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"
52 #include "loop_counter.h"
55 #include "vport-internal_dev.h"
57 int (*dp_ioctl_hook)(struct net_device *dev, struct ifreq *rq, int cmd);
58 EXPORT_SYMBOL(dp_ioctl_hook);
63 * Writes to device state (add/remove datapath, port, set operations on vports,
64 * etc.) are protected by RTNL.
66 * Writes to other state (flow table modifications, set miscellaneous datapath
67 * parameters such as drop frags, etc.) are protected by genl_mutex. The RTNL
68 * lock nests inside genl_mutex.
70 * Reads are protected by RCU.
72 * There are a few special cases (mostly stats) that have their own
73 * synchronization but they nest under all of above and don't interact with
77 /* Global list of datapaths to enable dumping them all out.
78 * Protected by genl_mutex.
80 static LIST_HEAD(dps);
82 static struct vport *new_vport(const struct vport_parms *);
83 static int queue_control_packets(struct datapath *, struct sk_buff *,
84 const struct dp_upcall_info *);
86 /* Must be called with rcu_read_lock, genl_mutex, or RTNL lock. */
87 struct datapath *get_dp(int dp_ifindex)
89 struct datapath *dp = NULL;
90 struct net_device *dev;
93 dev = dev_get_by_index_rcu(&init_net, dp_ifindex);
95 struct vport *vport = internal_dev_get_vport(dev);
103 EXPORT_SYMBOL_GPL(get_dp);
105 /* Must be called with genl_mutex. */
106 static struct tbl *get_table_protected(struct datapath *dp)
108 return rcu_dereference_protected(dp->table, lockdep_genl_is_held());
111 /* Must be called with rcu_read_lock or RTNL lock. */
112 static struct vport *get_vport_protected(struct datapath *dp, u16 port_no)
114 return rcu_dereference_rtnl(dp->ports[port_no]);
117 /* Must be called with rcu_read_lock or RTNL lock. */
118 const char *dp_name(const struct datapath *dp)
120 return vport_get_name(rcu_dereference_rtnl(dp->ports[ODPP_LOCAL]));
123 static inline size_t br_nlmsg_size(void)
125 return NLMSG_ALIGN(sizeof(struct ifinfomsg))
126 + nla_total_size(IFNAMSIZ) /* IFLA_IFNAME */
127 + nla_total_size(MAX_ADDR_LEN) /* IFLA_ADDRESS */
128 + nla_total_size(4) /* IFLA_MASTER */
129 + nla_total_size(4) /* IFLA_MTU */
130 + nla_total_size(4) /* IFLA_LINK */
131 + nla_total_size(1); /* IFLA_OPERSTATE */
134 /* Caller must hold RTNL lock. */
135 static int dp_fill_ifinfo(struct sk_buff *skb,
136 const struct vport *port,
137 int event, unsigned int flags)
139 struct datapath *dp = port->dp;
140 int ifindex = vport_get_ifindex(port);
141 int iflink = vport_get_iflink(port);
142 struct ifinfomsg *hdr;
143 struct nlmsghdr *nlh;
151 nlh = nlmsg_put(skb, 0, 0, event, sizeof(*hdr), flags);
155 hdr = nlmsg_data(nlh);
156 hdr->ifi_family = AF_BRIDGE;
158 hdr->ifi_type = ARPHRD_ETHER;
159 hdr->ifi_index = ifindex;
160 hdr->ifi_flags = vport_get_flags(port);
163 NLA_PUT_STRING(skb, IFLA_IFNAME, vport_get_name(port));
164 NLA_PUT_U32(skb, IFLA_MASTER,
165 vport_get_ifindex(get_vport_protected(dp, ODPP_LOCAL)));
166 NLA_PUT_U32(skb, IFLA_MTU, vport_get_mtu(port));
167 #ifdef IFLA_OPERSTATE
168 NLA_PUT_U8(skb, IFLA_OPERSTATE,
169 vport_is_running(port)
170 ? vport_get_operstate(port)
174 NLA_PUT(skb, IFLA_ADDRESS, ETH_ALEN, vport_get_addr(port));
176 if (ifindex != iflink)
177 NLA_PUT_U32(skb, IFLA_LINK,iflink);
179 return nlmsg_end(skb, nlh);
182 nlmsg_cancel(skb, nlh);
186 /* Caller must hold RTNL lock. */
187 static void dp_ifinfo_notify(int event, struct vport *port)
192 skb = nlmsg_new(br_nlmsg_size(), GFP_KERNEL);
196 err = dp_fill_ifinfo(skb, port, event, 0);
198 /* -EMSGSIZE implies BUG in br_nlmsg_size() */
199 WARN_ON(err == -EMSGSIZE);
203 rtnl_notify(skb, &init_net, 0, RTNLGRP_LINK, NULL, GFP_KERNEL);
207 rtnl_set_sk_err(&init_net, RTNLGRP_LINK, err);
210 static void release_dp(struct kobject *kobj)
212 struct datapath *dp = container_of(kobj, struct datapath, ifobj);
216 static struct kobj_type dp_ktype = {
217 .release = release_dp
220 static void destroy_dp_rcu(struct rcu_head *rcu)
222 struct datapath *dp = container_of(rcu, struct datapath, rcu);
224 tbl_destroy((struct tbl __force *)dp->table, flow_free_tbl);
225 free_percpu(dp->stats_percpu);
226 kobject_put(&dp->ifobj);
229 /* Called with RTNL lock and genl_lock. */
230 static struct vport *new_vport(const struct vport_parms *parms)
234 vport = vport_add(parms);
235 if (!IS_ERR(vport)) {
236 struct datapath *dp = parms->dp;
238 rcu_assign_pointer(dp->ports[parms->port_no], vport);
239 list_add(&vport->node, &dp->port_list);
241 dp_ifinfo_notify(RTM_NEWLINK, vport);
247 /* Called with RTNL lock. */
248 int dp_detach_port(struct vport *p)
252 if (p->port_no != ODPP_LOCAL)
254 dp_ifinfo_notify(RTM_DELLINK, p);
256 /* First drop references to device. */
258 rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
260 /* Then destroy it. */
264 /* Must be called with rcu_read_lock. */
265 void dp_process_received_packet(struct vport *p, struct sk_buff *skb)
267 struct datapath *dp = p->dp;
268 struct dp_stats_percpu *stats;
269 int stats_counter_off;
270 struct sw_flow_actions *acts;
271 struct loop_counter *loop;
274 OVS_CB(skb)->vport = p;
276 if (!OVS_CB(skb)->flow) {
277 struct sw_flow_key key;
278 struct tbl_node *flow_node;
281 /* Extract flow from 'skb' into 'key'. */
282 error = flow_extract(skb, p->port_no, &key, &is_frag);
283 if (unlikely(error)) {
288 if (is_frag && dp->drop_frags) {
290 stats_counter_off = offsetof(struct dp_stats_percpu, n_frags);
295 flow_node = tbl_lookup(rcu_dereference(dp->table), &key,
296 flow_hash(&key), flow_cmp);
297 if (unlikely(!flow_node)) {
298 struct dp_upcall_info upcall;
300 upcall.cmd = ODP_PACKET_CMD_MISS;
303 upcall.sample_pool = 0;
304 upcall.actions = NULL;
305 upcall.actions_len = 0;
306 dp_upcall(dp, skb, &upcall);
307 stats_counter_off = offsetof(struct dp_stats_percpu, n_missed);
311 OVS_CB(skb)->flow = flow_cast(flow_node);
314 stats_counter_off = offsetof(struct dp_stats_percpu, n_hit);
315 flow_used(OVS_CB(skb)->flow, skb);
317 acts = rcu_dereference(OVS_CB(skb)->flow->sf_acts);
319 /* Check whether we've looped too much. */
320 loop = loop_get_counter();
321 if (unlikely(++loop->count > MAX_LOOPS))
322 loop->looping = true;
323 if (unlikely(loop->looping)) {
324 loop_suppress(dp, acts);
329 /* Execute actions. */
330 execute_actions(dp, skb, &OVS_CB(skb)->flow->key, acts->actions,
333 /* Check whether sub-actions looped too much. */
334 if (unlikely(loop->looping))
335 loop_suppress(dp, acts);
338 /* Decrement loop counter. */
340 loop->looping = false;
344 /* Update datapath statistics. */
346 stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
348 write_seqcount_begin(&stats->seqlock);
349 (*(u64 *)((u8 *)stats + stats_counter_off))++;
350 write_seqcount_end(&stats->seqlock);
355 static void copy_and_csum_skb(struct sk_buff *skb, void *to)
357 u16 csum_start, csum_offset;
360 get_skb_csum_pointers(skb, &csum_start, &csum_offset);
361 csum_start -= skb_headroom(skb);
363 skb_copy_bits(skb, 0, to, csum_start);
365 csum = skb_copy_and_csum_bits(skb, csum_start, to + csum_start,
366 skb->len - csum_start, 0);
367 *(__sum16 *)(to + csum_start + csum_offset) = csum_fold(csum);
370 static struct genl_family dp_packet_genl_family = {
371 .id = GENL_ID_GENERATE,
372 .hdrsize = sizeof(struct odp_header),
373 .name = ODP_PACKET_FAMILY,
375 .maxattr = ODP_PACKET_ATTR_MAX
378 /* Generic Netlink multicast groups for upcalls.
380 * We really want three unique multicast groups per datapath, but we can't even
381 * get one, because genl_register_mc_group() takes genl_lock, which is also
382 * held during Generic Netlink message processing, so trying to acquire
383 * multicast groups during ODP_DP_NEW processing deadlocks. Instead, we
384 * preallocate a few groups and use them round-robin for datapaths. Collision
385 * isn't fatal--multicast listeners should check that the family is the one
386 * that they want and discard others--but it wastes time and memory to receive
389 #define PACKET_N_MC_GROUPS 16
390 static struct genl_multicast_group packet_mc_groups[PACKET_N_MC_GROUPS];
392 static u32 packet_mc_group(struct datapath *dp, u8 cmd)
395 BUILD_BUG_ON_NOT_POWER_OF_2(PACKET_N_MC_GROUPS);
397 idx = jhash_2words(dp->dp_ifindex, cmd, 0) & (PACKET_N_MC_GROUPS - 1);
398 return packet_mc_groups[idx].id;
401 static int packet_register_mc_groups(void)
405 for (i = 0; i < PACKET_N_MC_GROUPS; i++) {
406 struct genl_multicast_group *group = &packet_mc_groups[i];
409 sprintf(group->name, "packet%d", i);
410 error = genl_register_mc_group(&dp_packet_genl_family, group);
417 int dp_upcall(struct datapath *dp, struct sk_buff *skb, const struct dp_upcall_info *upcall_info)
419 struct dp_stats_percpu *stats;
422 WARN_ON_ONCE(skb_shared(skb));
424 forward_ip_summed(skb);
426 err = vswitch_skb_checksum_setup(skb);
430 /* Break apart GSO packets into their component pieces. Otherwise
431 * userspace may try to stuff a 64kB packet into a 1500-byte MTU. */
432 if (skb_is_gso(skb)) {
433 struct sk_buff *nskb = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
443 err = queue_control_packets(dp, skb, upcall_info);
453 stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
455 write_seqcount_begin(&stats->seqlock);
457 write_seqcount_end(&stats->seqlock);
464 /* Send each packet in the 'skb' list to userspace for 'dp' as directed by
465 * 'upcall_info'. There will be only one packet unless we broke up a GSO
468 static int queue_control_packets(struct datapath *dp, struct sk_buff *skb,
469 const struct dp_upcall_info *upcall_info)
471 u32 group = packet_mc_group(dp, upcall_info->cmd);
472 struct sk_buff *nskb;
476 if (OVS_CB(skb)->vport)
477 port_no = OVS_CB(skb)->vport->port_no;
479 port_no = ODPP_LOCAL;
482 struct odp_header *upcall;
483 struct sk_buff *user_skb; /* to be queued to userspace */
490 err = vlan_deaccel_tag(skb);
494 if (nla_attr_size(skb->len) > USHRT_MAX)
497 len = sizeof(struct odp_header);
498 len += nla_total_size(skb->len);
499 len += nla_total_size(FLOW_BUFSIZE);
500 if (upcall_info->userdata)
501 len += nla_total_size(8);
502 if (upcall_info->sample_pool)
503 len += nla_total_size(4);
504 if (upcall_info->actions_len)
505 len += nla_total_size(upcall_info->actions_len);
507 user_skb = genlmsg_new(len, GFP_ATOMIC);
509 netlink_set_err(INIT_NET_GENL_SOCK, 0, group, -ENOBUFS);
513 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family, 0, upcall_info->cmd);
514 upcall->dp_ifindex = dp->dp_ifindex;
516 nla = nla_nest_start(user_skb, ODP_PACKET_ATTR_KEY);
517 flow_to_nlattrs(upcall_info->key, user_skb);
518 nla_nest_end(user_skb, nla);
520 if (upcall_info->userdata)
521 nla_put_u64(user_skb, ODP_PACKET_ATTR_USERDATA, upcall_info->userdata);
522 if (upcall_info->sample_pool)
523 nla_put_u32(user_skb, ODP_PACKET_ATTR_SAMPLE_POOL, upcall_info->sample_pool);
524 if (upcall_info->actions_len) {
525 const struct nlattr *actions = upcall_info->actions;
526 u32 actions_len = upcall_info->actions_len;
528 nla = nla_nest_start(user_skb, ODP_PACKET_ATTR_ACTIONS);
529 memcpy(__skb_put(user_skb, actions_len), actions, actions_len);
530 nla_nest_end(user_skb, nla);
533 nla = __nla_reserve(user_skb, ODP_PACKET_ATTR_PACKET, skb->len);
534 if (skb->ip_summed == CHECKSUM_PARTIAL)
535 copy_and_csum_skb(skb, nla_data(nla));
537 skb_copy_bits(skb, 0, nla_data(nla), skb->len);
539 err = genlmsg_multicast(user_skb, 0, group, GFP_ATOMIC);
550 while ((skb = nskb) != NULL) {
557 /* Called with genl_mutex. */
558 static int flush_flows(int dp_ifindex)
560 struct tbl *old_table;
561 struct tbl *new_table;
564 dp = get_dp(dp_ifindex);
568 old_table = get_table_protected(dp);
569 new_table = tbl_create(TBL_MIN_BUCKETS);
573 rcu_assign_pointer(dp->table, new_table);
575 tbl_deferred_destroy(old_table, flow_free_tbl);
580 static int validate_actions(const struct nlattr *attr)
582 const struct nlattr *a;
585 nla_for_each_nested(a, attr, rem) {
586 static const u32 action_lens[ODP_ACTION_ATTR_MAX + 1] = {
587 [ODP_ACTION_ATTR_OUTPUT] = 4,
588 [ODP_ACTION_ATTR_CONTROLLER] = 8,
589 [ODP_ACTION_ATTR_SET_DL_TCI] = 2,
590 [ODP_ACTION_ATTR_STRIP_VLAN] = 0,
591 [ODP_ACTION_ATTR_SET_DL_SRC] = ETH_ALEN,
592 [ODP_ACTION_ATTR_SET_DL_DST] = ETH_ALEN,
593 [ODP_ACTION_ATTR_SET_NW_SRC] = 4,
594 [ODP_ACTION_ATTR_SET_NW_DST] = 4,
595 [ODP_ACTION_ATTR_SET_NW_TOS] = 1,
596 [ODP_ACTION_ATTR_SET_TP_SRC] = 2,
597 [ODP_ACTION_ATTR_SET_TP_DST] = 2,
598 [ODP_ACTION_ATTR_SET_TUNNEL] = 8,
599 [ODP_ACTION_ATTR_SET_PRIORITY] = 4,
600 [ODP_ACTION_ATTR_POP_PRIORITY] = 0,
601 [ODP_ACTION_ATTR_DROP_SPOOFED_ARP] = 0,
603 int type = nla_type(a);
605 if (type > ODP_ACTION_ATTR_MAX || nla_len(a) != action_lens[type])
609 case ODP_ACTION_ATTR_UNSPEC:
612 case ODP_ACTION_ATTR_CONTROLLER:
613 case ODP_ACTION_ATTR_STRIP_VLAN:
614 case ODP_ACTION_ATTR_SET_DL_SRC:
615 case ODP_ACTION_ATTR_SET_DL_DST:
616 case ODP_ACTION_ATTR_SET_NW_SRC:
617 case ODP_ACTION_ATTR_SET_NW_DST:
618 case ODP_ACTION_ATTR_SET_TP_SRC:
619 case ODP_ACTION_ATTR_SET_TP_DST:
620 case ODP_ACTION_ATTR_SET_TUNNEL:
621 case ODP_ACTION_ATTR_SET_PRIORITY:
622 case ODP_ACTION_ATTR_POP_PRIORITY:
623 case ODP_ACTION_ATTR_DROP_SPOOFED_ARP:
624 /* No validation needed. */
627 case ODP_ACTION_ATTR_OUTPUT:
628 if (nla_get_u32(a) >= DP_MAX_PORTS)
632 case ODP_ACTION_ATTR_SET_DL_TCI:
633 if (nla_get_be16(a) & htons(VLAN_CFI_MASK))
637 case ODP_ACTION_ATTR_SET_NW_TOS:
638 if (nla_get_u8(a) & INET_ECN_MASK)
652 static void clear_stats(struct sw_flow *flow)
656 flow->packet_count = 0;
657 flow->byte_count = 0;
660 /* Called with genl_mutex. */
661 static int expand_table(struct datapath *dp)
663 struct tbl *old_table = get_table_protected(dp);
664 struct tbl *new_table;
666 new_table = tbl_expand(old_table);
667 if (IS_ERR(new_table))
668 return PTR_ERR(new_table);
670 rcu_assign_pointer(dp->table, new_table);
671 tbl_deferred_destroy(old_table, NULL);
676 static int odp_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
678 struct odp_header *odp_header = info->userhdr;
679 struct nlattr **a = info->attrs;
680 struct sk_buff *packet;
681 struct sw_flow_key key;
689 if (!a[ODP_PACKET_ATTR_PACKET] || !a[ODP_PACKET_ATTR_ACTIONS] ||
690 nla_len(a[ODP_PACKET_ATTR_PACKET]) < ETH_HLEN)
693 err = validate_actions(a[ODP_PACKET_ATTR_ACTIONS]);
697 len = nla_len(a[ODP_PACKET_ATTR_PACKET]);
698 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
702 skb_reserve(packet, NET_IP_ALIGN);
704 memcpy(__skb_put(packet, len), nla_data(a[ODP_PACKET_ATTR_PACKET]), len);
706 skb_reset_mac_header(packet);
707 eth = eth_hdr(packet);
709 /* Normally, setting the skb 'protocol' field would be handled by a
710 * call to eth_type_trans(), but it assumes there's a sending
711 * device, which we may not have. */
712 if (ntohs(eth->h_proto) >= 1536)
713 packet->protocol = eth->h_proto;
715 packet->protocol = htons(ETH_P_802_2);
717 /* Initialize OVS_CB (it came from Netlink so might not be zeroed). */
718 memset(OVS_CB(packet), 0, sizeof(struct ovs_skb_cb));
720 err = flow_extract(packet, -1, &key, &is_frag);
725 dp = get_dp(odp_header->dp_ifindex);
729 err = execute_actions(dp, packet, &key,
730 nla_data(a[ODP_PACKET_ATTR_ACTIONS]),
731 nla_len(a[ODP_PACKET_ATTR_ACTIONS]));
743 static const struct nla_policy packet_policy[ODP_PACKET_ATTR_MAX + 1] = {
744 [ODP_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
745 [ODP_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
748 static struct genl_ops dp_packet_genl_ops[] = {
749 { .cmd = ODP_PACKET_CMD_EXECUTE,
750 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
751 .policy = packet_policy,
752 .doit = odp_packet_cmd_execute
756 static void get_dp_stats(struct datapath *dp, struct odp_stats *stats)
760 stats->n_frags = stats->n_hit = stats->n_missed = stats->n_lost = 0;
761 for_each_possible_cpu(i) {
762 const struct dp_stats_percpu *percpu_stats;
763 struct dp_stats_percpu local_stats;
766 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
769 seqcount = read_seqcount_begin(&percpu_stats->seqlock);
770 local_stats = *percpu_stats;
771 } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount));
773 stats->n_frags += local_stats.n_frags;
774 stats->n_hit += local_stats.n_hit;
775 stats->n_missed += local_stats.n_missed;
776 stats->n_lost += local_stats.n_lost;
780 /* MTU of the dp pseudo-device: ETH_DATA_LEN or the minimum of the ports.
781 * Called with RTNL lock.
783 int dp_min_mtu(const struct datapath *dp)
790 list_for_each_entry (p, &dp->port_list, node) {
793 /* Skip any internal ports, since that's what we're trying to
795 if (is_internal_vport(p))
798 dev_mtu = vport_get_mtu(p);
801 if (!mtu || dev_mtu < mtu)
805 return mtu ? mtu : ETH_DATA_LEN;
808 /* Sets the MTU of all datapath devices to the minimum of the ports
809 * Called with RTNL lock.
811 void set_internal_devs_mtu(const struct datapath *dp)
818 mtu = dp_min_mtu(dp);
820 list_for_each_entry (p, &dp->port_list, node) {
821 if (is_internal_vport(p))
822 vport_set_mtu(p, mtu);
826 static const struct nla_policy flow_policy[ODP_FLOW_ATTR_MAX + 1] = {
827 [ODP_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
828 [ODP_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
829 [ODP_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
832 static struct genl_family dp_flow_genl_family = {
833 .id = GENL_ID_GENERATE,
834 .hdrsize = sizeof(struct odp_header),
835 .name = ODP_FLOW_FAMILY,
837 .maxattr = ODP_FLOW_ATTR_MAX
840 static struct genl_multicast_group dp_flow_multicast_group = {
841 .name = ODP_FLOW_MCGROUP
844 /* Called with genl_lock. */
845 static int odp_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
846 struct sk_buff *skb, u32 pid, u32 seq, u32 flags, u8 cmd)
848 const int skb_orig_len = skb->len;
849 const struct sw_flow_actions *sf_acts;
850 struct odp_flow_stats stats;
851 struct odp_header *odp_header;
857 sf_acts = rcu_dereference_protected(flow->sf_acts,
858 lockdep_genl_is_held());
860 odp_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags, cmd);
864 odp_header->dp_ifindex = dp->dp_ifindex;
866 nla = nla_nest_start(skb, ODP_FLOW_ATTR_KEY);
868 goto nla_put_failure;
869 err = flow_to_nlattrs(&flow->key, skb);
872 nla_nest_end(skb, nla);
874 spin_lock_bh(&flow->lock);
876 stats.n_packets = flow->packet_count;
877 stats.n_bytes = flow->byte_count;
878 tcp_flags = flow->tcp_flags;
879 spin_unlock_bh(&flow->lock);
882 NLA_PUT_U64(skb, ODP_FLOW_ATTR_USED, flow_used_time(used));
885 NLA_PUT(skb, ODP_FLOW_ATTR_STATS, sizeof(struct odp_flow_stats), &stats);
888 NLA_PUT_U8(skb, ODP_FLOW_ATTR_TCP_FLAGS, tcp_flags);
890 /* If ODP_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
891 * this is the first flow to be dumped into 'skb'. This is unusual for
892 * Netlink but individual action lists can be longer than
893 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
894 * The userspace caller can always fetch the actions separately if it
895 * really wants them. (Most userspace callers in fact don't care.)
897 * This can only fail for dump operations because the skb is always
898 * properly sized for single flows.
900 err = nla_put(skb, ODP_FLOW_ATTR_ACTIONS, sf_acts->actions_len,
902 if (err < 0 && skb_orig_len)
905 return genlmsg_end(skb, odp_header);
910 genlmsg_cancel(skb, odp_header);
914 static struct sk_buff *odp_flow_cmd_alloc_info(struct sw_flow *flow)
916 const struct sw_flow_actions *sf_acts;
919 sf_acts = rcu_dereference_protected(flow->sf_acts,
920 lockdep_genl_is_held());
922 len = nla_total_size(FLOW_BUFSIZE); /* ODP_FLOW_ATTR_KEY */
923 len += nla_total_size(sf_acts->actions_len); /* ODP_FLOW_ATTR_ACTIONS */
924 len += nla_total_size(sizeof(struct odp_flow_stats)); /* ODP_FLOW_ATTR_STATS */
925 len += nla_total_size(1); /* ODP_FLOW_ATTR_TCP_FLAGS */
926 len += nla_total_size(8); /* ODP_FLOW_ATTR_USED */
927 return genlmsg_new(NLMSG_ALIGN(sizeof(struct odp_header)) + len, GFP_KERNEL);
930 static struct sk_buff *odp_flow_cmd_build_info(struct sw_flow *flow, struct datapath *dp,
931 u32 pid, u32 seq, u8 cmd)
936 skb = odp_flow_cmd_alloc_info(flow);
938 return ERR_PTR(-ENOMEM);
940 retval = odp_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd);
945 static int odp_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
947 struct nlattr **a = info->attrs;
948 struct odp_header *odp_header = info->userhdr;
949 struct tbl_node *flow_node;
950 struct sw_flow_key key;
951 struct sw_flow *flow;
952 struct sk_buff *reply;
960 if (!a[ODP_FLOW_ATTR_KEY])
962 error = flow_from_nlattrs(&key, a[ODP_FLOW_ATTR_KEY]);
966 /* Validate actions. */
967 if (a[ODP_FLOW_ATTR_ACTIONS]) {
968 error = validate_actions(a[ODP_FLOW_ATTR_ACTIONS]);
971 } else if (info->genlhdr->cmd == ODP_FLOW_CMD_NEW) {
976 dp = get_dp(odp_header->dp_ifindex);
981 hash = flow_hash(&key);
982 table = get_table_protected(dp);
983 flow_node = tbl_lookup(table, &key, hash, flow_cmp);
985 struct sw_flow_actions *acts;
987 /* Bail out if we're not allowed to create a new flow. */
989 if (info->genlhdr->cmd == ODP_FLOW_CMD_SET)
992 /* Expand table, if necessary, to make room. */
993 if (tbl_count(table) >= tbl_n_buckets(table)) {
994 error = expand_table(dp);
997 table = get_table_protected(dp);
1000 /* Allocate flow. */
1001 flow = flow_alloc();
1003 error = PTR_ERR(flow);
1009 /* Obtain actions. */
1010 acts = flow_actions_alloc(a[ODP_FLOW_ATTR_ACTIONS]);
1011 error = PTR_ERR(acts);
1013 goto error_free_flow;
1014 rcu_assign_pointer(flow->sf_acts, acts);
1016 /* Put flow in bucket. */
1017 error = tbl_insert(table, &flow->tbl_node, hash);
1019 goto error_free_flow;
1021 reply = odp_flow_cmd_build_info(flow, dp, info->snd_pid,
1022 info->snd_seq, ODP_FLOW_CMD_NEW);
1024 /* We found a matching flow. */
1025 struct sw_flow_actions *old_acts;
1027 /* Bail out if we're not allowed to modify an existing flow.
1028 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1029 * because Generic Netlink treats the latter as a dump
1030 * request. We also accept NLM_F_EXCL in case that bug ever
1034 if (info->genlhdr->cmd == ODP_FLOW_CMD_NEW &&
1035 info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1038 /* Update actions. */
1039 flow = flow_cast(flow_node);
1040 old_acts = rcu_dereference_protected(flow->sf_acts,
1041 lockdep_genl_is_held());
1042 if (a[ODP_FLOW_ATTR_ACTIONS] &&
1043 (old_acts->actions_len != nla_len(a[ODP_FLOW_ATTR_ACTIONS]) ||
1044 memcmp(old_acts->actions, nla_data(a[ODP_FLOW_ATTR_ACTIONS]),
1045 old_acts->actions_len))) {
1046 struct sw_flow_actions *new_acts;
1048 new_acts = flow_actions_alloc(a[ODP_FLOW_ATTR_ACTIONS]);
1049 error = PTR_ERR(new_acts);
1050 if (IS_ERR(new_acts))
1053 rcu_assign_pointer(flow->sf_acts, new_acts);
1054 flow_deferred_free_acts(old_acts);
1057 reply = odp_flow_cmd_build_info(flow, dp, info->snd_pid,
1058 info->snd_seq, ODP_FLOW_CMD_NEW);
1061 if (a[ODP_FLOW_ATTR_CLEAR]) {
1062 spin_lock_bh(&flow->lock);
1064 spin_unlock_bh(&flow->lock);
1069 genl_notify(reply, genl_info_net(info), info->snd_pid,
1070 dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1072 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1073 dp_flow_multicast_group.id, PTR_ERR(reply));
1082 static int odp_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1084 struct nlattr **a = info->attrs;
1085 struct odp_header *odp_header = info->userhdr;
1086 struct sw_flow_key key;
1087 struct tbl_node *flow_node;
1088 struct sk_buff *reply;
1089 struct sw_flow *flow;
1090 struct datapath *dp;
1094 if (!a[ODP_FLOW_ATTR_KEY])
1096 err = flow_from_nlattrs(&key, a[ODP_FLOW_ATTR_KEY]);
1100 dp = get_dp(odp_header->dp_ifindex);
1104 table = get_table_protected(dp);
1105 flow_node = tbl_lookup(table, &key, flow_hash(&key), flow_cmp);
1109 flow = flow_cast(flow_node);
1110 reply = odp_flow_cmd_build_info(flow, dp, info->snd_pid, info->snd_seq, ODP_FLOW_CMD_NEW);
1112 return PTR_ERR(reply);
1114 return genlmsg_reply(reply, info);
1117 static int odp_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1119 struct nlattr **a = info->attrs;
1120 struct odp_header *odp_header = info->userhdr;
1121 struct sw_flow_key key;
1122 struct tbl_node *flow_node;
1123 struct sk_buff *reply;
1124 struct sw_flow *flow;
1125 struct datapath *dp;
1129 if (!a[ODP_FLOW_ATTR_KEY])
1130 return flush_flows(odp_header->dp_ifindex);
1131 err = flow_from_nlattrs(&key, a[ODP_FLOW_ATTR_KEY]);
1135 dp = get_dp(odp_header->dp_ifindex);
1139 table = get_table_protected(dp);
1140 flow_node = tbl_lookup(table, &key, flow_hash(&key), flow_cmp);
1143 flow = flow_cast(flow_node);
1145 reply = odp_flow_cmd_alloc_info(flow);
1149 err = tbl_remove(table, flow_node);
1155 err = odp_flow_cmd_fill_info(flow, dp, reply, info->snd_pid,
1156 info->snd_seq, 0, ODP_FLOW_CMD_DEL);
1159 flow_deferred_free(flow);
1161 genl_notify(reply, genl_info_net(info), info->snd_pid,
1162 dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1166 static int odp_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1168 struct odp_header *odp_header = genlmsg_data(nlmsg_data(cb->nlh));
1169 struct datapath *dp;
1171 dp = get_dp(odp_header->dp_ifindex);
1176 struct tbl_node *flow_node;
1177 struct sw_flow *flow;
1180 bucket = cb->args[0];
1182 flow_node = tbl_next(get_table_protected(dp), &bucket, &obj);
1186 flow = flow_cast(flow_node);
1187 if (odp_flow_cmd_fill_info(flow, dp, skb, NETLINK_CB(cb->skb).pid,
1188 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1189 ODP_FLOW_CMD_NEW) < 0)
1192 cb->args[0] = bucket;
1198 static struct genl_ops dp_flow_genl_ops[] = {
1199 { .cmd = ODP_FLOW_CMD_NEW,
1200 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1201 .policy = flow_policy,
1202 .doit = odp_flow_cmd_new_or_set
1204 { .cmd = ODP_FLOW_CMD_DEL,
1205 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1206 .policy = flow_policy,
1207 .doit = odp_flow_cmd_del
1209 { .cmd = ODP_FLOW_CMD_GET,
1210 .flags = 0, /* OK for unprivileged users. */
1211 .policy = flow_policy,
1212 .doit = odp_flow_cmd_get,
1213 .dumpit = odp_flow_cmd_dump
1215 { .cmd = ODP_FLOW_CMD_SET,
1216 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1217 .policy = flow_policy,
1218 .doit = odp_flow_cmd_new_or_set,
1222 static const struct nla_policy datapath_policy[ODP_DP_ATTR_MAX + 1] = {
1223 #ifdef HAVE_NLA_NUL_STRING
1224 [ODP_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1226 [ODP_DP_ATTR_IPV4_FRAGS] = { .type = NLA_U32 },
1227 [ODP_DP_ATTR_SAMPLING] = { .type = NLA_U32 },
1230 static struct genl_family dp_datapath_genl_family = {
1231 .id = GENL_ID_GENERATE,
1232 .hdrsize = sizeof(struct odp_header),
1233 .name = ODP_DATAPATH_FAMILY,
1235 .maxattr = ODP_DP_ATTR_MAX
1238 static struct genl_multicast_group dp_datapath_multicast_group = {
1239 .name = ODP_DATAPATH_MCGROUP
1242 static int odp_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1243 u32 pid, u32 seq, u32 flags, u8 cmd)
1245 struct odp_header *odp_header;
1249 odp_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family,
1254 odp_header->dp_ifindex = dp->dp_ifindex;
1257 err = nla_put_string(skb, ODP_DP_ATTR_NAME, dp_name(dp));
1260 goto nla_put_failure;
1262 nla = nla_reserve(skb, ODP_DP_ATTR_STATS, sizeof(struct odp_stats));
1264 goto nla_put_failure;
1265 get_dp_stats(dp, nla_data(nla));
1267 NLA_PUT_U32(skb, ODP_DP_ATTR_IPV4_FRAGS,
1268 dp->drop_frags ? ODP_DP_FRAG_DROP : ODP_DP_FRAG_ZERO);
1270 if (dp->sflow_probability)
1271 NLA_PUT_U32(skb, ODP_DP_ATTR_SAMPLING, dp->sflow_probability);
1273 nla = nla_nest_start(skb, ODP_DP_ATTR_MCGROUPS);
1275 goto nla_put_failure;
1276 NLA_PUT_U32(skb, ODP_PACKET_CMD_MISS, packet_mc_group(dp, ODP_PACKET_CMD_MISS));
1277 NLA_PUT_U32(skb, ODP_PACKET_CMD_ACTION, packet_mc_group(dp, ODP_PACKET_CMD_ACTION));
1278 NLA_PUT_U32(skb, ODP_PACKET_CMD_SAMPLE, packet_mc_group(dp, ODP_PACKET_CMD_SAMPLE));
1279 nla_nest_end(skb, nla);
1281 return genlmsg_end(skb, odp_header);
1284 genlmsg_cancel(skb, odp_header);
1289 static struct sk_buff *odp_dp_cmd_build_info(struct datapath *dp, u32 pid,
1292 struct sk_buff *skb;
1295 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1297 return ERR_PTR(-ENOMEM);
1299 retval = odp_dp_cmd_fill_info(dp, skb, pid, seq, 0, cmd);
1302 return ERR_PTR(retval);
1307 static int odp_dp_cmd_validate(struct nlattr *a[ODP_DP_ATTR_MAX + 1])
1309 if (a[ODP_DP_ATTR_IPV4_FRAGS]) {
1310 u32 frags = nla_get_u32(a[ODP_DP_ATTR_IPV4_FRAGS]);
1312 if (frags != ODP_DP_FRAG_ZERO && frags != ODP_DP_FRAG_DROP)
1316 return CHECK_NUL_STRING(a[ODP_DP_ATTR_NAME], IFNAMSIZ - 1);
1319 /* Called with genl_mutex and optionally with RTNL lock also. */
1320 static struct datapath *lookup_datapath(struct odp_header *odp_header, struct nlattr *a[ODP_DP_ATTR_MAX + 1])
1322 struct datapath *dp;
1324 if (!a[ODP_DP_ATTR_NAME])
1325 dp = get_dp(odp_header->dp_ifindex);
1327 struct vport *vport;
1330 vport = vport_locate(nla_data(a[ODP_DP_ATTR_NAME]));
1331 dp = vport && vport->port_no == ODPP_LOCAL ? vport->dp : NULL;
1334 return dp ? dp : ERR_PTR(-ENODEV);
1337 /* Called with genl_mutex. */
1338 static void change_datapath(struct datapath *dp, struct nlattr *a[ODP_DP_ATTR_MAX + 1])
1340 if (a[ODP_DP_ATTR_IPV4_FRAGS])
1341 dp->drop_frags = nla_get_u32(a[ODP_DP_ATTR_IPV4_FRAGS]) == ODP_DP_FRAG_DROP;
1342 if (a[ODP_DP_ATTR_SAMPLING])
1343 dp->sflow_probability = nla_get_u32(a[ODP_DP_ATTR_SAMPLING]);
1346 static int odp_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1348 struct nlattr **a = info->attrs;
1349 struct vport_parms parms;
1350 struct sk_buff *reply;
1351 struct datapath *dp;
1352 struct vport *vport;
1356 if (!a[ODP_DP_ATTR_NAME])
1359 err = odp_dp_cmd_validate(a);
1365 if (!try_module_get(THIS_MODULE))
1366 goto err_unlock_rtnl;
1369 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1371 goto err_put_module;
1372 INIT_LIST_HEAD(&dp->port_list);
1374 /* Initialize kobject for bridge. This will be added as
1375 * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
1376 dp->ifobj.kset = NULL;
1377 kobject_init(&dp->ifobj, &dp_ktype);
1379 /* Allocate table. */
1381 rcu_assign_pointer(dp->table, tbl_create(TBL_MIN_BUCKETS));
1385 /* Set up our datapath device. */
1386 parms.name = nla_data(a[ODP_DP_ATTR_NAME]);
1387 parms.type = ODP_VPORT_TYPE_INTERNAL;
1388 parms.options = NULL;
1390 parms.port_no = ODPP_LOCAL;
1391 vport = new_vport(&parms);
1392 if (IS_ERR(vport)) {
1393 err = PTR_ERR(vport);
1397 goto err_destroy_table;
1399 dp->dp_ifindex = vport_get_ifindex(vport);
1402 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1403 if (!dp->stats_percpu) {
1405 goto err_destroy_local_port;
1408 change_datapath(dp, a);
1410 reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_NEW);
1411 err = PTR_ERR(reply);
1413 goto err_destroy_local_port;
1415 list_add_tail(&dp->list_node, &dps);
1416 dp_sysfs_add_dp(dp);
1420 genl_notify(reply, genl_info_net(info), info->snd_pid,
1421 dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1424 err_destroy_local_port:
1425 dp_detach_port(get_vport_protected(dp, ODPP_LOCAL));
1427 tbl_destroy(get_table_protected(dp), NULL);
1431 module_put(THIS_MODULE);
1438 static int odp_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1440 struct vport *vport, *next_vport;
1441 struct sk_buff *reply;
1442 struct datapath *dp;
1445 err = odp_dp_cmd_validate(info->attrs);
1450 dp = lookup_datapath(info->userhdr, info->attrs);
1455 reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_DEL);
1456 err = PTR_ERR(reply);
1460 list_for_each_entry_safe (vport, next_vport, &dp->port_list, node)
1461 if (vport->port_no != ODPP_LOCAL)
1462 dp_detach_port(vport);
1464 dp_sysfs_del_dp(dp);
1465 list_del(&dp->list_node);
1466 dp_detach_port(get_vport_protected(dp, ODPP_LOCAL));
1468 /* rtnl_unlock() will wait until all the references to devices that
1469 * are pending unregistration have been dropped. We do it here to
1470 * ensure that any internal devices (which contain DP pointers) are
1471 * fully destroyed before freeing the datapath.
1475 call_rcu(&dp->rcu, destroy_dp_rcu);
1476 module_put(THIS_MODULE);
1478 genl_notify(reply, genl_info_net(info), info->snd_pid,
1479 dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1489 static int odp_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1491 struct sk_buff *reply;
1492 struct datapath *dp;
1495 err = odp_dp_cmd_validate(info->attrs);
1499 dp = lookup_datapath(info->userhdr, info->attrs);
1503 change_datapath(dp, info->attrs);
1505 reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_NEW);
1506 if (IS_ERR(reply)) {
1507 err = PTR_ERR(reply);
1508 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1509 dp_datapath_multicast_group.id, err);
1513 genl_notify(reply, genl_info_net(info), info->snd_pid,
1514 dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1518 static int odp_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1520 struct sk_buff *reply;
1521 struct datapath *dp;
1524 err = odp_dp_cmd_validate(info->attrs);
1528 dp = lookup_datapath(info->userhdr, info->attrs);
1532 reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_NEW);
1534 return PTR_ERR(reply);
1536 return genlmsg_reply(reply, info);
1539 static int odp_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1541 struct datapath *dp;
1542 int skip = cb->args[0];
1545 list_for_each_entry (dp, &dps, list_node) {
1548 if (odp_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).pid,
1549 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1550 ODP_DP_CMD_NEW) < 0)
1560 static struct genl_ops dp_datapath_genl_ops[] = {
1561 { .cmd = ODP_DP_CMD_NEW,
1562 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1563 .policy = datapath_policy,
1564 .doit = odp_dp_cmd_new
1566 { .cmd = ODP_DP_CMD_DEL,
1567 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1568 .policy = datapath_policy,
1569 .doit = odp_dp_cmd_del
1571 { .cmd = ODP_DP_CMD_GET,
1572 .flags = 0, /* OK for unprivileged users. */
1573 .policy = datapath_policy,
1574 .doit = odp_dp_cmd_get,
1575 .dumpit = odp_dp_cmd_dump
1577 { .cmd = ODP_DP_CMD_SET,
1578 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1579 .policy = datapath_policy,
1580 .doit = odp_dp_cmd_set,
1584 static const struct nla_policy vport_policy[ODP_VPORT_ATTR_MAX + 1] = {
1585 #ifdef HAVE_NLA_NUL_STRING
1586 [ODP_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1587 [ODP_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1588 [ODP_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1589 [ODP_VPORT_ATTR_STATS] = { .len = sizeof(struct rtnl_link_stats64) },
1590 [ODP_VPORT_ATTR_ADDRESS] = { .len = ETH_ALEN },
1592 [ODP_VPORT_ATTR_STATS] = { .minlen = sizeof(struct rtnl_link_stats64) },
1593 [ODP_VPORT_ATTR_ADDRESS] = { .minlen = ETH_ALEN },
1595 [ODP_VPORT_ATTR_MTU] = { .type = NLA_U32 },
1596 [ODP_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1599 static struct genl_family dp_vport_genl_family = {
1600 .id = GENL_ID_GENERATE,
1601 .hdrsize = sizeof(struct odp_header),
1602 .name = ODP_VPORT_FAMILY,
1604 .maxattr = ODP_VPORT_ATTR_MAX
1607 static struct genl_multicast_group dp_vport_multicast_group = {
1608 .name = ODP_VPORT_MCGROUP
1611 /* Called with RTNL lock or RCU read lock. */
1612 static int odp_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1613 u32 pid, u32 seq, u32 flags, u8 cmd)
1615 struct odp_header *odp_header;
1617 int ifindex, iflink;
1621 odp_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family,
1626 odp_header->dp_ifindex = vport->dp->dp_ifindex;
1628 NLA_PUT_U32(skb, ODP_VPORT_ATTR_PORT_NO, vport->port_no);
1629 NLA_PUT_U32(skb, ODP_VPORT_ATTR_TYPE, vport_get_type(vport));
1630 NLA_PUT_STRING(skb, ODP_VPORT_ATTR_NAME, vport_get_name(vport));
1632 nla = nla_reserve(skb, ODP_VPORT_ATTR_STATS, sizeof(struct rtnl_link_stats64));
1634 goto nla_put_failure;
1635 if (vport_get_stats(vport, nla_data(nla)))
1636 __skb_trim(skb, skb->len - nla->nla_len);
1638 NLA_PUT(skb, ODP_VPORT_ATTR_ADDRESS, ETH_ALEN, vport_get_addr(vport));
1640 mtu = vport_get_mtu(vport);
1642 NLA_PUT_U32(skb, ODP_VPORT_ATTR_MTU, mtu);
1644 err = vport_get_options(vport, skb);
1645 if (err == -EMSGSIZE)
1648 ifindex = vport_get_ifindex(vport);
1650 NLA_PUT_U32(skb, ODP_VPORT_ATTR_IFINDEX, ifindex);
1652 iflink = vport_get_iflink(vport);
1654 NLA_PUT_U32(skb, ODP_VPORT_ATTR_IFLINK, iflink);
1656 return genlmsg_end(skb, odp_header);
1661 genlmsg_cancel(skb, odp_header);
1665 /* Called with RTNL lock or RCU read lock. */
1666 static struct sk_buff *odp_vport_cmd_build_info(struct vport *vport, u32 pid,
1669 struct sk_buff *skb;
1672 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1674 return ERR_PTR(-ENOMEM);
1676 retval = odp_vport_cmd_fill_info(vport, skb, pid, seq, 0, cmd);
1679 return ERR_PTR(retval);
1684 static int odp_vport_cmd_validate(struct nlattr *a[ODP_VPORT_ATTR_MAX + 1])
1686 return CHECK_NUL_STRING(a[ODP_VPORT_ATTR_NAME], IFNAMSIZ - 1);
1689 /* Called with RTNL lock or RCU read lock. */
1690 static struct vport *lookup_vport(struct odp_header *odp_header,
1691 struct nlattr *a[ODP_VPORT_ATTR_MAX + 1])
1693 struct datapath *dp;
1694 struct vport *vport;
1696 if (a[ODP_VPORT_ATTR_NAME]) {
1697 vport = vport_locate(nla_data(a[ODP_VPORT_ATTR_NAME]));
1699 return ERR_PTR(-ENODEV);
1701 } else if (a[ODP_VPORT_ATTR_PORT_NO]) {
1702 u32 port_no = nla_get_u32(a[ODP_VPORT_ATTR_PORT_NO]);
1704 if (port_no >= DP_MAX_PORTS)
1705 return ERR_PTR(-EFBIG);
1707 dp = get_dp(odp_header->dp_ifindex);
1709 return ERR_PTR(-ENODEV);
1711 vport = get_vport_protected(dp, port_no);
1713 return ERR_PTR(-ENOENT);
1716 return ERR_PTR(-EINVAL);
1719 /* Called with RTNL lock. */
1720 static int change_vport(struct vport *vport, struct nlattr *a[ODP_VPORT_ATTR_MAX + 1])
1723 if (a[ODP_VPORT_ATTR_STATS])
1724 err = vport_set_stats(vport, nla_data(a[ODP_VPORT_ATTR_STATS]));
1725 if (!err && a[ODP_VPORT_ATTR_ADDRESS])
1726 err = vport_set_addr(vport, nla_data(a[ODP_VPORT_ATTR_ADDRESS]));
1727 if (!err && a[ODP_VPORT_ATTR_MTU])
1728 err = vport_set_mtu(vport, nla_get_u32(a[ODP_VPORT_ATTR_MTU]));
1732 static int odp_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1734 struct nlattr **a = info->attrs;
1735 struct odp_header *odp_header = info->userhdr;
1736 struct vport_parms parms;
1737 struct sk_buff *reply;
1738 struct vport *vport;
1739 struct datapath *dp;
1744 if (!a[ODP_VPORT_ATTR_NAME] || !a[ODP_VPORT_ATTR_TYPE])
1747 err = odp_vport_cmd_validate(a);
1752 dp = get_dp(odp_header->dp_ifindex);
1757 if (a[ODP_VPORT_ATTR_PORT_NO]) {
1758 port_no = nla_get_u32(a[ODP_VPORT_ATTR_PORT_NO]);
1761 if (port_no >= DP_MAX_PORTS)
1764 vport = get_vport_protected(dp, port_no);
1769 for (port_no = 1; ; port_no++) {
1770 if (port_no >= DP_MAX_PORTS) {
1774 vport = get_vport_protected(dp, port_no);
1780 parms.name = nla_data(a[ODP_VPORT_ATTR_NAME]);
1781 parms.type = nla_get_u32(a[ODP_VPORT_ATTR_TYPE]);
1782 parms.options = a[ODP_VPORT_ATTR_OPTIONS];
1784 parms.port_no = port_no;
1786 vport = new_vport(&parms);
1787 err = PTR_ERR(vport);
1791 set_internal_devs_mtu(dp);
1792 dp_sysfs_add_if(vport);
1794 err = change_vport(vport, a);
1796 reply = odp_vport_cmd_build_info(vport, info->snd_pid,
1797 info->snd_seq, ODP_VPORT_CMD_NEW);
1799 err = PTR_ERR(reply);
1802 dp_detach_port(vport);
1805 genl_notify(reply, genl_info_net(info), info->snd_pid,
1806 dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1815 static int odp_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1817 struct nlattr **a = info->attrs;
1818 struct sk_buff *reply;
1819 struct vport *vport;
1822 err = odp_vport_cmd_validate(a);
1827 vport = lookup_vport(info->userhdr, a);
1828 err = PTR_ERR(vport);
1833 if (a[ODP_VPORT_ATTR_OPTIONS])
1834 err = vport_set_options(vport, a[ODP_VPORT_ATTR_OPTIONS]);
1836 err = change_vport(vport, a);
1838 reply = odp_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1840 if (IS_ERR(reply)) {
1841 err = PTR_ERR(reply);
1842 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1843 dp_vport_multicast_group.id, err);
1847 genl_notify(reply, genl_info_net(info), info->snd_pid,
1848 dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1856 static int odp_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1858 struct nlattr **a = info->attrs;
1859 struct sk_buff *reply;
1860 struct vport *vport;
1863 err = odp_vport_cmd_validate(a);
1868 vport = lookup_vport(info->userhdr, a);
1869 err = PTR_ERR(vport);
1873 if (vport->port_no == ODPP_LOCAL) {
1878 reply = odp_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1880 err = PTR_ERR(reply);
1884 err = dp_detach_port(vport);
1886 genl_notify(reply, genl_info_net(info), info->snd_pid,
1887 dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1895 static int odp_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1897 struct nlattr **a = info->attrs;
1898 struct odp_header *odp_header = info->userhdr;
1899 struct sk_buff *reply;
1900 struct vport *vport;
1903 err = odp_vport_cmd_validate(a);
1908 vport = lookup_vport(odp_header, a);
1909 err = PTR_ERR(vport);
1913 reply = odp_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1915 err = PTR_ERR(reply);
1919 err = genlmsg_reply(reply, info);
1927 static int odp_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1929 struct odp_header *odp_header = genlmsg_data(nlmsg_data(cb->nlh));
1930 struct datapath *dp;
1934 dp = get_dp(odp_header->dp_ifindex);
1939 for (port_no = cb->args[0]; port_no < DP_MAX_PORTS; port_no++) {
1940 struct vport *vport;
1942 vport = get_vport_protected(dp, port_no);
1946 if (odp_vport_cmd_fill_info(vport, skb, NETLINK_CB(cb->skb).pid,
1947 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1948 ODP_VPORT_CMD_NEW) < 0)
1953 cb->args[0] = port_no;
1959 static struct genl_ops dp_vport_genl_ops[] = {
1960 { .cmd = ODP_VPORT_CMD_NEW,
1961 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1962 .policy = vport_policy,
1963 .doit = odp_vport_cmd_new
1965 { .cmd = ODP_VPORT_CMD_DEL,
1966 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1967 .policy = vport_policy,
1968 .doit = odp_vport_cmd_del
1970 { .cmd = ODP_VPORT_CMD_GET,
1971 .flags = 0, /* OK for unprivileged users. */
1972 .policy = vport_policy,
1973 .doit = odp_vport_cmd_get,
1974 .dumpit = odp_vport_cmd_dump
1976 { .cmd = ODP_VPORT_CMD_SET,
1977 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1978 .policy = vport_policy,
1979 .doit = odp_vport_cmd_set,
1983 struct genl_family_and_ops {
1984 struct genl_family *family;
1985 struct genl_ops *ops;
1987 struct genl_multicast_group *group;
1990 static const struct genl_family_and_ops dp_genl_families[] = {
1991 { &dp_datapath_genl_family,
1992 dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1993 &dp_datapath_multicast_group },
1994 { &dp_vport_genl_family,
1995 dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1996 &dp_vport_multicast_group },
1997 { &dp_flow_genl_family,
1998 dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1999 &dp_flow_multicast_group },
2000 { &dp_packet_genl_family,
2001 dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
2005 static void dp_unregister_genl(int n_families)
2009 for (i = 0; i < n_families; i++) {
2010 genl_unregister_family(dp_genl_families[i].family);
2014 static int dp_register_genl(void)
2021 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2022 const struct genl_family_and_ops *f = &dp_genl_families[i];
2024 err = genl_register_family_with_ops(f->family, f->ops,
2031 err = genl_register_mc_group(f->family, f->group);
2037 err = packet_register_mc_groups();
2043 dp_unregister_genl(n_registered);
2047 static int __init dp_init(void)
2049 struct sk_buff *dummy_skb;
2052 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
2054 printk("Open vSwitch %s, built "__DATE__" "__TIME__"\n", VERSION BUILDNR);
2062 goto error_flow_exit;
2064 err = register_netdevice_notifier(&dp_device_notifier);
2066 goto error_vport_exit;
2068 err = dp_register_genl();
2070 goto error_unreg_notifier;
2074 error_unreg_notifier:
2075 unregister_netdevice_notifier(&dp_device_notifier);
2084 static void dp_cleanup(void)
2087 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2088 unregister_netdevice_notifier(&dp_device_notifier);
2093 module_init(dp_init);
2094 module_exit(dp_cleanup);
2096 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2097 MODULE_LICENSE("GPL");