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 sw_flow_actions *acts;
681 struct sk_buff *packet;
682 struct sw_flow *flow;
690 if (!a[ODP_PACKET_ATTR_PACKET] || !a[ODP_PACKET_ATTR_ACTIONS] ||
691 nla_len(a[ODP_PACKET_ATTR_PACKET]) < ETH_HLEN)
694 err = validate_actions(a[ODP_PACKET_ATTR_ACTIONS]);
698 len = nla_len(a[ODP_PACKET_ATTR_PACKET]);
699 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
703 skb_reserve(packet, NET_IP_ALIGN);
705 memcpy(__skb_put(packet, len), nla_data(a[ODP_PACKET_ATTR_PACKET]), len);
707 skb_reset_mac_header(packet);
708 eth = eth_hdr(packet);
710 /* Normally, setting the skb 'protocol' field would be handled by a
711 * call to eth_type_trans(), but it assumes there's a sending
712 * device, which we may not have. */
713 if (ntohs(eth->h_proto) >= 1536)
714 packet->protocol = eth->h_proto;
716 packet->protocol = htons(ETH_P_802_2);
718 /* Build an sw_flow for sending this packet. */
724 err = flow_extract(packet, -1, &flow->key, &is_frag);
727 flow->tbl_node.hash = flow_hash(&flow->key);
729 acts = flow_actions_alloc(a[ODP_PACKET_ATTR_ACTIONS]);
733 rcu_assign_pointer(flow->sf_acts, acts);
735 OVS_CB(packet)->flow = flow;
738 dp = get_dp(odp_header->dp_ifindex);
742 err = execute_actions(dp, packet, &flow->key,
743 nla_data(a[ODP_PACKET_ATTR_ACTIONS]),
744 nla_len(a[ODP_PACKET_ATTR_ACTIONS]));
760 static const struct nla_policy packet_policy[ODP_PACKET_ATTR_MAX + 1] = {
761 [ODP_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
762 [ODP_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
765 static struct genl_ops dp_packet_genl_ops[] = {
766 { .cmd = ODP_PACKET_CMD_EXECUTE,
767 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
768 .policy = packet_policy,
769 .doit = odp_packet_cmd_execute
773 static void get_dp_stats(struct datapath *dp, struct odp_stats *stats)
777 stats->n_frags = stats->n_hit = stats->n_missed = stats->n_lost = 0;
778 for_each_possible_cpu(i) {
779 const struct dp_stats_percpu *percpu_stats;
780 struct dp_stats_percpu local_stats;
783 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
786 seqcount = read_seqcount_begin(&percpu_stats->seqlock);
787 local_stats = *percpu_stats;
788 } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount));
790 stats->n_frags += local_stats.n_frags;
791 stats->n_hit += local_stats.n_hit;
792 stats->n_missed += local_stats.n_missed;
793 stats->n_lost += local_stats.n_lost;
797 /* MTU of the dp pseudo-device: ETH_DATA_LEN or the minimum of the ports.
798 * Called with RTNL lock.
800 int dp_min_mtu(const struct datapath *dp)
807 list_for_each_entry (p, &dp->port_list, node) {
810 /* Skip any internal ports, since that's what we're trying to
812 if (is_internal_vport(p))
815 dev_mtu = vport_get_mtu(p);
818 if (!mtu || dev_mtu < mtu)
822 return mtu ? mtu : ETH_DATA_LEN;
825 /* Sets the MTU of all datapath devices to the minimum of the ports
826 * Called with RTNL lock.
828 void set_internal_devs_mtu(const struct datapath *dp)
835 mtu = dp_min_mtu(dp);
837 list_for_each_entry (p, &dp->port_list, node) {
838 if (is_internal_vport(p))
839 vport_set_mtu(p, mtu);
843 static const struct nla_policy flow_policy[ODP_FLOW_ATTR_MAX + 1] = {
844 [ODP_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
845 [ODP_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
846 [ODP_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
849 static struct genl_family dp_flow_genl_family = {
850 .id = GENL_ID_GENERATE,
851 .hdrsize = sizeof(struct odp_header),
852 .name = ODP_FLOW_FAMILY,
854 .maxattr = ODP_FLOW_ATTR_MAX
857 static struct genl_multicast_group dp_flow_multicast_group = {
858 .name = ODP_FLOW_MCGROUP
861 /* Called with genl_lock. */
862 static int odp_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
863 struct sk_buff *skb, u32 pid, u32 seq, u32 flags, u8 cmd)
865 const int skb_orig_len = skb->len;
866 const struct sw_flow_actions *sf_acts;
867 struct odp_flow_stats stats;
868 struct odp_header *odp_header;
874 sf_acts = rcu_dereference_protected(flow->sf_acts,
875 lockdep_genl_is_held());
877 odp_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags, cmd);
881 odp_header->dp_ifindex = dp->dp_ifindex;
883 nla = nla_nest_start(skb, ODP_FLOW_ATTR_KEY);
885 goto nla_put_failure;
886 err = flow_to_nlattrs(&flow->key, skb);
889 nla_nest_end(skb, nla);
891 spin_lock_bh(&flow->lock);
893 stats.n_packets = flow->packet_count;
894 stats.n_bytes = flow->byte_count;
895 tcp_flags = flow->tcp_flags;
896 spin_unlock_bh(&flow->lock);
899 NLA_PUT_U64(skb, ODP_FLOW_ATTR_USED, flow_used_time(used));
902 NLA_PUT(skb, ODP_FLOW_ATTR_STATS, sizeof(struct odp_flow_stats), &stats);
905 NLA_PUT_U8(skb, ODP_FLOW_ATTR_TCP_FLAGS, tcp_flags);
907 /* If ODP_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
908 * this is the first flow to be dumped into 'skb'. This is unusual for
909 * Netlink but individual action lists can be longer than
910 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
911 * The userspace caller can always fetch the actions separately if it
912 * really wants them. (Most userspace callers in fact don't care.)
914 * This can only fail for dump operations because the skb is always
915 * properly sized for single flows.
917 err = nla_put(skb, ODP_FLOW_ATTR_ACTIONS, sf_acts->actions_len,
919 if (err < 0 && skb_orig_len)
922 return genlmsg_end(skb, odp_header);
927 genlmsg_cancel(skb, odp_header);
931 static struct sk_buff *odp_flow_cmd_alloc_info(struct sw_flow *flow)
933 const struct sw_flow_actions *sf_acts;
936 sf_acts = rcu_dereference_protected(flow->sf_acts,
937 lockdep_genl_is_held());
939 len = nla_total_size(FLOW_BUFSIZE); /* ODP_FLOW_ATTR_KEY */
940 len += nla_total_size(sf_acts->actions_len); /* ODP_FLOW_ATTR_ACTIONS */
941 len += nla_total_size(sizeof(struct odp_flow_stats)); /* ODP_FLOW_ATTR_STATS */
942 len += nla_total_size(1); /* ODP_FLOW_ATTR_TCP_FLAGS */
943 len += nla_total_size(8); /* ODP_FLOW_ATTR_USED */
944 return genlmsg_new(NLMSG_ALIGN(sizeof(struct odp_header)) + len, GFP_KERNEL);
947 static struct sk_buff *odp_flow_cmd_build_info(struct sw_flow *flow, struct datapath *dp,
948 u32 pid, u32 seq, u8 cmd)
953 skb = odp_flow_cmd_alloc_info(flow);
955 return ERR_PTR(-ENOMEM);
957 retval = odp_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd);
962 static int odp_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
964 struct nlattr **a = info->attrs;
965 struct odp_header *odp_header = info->userhdr;
966 struct tbl_node *flow_node;
967 struct sw_flow_key key;
968 struct sw_flow *flow;
969 struct sk_buff *reply;
977 if (!a[ODP_FLOW_ATTR_KEY])
979 error = flow_from_nlattrs(&key, a[ODP_FLOW_ATTR_KEY]);
983 /* Validate actions. */
984 if (a[ODP_FLOW_ATTR_ACTIONS]) {
985 error = validate_actions(a[ODP_FLOW_ATTR_ACTIONS]);
988 } else if (info->genlhdr->cmd == ODP_FLOW_CMD_NEW) {
993 dp = get_dp(odp_header->dp_ifindex);
998 hash = flow_hash(&key);
999 table = get_table_protected(dp);
1000 flow_node = tbl_lookup(table, &key, hash, flow_cmp);
1002 struct sw_flow_actions *acts;
1004 /* Bail out if we're not allowed to create a new flow. */
1006 if (info->genlhdr->cmd == ODP_FLOW_CMD_SET)
1009 /* Expand table, if necessary, to make room. */
1010 if (tbl_count(table) >= tbl_n_buckets(table)) {
1011 error = expand_table(dp);
1014 table = get_table_protected(dp);
1017 /* Allocate flow. */
1018 flow = flow_alloc();
1020 error = PTR_ERR(flow);
1026 /* Obtain actions. */
1027 acts = flow_actions_alloc(a[ODP_FLOW_ATTR_ACTIONS]);
1028 error = PTR_ERR(acts);
1030 goto error_free_flow;
1031 rcu_assign_pointer(flow->sf_acts, acts);
1033 /* Put flow in bucket. */
1034 error = tbl_insert(table, &flow->tbl_node, hash);
1036 goto error_free_flow;
1038 reply = odp_flow_cmd_build_info(flow, dp, info->snd_pid,
1039 info->snd_seq, ODP_FLOW_CMD_NEW);
1041 /* We found a matching flow. */
1042 struct sw_flow_actions *old_acts;
1044 /* Bail out if we're not allowed to modify an existing flow.
1045 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1046 * because Generic Netlink treats the latter as a dump
1047 * request. We also accept NLM_F_EXCL in case that bug ever
1051 if (info->genlhdr->cmd == ODP_FLOW_CMD_NEW &&
1052 info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1055 /* Update actions. */
1056 flow = flow_cast(flow_node);
1057 old_acts = rcu_dereference_protected(flow->sf_acts,
1058 lockdep_genl_is_held());
1059 if (a[ODP_FLOW_ATTR_ACTIONS] &&
1060 (old_acts->actions_len != nla_len(a[ODP_FLOW_ATTR_ACTIONS]) ||
1061 memcmp(old_acts->actions, nla_data(a[ODP_FLOW_ATTR_ACTIONS]),
1062 old_acts->actions_len))) {
1063 struct sw_flow_actions *new_acts;
1065 new_acts = flow_actions_alloc(a[ODP_FLOW_ATTR_ACTIONS]);
1066 error = PTR_ERR(new_acts);
1067 if (IS_ERR(new_acts))
1070 rcu_assign_pointer(flow->sf_acts, new_acts);
1071 flow_deferred_free_acts(old_acts);
1074 reply = odp_flow_cmd_build_info(flow, dp, info->snd_pid,
1075 info->snd_seq, ODP_FLOW_CMD_NEW);
1078 if (a[ODP_FLOW_ATTR_CLEAR]) {
1079 spin_lock_bh(&flow->lock);
1081 spin_unlock_bh(&flow->lock);
1086 genl_notify(reply, genl_info_net(info), info->snd_pid,
1087 dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1089 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1090 dp_flow_multicast_group.id, PTR_ERR(reply));
1099 static int odp_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1101 struct nlattr **a = info->attrs;
1102 struct odp_header *odp_header = info->userhdr;
1103 struct sw_flow_key key;
1104 struct tbl_node *flow_node;
1105 struct sk_buff *reply;
1106 struct sw_flow *flow;
1107 struct datapath *dp;
1111 if (!a[ODP_FLOW_ATTR_KEY])
1113 err = flow_from_nlattrs(&key, a[ODP_FLOW_ATTR_KEY]);
1117 dp = get_dp(odp_header->dp_ifindex);
1121 table = get_table_protected(dp);
1122 flow_node = tbl_lookup(table, &key, flow_hash(&key), flow_cmp);
1126 flow = flow_cast(flow_node);
1127 reply = odp_flow_cmd_build_info(flow, dp, info->snd_pid, info->snd_seq, ODP_FLOW_CMD_NEW);
1129 return PTR_ERR(reply);
1131 return genlmsg_reply(reply, info);
1134 static int odp_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1136 struct nlattr **a = info->attrs;
1137 struct odp_header *odp_header = info->userhdr;
1138 struct sw_flow_key key;
1139 struct tbl_node *flow_node;
1140 struct sk_buff *reply;
1141 struct sw_flow *flow;
1142 struct datapath *dp;
1146 if (!a[ODP_FLOW_ATTR_KEY])
1147 return flush_flows(odp_header->dp_ifindex);
1148 err = flow_from_nlattrs(&key, a[ODP_FLOW_ATTR_KEY]);
1152 dp = get_dp(odp_header->dp_ifindex);
1156 table = get_table_protected(dp);
1157 flow_node = tbl_lookup(table, &key, flow_hash(&key), flow_cmp);
1160 flow = flow_cast(flow_node);
1162 reply = odp_flow_cmd_alloc_info(flow);
1166 err = tbl_remove(table, flow_node);
1172 err = odp_flow_cmd_fill_info(flow, dp, reply, info->snd_pid,
1173 info->snd_seq, 0, ODP_FLOW_CMD_DEL);
1176 flow_deferred_free(flow);
1178 genl_notify(reply, genl_info_net(info), info->snd_pid,
1179 dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1183 static int odp_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1185 struct odp_header *odp_header = genlmsg_data(nlmsg_data(cb->nlh));
1186 struct datapath *dp;
1188 dp = get_dp(odp_header->dp_ifindex);
1193 struct tbl_node *flow_node;
1194 struct sw_flow *flow;
1197 bucket = cb->args[0];
1199 flow_node = tbl_next(get_table_protected(dp), &bucket, &obj);
1203 flow = flow_cast(flow_node);
1204 if (odp_flow_cmd_fill_info(flow, dp, skb, NETLINK_CB(cb->skb).pid,
1205 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1206 ODP_FLOW_CMD_NEW) < 0)
1209 cb->args[0] = bucket;
1215 static struct genl_ops dp_flow_genl_ops[] = {
1216 { .cmd = ODP_FLOW_CMD_NEW,
1217 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1218 .policy = flow_policy,
1219 .doit = odp_flow_cmd_new_or_set
1221 { .cmd = ODP_FLOW_CMD_DEL,
1222 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1223 .policy = flow_policy,
1224 .doit = odp_flow_cmd_del
1226 { .cmd = ODP_FLOW_CMD_GET,
1227 .flags = 0, /* OK for unprivileged users. */
1228 .policy = flow_policy,
1229 .doit = odp_flow_cmd_get,
1230 .dumpit = odp_flow_cmd_dump
1232 { .cmd = ODP_FLOW_CMD_SET,
1233 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1234 .policy = flow_policy,
1235 .doit = odp_flow_cmd_new_or_set,
1239 static const struct nla_policy datapath_policy[ODP_DP_ATTR_MAX + 1] = {
1240 #ifdef HAVE_NLA_NUL_STRING
1241 [ODP_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1243 [ODP_DP_ATTR_IPV4_FRAGS] = { .type = NLA_U32 },
1244 [ODP_DP_ATTR_SAMPLING] = { .type = NLA_U32 },
1247 static struct genl_family dp_datapath_genl_family = {
1248 .id = GENL_ID_GENERATE,
1249 .hdrsize = sizeof(struct odp_header),
1250 .name = ODP_DATAPATH_FAMILY,
1252 .maxattr = ODP_DP_ATTR_MAX
1255 static struct genl_multicast_group dp_datapath_multicast_group = {
1256 .name = ODP_DATAPATH_MCGROUP
1259 static int odp_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1260 u32 pid, u32 seq, u32 flags, u8 cmd)
1262 struct odp_header *odp_header;
1266 odp_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family,
1271 odp_header->dp_ifindex = dp->dp_ifindex;
1274 err = nla_put_string(skb, ODP_DP_ATTR_NAME, dp_name(dp));
1277 goto nla_put_failure;
1279 nla = nla_reserve(skb, ODP_DP_ATTR_STATS, sizeof(struct odp_stats));
1281 goto nla_put_failure;
1282 get_dp_stats(dp, nla_data(nla));
1284 NLA_PUT_U32(skb, ODP_DP_ATTR_IPV4_FRAGS,
1285 dp->drop_frags ? ODP_DP_FRAG_DROP : ODP_DP_FRAG_ZERO);
1287 if (dp->sflow_probability)
1288 NLA_PUT_U32(skb, ODP_DP_ATTR_SAMPLING, dp->sflow_probability);
1290 nla = nla_nest_start(skb, ODP_DP_ATTR_MCGROUPS);
1292 goto nla_put_failure;
1293 NLA_PUT_U32(skb, ODP_PACKET_CMD_MISS, packet_mc_group(dp, ODP_PACKET_CMD_MISS));
1294 NLA_PUT_U32(skb, ODP_PACKET_CMD_ACTION, packet_mc_group(dp, ODP_PACKET_CMD_ACTION));
1295 NLA_PUT_U32(skb, ODP_PACKET_CMD_SAMPLE, packet_mc_group(dp, ODP_PACKET_CMD_SAMPLE));
1296 nla_nest_end(skb, nla);
1298 return genlmsg_end(skb, odp_header);
1301 genlmsg_cancel(skb, odp_header);
1306 static struct sk_buff *odp_dp_cmd_build_info(struct datapath *dp, u32 pid,
1309 struct sk_buff *skb;
1312 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1314 return ERR_PTR(-ENOMEM);
1316 retval = odp_dp_cmd_fill_info(dp, skb, pid, seq, 0, cmd);
1319 return ERR_PTR(retval);
1324 static int odp_dp_cmd_validate(struct nlattr *a[ODP_DP_ATTR_MAX + 1])
1326 if (a[ODP_DP_ATTR_IPV4_FRAGS]) {
1327 u32 frags = nla_get_u32(a[ODP_DP_ATTR_IPV4_FRAGS]);
1329 if (frags != ODP_DP_FRAG_ZERO && frags != ODP_DP_FRAG_DROP)
1333 return CHECK_NUL_STRING(a[ODP_DP_ATTR_NAME], IFNAMSIZ - 1);
1336 /* Called with genl_mutex and optionally with RTNL lock also. */
1337 static struct datapath *lookup_datapath(struct odp_header *odp_header, struct nlattr *a[ODP_DP_ATTR_MAX + 1])
1339 struct datapath *dp;
1341 if (!a[ODP_DP_ATTR_NAME])
1342 dp = get_dp(odp_header->dp_ifindex);
1344 struct vport *vport;
1347 vport = vport_locate(nla_data(a[ODP_DP_ATTR_NAME]));
1348 dp = vport && vport->port_no == ODPP_LOCAL ? vport->dp : NULL;
1351 return dp ? dp : ERR_PTR(-ENODEV);
1354 /* Called with genl_mutex. */
1355 static void change_datapath(struct datapath *dp, struct nlattr *a[ODP_DP_ATTR_MAX + 1])
1357 if (a[ODP_DP_ATTR_IPV4_FRAGS])
1358 dp->drop_frags = nla_get_u32(a[ODP_DP_ATTR_IPV4_FRAGS]) == ODP_DP_FRAG_DROP;
1359 if (a[ODP_DP_ATTR_SAMPLING])
1360 dp->sflow_probability = nla_get_u32(a[ODP_DP_ATTR_SAMPLING]);
1363 static int odp_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1365 struct nlattr **a = info->attrs;
1366 struct vport_parms parms;
1367 struct sk_buff *reply;
1368 struct datapath *dp;
1369 struct vport *vport;
1373 if (!a[ODP_DP_ATTR_NAME])
1376 err = odp_dp_cmd_validate(a);
1382 if (!try_module_get(THIS_MODULE))
1383 goto err_unlock_rtnl;
1386 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1388 goto err_put_module;
1389 INIT_LIST_HEAD(&dp->port_list);
1391 /* Initialize kobject for bridge. This will be added as
1392 * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
1393 dp->ifobj.kset = NULL;
1394 kobject_init(&dp->ifobj, &dp_ktype);
1396 /* Allocate table. */
1398 rcu_assign_pointer(dp->table, tbl_create(TBL_MIN_BUCKETS));
1402 /* Set up our datapath device. */
1403 parms.name = nla_data(a[ODP_DP_ATTR_NAME]);
1404 parms.type = ODP_VPORT_TYPE_INTERNAL;
1405 parms.options = NULL;
1407 parms.port_no = ODPP_LOCAL;
1408 vport = new_vport(&parms);
1409 if (IS_ERR(vport)) {
1410 err = PTR_ERR(vport);
1414 goto err_destroy_table;
1416 dp->dp_ifindex = vport_get_ifindex(vport);
1419 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1420 if (!dp->stats_percpu) {
1422 goto err_destroy_local_port;
1425 change_datapath(dp, a);
1427 reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_NEW);
1428 err = PTR_ERR(reply);
1430 goto err_destroy_local_port;
1432 list_add_tail(&dp->list_node, &dps);
1433 dp_sysfs_add_dp(dp);
1437 genl_notify(reply, genl_info_net(info), info->snd_pid,
1438 dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1441 err_destroy_local_port:
1442 dp_detach_port(get_vport_protected(dp, ODPP_LOCAL));
1444 tbl_destroy(get_table_protected(dp), NULL);
1448 module_put(THIS_MODULE);
1455 static int odp_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1457 struct vport *vport, *next_vport;
1458 struct sk_buff *reply;
1459 struct datapath *dp;
1462 err = odp_dp_cmd_validate(info->attrs);
1467 dp = lookup_datapath(info->userhdr, info->attrs);
1472 reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_DEL);
1473 err = PTR_ERR(reply);
1477 list_for_each_entry_safe (vport, next_vport, &dp->port_list, node)
1478 if (vport->port_no != ODPP_LOCAL)
1479 dp_detach_port(vport);
1481 dp_sysfs_del_dp(dp);
1482 list_del(&dp->list_node);
1483 dp_detach_port(get_vport_protected(dp, ODPP_LOCAL));
1485 /* rtnl_unlock() will wait until all the references to devices that
1486 * are pending unregistration have been dropped. We do it here to
1487 * ensure that any internal devices (which contain DP pointers) are
1488 * fully destroyed before freeing the datapath.
1492 call_rcu(&dp->rcu, destroy_dp_rcu);
1493 module_put(THIS_MODULE);
1495 genl_notify(reply, genl_info_net(info), info->snd_pid,
1496 dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1506 static int odp_dp_cmd_set(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 change_datapath(dp, info->attrs);
1522 reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_NEW);
1523 if (IS_ERR(reply)) {
1524 err = PTR_ERR(reply);
1525 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1526 dp_datapath_multicast_group.id, err);
1530 genl_notify(reply, genl_info_net(info), info->snd_pid,
1531 dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1535 static int odp_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1537 struct sk_buff *reply;
1538 struct datapath *dp;
1541 err = odp_dp_cmd_validate(info->attrs);
1545 dp = lookup_datapath(info->userhdr, info->attrs);
1549 reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_NEW);
1551 return PTR_ERR(reply);
1553 return genlmsg_reply(reply, info);
1556 static int odp_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1558 struct datapath *dp;
1559 int skip = cb->args[0];
1562 list_for_each_entry (dp, &dps, list_node) {
1565 if (odp_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).pid,
1566 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1567 ODP_DP_CMD_NEW) < 0)
1577 static struct genl_ops dp_datapath_genl_ops[] = {
1578 { .cmd = ODP_DP_CMD_NEW,
1579 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1580 .policy = datapath_policy,
1581 .doit = odp_dp_cmd_new
1583 { .cmd = ODP_DP_CMD_DEL,
1584 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1585 .policy = datapath_policy,
1586 .doit = odp_dp_cmd_del
1588 { .cmd = ODP_DP_CMD_GET,
1589 .flags = 0, /* OK for unprivileged users. */
1590 .policy = datapath_policy,
1591 .doit = odp_dp_cmd_get,
1592 .dumpit = odp_dp_cmd_dump
1594 { .cmd = ODP_DP_CMD_SET,
1595 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1596 .policy = datapath_policy,
1597 .doit = odp_dp_cmd_set,
1601 static const struct nla_policy vport_policy[ODP_VPORT_ATTR_MAX + 1] = {
1602 #ifdef HAVE_NLA_NUL_STRING
1603 [ODP_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1604 [ODP_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1605 [ODP_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1606 [ODP_VPORT_ATTR_STATS] = { .len = sizeof(struct rtnl_link_stats64) },
1607 [ODP_VPORT_ATTR_ADDRESS] = { .len = ETH_ALEN },
1609 [ODP_VPORT_ATTR_STATS] = { .minlen = sizeof(struct rtnl_link_stats64) },
1610 [ODP_VPORT_ATTR_ADDRESS] = { .minlen = ETH_ALEN },
1612 [ODP_VPORT_ATTR_MTU] = { .type = NLA_U32 },
1613 [ODP_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1616 static struct genl_family dp_vport_genl_family = {
1617 .id = GENL_ID_GENERATE,
1618 .hdrsize = sizeof(struct odp_header),
1619 .name = ODP_VPORT_FAMILY,
1621 .maxattr = ODP_VPORT_ATTR_MAX
1624 static struct genl_multicast_group dp_vport_multicast_group = {
1625 .name = ODP_VPORT_MCGROUP
1628 /* Called with RTNL lock or RCU read lock. */
1629 static int odp_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1630 u32 pid, u32 seq, u32 flags, u8 cmd)
1632 struct odp_header *odp_header;
1634 int ifindex, iflink;
1638 odp_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family,
1643 odp_header->dp_ifindex = vport->dp->dp_ifindex;
1645 NLA_PUT_U32(skb, ODP_VPORT_ATTR_PORT_NO, vport->port_no);
1646 NLA_PUT_U32(skb, ODP_VPORT_ATTR_TYPE, vport_get_type(vport));
1647 NLA_PUT_STRING(skb, ODP_VPORT_ATTR_NAME, vport_get_name(vport));
1649 nla = nla_reserve(skb, ODP_VPORT_ATTR_STATS, sizeof(struct rtnl_link_stats64));
1651 goto nla_put_failure;
1652 if (vport_get_stats(vport, nla_data(nla)))
1653 __skb_trim(skb, skb->len - nla->nla_len);
1655 NLA_PUT(skb, ODP_VPORT_ATTR_ADDRESS, ETH_ALEN, vport_get_addr(vport));
1657 mtu = vport_get_mtu(vport);
1659 NLA_PUT_U32(skb, ODP_VPORT_ATTR_MTU, mtu);
1661 err = vport_get_options(vport, skb);
1662 if (err == -EMSGSIZE)
1665 ifindex = vport_get_ifindex(vport);
1667 NLA_PUT_U32(skb, ODP_VPORT_ATTR_IFINDEX, ifindex);
1669 iflink = vport_get_iflink(vport);
1671 NLA_PUT_U32(skb, ODP_VPORT_ATTR_IFLINK, iflink);
1673 return genlmsg_end(skb, odp_header);
1678 genlmsg_cancel(skb, odp_header);
1682 /* Called with RTNL lock or RCU read lock. */
1683 static struct sk_buff *odp_vport_cmd_build_info(struct vport *vport, u32 pid,
1686 struct sk_buff *skb;
1689 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1691 return ERR_PTR(-ENOMEM);
1693 retval = odp_vport_cmd_fill_info(vport, skb, pid, seq, 0, cmd);
1696 return ERR_PTR(retval);
1701 static int odp_vport_cmd_validate(struct nlattr *a[ODP_VPORT_ATTR_MAX + 1])
1703 return CHECK_NUL_STRING(a[ODP_VPORT_ATTR_NAME], IFNAMSIZ - 1);
1706 /* Called with RTNL lock or RCU read lock. */
1707 static struct vport *lookup_vport(struct odp_header *odp_header,
1708 struct nlattr *a[ODP_VPORT_ATTR_MAX + 1])
1710 struct datapath *dp;
1711 struct vport *vport;
1713 if (a[ODP_VPORT_ATTR_NAME]) {
1714 vport = vport_locate(nla_data(a[ODP_VPORT_ATTR_NAME]));
1716 return ERR_PTR(-ENODEV);
1718 } else if (a[ODP_VPORT_ATTR_PORT_NO]) {
1719 u32 port_no = nla_get_u32(a[ODP_VPORT_ATTR_PORT_NO]);
1721 if (port_no >= DP_MAX_PORTS)
1722 return ERR_PTR(-EFBIG);
1724 dp = get_dp(odp_header->dp_ifindex);
1726 return ERR_PTR(-ENODEV);
1728 vport = get_vport_protected(dp, port_no);
1730 return ERR_PTR(-ENOENT);
1733 return ERR_PTR(-EINVAL);
1736 /* Called with RTNL lock. */
1737 static int change_vport(struct vport *vport, struct nlattr *a[ODP_VPORT_ATTR_MAX + 1])
1740 if (a[ODP_VPORT_ATTR_STATS])
1741 err = vport_set_stats(vport, nla_data(a[ODP_VPORT_ATTR_STATS]));
1742 if (!err && a[ODP_VPORT_ATTR_ADDRESS])
1743 err = vport_set_addr(vport, nla_data(a[ODP_VPORT_ATTR_ADDRESS]));
1744 if (!err && a[ODP_VPORT_ATTR_MTU])
1745 err = vport_set_mtu(vport, nla_get_u32(a[ODP_VPORT_ATTR_MTU]));
1749 static int odp_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1751 struct nlattr **a = info->attrs;
1752 struct odp_header *odp_header = info->userhdr;
1753 struct vport_parms parms;
1754 struct sk_buff *reply;
1755 struct vport *vport;
1756 struct datapath *dp;
1761 if (!a[ODP_VPORT_ATTR_NAME] || !a[ODP_VPORT_ATTR_TYPE])
1764 err = odp_vport_cmd_validate(a);
1769 dp = get_dp(odp_header->dp_ifindex);
1774 if (a[ODP_VPORT_ATTR_PORT_NO]) {
1775 port_no = nla_get_u32(a[ODP_VPORT_ATTR_PORT_NO]);
1778 if (port_no >= DP_MAX_PORTS)
1781 vport = get_vport_protected(dp, port_no);
1786 for (port_no = 1; ; port_no++) {
1787 if (port_no >= DP_MAX_PORTS) {
1791 vport = get_vport_protected(dp, port_no);
1797 parms.name = nla_data(a[ODP_VPORT_ATTR_NAME]);
1798 parms.type = nla_get_u32(a[ODP_VPORT_ATTR_TYPE]);
1799 parms.options = a[ODP_VPORT_ATTR_OPTIONS];
1801 parms.port_no = port_no;
1803 vport = new_vport(&parms);
1804 err = PTR_ERR(vport);
1808 set_internal_devs_mtu(dp);
1809 dp_sysfs_add_if(vport);
1811 err = change_vport(vport, a);
1813 reply = odp_vport_cmd_build_info(vport, info->snd_pid,
1814 info->snd_seq, ODP_VPORT_CMD_NEW);
1816 err = PTR_ERR(reply);
1819 dp_detach_port(vport);
1822 genl_notify(reply, genl_info_net(info), info->snd_pid,
1823 dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1832 static int odp_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1834 struct nlattr **a = info->attrs;
1835 struct sk_buff *reply;
1836 struct vport *vport;
1839 err = odp_vport_cmd_validate(a);
1844 vport = lookup_vport(info->userhdr, a);
1845 err = PTR_ERR(vport);
1850 if (a[ODP_VPORT_ATTR_OPTIONS])
1851 err = vport_set_options(vport, a[ODP_VPORT_ATTR_OPTIONS]);
1853 err = change_vport(vport, a);
1855 reply = odp_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1857 if (IS_ERR(reply)) {
1858 err = PTR_ERR(reply);
1859 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1860 dp_vport_multicast_group.id, err);
1864 genl_notify(reply, genl_info_net(info), info->snd_pid,
1865 dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1873 static int odp_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1875 struct nlattr **a = info->attrs;
1876 struct sk_buff *reply;
1877 struct vport *vport;
1880 err = odp_vport_cmd_validate(a);
1885 vport = lookup_vport(info->userhdr, a);
1886 err = PTR_ERR(vport);
1890 if (vport->port_no == ODPP_LOCAL) {
1895 reply = odp_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1897 err = PTR_ERR(reply);
1901 err = dp_detach_port(vport);
1903 genl_notify(reply, genl_info_net(info), info->snd_pid,
1904 dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1912 static int odp_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1914 struct nlattr **a = info->attrs;
1915 struct odp_header *odp_header = info->userhdr;
1916 struct sk_buff *reply;
1917 struct vport *vport;
1920 err = odp_vport_cmd_validate(a);
1925 vport = lookup_vport(odp_header, a);
1926 err = PTR_ERR(vport);
1930 reply = odp_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1932 err = PTR_ERR(reply);
1936 err = genlmsg_reply(reply, info);
1944 static int odp_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1946 struct odp_header *odp_header = genlmsg_data(nlmsg_data(cb->nlh));
1947 struct datapath *dp;
1951 dp = get_dp(odp_header->dp_ifindex);
1956 for (port_no = cb->args[0]; port_no < DP_MAX_PORTS; port_no++) {
1957 struct vport *vport;
1959 vport = get_vport_protected(dp, port_no);
1963 if (odp_vport_cmd_fill_info(vport, skb, NETLINK_CB(cb->skb).pid,
1964 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1965 ODP_VPORT_CMD_NEW) < 0)
1970 cb->args[0] = port_no;
1976 static struct genl_ops dp_vport_genl_ops[] = {
1977 { .cmd = ODP_VPORT_CMD_NEW,
1978 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1979 .policy = vport_policy,
1980 .doit = odp_vport_cmd_new
1982 { .cmd = ODP_VPORT_CMD_DEL,
1983 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1984 .policy = vport_policy,
1985 .doit = odp_vport_cmd_del
1987 { .cmd = ODP_VPORT_CMD_GET,
1988 .flags = 0, /* OK for unprivileged users. */
1989 .policy = vport_policy,
1990 .doit = odp_vport_cmd_get,
1991 .dumpit = odp_vport_cmd_dump
1993 { .cmd = ODP_VPORT_CMD_SET,
1994 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1995 .policy = vport_policy,
1996 .doit = odp_vport_cmd_set,
2000 struct genl_family_and_ops {
2001 struct genl_family *family;
2002 struct genl_ops *ops;
2004 struct genl_multicast_group *group;
2007 static const struct genl_family_and_ops dp_genl_families[] = {
2008 { &dp_datapath_genl_family,
2009 dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
2010 &dp_datapath_multicast_group },
2011 { &dp_vport_genl_family,
2012 dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
2013 &dp_vport_multicast_group },
2014 { &dp_flow_genl_family,
2015 dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
2016 &dp_flow_multicast_group },
2017 { &dp_packet_genl_family,
2018 dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
2022 static void dp_unregister_genl(int n_families)
2026 for (i = 0; i < n_families; i++) {
2027 genl_unregister_family(dp_genl_families[i].family);
2031 static int dp_register_genl(void)
2038 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2039 const struct genl_family_and_ops *f = &dp_genl_families[i];
2041 err = genl_register_family_with_ops(f->family, f->ops,
2048 err = genl_register_mc_group(f->family, f->group);
2054 err = packet_register_mc_groups();
2060 dp_unregister_genl(n_registered);
2064 static int __init dp_init(void)
2066 struct sk_buff *dummy_skb;
2069 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
2071 printk("Open vSwitch %s, built "__DATE__" "__TIME__"\n", VERSION BUILDNR);
2079 goto error_flow_exit;
2081 err = register_netdevice_notifier(&dp_device_notifier);
2083 goto error_vport_exit;
2085 err = dp_register_genl();
2087 goto error_unreg_notifier;
2091 error_unreg_notifier:
2092 unregister_netdevice_notifier(&dp_device_notifier);
2101 static void dp_cleanup(void)
2104 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2105 unregister_netdevice_notifier(&dp_device_notifier);
2110 module_init(dp_init);
2111 module_exit(dp_cleanup);
2113 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2114 MODULE_LICENSE("GPL");