO, vport->port_no);
1617 NLA_PUT_U32(skb, ODP_VPORT_ATTR_TYPE, vport_get_type(vport));
1618 NLA_PUT_STRING(skb, ODP_VPORT_ATTR_NAME, vport_get_name(vport));
1620 nla = nla_reserve(skb, ODP_VPORT_ATTR_STATS, sizeof(struct rtnl_link_stats64));
1622 goto nla_put_failure;
1623 if (vport_get_stats(vport, nla_data(nla)))
1624 __skb_trim(skb, skb->len - nla->nla_len);
1626 NLA_PUT(skb, ODP_VPORT_ATTR_ADDRESS, ETH_ALEN, vport_get_addr(vport));
1628 mtu = vport_get_mtu(vport);
1630 NLA_PUT_U32(skb, ODP_VPORT_ATTR_MTU, mtu);
1632 err = vport_get_options(vport, skb);
1633 if (err == -EMSGSIZE)
1636 ifindex = vport_get_ifindex(vport);
1638 NLA_PUT_U32(skb, ODP_VPORT_ATTR_IFINDEX, ifindex);
1640 iflink = vport_get_iflink(vport);
1642 NLA_PUT_U32(skb, ODP_VPORT_ATTR_IFLINK, iflink);
1644 return genlmsg_end(skb, odp_header);
1649 genlmsg_cancel(skb, odp_header);
1653 /* Called with RTNL lock or RCU read lock. */
1654 static struct sk_buff *odp_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 = odp_vport_cmd_fill_info(vport, skb, pid, seq, 0, cmd);
1667 return ERR_PTR(retval);
1672 static int odp_vport_cmd_validate(struct nlattr *a[ODP_VPORT_ATTR_MAX + 1])
1674 return CHECK_NUL_STRING(a[ODP_VPORT_ATTR_NAME], IFNAMSIZ - 1);
1677 /* Called with RTNL lock or RCU read lock. */
1678 static struct vport *lookup_vport(struct odp_header *odp_header,
1679 struct nlattr *a[ODP_VPORT_ATTR_MAX + 1])
1681 struct datapath *dp;
1682 struct vport *vport;
1684 if (a[ODP_VPORT_ATTR_NAME]) {
1685 vport = vport_locate(nla_data(a[ODP_VPORT_ATTR_NAME]));
1687 return ERR_PTR(-ENODEV);
1689 } else if (a[ODP_VPORT_ATTR_PORT_NO]) {
1690 u32 port_no = nla_get_u32(a[ODP_VPORT_ATTR_PORT_NO]);
1692 if (port_no >= DP_MAX_PORTS)
1693 return ERR_PTR(-EFBIG);
1695 dp = get_dp(odp_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[ODP_VPORT_ATTR_MAX + 1])
1711 if (a[ODP_VPORT_ATTR_STATS])
1712 err = vport_set_stats(vport, nla_data(a[ODP_VPORT_ATTR_STATS]));
1713 if (!err && a[ODP_VPORT_ATTR_ADDRESS])
1714 err = vport_set_addr(vport, nla_data(a[ODP_VPORT_ATTR_ADDRESS]));
1715 if (!err && a[ODP_VPORT_ATTR_MTU])
1716 err = vport_set_mtu(vport, nla_get_u32(a[ODP_VPORT_ATTR_MTU]));
1720 static int odp_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1722 struct nlattr **a = info->attrs;
1723 struct odp_header *odp_header = info->userhdr;
1724 struct vport_parms parms;
1725 struct sk_buff *reply;
1726 struct vport *vport;
1727 struct datapath *dp;
1732 if (!a[ODP_VPORT_ATTR_NAME] || !a[ODP_VPORT_ATTR_TYPE])
1735 err = odp_vport_cmd_validate(a);
1740 dp = get_dp(odp_header->dp_ifindex);
1745 if (a[ODP_VPORT_ATTR_PORT_NO]) {
1746 port_no = nla_get_u32(a[ODP_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[ODP_VPORT_ATTR_NAME]);
1769 parms.type = nla_get_u32(a[ODP_VPORT_ATTR_TYPE]);
1770 parms.options = a[ODP_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 = odp_vport_cmd_build_info(vport, info->snd_pid,
1785 info->snd_seq, ODP_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 odp_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 = odp_vport_cmd_validate(a);
1815 vport = lookup_vport(info->userhdr, a);
1816 err = PTR_ERR(vport);
1821 if (a[ODP_VPORT_ATTR_OPTIONS])
1822 err = vport_set_options(vport, a[ODP_VPORT_ATTR_OPTIONS]);
1824 err = change_vport(vport, a);
1826 reply = odp_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 odp_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 = odp_vport_cmd_validate(a);
1856 vport = lookup_vport(info->userhdr, a);
1857 err = PTR_ERR(vport);
1861 if (vport->port_no == ODPP_LOCAL) {
1866 reply = odp_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 odp_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1885 struct nlattr **a = info->attrs;
1886 struct odp_header *odp_header = info->userhdr;
1887 struct sk_buff *reply;
1888 struct vport *vport;
1891 err = odp_vport_cmd_validate(a);
1896 vport = lookup_vport(odp_header, a);
1897 err = PTR_ERR(vport);
1901 reply = odp_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 odp_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1919 struct odp_header *odp_header = genlmsg_data(nlmsg_data(cb->nlh));
1920 struct datapath *dp;
1924 dp = get_dp(odp_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 (odp_vport_cmd_fill_info(vport, skb, NETLINK_CB(cb->skb).pid,
1937 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1938 ODP_VPORT_CMD_NEW) < 0)
1943 cb->args[0] = port_no;
1949 static struct genl_ops dp_vport_genl_ops[] = {
1950 { .cmd = ODP_VPORT_CMD_NEW,
1951 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1952 .policy = vport_policy,
1953 .doit = odp_vport_cmd_new
1955 { .cmd = ODP_VPORT_CMD_DEL,
1956 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1957 .policy = vport_policy,
1958 .doit = odp_vport_cmd_del
1960 { .cmd = ODP_VPORT_CMD_GET,
1961 .flags = 0, /* OK for unprivileged users. */
1962 .policy = vport_policy,
1963 .doit = odp_vport_cmd_get,
1964 .dumpit = odp_vport_cmd_dump
1966 { .cmd = ODP_VPORT_CMD_SET,
1967 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1968 .policy = vport_policy,
1969 .doit = odp_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);
2004 static int dp_register_genl(void)
2011 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2012 const struct genl_family_and_ops *f = &dp_genl_families[i];
2014 err = genl_register_family_with_ops(f->family, f->ops,
2021 err = genl_register_mc_group(f->family, f->group);
2027 err = packet_register_mc_groups();
2033 dp_unregister_genl(n_registered);
2037 static int __init dp_init(void)
2039 struct sk_buff *dummy_skb;
2042 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
2044 printk("Open vSwitch %s, built "__DATE__" "__TIME__"\n", VERSION BUILDNR);
2052 goto error_flow_exit;
2054 err = register_netdevice_notifier(&dp_device_notifier);
2056 goto error_vport_exit;
2058 err = dp_register_genl();
2060 goto error_unreg_notifier;
2064 error_unreg_notifier:
2065 unregister_netdevice_notifier(&dp_device_notifier);
2074 static void dp_cleanup(void)
2077 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2078 unregister_netdevice_notifier(&dp_device_notifier);
2083 module_init(dp_init);
2084 module_exit(dp_cleanup);
2086 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2087 MODULE_LICENSE("GPL");