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);
362 BUG_ON(csum_start >= skb_headlen(skb));
364 skb_copy_bits(skb, 0, to, csum_start);
366 csum = skb_copy_and_csum_bits(skb, csum_start, to + csum_start,
367 skb->len - csum_start, 0);
368 *(__sum16 *)(to + csum_start + csum_offset) = csum_fold(csum);
371 static struct genl_family dp_packet_genl_family = {
372 .id = GENL_ID_GENERATE,
373 .hdrsize = sizeof(struct odp_header),
374 .name = ODP_PACKET_FAMILY,
376 .maxattr = ODP_PACKET_ATTR_MAX
379 /* Generic Netlink multicast groups for upcalls.
381 * We really want three unique multicast groups per datapath, but we can't even
382 * get one, because genl_register_mc_group() takes genl_lock, which is also
383 * held during Generic Netlink message processing, so trying to acquire
384 * multicast groups during ODP_DP_NEW processing deadlocks. Instead, we
385 * preallocate a few groups and use them round-robin for datapaths. Collision
386 * isn't fatal--multicast listeners should check that the family is the one
387 * that they want and discard others--but it wastes time and memory to receive
390 #define PACKET_N_MC_GROUPS 16
391 static struct genl_multicast_group packet_mc_groups[PACKET_N_MC_GROUPS];
393 static u32 packet_mc_group(struct datapath *dp, u8 cmd)
396 BUILD_BUG_ON_NOT_POWER_OF_2(PACKET_N_MC_GROUPS);
398 idx = jhash_2words(dp->dp_ifindex, cmd, 0) & (PACKET_N_MC_GROUPS - 1);
399 return packet_mc_groups[idx].id;
402 static int packet_register_mc_groups(void)
406 for (i = 0; i < PACKET_N_MC_GROUPS; i++) {
407 struct genl_multicast_group *group = &packet_mc_groups[i];
410 sprintf(group->name, "packet%d", i);
411 error = genl_register_mc_group(&dp_packet_genl_family, group);
418 int dp_upcall(struct datapath *dp, struct sk_buff *skb, const struct dp_upcall_info *upcall_info)
420 struct dp_stats_percpu *stats;
423 WARN_ON_ONCE(skb_shared(skb));
425 forward_ip_summed(skb);
427 err = vswitch_skb_checksum_setup(skb);
431 /* Break apart GSO packets into their component pieces. Otherwise
432 * userspace may try to stuff a 64kB packet into a 1500-byte MTU. */
433 if (skb_is_gso(skb)) {
434 struct sk_buff *nskb = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
444 err = queue_control_packets(dp, skb, upcall_info);
454 stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
456 write_seqcount_begin(&stats->seqlock);
458 write_seqcount_end(&stats->seqlock);
465 /* Send each packet in the 'skb' list to userspace for 'dp' as directed by
466 * 'upcall_info'. There will be only one packet unless we broke up a GSO
469 static int queue_control_packets(struct datapath *dp, struct sk_buff *skb,
470 const struct dp_upcall_info *upcall_info)
472 u32 group = packet_mc_group(dp, upcall_info->cmd);
473 struct sk_buff *nskb;
477 if (OVS_CB(skb)->vport)
478 port_no = OVS_CB(skb)->vport->port_no;
480 port_no = ODPP_LOCAL;
483 struct odp_header *upcall;
484 struct sk_buff *user_skb; /* to be queued to userspace */
491 err = vlan_deaccel_tag(skb);
495 len = sizeof(struct odp_header);
496 len += nla_total_size(skb->len);
497 len += nla_total_size(FLOW_BUFSIZE);
498 if (upcall_info->userdata)
499 len += nla_total_size(8);
500 if (upcall_info->sample_pool)
501 len += nla_total_size(4);
502 if (upcall_info->actions_len)
503 len += nla_total_size(upcall_info->actions_len);
505 user_skb = genlmsg_new(len, GFP_ATOMIC);
507 netlink_set_err(INIT_NET_GENL_SOCK, 0, group, -ENOBUFS);
511 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family, 0, upcall_info->cmd);
512 upcall->dp_ifindex = dp->dp_ifindex;
514 nla = nla_nest_start(user_skb, ODP_PACKET_ATTR_KEY);
515 flow_to_nlattrs(upcall_info->key, user_skb);
516 nla_nest_end(user_skb, nla);
518 if (upcall_info->userdata)
519 nla_put_u64(user_skb, ODP_PACKET_ATTR_USERDATA, upcall_info->userdata);
520 if (upcall_info->sample_pool)
521 nla_put_u32(user_skb, ODP_PACKET_ATTR_SAMPLE_POOL, upcall_info->sample_pool);
522 if (upcall_info->actions_len) {
523 const struct nlattr *actions = upcall_info->actions;
524 u32 actions_len = upcall_info->actions_len;
526 nla = nla_nest_start(user_skb, ODP_PACKET_ATTR_ACTIONS);
527 memcpy(__skb_put(user_skb, actions_len), actions, actions_len);
528 nla_nest_end(user_skb, nla);
531 nla = __nla_reserve(user_skb, ODP_PACKET_ATTR_PACKET, skb->len);
532 if (skb->ip_summed == CHECKSUM_PARTIAL)
533 copy_and_csum_skb(skb, nla_data(nla));
535 skb_copy_bits(skb, 0, nla_data(nla), skb->len);
537 err = genlmsg_multicast(user_skb, 0, group, GFP_ATOMIC);
548 while ((skb = nskb) != NULL) {
555 /* Called with genl_mutex. */
556 static int flush_flows(int dp_ifindex)
558 struct tbl *old_table;
559 struct tbl *new_table;
562 dp = get_dp(dp_ifindex);
566 old_table = get_table_protected(dp);
567 new_table = tbl_create(TBL_MIN_BUCKETS);
571 rcu_assign_pointer(dp->table, new_table);
573 tbl_deferred_destroy(old_table, flow_free_tbl);
578 static int validate_actions(const struct nlattr *attr)
580 const struct nlattr *a;
583 nla_for_each_nested(a, attr, rem) {
584 static const u32 action_lens[ODP_ACTION_ATTR_MAX + 1] = {
585 [ODP_ACTION_ATTR_OUTPUT] = 4,
586 [ODP_ACTION_ATTR_CONTROLLER] = 8,
587 [ODP_ACTION_ATTR_SET_DL_TCI] = 2,
588 [ODP_ACTION_ATTR_STRIP_VLAN] = 0,
589 [ODP_ACTION_ATTR_SET_DL_SRC] = ETH_ALEN,
590 [ODP_ACTION_ATTR_SET_DL_DST] = ETH_ALEN,
591 [ODP_ACTION_ATTR_SET_NW_SRC] = 4,
592 [ODP_ACTION_ATTR_SET_NW_DST] = 4,
593 [ODP_ACTION_ATTR_SET_NW_TOS] = 1,
594 [ODP_ACTION_ATTR_SET_TP_SRC] = 2,
595 [ODP_ACTION_ATTR_SET_TP_DST] = 2,
596 [ODP_ACTION_ATTR_SET_TUNNEL] = 8,
597 [ODP_ACTION_ATTR_SET_PRIORITY] = 4,
598 [ODP_ACTION_ATTR_POP_PRIORITY] = 0,
599 [ODP_ACTION_ATTR_DROP_SPOOFED_ARP] = 0,
601 int type = nla_type(a);
603 if (type > ODP_ACTION_ATTR_MAX || nla_len(a) != action_lens[type])
607 case ODP_ACTION_ATTR_UNSPEC:
610 case ODP_ACTION_ATTR_CONTROLLER:
611 case ODP_ACTION_ATTR_STRIP_VLAN:
612 case ODP_ACTION_ATTR_SET_DL_SRC:
613 case ODP_ACTION_ATTR_SET_DL_DST:
614 case ODP_ACTION_ATTR_SET_NW_SRC:
615 case ODP_ACTION_ATTR_SET_NW_DST:
616 case ODP_ACTION_ATTR_SET_TP_SRC:
617 case ODP_ACTION_ATTR_SET_TP_DST:
618 case ODP_ACTION_ATTR_SET_TUNNEL:
619 case ODP_ACTION_ATTR_SET_PRIORITY:
620 case ODP_ACTION_ATTR_POP_PRIORITY:
621 case ODP_ACTION_ATTR_DROP_SPOOFED_ARP:
622 /* No validation needed. */
625 case ODP_ACTION_ATTR_OUTPUT:
626 if (nla_get_u32(a) >= DP_MAX_PORTS)
630 case ODP_ACTION_ATTR_SET_DL_TCI:
631 if (nla_get_be16(a) & htons(VLAN_CFI_MASK))
635 case ODP_ACTION_ATTR_SET_NW_TOS:
636 if (nla_get_u8(a) & INET_ECN_MASK)
650 static void clear_stats(struct sw_flow *flow)
654 flow->packet_count = 0;
655 flow->byte_count = 0;
658 /* Called with genl_mutex. */
659 static int expand_table(struct datapath *dp)
661 struct tbl *old_table = get_table_protected(dp);
662 struct tbl *new_table;
664 new_table = tbl_expand(old_table);
665 if (IS_ERR(new_table))
666 return PTR_ERR(new_table);
668 rcu_assign_pointer(dp->table, new_table);
669 tbl_deferred_destroy(old_table, NULL);
674 static int odp_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
676 struct odp_header *odp_header = info->userhdr;
677 struct nlattr **a = info->attrs;
678 struct sk_buff *packet;
679 struct sw_flow_key key;
686 if (!a[ODP_PACKET_ATTR_PACKET] || !a[ODP_PACKET_ATTR_ACTIONS] ||
687 nla_len(a[ODP_PACKET_ATTR_PACKET]) < ETH_HLEN)
690 err = validate_actions(a[ODP_PACKET_ATTR_ACTIONS]);
694 packet = skb_clone(skb, GFP_KERNEL);
698 packet->data = nla_data(a[ODP_PACKET_ATTR_PACKET]);
699 packet->len = nla_len(a[ODP_PACKET_ATTR_PACKET]);
701 skb_reset_mac_header(packet);
702 eth = eth_hdr(packet);
704 /* Normally, setting the skb 'protocol' field would be handled by a
705 * call to eth_type_trans(), but it assumes there's a sending
706 * device, which we may not have. */
707 if (ntohs(eth->h_proto) >= 1536)
708 packet->protocol = eth->h_proto;
710 packet->protocol = htons(ETH_P_802_2);
712 err = flow_extract(packet, -1, &key, &is_frag);
716 /* Initialize OVS_CB (it came from Netlink so might not be zeroed). */
717 memset(OVS_CB(packet), 0, sizeof(struct ovs_skb_cb));
720 dp = get_dp(odp_header->dp_ifindex);
723 err = execute_actions(dp, packet, &key,
724 nla_data(a[ODP_PACKET_ATTR_ACTIONS]),
725 nla_len(a[ODP_PACKET_ATTR_ACTIONS]));
732 static const struct nla_policy packet_policy[ODP_PACKET_ATTR_MAX + 1] = {
733 [ODP_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
734 [ODP_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
737 static struct genl_ops dp_packet_genl_ops[] = {
738 { .cmd = ODP_PACKET_CMD_EXECUTE,
739 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
740 .policy = packet_policy,
741 .doit = odp_packet_cmd_execute
745 static void get_dp_stats(struct datapath *dp, struct odp_stats *stats)
749 stats->n_frags = stats->n_hit = stats->n_missed = stats->n_lost = 0;
750 for_each_possible_cpu(i) {
751 const struct dp_stats_percpu *percpu_stats;
752 struct dp_stats_percpu local_stats;
755 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
758 seqcount = read_seqcount_begin(&percpu_stats->seqlock);
759 local_stats = *percpu_stats;
760 } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount));
762 stats->n_frags += local_stats.n_frags;
763 stats->n_hit += local_stats.n_hit;
764 stats->n_missed += local_stats.n_missed;
765 stats->n_lost += local_stats.n_lost;
769 /* MTU of the dp pseudo-device: ETH_DATA_LEN or the minimum of the ports.
770 * Called with RTNL lock.
772 int dp_min_mtu(const struct datapath *dp)
779 list_for_each_entry (p, &dp->port_list, node) {
782 /* Skip any internal ports, since that's what we're trying to
784 if (is_internal_vport(p))
787 dev_mtu = vport_get_mtu(p);
790 if (!mtu || dev_mtu < mtu)
794 return mtu ? mtu : ETH_DATA_LEN;
797 /* Sets the MTU of all datapath devices to the minimum of the ports
798 * Called with RTNL lock.
800 void set_internal_devs_mtu(const struct datapath *dp)
807 mtu = dp_min_mtu(dp);
809 list_for_each_entry (p, &dp->port_list, node) {
810 if (is_internal_vport(p))
811 vport_set_mtu(p, mtu);
815 static const struct nla_policy flow_policy[ODP_FLOW_ATTR_MAX + 1] = {
816 [ODP_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
817 [ODP_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
818 [ODP_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
821 static struct genl_family dp_flow_genl_family = {
822 .id = GENL_ID_GENERATE,
823 .hdrsize = sizeof(struct odp_header),
824 .name = ODP_FLOW_FAMILY,
826 .maxattr = ODP_FLOW_ATTR_MAX
829 static struct genl_multicast_group dp_flow_multicast_group = {
830 .name = ODP_FLOW_MCGROUP
833 /* Called with genl_lock. */
834 static int odp_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
835 struct sk_buff *skb, u32 pid, u32 seq, u32 flags, u8 cmd)
837 const int skb_orig_len = skb->len;
838 const struct sw_flow_actions *sf_acts;
839 struct odp_flow_stats stats;
840 struct odp_header *odp_header;
846 sf_acts = rcu_dereference_protected(flow->sf_acts,
847 lockdep_genl_is_held());
849 odp_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags, cmd);
853 odp_header->dp_ifindex = dp->dp_ifindex;
855 nla = nla_nest_start(skb, ODP_FLOW_ATTR_KEY);
857 goto nla_put_failure;
858 err = flow_to_nlattrs(&flow->key, skb);
861 nla_nest_end(skb, nla);
863 spin_lock_bh(&flow->lock);
865 stats.n_packets = flow->packet_count;
866 stats.n_bytes = flow->byte_count;
867 tcp_flags = flow->tcp_flags;
868 spin_unlock_bh(&flow->lock);
871 NLA_PUT_U64(skb, ODP_FLOW_ATTR_USED, flow_used_time(used));
874 NLA_PUT(skb, ODP_FLOW_ATTR_STATS, sizeof(struct odp_flow_stats), &stats);
877 NLA_PUT_U8(skb, ODP_FLOW_ATTR_TCP_FLAGS, tcp_flags);
879 /* If ODP_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
880 * this is the first flow to be dumped into 'skb'. This is unusual for
881 * Netlink but individual action lists can be longer than
882 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
883 * The userspace caller can always fetch the actions separately if it
884 * really wants them. (Most userspace callers in fact don't care.)
886 * This can only fail for dump operations because the skb is always
887 * properly sized for single flows.
889 err = nla_put(skb, ODP_FLOW_ATTR_ACTIONS, sf_acts->actions_len,
891 if (err < 0 && skb_orig_len)
894 return genlmsg_end(skb, odp_header);
899 genlmsg_cancel(skb, odp_header);
903 static struct sk_buff *odp_flow_cmd_alloc_info(struct sw_flow *flow)
905 const struct sw_flow_actions *sf_acts;
908 sf_acts = rcu_dereference_protected(flow->sf_acts,
909 lockdep_genl_is_held());
911 len = nla_total_size(FLOW_BUFSIZE); /* ODP_FLOW_ATTR_KEY */
912 len += nla_total_size(sf_acts->actions_len); /* ODP_FLOW_ATTR_ACTIONS */
913 len += nla_total_size(sizeof(struct odp_flow_stats)); /* ODP_FLOW_ATTR_STATS */
914 len += nla_total_size(1); /* ODP_FLOW_ATTR_TCP_FLAGS */
915 len += nla_total_size(8); /* ODP_FLOW_ATTR_USED */
916 return genlmsg_new(NLMSG_ALIGN(sizeof(struct odp_header)) + len, GFP_KERNEL);
919 static struct sk_buff *odp_flow_cmd_build_info(struct sw_flow *flow, struct datapath *dp,
920 u32 pid, u32 seq, u8 cmd)
925 skb = odp_flow_cmd_alloc_info(flow);
927 return ERR_PTR(-ENOMEM);
929 retval = odp_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd);
934 static int odp_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
936 struct nlattr **a = info->attrs;
937 struct odp_header *odp_header = info->userhdr;
938 struct tbl_node *flow_node;
939 struct sw_flow_key key;
940 struct sw_flow *flow;
941 struct sk_buff *reply;
949 if (!a[ODP_FLOW_ATTR_KEY])
951 error = flow_from_nlattrs(&key, a[ODP_FLOW_ATTR_KEY]);
955 /* Validate actions. */
956 if (a[ODP_FLOW_ATTR_ACTIONS]) {
957 error = validate_actions(a[ODP_FLOW_ATTR_ACTIONS]);
960 } else if (info->genlhdr->cmd == ODP_FLOW_CMD_NEW) {
965 dp = get_dp(odp_header->dp_ifindex);
970 hash = flow_hash(&key);
971 table = get_table_protected(dp);
972 flow_node = tbl_lookup(table, &key, hash, flow_cmp);
974 struct sw_flow_actions *acts;
976 /* Bail out if we're not allowed to create a new flow. */
978 if (info->genlhdr->cmd == ODP_FLOW_CMD_SET)
981 /* Expand table, if necessary, to make room. */
982 if (tbl_count(table) >= tbl_n_buckets(table)) {
983 error = expand_table(dp);
986 table = get_table_protected(dp);
992 error = PTR_ERR(flow);
998 /* Obtain actions. */
999 acts = flow_actions_alloc(a[ODP_FLOW_ATTR_ACTIONS]);
1000 error = PTR_ERR(acts);
1002 goto error_free_flow;
1003 rcu_assign_pointer(flow->sf_acts, acts);
1005 /* Put flow in bucket. */
1006 error = tbl_insert(table, &flow->tbl_node, hash);
1008 goto error_free_flow;
1010 reply = odp_flow_cmd_build_info(flow, dp, info->snd_pid,
1011 info->snd_seq, ODP_FLOW_CMD_NEW);
1013 /* We found a matching flow. */
1014 struct sw_flow_actions *old_acts;
1016 /* Bail out if we're not allowed to modify an existing flow.
1017 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1018 * because Generic Netlink treats the latter as a dump
1019 * request. We also accept NLM_F_EXCL in case that bug ever
1023 if (info->genlhdr->cmd == ODP_FLOW_CMD_NEW &&
1024 info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1027 /* Update actions. */
1028 flow = flow_cast(flow_node);
1029 old_acts = rcu_dereference_protected(flow->sf_acts,
1030 lockdep_genl_is_held());
1031 if (a[ODP_FLOW_ATTR_ACTIONS] &&
1032 (old_acts->actions_len != nla_len(a[ODP_FLOW_ATTR_ACTIONS]) ||
1033 memcmp(old_acts->actions, nla_data(a[ODP_FLOW_ATTR_ACTIONS]),
1034 old_acts->actions_len))) {
1035 struct sw_flow_actions *new_acts;
1037 new_acts = flow_actions_alloc(a[ODP_FLOW_ATTR_ACTIONS]);
1038 error = PTR_ERR(new_acts);
1039 if (IS_ERR(new_acts))
1042 rcu_assign_pointer(flow->sf_acts, new_acts);
1043 flow_deferred_free_acts(old_acts);
1046 reply = odp_flow_cmd_build_info(flow, dp, info->snd_pid,
1047 info->snd_seq, ODP_FLOW_CMD_NEW);
1050 if (a[ODP_FLOW_ATTR_CLEAR]) {
1051 spin_lock_bh(&flow->lock);
1053 spin_unlock_bh(&flow->lock);
1058 genl_notify(reply, genl_info_net(info), info->snd_pid,
1059 dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1061 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1062 dp_flow_multicast_group.id, PTR_ERR(reply));
1071 static int odp_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1073 struct nlattr **a = info->attrs;
1074 struct odp_header *odp_header = info->userhdr;
1075 struct sw_flow_key key;
1076 struct tbl_node *flow_node;
1077 struct sk_buff *reply;
1078 struct sw_flow *flow;
1079 struct datapath *dp;
1083 if (!a[ODP_FLOW_ATTR_KEY])
1085 err = flow_from_nlattrs(&key, a[ODP_FLOW_ATTR_KEY]);
1089 dp = get_dp(odp_header->dp_ifindex);
1093 table = get_table_protected(dp);
1094 flow_node = tbl_lookup(table, &key, flow_hash(&key), flow_cmp);
1098 flow = flow_cast(flow_node);
1099 reply = odp_flow_cmd_build_info(flow, dp, info->snd_pid, info->snd_seq, ODP_FLOW_CMD_NEW);
1101 return PTR_ERR(reply);
1103 return genlmsg_reply(reply, info);
1106 static int odp_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1108 struct nlattr **a = info->attrs;
1109 struct odp_header *odp_header = info->userhdr;
1110 struct sw_flow_key key;
1111 struct tbl_node *flow_node;
1112 struct sk_buff *reply;
1113 struct sw_flow *flow;
1114 struct datapath *dp;
1118 if (!a[ODP_FLOW_ATTR_KEY])
1119 return flush_flows(odp_header->dp_ifindex);
1120 err = flow_from_nlattrs(&key, a[ODP_FLOW_ATTR_KEY]);
1124 dp = get_dp(odp_header->dp_ifindex);
1128 table = get_table_protected(dp);
1129 flow_node = tbl_lookup(table, &key, flow_hash(&key), flow_cmp);
1132 flow = flow_cast(flow_node);
1134 reply = odp_flow_cmd_alloc_info(flow);
1138 err = tbl_remove(table, flow_node);
1144 err = odp_flow_cmd_fill_info(flow, dp, reply, info->snd_pid,
1145 info->snd_seq, 0, ODP_FLOW_CMD_DEL);
1148 flow_deferred_free(flow);
1150 genl_notify(reply, genl_info_net(info), info->snd_pid,
1151 dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1155 static int odp_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1157 struct odp_header *odp_header = genlmsg_data(nlmsg_data(cb->nlh));
1158 struct datapath *dp;
1160 dp = get_dp(odp_header->dp_ifindex);
1165 struct tbl_node *flow_node;
1166 struct sw_flow *flow;
1169 bucket = cb->args[0];
1171 flow_node = tbl_next(get_table_protected(dp), &bucket, &obj);
1175 flow = flow_cast(flow_node);
1176 if (odp_flow_cmd_fill_info(flow, dp, skb, NETLINK_CB(cb->skb).pid,
1177 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1178 ODP_FLOW_CMD_NEW) < 0)
1181 cb->args[0] = bucket;
1187 static struct genl_ops dp_flow_genl_ops[] = {
1188 { .cmd = ODP_FLOW_CMD_NEW,
1189 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1190 .policy = flow_policy,
1191 .doit = odp_flow_cmd_new_or_set
1193 { .cmd = ODP_FLOW_CMD_DEL,
1194 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1195 .policy = flow_policy,
1196 .doit = odp_flow_cmd_del
1198 { .cmd = ODP_FLOW_CMD_GET,
1199 .flags = 0, /* OK for unprivileged users. */
1200 .policy = flow_policy,
1201 .doit = odp_flow_cmd_get,
1202 .dumpit = odp_flow_cmd_dump
1204 { .cmd = ODP_FLOW_CMD_SET,
1205 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1206 .policy = flow_policy,
1207 .doit = odp_flow_cmd_new_or_set,
1211 static const struct nla_policy datapath_policy[ODP_DP_ATTR_MAX + 1] = {
1212 #ifdef HAVE_NLA_NUL_STRING
1213 [ODP_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1215 [ODP_DP_ATTR_IPV4_FRAGS] = { .type = NLA_U32 },
1216 [ODP_DP_ATTR_SAMPLING] = { .type = NLA_U32 },
1219 static struct genl_family dp_datapath_genl_family = {
1220 .id = GENL_ID_GENERATE,
1221 .hdrsize = sizeof(struct odp_header),
1222 .name = ODP_DATAPATH_FAMILY,
1224 .maxattr = ODP_DP_ATTR_MAX
1227 static struct genl_multicast_group dp_datapath_multicast_group = {
1228 .name = ODP_DATAPATH_MCGROUP
1231 static int odp_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1232 u32 pid, u32 seq, u32 flags, u8 cmd)
1234 struct odp_header *odp_header;
1238 odp_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family,
1243 odp_header->dp_ifindex = dp->dp_ifindex;
1246 err = nla_put_string(skb, ODP_DP_ATTR_NAME, dp_name(dp));
1249 goto nla_put_failure;
1251 nla = nla_reserve(skb, ODP_DP_ATTR_STATS, sizeof(struct odp_stats));
1253 goto nla_put_failure;
1254 get_dp_stats(dp, nla_data(nla));
1256 NLA_PUT_U32(skb, ODP_DP_ATTR_IPV4_FRAGS,
1257 dp->drop_frags ? ODP_DP_FRAG_DROP : ODP_DP_FRAG_ZERO);
1259 if (dp->sflow_probability)
1260 NLA_PUT_U32(skb, ODP_DP_ATTR_SAMPLING, dp->sflow_probability);
1262 nla = nla_nest_start(skb, ODP_DP_ATTR_MCGROUPS);
1264 goto nla_put_failure;
1265 NLA_PUT_U32(skb, ODP_PACKET_CMD_MISS, packet_mc_group(dp, ODP_PACKET_CMD_MISS));
1266 NLA_PUT_U32(skb, ODP_PACKET_CMD_ACTION, packet_mc_group(dp, ODP_PACKET_CMD_ACTION));
1267 NLA_PUT_U32(skb, ODP_PACKET_CMD_SAMPLE, packet_mc_group(dp, ODP_PACKET_CMD_SAMPLE));
1268 nla_nest_end(skb, nla);
1270 return genlmsg_end(skb, odp_header);
1273 genlmsg_cancel(skb, odp_header);
1278 static struct sk_buff *odp_dp_cmd_build_info(struct datapath *dp, u32 pid,
1281 struct sk_buff *skb;
1284 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1286 return ERR_PTR(-ENOMEM);
1288 retval = odp_dp_cmd_fill_info(dp, skb, pid, seq, 0, cmd);
1291 return ERR_PTR(retval);
1296 static int odp_dp_cmd_validate(struct nlattr *a[ODP_DP_ATTR_MAX + 1])
1298 if (a[ODP_DP_ATTR_IPV4_FRAGS]) {
1299 u32 frags = nla_get_u32(a[ODP_DP_ATTR_IPV4_FRAGS]);
1301 if (frags != ODP_DP_FRAG_ZERO && frags != ODP_DP_FRAG_DROP)
1305 return VERIFY_NUL_STRING(a[ODP_DP_ATTR_NAME], IFNAMSIZ - 1);
1308 /* Called with genl_mutex and optionally with RTNL lock also. */
1309 static struct datapath *lookup_datapath(struct odp_header *odp_header, struct nlattr *a[ODP_DP_ATTR_MAX + 1])
1311 struct datapath *dp;
1313 if (!a[ODP_DP_ATTR_NAME])
1314 dp = get_dp(odp_header->dp_ifindex);
1316 struct vport *vport;
1319 vport = vport_locate(nla_data(a[ODP_DP_ATTR_NAME]));
1320 dp = vport && vport->port_no == ODPP_LOCAL ? vport->dp : NULL;
1323 return dp ? dp : ERR_PTR(-ENODEV);
1326 /* Called with genl_mutex. */
1327 static void change_datapath(struct datapath *dp, struct nlattr *a[ODP_DP_ATTR_MAX + 1])
1329 if (a[ODP_DP_ATTR_IPV4_FRAGS])
1330 dp->drop_frags = nla_get_u32(a[ODP_DP_ATTR_IPV4_FRAGS]) == ODP_DP_FRAG_DROP;
1331 if (a[ODP_DP_ATTR_SAMPLING])
1332 dp->sflow_probability = nla_get_u32(a[ODP_DP_ATTR_SAMPLING]);
1335 static int odp_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1337 struct nlattr **a = info->attrs;
1338 struct vport_parms parms;
1339 struct sk_buff *reply;
1340 struct datapath *dp;
1341 struct vport *vport;
1345 if (!a[ODP_DP_ATTR_NAME])
1348 err = odp_dp_cmd_validate(a);
1354 if (!try_module_get(THIS_MODULE))
1355 goto err_unlock_rtnl;
1358 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1360 goto err_put_module;
1361 INIT_LIST_HEAD(&dp->port_list);
1363 /* Initialize kobject for bridge. This will be added as
1364 * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
1365 dp->ifobj.kset = NULL;
1366 kobject_init(&dp->ifobj, &dp_ktype);
1368 /* Allocate table. */
1370 rcu_assign_pointer(dp->table, tbl_create(TBL_MIN_BUCKETS));
1374 /* Set up our datapath device. */
1375 parms.name = nla_data(a[ODP_DP_ATTR_NAME]);
1376 parms.type = ODP_VPORT_TYPE_INTERNAL;
1377 parms.options = NULL;
1379 parms.port_no = ODPP_LOCAL;
1380 vport = new_vport(&parms);
1381 if (IS_ERR(vport)) {
1382 err = PTR_ERR(vport);
1386 goto err_destroy_table;
1388 dp->dp_ifindex = vport_get_ifindex(vport);
1391 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1392 if (!dp->stats_percpu) {
1394 goto err_destroy_local_port;
1397 change_datapath(dp, a);
1399 reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_NEW);
1400 err = PTR_ERR(reply);
1402 goto err_destroy_local_port;
1404 list_add_tail(&dp->list_node, &dps);
1405 dp_sysfs_add_dp(dp);
1409 genl_notify(reply, genl_info_net(info), info->snd_pid,
1410 dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1413 err_destroy_local_port:
1414 dp_detach_port(get_vport_protected(dp, ODPP_LOCAL));
1416 tbl_destroy(get_table_protected(dp), NULL);
1420 module_put(THIS_MODULE);
1427 static int odp_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1429 struct vport *vport, *next_vport;
1430 struct sk_buff *reply;
1431 struct datapath *dp;
1434 err = odp_dp_cmd_validate(info->attrs);
1439 dp = lookup_datapath(info->userhdr, info->attrs);
1444 reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_DEL);
1445 err = PTR_ERR(reply);
1449 list_for_each_entry_safe (vport, next_vport, &dp->port_list, node)
1450 if (vport->port_no != ODPP_LOCAL)
1451 dp_detach_port(vport);
1453 dp_sysfs_del_dp(dp);
1454 list_del(&dp->list_node);
1455 dp_detach_port(get_vport_protected(dp, ODPP_LOCAL));
1457 /* rtnl_unlock() will wait until all the references to devices that
1458 * are pending unregistration have been dropped. We do it here to
1459 * ensure that any internal devices (which contain DP pointers) are
1460 * fully destroyed before freeing the datapath.
1464 call_rcu(&dp->rcu, destroy_dp_rcu);
1465 module_put(THIS_MODULE);
1467 genl_notify(reply, genl_info_net(info), info->snd_pid,
1468 dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1478 static int odp_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1480 struct sk_buff *reply;
1481 struct datapath *dp;
1484 err = odp_dp_cmd_validate(info->attrs);
1488 dp = lookup_datapath(info->userhdr, info->attrs);
1492 change_datapath(dp, info->attrs);
1494 reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_NEW);
1495 if (IS_ERR(reply)) {
1496 err = PTR_ERR(reply);
1497 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1498 dp_datapath_multicast_group.id, err);
1502 genl_notify(reply, genl_info_net(info), info->snd_pid,
1503 dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1507 static int odp_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1509 struct sk_buff *reply;
1510 struct datapath *dp;
1513 err = odp_dp_cmd_validate(info->attrs);
1517 dp = lookup_datapath(info->userhdr, info->attrs);
1521 reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_NEW);
1523 return PTR_ERR(reply);
1525 return genlmsg_reply(reply, info);
1528 static int odp_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1530 struct datapath *dp;
1531 int skip = cb->args[0];
1534 list_for_each_entry (dp, &dps, list_node) {
1537 if (odp_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).pid,
1538 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1539 ODP_DP_CMD_NEW) < 0)
1549 static struct genl_ops dp_datapath_genl_ops[] = {
1550 { .cmd = ODP_DP_CMD_NEW,
1551 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1552 .policy = datapath_policy,
1553 .doit = odp_dp_cmd_new
1555 { .cmd = ODP_DP_CMD_DEL,
1556 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1557 .policy = datapath_policy,
1558 .doit = odp_dp_cmd_del
1560 { .cmd = ODP_DP_CMD_GET,
1561 .flags = 0, /* OK for unprivileged users. */
1562 .policy = datapath_policy,
1563 .doit = odp_dp_cmd_get,
1564 .dumpit = odp_dp_cmd_dump
1566 { .cmd = ODP_DP_CMD_SET,
1567 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1568 .policy = datapath_policy,
1569 .doit = odp_dp_cmd_set,
1573 static const struct nla_policy vport_policy[ODP_VPORT_ATTR_MAX + 1] = {
1574 #ifdef HAVE_NLA_NUL_STRING
1575 [ODP_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1576 [ODP_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1577 [ODP_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1578 [ODP_VPORT_ATTR_STATS] = { .len = sizeof(struct rtnl_link_stats64) },
1579 [ODP_VPORT_ATTR_ADDRESS] = { .len = ETH_ALEN },
1581 [ODP_VPORT_ATTR_STATS] = { .minlen = sizeof(struct rtnl_link_stats64) },
1582 [ODP_VPORT_ATTR_ADDRESS] = { .minlen = ETH_ALEN },
1584 [ODP_VPORT_ATTR_MTU] = { .type = NLA_U32 },
1585 [ODP_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1588 static struct genl_family dp_vport_genl_family = {
1589 .id = GENL_ID_GENERATE,
1590 .hdrsize = sizeof(struct odp_header),
1591 .name = ODP_VPORT_FAMILY,
1593 .maxattr = ODP_VPORT_ATTR_MAX
1596 static struct genl_multicast_group dp_vport_multicast_group = {
1597 .name = ODP_VPORT_MCGROUP
1600 /* Called with RTNL lock or RCU read lock. */
1601 static int odp_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1602 u32 pid, u32 seq, u32 flags, u8 cmd)
1604 struct odp_header *odp_header;
1606 int ifindex, iflink;
1610 odp_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family,
1615 odp_header->dp_ifindex = vport->dp->dp_ifindex;
1617 NLA_PUT_U32(skb, ODP_VPORT_ATTR_PORT_NO, vport->port_no);
1618 NLA_PUT_U32(skb, ODP_VPORT_ATTR_TYPE, vport_get_type(vport));
1619 NLA_PUT_STRING(skb, ODP_VPORT_ATTR_NAME, vport_get_name(vport));
1621 nla = nla_reserve(skb, ODP_VPORT_ATTR_STATS, sizeof(struct rtnl_link_stats64));
1623 goto nla_put_failure;
1624 if (vport_get_stats(vport, nla_data(nla)))
1625 __skb_trim(skb, skb->len - nla->nla_len);
1627 NLA_PUT(skb, ODP_VPORT_ATTR_ADDRESS, ETH_ALEN, vport_get_addr(vport));
1629 mtu = vport_get_mtu(vport);
1631 NLA_PUT_U32(skb, ODP_VPORT_ATTR_MTU, mtu);
1633 err = vport_get_options(vport, skb);
1634 if (err == -EMSGSIZE)
1637 ifindex = vport_get_ifindex(vport);
1639 NLA_PUT_U32(skb, ODP_VPORT_ATTR_IFINDEX, ifindex);
1641 iflink = vport_get_iflink(vport);
1643 NLA_PUT_U32(skb, ODP_VPORT_ATTR_IFLINK, iflink);
1645 return genlmsg_end(skb, odp_header);
1650 genlmsg_cancel(skb, odp_header);
1654 /* Called with RTNL lock or RCU read lock. */
1655 static struct sk_buff *odp_vport_cmd_build_info(struct vport *vport, u32 pid,
1658 struct sk_buff *skb;
1661 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1663 return ERR_PTR(-ENOMEM);
1665 retval = odp_vport_cmd_fill_info(vport, skb, pid, seq, 0, cmd);
1668 return ERR_PTR(retval);
1673 static int odp_vport_cmd_validate(struct nlattr *a[ODP_VPORT_ATTR_MAX + 1])
1675 return VERIFY_NUL_STRING(a[ODP_VPORT_ATTR_NAME], IFNAMSIZ - 1);
1678 /* Called with RTNL lock or RCU read lock. */
1679 static struct vport *lookup_vport(struct odp_header *odp_header,
1680 struct nlattr *a[ODP_VPORT_ATTR_MAX + 1])
1682 struct datapath *dp;
1683 struct vport *vport;
1685 if (a[ODP_VPORT_ATTR_NAME]) {
1686 vport = vport_locate(nla_data(a[ODP_VPORT_ATTR_NAME]));
1688 return ERR_PTR(-ENODEV);
1690 } else if (a[ODP_VPORT_ATTR_PORT_NO]) {
1691 u32 port_no = nla_get_u32(a[ODP_VPORT_ATTR_PORT_NO]);
1693 if (port_no >= DP_MAX_PORTS)
1694 return ERR_PTR(-EFBIG);
1696 dp = get_dp(odp_header->dp_ifindex);
1698 return ERR_PTR(-ENODEV);
1700 vport = get_vport_protected(dp, port_no);
1702 return ERR_PTR(-ENOENT);
1705 return ERR_PTR(-EINVAL);
1708 /* Called with RTNL lock. */
1709 static int change_vport(struct vport *vport, struct nlattr *a[ODP_VPORT_ATTR_MAX + 1])
1712 if (a[ODP_VPORT_ATTR_STATS])
1713 err = vport_set_stats(vport, nla_data(a[ODP_VPORT_ATTR_STATS]));
1714 if (!err && a[ODP_VPORT_ATTR_ADDRESS])
1715 err = vport_set_addr(vport, nla_data(a[ODP_VPORT_ATTR_ADDRESS]));
1716 if (!err && a[ODP_VPORT_ATTR_MTU])
1717 err = vport_set_mtu(vport, nla_get_u32(a[ODP_VPORT_ATTR_MTU]));
1721 static int odp_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1723 struct nlattr **a = info->attrs;
1724 struct odp_header *odp_header = info->userhdr;
1725 struct vport_parms parms;
1726 struct sk_buff *reply;
1727 struct vport *vport;
1728 struct datapath *dp;
1733 if (!a[ODP_VPORT_ATTR_NAME] || !a[ODP_VPORT_ATTR_TYPE])
1736 err = odp_vport_cmd_validate(a);
1741 dp = get_dp(odp_header->dp_ifindex);
1746 if (a[ODP_VPORT_ATTR_PORT_NO]) {
1747 port_no = nla_get_u32(a[ODP_VPORT_ATTR_PORT_NO]);
1750 if (port_no >= DP_MAX_PORTS)
1753 vport = get_vport_protected(dp, port_no);
1758 for (port_no = 1; ; port_no++) {
1759 if (port_no >= DP_MAX_PORTS) {
1763 vport = get_vport_protected(dp, port_no);
1769 parms.name = nla_data(a[ODP_VPORT_ATTR_NAME]);
1770 parms.type = nla_get_u32(a[ODP_VPORT_ATTR_TYPE]);
1771 parms.options = a[ODP_VPORT_ATTR_OPTIONS];
1773 parms.port_no = port_no;
1775 vport = new_vport(&parms);
1776 err = PTR_ERR(vport);
1780 set_internal_devs_mtu(dp);
1781 dp_sysfs_add_if(vport);
1783 err = change_vport(vport, a);
1785 reply = odp_vport_cmd_build_info(vport, info->snd_pid,
1786 info->snd_seq, ODP_VPORT_CMD_NEW);
1788 err = PTR_ERR(reply);
1791 dp_detach_port(vport);
1794 genl_notify(reply, genl_info_net(info), info->snd_pid,
1795 dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1804 static int odp_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1806 struct nlattr **a = info->attrs;
1807 struct sk_buff *reply;
1808 struct vport *vport;
1811 err = odp_vport_cmd_validate(a);
1816 vport = lookup_vport(info->userhdr, a);
1817 err = PTR_ERR(vport);
1822 if (a[ODP_VPORT_ATTR_OPTIONS])
1823 err = vport_set_options(vport, a[ODP_VPORT_ATTR_OPTIONS]);
1825 err = change_vport(vport, a);
1827 reply = odp_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1829 if (IS_ERR(reply)) {
1830 err = PTR_ERR(reply);
1831 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1832 dp_vport_multicast_group.id, err);
1836 genl_notify(reply, genl_info_net(info), info->snd_pid,
1837 dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1845 static int odp_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1847 struct nlattr **a = info->attrs;
1848 struct sk_buff *reply;
1849 struct vport *vport;
1852 err = odp_vport_cmd_validate(a);
1857 vport = lookup_vport(info->userhdr, a);
1858 err = PTR_ERR(vport);
1862 if (vport->port_no == ODPP_LOCAL) {
1867 reply = odp_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1869 err = PTR_ERR(reply);
1873 err = dp_detach_port(vport);
1875 genl_notify(reply, genl_info_net(info), info->snd_pid,
1876 dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1884 static int odp_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1886 struct nlattr **a = info->attrs;
1887 struct odp_header *odp_header = info->userhdr;
1888 struct sk_buff *reply;
1889 struct vport *vport;
1892 err = odp_vport_cmd_validate(a);
1897 vport = lookup_vport(odp_header, a);
1898 err = PTR_ERR(vport);
1902 reply = odp_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1904 err = PTR_ERR(reply);
1908 err = genlmsg_reply(reply, info);
1916 static int odp_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1918 struct odp_header *odp_header = genlmsg_data(nlmsg_data(cb->nlh));
1919 struct datapath *dp;
1923 dp = get_dp(odp_header->dp_ifindex);
1928 for (port_no = cb->args[0]; port_no < DP_MAX_PORTS; port_no++) {
1929 struct vport *vport;
1931 vport = get_vport_protected(dp, port_no);
1935 if (odp_vport_cmd_fill_info(vport, skb, NETLINK_CB(cb->skb).pid,
1936 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1937 ODP_VPORT_CMD_NEW) < 0)
1942 cb->args[0] = port_no;
1948 static struct genl_ops dp_vport_genl_ops[] = {
1949 { .cmd = ODP_VPORT_CMD_NEW,
1950 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1951 .policy = vport_policy,
1952 .doit = odp_vport_cmd_new
1954 { .cmd = ODP_VPORT_CMD_DEL,
1955 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1956 .policy = vport_policy,
1957 .doit = odp_vport_cmd_del
1959 { .cmd = ODP_VPORT_CMD_GET,
1960 .flags = 0, /* OK for unprivileged users. */
1961 .policy = vport_policy,
1962 .doit = odp_vport_cmd_get,
1963 .dumpit = odp_vport_cmd_dump
1965 { .cmd = ODP_VPORT_CMD_SET,
1966 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1967 .policy = vport_policy,
1968 .doit = odp_vport_cmd_set,
1972 struct genl_family_and_ops {
1973 struct genl_family *family;
1974 struct genl_ops *ops;
1976 struct genl_multicast_group *group;
1979 static const struct genl_family_and_ops dp_genl_families[] = {
1980 { &dp_datapath_genl_family,
1981 dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1982 &dp_datapath_multicast_group },
1983 { &dp_vport_genl_family,
1984 dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1985 &dp_vport_multicast_group },
1986 { &dp_flow_genl_family,
1987 dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1988 &dp_flow_multicast_group },
1989 { &dp_packet_genl_family,
1990 dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1994 static void dp_unregister_genl(int n_families)
1998 for (i = 0; i < n_families; i++) {
1999 genl_unregister_family(dp_genl_families[i].family);
2003 static int dp_register_genl(void)
2010 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2011 const struct genl_family_and_ops *f = &dp_genl_families[i];
2013 err = genl_register_family_with_ops(f->family, f->ops,
2020 err = genl_register_mc_group(f->family, f->group);
2026 err = packet_register_mc_groups();
2032 dp_unregister_genl(n_registered);
2036 static int __init dp_init(void)
2038 struct sk_buff *dummy_skb;
2041 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
2043 printk("Open vSwitch %s, built "__DATE__" "__TIME__"\n", VERSION BUILDNR);
2051 goto error_flow_exit;
2053 err = register_netdevice_notifier(&dp_device_notifier);
2055 goto error_vport_exit;
2057 err = dp_register_genl();
2059 goto error_unreg_notifier;
2063 error_unreg_notifier:
2064 unregister_netdevice_notifier(&dp_device_notifier);
2073 static void dp_cleanup(void)
2076 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2077 unregister_netdevice_notifier(&dp_device_notifier);
2082 module_init(dp_init);
2083 module_exit(dp_cleanup);
2085 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2086 MODULE_LICENSE("GPL");