2 * Copyright (c) 2007, 2008, 2009, 2010, 2011 Nicira Networks.
3 * Distributed under the terms of the GNU GPL version 2.
5 * Significant portions of this file may be copied from parts of the Linux
6 * kernel, by Linus Torvalds and others.
9 /* Functions for managing the dp interface/device. */
11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
13 #include <linux/init.h>
14 #include <linux/module.h>
15 #include <linux/if_arp.h>
16 #include <linux/if_vlan.h>
19 #include <linux/jhash.h>
20 #include <linux/delay.h>
21 #include <linux/time.h>
22 #include <linux/etherdevice.h>
23 #include <linux/genetlink.h>
24 #include <linux/kernel.h>
25 #include <linux/kthread.h>
26 #include <linux/mutex.h>
27 #include <linux/percpu.h>
28 #include <linux/rcupdate.h>
29 #include <linux/tcp.h>
30 #include <linux/udp.h>
31 #include <linux/version.h>
32 #include <linux/ethtool.h>
33 #include <linux/wait.h>
34 #include <asm/system.h>
35 #include <asm/div64.h>
37 #include <linux/highmem.h>
38 #include <linux/netfilter_bridge.h>
39 #include <linux/netfilter_ipv4.h>
40 #include <linux/inetdevice.h>
41 #include <linux/list.h>
42 #include <linux/rculist.h>
43 #include <linux/dmi.h>
44 #include <net/inet_ecn.h>
45 #include <net/genetlink.h>
47 #include "openvswitch/datapath-protocol.h"
54 #include "vport-internal_dev.h"
56 int (*dp_ioctl_hook)(struct net_device *dev, struct ifreq *rq, int cmd);
57 EXPORT_SYMBOL(dp_ioctl_hook);
62 * Writes to device state (add/remove datapath, port, set operations on vports,
63 * etc.) are protected by RTNL.
65 * Writes to other state (flow table modifications, set miscellaneous datapath
66 * parameters such as drop frags, etc.) are protected by genl_mutex. The RTNL
67 * lock nests inside genl_mutex.
69 * Reads are protected by RCU.
71 * There are a few special cases (mostly stats) that have their own
72 * synchronization but they nest under all of above and don't interact with
76 /* Global list of datapaths to enable dumping them all out.
77 * Protected by genl_mutex.
79 static LIST_HEAD(dps);
81 static struct vport *new_vport(const struct vport_parms *);
82 static int queue_control_packets(struct datapath *, struct sk_buff *,
83 const struct dp_upcall_info *);
85 /* Must be called with rcu_read_lock, genl_mutex, or RTNL lock. */
86 struct datapath *get_dp(int dp_ifindex)
88 struct datapath *dp = NULL;
89 struct net_device *dev;
92 dev = dev_get_by_index_rcu(&init_net, dp_ifindex);
94 struct vport *vport = internal_dev_get_vport(dev);
102 EXPORT_SYMBOL_GPL(get_dp);
104 /* Must be called with genl_mutex. */
105 static struct tbl *get_table_protected(struct datapath *dp)
107 return rcu_dereference_protected(dp->table, lockdep_genl_is_held());
110 /* Must be called with rcu_read_lock or RTNL lock. */
111 static struct vport *get_vport_protected(struct datapath *dp, u16 port_no)
113 return rcu_dereference_rtnl(dp->ports[port_no]);
116 /* Must be called with rcu_read_lock or RTNL lock. */
117 const char *dp_name(const struct datapath *dp)
119 return vport_get_name(rcu_dereference_rtnl(dp->ports[ODPP_LOCAL]));
122 static inline size_t br_nlmsg_size(void)
124 return NLMSG_ALIGN(sizeof(struct ifinfomsg))
125 + nla_total_size(IFNAMSIZ) /* IFLA_IFNAME */
126 + nla_total_size(MAX_ADDR_LEN) /* IFLA_ADDRESS */
127 + nla_total_size(4) /* IFLA_MASTER */
128 + nla_total_size(4) /* IFLA_MTU */
129 + nla_total_size(4) /* IFLA_LINK */
130 + nla_total_size(1); /* IFLA_OPERSTATE */
133 /* Caller must hold RTNL lock. */
134 static int dp_fill_ifinfo(struct sk_buff *skb,
135 const struct vport *port,
136 int event, unsigned int flags)
138 struct datapath *dp = port->dp;
139 int ifindex = vport_get_ifindex(port);
140 int iflink = vport_get_iflink(port);
141 struct ifinfomsg *hdr;
142 struct nlmsghdr *nlh;
150 nlh = nlmsg_put(skb, 0, 0, event, sizeof(*hdr), flags);
154 hdr = nlmsg_data(nlh);
155 hdr->ifi_family = AF_BRIDGE;
157 hdr->ifi_type = ARPHRD_ETHER;
158 hdr->ifi_index = ifindex;
159 hdr->ifi_flags = vport_get_flags(port);
162 NLA_PUT_STRING(skb, IFLA_IFNAME, vport_get_name(port));
163 NLA_PUT_U32(skb, IFLA_MASTER,
164 vport_get_ifindex(get_vport_protected(dp, ODPP_LOCAL)));
165 NLA_PUT_U32(skb, IFLA_MTU, vport_get_mtu(port));
166 #ifdef IFLA_OPERSTATE
167 NLA_PUT_U8(skb, IFLA_OPERSTATE,
168 vport_is_running(port)
169 ? vport_get_operstate(port)
173 NLA_PUT(skb, IFLA_ADDRESS, ETH_ALEN, vport_get_addr(port));
175 if (ifindex != iflink)
176 NLA_PUT_U32(skb, IFLA_LINK,iflink);
178 return nlmsg_end(skb, nlh);
181 nlmsg_cancel(skb, nlh);
185 /* Caller must hold RTNL lock. */
186 static void dp_ifinfo_notify(int event, struct vport *port)
191 skb = nlmsg_new(br_nlmsg_size(), GFP_KERNEL);
195 err = dp_fill_ifinfo(skb, port, event, 0);
197 /* -EMSGSIZE implies BUG in br_nlmsg_size() */
198 WARN_ON(err == -EMSGSIZE);
202 rtnl_notify(skb, &init_net, 0, RTNLGRP_LINK, NULL, GFP_KERNEL);
206 rtnl_set_sk_err(&init_net, RTNLGRP_LINK, err);
209 static void release_dp(struct kobject *kobj)
211 struct datapath *dp = container_of(kobj, struct datapath, ifobj);
215 static struct kobj_type dp_ktype = {
216 .release = release_dp
219 static void destroy_dp_rcu(struct rcu_head *rcu)
221 struct datapath *dp = container_of(rcu, struct datapath, rcu);
223 tbl_destroy((struct tbl __force *)dp->table, flow_free_tbl);
224 free_percpu(dp->stats_percpu);
225 kobject_put(&dp->ifobj);
228 /* Called with RTNL lock and genl_lock. */
229 static struct vport *new_vport(const struct vport_parms *parms)
233 vport = vport_add(parms);
234 if (!IS_ERR(vport)) {
235 struct datapath *dp = parms->dp;
237 rcu_assign_pointer(dp->ports[parms->port_no], vport);
238 list_add(&vport->node, &dp->port_list);
240 dp_ifinfo_notify(RTM_NEWLINK, vport);
246 /* Called with RTNL lock. */
247 int dp_detach_port(struct vport *p)
251 if (p->port_no != ODPP_LOCAL)
253 dp_ifinfo_notify(RTM_DELLINK, p);
255 /* First drop references to device. */
257 rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
259 /* Then destroy it. */
263 /* Must be called with rcu_read_lock. */
264 void dp_process_received_packet(struct vport *p, struct sk_buff *skb)
266 struct datapath *dp = p->dp;
267 struct dp_stats_percpu *stats;
268 int stats_counter_off;
271 OVS_CB(skb)->vport = p;
273 if (!OVS_CB(skb)->flow) {
274 struct sw_flow_key key;
275 struct tbl_node *flow_node;
278 /* Extract flow from 'skb' into 'key'. */
279 error = flow_extract(skb, p->port_no, &key, &is_frag);
280 if (unlikely(error)) {
285 if (is_frag && dp->drop_frags) {
287 stats_counter_off = offsetof(struct dp_stats_percpu, n_frags);
292 flow_node = tbl_lookup(rcu_dereference(dp->table), &key,
293 flow_hash(&key), flow_cmp);
294 if (unlikely(!flow_node)) {
295 struct dp_upcall_info upcall;
297 upcall.cmd = ODP_PACKET_CMD_MISS;
300 upcall.sample_pool = 0;
301 upcall.actions = NULL;
302 upcall.actions_len = 0;
303 dp_upcall(dp, skb, &upcall);
304 stats_counter_off = offsetof(struct dp_stats_percpu, n_missed);
308 OVS_CB(skb)->flow = flow_cast(flow_node);
311 stats_counter_off = offsetof(struct dp_stats_percpu, n_hit);
312 flow_used(OVS_CB(skb)->flow, skb);
313 execute_actions(dp, skb);
316 /* Update datapath statistics. */
318 stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
320 write_seqcount_begin(&stats->seqlock);
321 (*(u64 *)((u8 *)stats + stats_counter_off))++;
322 write_seqcount_end(&stats->seqlock);
327 static void copy_and_csum_skb(struct sk_buff *skb, void *to)
329 u16 csum_start, csum_offset;
332 get_skb_csum_pointers(skb, &csum_start, &csum_offset);
333 csum_start -= skb_headroom(skb);
335 skb_copy_bits(skb, 0, to, csum_start);
337 csum = skb_copy_and_csum_bits(skb, csum_start, to + csum_start,
338 skb->len - csum_start, 0);
339 *(__sum16 *)(to + csum_start + csum_offset) = csum_fold(csum);
342 static struct genl_family dp_packet_genl_family = {
343 .id = GENL_ID_GENERATE,
344 .hdrsize = sizeof(struct odp_header),
345 .name = ODP_PACKET_FAMILY,
347 .maxattr = ODP_PACKET_ATTR_MAX
350 /* Generic Netlink multicast groups for upcalls.
352 * We really want three unique multicast groups per datapath, but we can't even
353 * get one, because genl_register_mc_group() takes genl_lock, which is also
354 * held during Generic Netlink message processing, so trying to acquire
355 * multicast groups during ODP_DP_NEW processing deadlocks. Instead, we
356 * preallocate a few groups and use them round-robin for datapaths. Collision
357 * isn't fatal--multicast listeners should check that the family is the one
358 * that they want and discard others--but it wastes time and memory to receive
361 #define PACKET_N_MC_GROUPS 16
362 static struct genl_multicast_group packet_mc_groups[PACKET_N_MC_GROUPS];
364 static u32 packet_mc_group(struct datapath *dp, u8 cmd)
367 BUILD_BUG_ON_NOT_POWER_OF_2(PACKET_N_MC_GROUPS);
369 idx = jhash_2words(dp->dp_ifindex, cmd, 0) & (PACKET_N_MC_GROUPS - 1);
370 return packet_mc_groups[idx].id;
373 static int packet_register_mc_groups(void)
377 for (i = 0; i < PACKET_N_MC_GROUPS; i++) {
378 struct genl_multicast_group *group = &packet_mc_groups[i];
381 sprintf(group->name, "packet%d", i);
382 error = genl_register_mc_group(&dp_packet_genl_family, group);
389 int dp_upcall(struct datapath *dp, struct sk_buff *skb, const struct dp_upcall_info *upcall_info)
391 struct dp_stats_percpu *stats;
394 WARN_ON_ONCE(skb_shared(skb));
396 forward_ip_summed(skb);
398 err = vswitch_skb_checksum_setup(skb);
402 /* Break apart GSO packets into their component pieces. Otherwise
403 * userspace may try to stuff a 64kB packet into a 1500-byte MTU. */
404 if (skb_is_gso(skb)) {
405 struct sk_buff *nskb = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
415 err = queue_control_packets(dp, skb, upcall_info);
425 stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
427 write_seqcount_begin(&stats->seqlock);
429 write_seqcount_end(&stats->seqlock);
436 /* Send each packet in the 'skb' list to userspace for 'dp' as directed by
437 * 'upcall_info'. There will be only one packet unless we broke up a GSO
440 static int queue_control_packets(struct datapath *dp, struct sk_buff *skb,
441 const struct dp_upcall_info *upcall_info)
443 u32 group = packet_mc_group(dp, upcall_info->cmd);
444 struct sk_buff *nskb;
448 struct odp_header *upcall;
449 struct sk_buff *user_skb; /* to be queued to userspace */
456 err = vlan_deaccel_tag(skb);
460 if (nla_attr_size(skb->len) > USHRT_MAX)
463 len = sizeof(struct odp_header);
464 len += nla_total_size(skb->len);
465 len += nla_total_size(FLOW_BUFSIZE);
466 if (upcall_info->userdata)
467 len += nla_total_size(8);
468 if (upcall_info->sample_pool)
469 len += nla_total_size(4);
470 if (upcall_info->actions_len)
471 len += nla_total_size(upcall_info->actions_len);
473 user_skb = genlmsg_new(len, GFP_ATOMIC);
475 netlink_set_err(INIT_NET_GENL_SOCK, 0, group, -ENOBUFS);
479 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family, 0, upcall_info->cmd);
480 upcall->dp_ifindex = dp->dp_ifindex;
482 nla = nla_nest_start(user_skb, ODP_PACKET_ATTR_KEY);
483 flow_to_nlattrs(upcall_info->key, user_skb);
484 nla_nest_end(user_skb, nla);
486 if (upcall_info->userdata)
487 nla_put_u64(user_skb, ODP_PACKET_ATTR_USERDATA, upcall_info->userdata);
488 if (upcall_info->sample_pool)
489 nla_put_u32(user_skb, ODP_PACKET_ATTR_SAMPLE_POOL, upcall_info->sample_pool);
490 if (upcall_info->actions_len) {
491 const struct nlattr *actions = upcall_info->actions;
492 u32 actions_len = upcall_info->actions_len;
494 nla = nla_nest_start(user_skb, ODP_PACKET_ATTR_ACTIONS);
495 memcpy(__skb_put(user_skb, actions_len), actions, actions_len);
496 nla_nest_end(user_skb, nla);
499 nla = __nla_reserve(user_skb, ODP_PACKET_ATTR_PACKET, skb->len);
500 if (skb->ip_summed == CHECKSUM_PARTIAL)
501 copy_and_csum_skb(skb, nla_data(nla));
503 skb_copy_bits(skb, 0, nla_data(nla), skb->len);
505 err = genlmsg_multicast(user_skb, 0, group, GFP_ATOMIC);
516 while ((skb = nskb) != NULL) {
523 /* Called with genl_mutex. */
524 static int flush_flows(int dp_ifindex)
526 struct tbl *old_table;
527 struct tbl *new_table;
530 dp = get_dp(dp_ifindex);
534 old_table = get_table_protected(dp);
535 new_table = tbl_create(TBL_MIN_BUCKETS);
539 rcu_assign_pointer(dp->table, new_table);
541 tbl_deferred_destroy(old_table, flow_free_tbl);
546 static int validate_actions(const struct nlattr *attr)
548 const struct nlattr *a;
551 nla_for_each_nested(a, attr, rem) {
552 static const u32 action_lens[ODP_ACTION_ATTR_MAX + 1] = {
553 [ODP_ACTION_ATTR_OUTPUT] = 4,
554 [ODP_ACTION_ATTR_CONTROLLER] = 8,
555 [ODP_ACTION_ATTR_SET_DL_TCI] = 2,
556 [ODP_ACTION_ATTR_STRIP_VLAN] = 0,
557 [ODP_ACTION_ATTR_SET_DL_SRC] = ETH_ALEN,
558 [ODP_ACTION_ATTR_SET_DL_DST] = ETH_ALEN,
559 [ODP_ACTION_ATTR_SET_NW_SRC] = 4,
560 [ODP_ACTION_ATTR_SET_NW_DST] = 4,
561 [ODP_ACTION_ATTR_SET_NW_TOS] = 1,
562 [ODP_ACTION_ATTR_SET_TP_SRC] = 2,
563 [ODP_ACTION_ATTR_SET_TP_DST] = 2,
564 [ODP_ACTION_ATTR_SET_TUNNEL] = 8,
565 [ODP_ACTION_ATTR_SET_PRIORITY] = 4,
566 [ODP_ACTION_ATTR_POP_PRIORITY] = 0,
567 [ODP_ACTION_ATTR_DROP_SPOOFED_ARP] = 0,
569 int type = nla_type(a);
571 if (type > ODP_ACTION_ATTR_MAX || nla_len(a) != action_lens[type])
575 case ODP_ACTION_ATTR_UNSPEC:
578 case ODP_ACTION_ATTR_CONTROLLER:
579 case ODP_ACTION_ATTR_STRIP_VLAN:
580 case ODP_ACTION_ATTR_SET_DL_SRC:
581 case ODP_ACTION_ATTR_SET_DL_DST:
582 case ODP_ACTION_ATTR_SET_NW_SRC:
583 case ODP_ACTION_ATTR_SET_NW_DST:
584 case ODP_ACTION_ATTR_SET_TP_SRC:
585 case ODP_ACTION_ATTR_SET_TP_DST:
586 case ODP_ACTION_ATTR_SET_TUNNEL:
587 case ODP_ACTION_ATTR_SET_PRIORITY:
588 case ODP_ACTION_ATTR_POP_PRIORITY:
589 case ODP_ACTION_ATTR_DROP_SPOOFED_ARP:
590 /* No validation needed. */
593 case ODP_ACTION_ATTR_OUTPUT:
594 if (nla_get_u32(a) >= DP_MAX_PORTS)
598 case ODP_ACTION_ATTR_SET_DL_TCI:
599 if (nla_get_be16(a) & htons(VLAN_CFI_MASK))
603 case ODP_ACTION_ATTR_SET_NW_TOS:
604 if (nla_get_u8(a) & INET_ECN_MASK)
618 static void clear_stats(struct sw_flow *flow)
622 flow->packet_count = 0;
623 flow->byte_count = 0;
626 /* Called with genl_mutex. */
627 static int expand_table(struct datapath *dp)
629 struct tbl *old_table = get_table_protected(dp);
630 struct tbl *new_table;
632 new_table = tbl_expand(old_table);
633 if (IS_ERR(new_table))
634 return PTR_ERR(new_table);
636 rcu_assign_pointer(dp->table, new_table);
637 tbl_deferred_destroy(old_table, NULL);
642 static int odp_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
644 struct odp_header *odp_header = info->userhdr;
645 struct nlattr **a = info->attrs;
646 struct sw_flow_actions *acts;
647 struct sk_buff *packet;
648 struct sw_flow *flow;
656 if (!a[ODP_PACKET_ATTR_PACKET] || !a[ODP_PACKET_ATTR_ACTIONS] ||
657 nla_len(a[ODP_PACKET_ATTR_PACKET]) < ETH_HLEN)
660 err = validate_actions(a[ODP_PACKET_ATTR_ACTIONS]);
664 len = nla_len(a[ODP_PACKET_ATTR_PACKET]);
665 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
669 skb_reserve(packet, NET_IP_ALIGN);
671 memcpy(__skb_put(packet, len), nla_data(a[ODP_PACKET_ATTR_PACKET]), len);
673 skb_reset_mac_header(packet);
674 eth = eth_hdr(packet);
676 /* Normally, setting the skb 'protocol' field would be handled by a
677 * call to eth_type_trans(), but it assumes there's a sending
678 * device, which we may not have. */
679 if (ntohs(eth->h_proto) >= 1536)
680 packet->protocol = eth->h_proto;
682 packet->protocol = htons(ETH_P_802_2);
684 /* Build an sw_flow for sending this packet. */
690 err = flow_extract(packet, -1, &flow->key, &is_frag);
693 flow->tbl_node.hash = flow_hash(&flow->key);
695 acts = flow_actions_alloc(a[ODP_PACKET_ATTR_ACTIONS]);
699 rcu_assign_pointer(flow->sf_acts, acts);
701 OVS_CB(packet)->flow = flow;
704 dp = get_dp(odp_header->dp_ifindex);
708 err = execute_actions(dp, packet);
724 static const struct nla_policy packet_policy[ODP_PACKET_ATTR_MAX + 1] = {
725 [ODP_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
726 [ODP_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
729 static struct genl_ops dp_packet_genl_ops[] = {
730 { .cmd = ODP_PACKET_CMD_EXECUTE,
731 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
732 .policy = packet_policy,
733 .doit = odp_packet_cmd_execute
737 static void get_dp_stats(struct datapath *dp, struct odp_stats *stats)
741 stats->n_frags = stats->n_hit = stats->n_missed = stats->n_lost = 0;
742 for_each_possible_cpu(i) {
743 const struct dp_stats_percpu *percpu_stats;
744 struct dp_stats_percpu local_stats;
747 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
750 seqcount = read_seqcount_begin(&percpu_stats->seqlock);
751 local_stats = *percpu_stats;
752 } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount));
754 stats->n_frags += local_stats.n_frags;
755 stats->n_hit += local_stats.n_hit;
756 stats->n_missed += local_stats.n_missed;
757 stats->n_lost += local_stats.n_lost;
761 /* MTU of the dp pseudo-device: ETH_DATA_LEN or the minimum of the ports.
762 * Called with RTNL lock.
764 int dp_min_mtu(const struct datapath *dp)
771 list_for_each_entry (p, &dp->port_list, node) {
774 /* Skip any internal ports, since that's what we're trying to
776 if (is_internal_vport(p))
779 dev_mtu = vport_get_mtu(p);
782 if (!mtu || dev_mtu < mtu)
786 return mtu ? mtu : ETH_DATA_LEN;
789 /* Sets the MTU of all datapath devices to the minimum of the ports
790 * Called with RTNL lock.
792 void set_internal_devs_mtu(const struct datapath *dp)
799 mtu = dp_min_mtu(dp);
801 list_for_each_entry (p, &dp->port_list, node) {
802 if (is_internal_vport(p))
803 vport_set_mtu(p, mtu);
807 static const struct nla_policy flow_policy[ODP_FLOW_ATTR_MAX + 1] = {
808 [ODP_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
809 [ODP_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
810 [ODP_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
813 static struct genl_family dp_flow_genl_family = {
814 .id = GENL_ID_GENERATE,
815 .hdrsize = sizeof(struct odp_header),
816 .name = ODP_FLOW_FAMILY,
818 .maxattr = ODP_FLOW_ATTR_MAX
821 static struct genl_multicast_group dp_flow_multicast_group = {
822 .name = ODP_FLOW_MCGROUP
825 /* Called with genl_lock. */
826 static int odp_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
827 struct sk_buff *skb, u32 pid, u32 seq, u32 flags, u8 cmd)
829 const int skb_orig_len = skb->len;
830 const struct sw_flow_actions *sf_acts;
831 struct odp_flow_stats stats;
832 struct odp_header *odp_header;
838 sf_acts = rcu_dereference_protected(flow->sf_acts,
839 lockdep_genl_is_held());
841 odp_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags, cmd);
845 odp_header->dp_ifindex = dp->dp_ifindex;
847 nla = nla_nest_start(skb, ODP_FLOW_ATTR_KEY);
849 goto nla_put_failure;
850 err = flow_to_nlattrs(&flow->key, skb);
853 nla_nest_end(skb, nla);
855 spin_lock_bh(&flow->lock);
857 stats.n_packets = flow->packet_count;
858 stats.n_bytes = flow->byte_count;
859 tcp_flags = flow->tcp_flags;
860 spin_unlock_bh(&flow->lock);
863 NLA_PUT_U64(skb, ODP_FLOW_ATTR_USED, flow_used_time(used));
866 NLA_PUT(skb, ODP_FLOW_ATTR_STATS, sizeof(struct odp_flow_stats), &stats);
869 NLA_PUT_U8(skb, ODP_FLOW_ATTR_TCP_FLAGS, tcp_flags);
871 /* If ODP_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
872 * this is the first flow to be dumped into 'skb'. This is unusual for
873 * Netlink but individual action lists can be longer than
874 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
875 * The userspace caller can always fetch the actions separately if it
876 * really wants them. (Most userspace callers in fact don't care.)
878 * This can only fail for dump operations because the skb is always
879 * properly sized for single flows.
881 err = nla_put(skb, ODP_FLOW_ATTR_ACTIONS, sf_acts->actions_len,
883 if (err < 0 && skb_orig_len)
886 return genlmsg_end(skb, odp_header);
891 genlmsg_cancel(skb, odp_header);
895 static struct sk_buff *odp_flow_cmd_alloc_info(struct sw_flow *flow)
897 const struct sw_flow_actions *sf_acts;
900 sf_acts = rcu_dereference_protected(flow->sf_acts,
901 lockdep_genl_is_held());
903 len = nla_total_size(FLOW_BUFSIZE); /* ODP_FLOW_ATTR_KEY */
904 len += nla_total_size(sf_acts->actions_len); /* ODP_FLOW_ATTR_ACTIONS */
905 len += nla_total_size(sizeof(struct odp_flow_stats)); /* ODP_FLOW_ATTR_STATS */
906 len += nla_total_size(1); /* ODP_FLOW_ATTR_TCP_FLAGS */
907 len += nla_total_size(8); /* ODP_FLOW_ATTR_USED */
908 return genlmsg_new(NLMSG_ALIGN(sizeof(struct odp_header)) + len, GFP_KERNEL);
911 static struct sk_buff *odp_flow_cmd_build_info(struct sw_flow *flow, struct datapath *dp,
912 u32 pid, u32 seq, u8 cmd)
917 skb = odp_flow_cmd_alloc_info(flow);
919 return ERR_PTR(-ENOMEM);
921 retval = odp_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd);
926 static int odp_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
928 struct nlattr **a = info->attrs;
929 struct odp_header *odp_header = info->userhdr;
930 struct tbl_node *flow_node;
931 struct sw_flow_key key;
932 struct sw_flow *flow;
933 struct sk_buff *reply;
941 if (!a[ODP_FLOW_ATTR_KEY])
943 error = flow_from_nlattrs(&key, a[ODP_FLOW_ATTR_KEY]);
947 /* Validate actions. */
948 if (a[ODP_FLOW_ATTR_ACTIONS]) {
949 error = validate_actions(a[ODP_FLOW_ATTR_ACTIONS]);
952 } else if (info->genlhdr->cmd == ODP_FLOW_CMD_NEW) {
957 dp = get_dp(odp_header->dp_ifindex);
962 hash = flow_hash(&key);
963 table = get_table_protected(dp);
964 flow_node = tbl_lookup(table, &key, hash, flow_cmp);
966 struct sw_flow_actions *acts;
968 /* Bail out if we're not allowed to create a new flow. */
970 if (info->genlhdr->cmd == ODP_FLOW_CMD_SET)
973 /* Expand table, if necessary, to make room. */
974 if (tbl_count(table) >= tbl_n_buckets(table)) {
975 error = expand_table(dp);
978 table = get_table_protected(dp);
984 error = PTR_ERR(flow);
990 /* Obtain actions. */
991 acts = flow_actions_alloc(a[ODP_FLOW_ATTR_ACTIONS]);
992 error = PTR_ERR(acts);
994 goto error_free_flow;
995 rcu_assign_pointer(flow->sf_acts, acts);
997 /* Put flow in bucket. */
998 error = tbl_insert(table, &flow->tbl_node, hash);
1000 goto error_free_flow;
1002 reply = odp_flow_cmd_build_info(flow, dp, info->snd_pid,
1003 info->snd_seq, ODP_FLOW_CMD_NEW);
1005 /* We found a matching flow. */
1006 struct sw_flow_actions *old_acts;
1008 /* Bail out if we're not allowed to modify an existing flow.
1009 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1010 * because Generic Netlink treats the latter as a dump
1011 * request. We also accept NLM_F_EXCL in case that bug ever
1015 if (info->genlhdr->cmd == ODP_FLOW_CMD_NEW &&
1016 info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1019 /* Update actions. */
1020 flow = flow_cast(flow_node);
1021 old_acts = rcu_dereference_protected(flow->sf_acts,
1022 lockdep_genl_is_held());
1023 if (a[ODP_FLOW_ATTR_ACTIONS] &&
1024 (old_acts->actions_len != nla_len(a[ODP_FLOW_ATTR_ACTIONS]) ||
1025 memcmp(old_acts->actions, nla_data(a[ODP_FLOW_ATTR_ACTIONS]),
1026 old_acts->actions_len))) {
1027 struct sw_flow_actions *new_acts;
1029 new_acts = flow_actions_alloc(a[ODP_FLOW_ATTR_ACTIONS]);
1030 error = PTR_ERR(new_acts);
1031 if (IS_ERR(new_acts))
1034 rcu_assign_pointer(flow->sf_acts, new_acts);
1035 flow_deferred_free_acts(old_acts);
1038 reply = odp_flow_cmd_build_info(flow, dp, info->snd_pid,
1039 info->snd_seq, ODP_FLOW_CMD_NEW);
1042 if (a[ODP_FLOW_ATTR_CLEAR]) {
1043 spin_lock_bh(&flow->lock);
1045 spin_unlock_bh(&flow->lock);
1050 genl_notify(reply, genl_info_net(info), info->snd_pid,
1051 dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1053 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1054 dp_flow_multicast_group.id, PTR_ERR(reply));
1063 static int odp_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1065 struct nlattr **a = info->attrs;
1066 struct odp_header *odp_header = info->userhdr;
1067 struct sw_flow_key key;
1068 struct tbl_node *flow_node;
1069 struct sk_buff *reply;
1070 struct sw_flow *flow;
1071 struct datapath *dp;
1075 if (!a[ODP_FLOW_ATTR_KEY])
1077 err = flow_from_nlattrs(&key, a[ODP_FLOW_ATTR_KEY]);
1081 dp = get_dp(odp_header->dp_ifindex);
1085 table = get_table_protected(dp);
1086 flow_node = tbl_lookup(table, &key, flow_hash(&key), flow_cmp);
1090 flow = flow_cast(flow_node);
1091 reply = odp_flow_cmd_build_info(flow, dp, info->snd_pid, info->snd_seq, ODP_FLOW_CMD_NEW);
1093 return PTR_ERR(reply);
1095 return genlmsg_reply(reply, info);
1098 static int odp_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1100 struct nlattr **a = info->attrs;
1101 struct odp_header *odp_header = info->userhdr;
1102 struct sw_flow_key key;
1103 struct tbl_node *flow_node;
1104 struct sk_buff *reply;
1105 struct sw_flow *flow;
1106 struct datapath *dp;
1110 if (!a[ODP_FLOW_ATTR_KEY])
1111 return flush_flows(odp_header->dp_ifindex);
1112 err = flow_from_nlattrs(&key, a[ODP_FLOW_ATTR_KEY]);
1116 dp = get_dp(odp_header->dp_ifindex);
1120 table = get_table_protected(dp);
1121 flow_node = tbl_lookup(table, &key, flow_hash(&key), flow_cmp);
1124 flow = flow_cast(flow_node);
1126 reply = odp_flow_cmd_alloc_info(flow);
1130 err = tbl_remove(table, flow_node);
1136 err = odp_flow_cmd_fill_info(flow, dp, reply, info->snd_pid,
1137 info->snd_seq, 0, ODP_FLOW_CMD_DEL);
1140 flow_deferred_free(flow);
1142 genl_notify(reply, genl_info_net(info), info->snd_pid,
1143 dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1147 static int odp_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1149 struct odp_header *odp_header = genlmsg_data(nlmsg_data(cb->nlh));
1150 struct datapath *dp;
1152 dp = get_dp(odp_header->dp_ifindex);
1157 struct tbl_node *flow_node;
1158 struct sw_flow *flow;
1161 bucket = cb->args[0];
1163 flow_node = tbl_next(get_table_protected(dp), &bucket, &obj);
1167 flow = flow_cast(flow_node);
1168 if (odp_flow_cmd_fill_info(flow, dp, skb, NETLINK_CB(cb->skb).pid,
1169 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1170 ODP_FLOW_CMD_NEW) < 0)
1173 cb->args[0] = bucket;
1179 static struct genl_ops dp_flow_genl_ops[] = {
1180 { .cmd = ODP_FLOW_CMD_NEW,
1181 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1182 .policy = flow_policy,
1183 .doit = odp_flow_cmd_new_or_set
1185 { .cmd = ODP_FLOW_CMD_DEL,
1186 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1187 .policy = flow_policy,
1188 .doit = odp_flow_cmd_del
1190 { .cmd = ODP_FLOW_CMD_GET,
1191 .flags = 0, /* OK for unprivileged users. */
1192 .policy = flow_policy,
1193 .doit = odp_flow_cmd_get,
1194 .dumpit = odp_flow_cmd_dump
1196 { .cmd = ODP_FLOW_CMD_SET,
1197 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1198 .policy = flow_policy,
1199 .doit = odp_flow_cmd_new_or_set,
1203 static const struct nla_policy datapath_policy[ODP_DP_ATTR_MAX + 1] = {
1204 #ifdef HAVE_NLA_NUL_STRING
1205 [ODP_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1207 [ODP_DP_ATTR_IPV4_FRAGS] = { .type = NLA_U32 },
1208 [ODP_DP_ATTR_SAMPLING] = { .type = NLA_U32 },
1211 static struct genl_family dp_datapath_genl_family = {
1212 .id = GENL_ID_GENERATE,
1213 .hdrsize = sizeof(struct odp_header),
1214 .name = ODP_DATAPATH_FAMILY,
1216 .maxattr = ODP_DP_ATTR_MAX
1219 static struct genl_multicast_group dp_datapath_multicast_group = {
1220 .name = ODP_DATAPATH_MCGROUP
1223 static int odp_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1224 u32 pid, u32 seq, u32 flags, u8 cmd)
1226 struct odp_header *odp_header;
1230 odp_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family,
1235 odp_header->dp_ifindex = dp->dp_ifindex;
1238 err = nla_put_string(skb, ODP_DP_ATTR_NAME, dp_name(dp));
1241 goto nla_put_failure;
1243 nla = nla_reserve(skb, ODP_DP_ATTR_STATS, sizeof(struct odp_stats));
1245 goto nla_put_failure;
1246 get_dp_stats(dp, nla_data(nla));
1248 NLA_PUT_U32(skb, ODP_DP_ATTR_IPV4_FRAGS,
1249 dp->drop_frags ? ODP_DP_FRAG_DROP : ODP_DP_FRAG_ZERO);
1251 if (dp->sflow_probability)
1252 NLA_PUT_U32(skb, ODP_DP_ATTR_SAMPLING, dp->sflow_probability);
1254 nla = nla_nest_start(skb, ODP_DP_ATTR_MCGROUPS);
1256 goto nla_put_failure;
1257 NLA_PUT_U32(skb, ODP_PACKET_CMD_MISS, packet_mc_group(dp, ODP_PACKET_CMD_MISS));
1258 NLA_PUT_U32(skb, ODP_PACKET_CMD_ACTION, packet_mc_group(dp, ODP_PACKET_CMD_ACTION));
1259 NLA_PUT_U32(skb, ODP_PACKET_CMD_SAMPLE, packet_mc_group(dp, ODP_PACKET_CMD_SAMPLE));
1260 nla_nest_end(skb, nla);
1262 return genlmsg_end(skb, odp_header);
1265 genlmsg_cancel(skb, odp_header);
1270 static struct sk_buff *odp_dp_cmd_build_info(struct datapath *dp, u32 pid,
1273 struct sk_buff *skb;
1276 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1278 return ERR_PTR(-ENOMEM);
1280 retval = odp_dp_cmd_fill_info(dp, skb, pid, seq, 0, cmd);
1283 return ERR_PTR(retval);
1288 static int odp_dp_cmd_validate(struct nlattr *a[ODP_DP_ATTR_MAX + 1])
1290 if (a[ODP_DP_ATTR_IPV4_FRAGS]) {
1291 u32 frags = nla_get_u32(a[ODP_DP_ATTR_IPV4_FRAGS]);
1293 if (frags != ODP_DP_FRAG_ZERO && frags != ODP_DP_FRAG_DROP)
1297 return CHECK_NUL_STRING(a[ODP_DP_ATTR_NAME], IFNAMSIZ - 1);
1300 /* Called with genl_mutex and optionally with RTNL lock also. */
1301 static struct datapath *lookup_datapath(struct odp_header *odp_header, struct nlattr *a[ODP_DP_ATTR_MAX + 1])
1303 struct datapath *dp;
1305 if (!a[ODP_DP_ATTR_NAME])
1306 dp = get_dp(odp_header->dp_ifindex);
1308 struct vport *vport;
1311 vport = vport_locate(nla_data(a[ODP_DP_ATTR_NAME]));
1312 dp = vport && vport->port_no == ODPP_LOCAL ? vport->dp : NULL;
1315 return dp ? dp : ERR_PTR(-ENODEV);
1318 /* Called with genl_mutex. */
1319 static void change_datapath(struct datapath *dp, struct nlattr *a[ODP_DP_ATTR_MAX + 1])
1321 if (a[ODP_DP_ATTR_IPV4_FRAGS])
1322 dp->drop_frags = nla_get_u32(a[ODP_DP_ATTR_IPV4_FRAGS]) == ODP_DP_FRAG_DROP;
1323 if (a[ODP_DP_ATTR_SAMPLING])
1324 dp->sflow_probability = nla_get_u32(a[ODP_DP_ATTR_SAMPLING]);
1327 static int odp_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1329 struct nlattr **a = info->attrs;
1330 struct vport_parms parms;
1331 struct sk_buff *reply;
1332 struct datapath *dp;
1333 struct vport *vport;
1337 if (!a[ODP_DP_ATTR_NAME])
1340 err = odp_dp_cmd_validate(a);
1346 if (!try_module_get(THIS_MODULE))
1347 goto err_unlock_rtnl;
1350 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1352 goto err_put_module;
1353 INIT_LIST_HEAD(&dp->port_list);
1355 /* Initialize kobject for bridge. This will be added as
1356 * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
1357 dp->ifobj.kset = NULL;
1358 kobject_init(&dp->ifobj, &dp_ktype);
1360 /* Allocate table. */
1362 rcu_assign_pointer(dp->table, tbl_create(TBL_MIN_BUCKETS));
1366 /* Set up our datapath device. */
1367 parms.name = nla_data(a[ODP_DP_ATTR_NAME]);
1368 parms.type = ODP_VPORT_TYPE_INTERNAL;
1369 parms.options = NULL;
1371 parms.port_no = ODPP_LOCAL;
1372 vport = new_vport(&parms);
1373 if (IS_ERR(vport)) {
1374 err = PTR_ERR(vport);
1378 goto err_destroy_table;
1380 dp->dp_ifindex = vport_get_ifindex(vport);
1383 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1384 if (!dp->stats_percpu) {
1386 goto err_destroy_local_port;
1389 change_datapath(dp, a);
1391 reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_NEW);
1392 err = PTR_ERR(reply);
1394 goto err_destroy_local_port;
1396 list_add_tail(&dp->list_node, &dps);
1397 dp_sysfs_add_dp(dp);
1401 genl_notify(reply, genl_info_net(info), info->snd_pid,
1402 dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1405 err_destroy_local_port:
1406 dp_detach_port(get_vport_protected(dp, ODPP_LOCAL));
1408 tbl_destroy(get_table_protected(dp), NULL);
1412 module_put(THIS_MODULE);
1419 static int odp_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1421 struct vport *vport, *next_vport;
1422 struct sk_buff *reply;
1423 struct datapath *dp;
1426 err = odp_dp_cmd_validate(info->attrs);
1431 dp = lookup_datapath(info->userhdr, info->attrs);
1436 reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_DEL);
1437 err = PTR_ERR(reply);
1441 list_for_each_entry_safe (vport, next_vport, &dp->port_list, node)
1442 if (vport->port_no != ODPP_LOCAL)
1443 dp_detach_port(vport);
1445 dp_sysfs_del_dp(dp);
1446 list_del(&dp->list_node);
1447 dp_detach_port(get_vport_protected(dp, ODPP_LOCAL));
1449 /* rtnl_unlock() will wait until all the references to devices that
1450 * are pending unregistration have been dropped. We do it here to
1451 * ensure that any internal devices (which contain DP pointers) are
1452 * fully destroyed before freeing the datapath.
1456 call_rcu(&dp->rcu, destroy_dp_rcu);
1457 module_put(THIS_MODULE);
1459 genl_notify(reply, genl_info_net(info), info->snd_pid,
1460 dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1470 static int odp_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1472 struct sk_buff *reply;
1473 struct datapath *dp;
1476 err = odp_dp_cmd_validate(info->attrs);
1480 dp = lookup_datapath(info->userhdr, info->attrs);
1484 change_datapath(dp, info->attrs);
1486 reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_NEW);
1487 if (IS_ERR(reply)) {
1488 err = PTR_ERR(reply);
1489 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1490 dp_datapath_multicast_group.id, err);
1494 genl_notify(reply, genl_info_net(info), info->snd_pid,
1495 dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1499 static int odp_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1501 struct sk_buff *reply;
1502 struct datapath *dp;
1505 err = odp_dp_cmd_validate(info->attrs);
1509 dp = lookup_datapath(info->userhdr, info->attrs);
1513 reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_NEW);
1515 return PTR_ERR(reply);
1517 return genlmsg_reply(reply, info);
1520 static int odp_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1522 struct datapath *dp;
1523 int skip = cb->args[0];
1526 list_for_each_entry (dp, &dps, list_node) {
1529 if (odp_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).pid,
1530 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1531 ODP_DP_CMD_NEW) < 0)
1541 static struct genl_ops dp_datapath_genl_ops[] = {
1542 { .cmd = ODP_DP_CMD_NEW,
1543 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1544 .policy = datapath_policy,
1545 .doit = odp_dp_cmd_new
1547 { .cmd = ODP_DP_CMD_DEL,
1548 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1549 .policy = datapath_policy,
1550 .doit = odp_dp_cmd_del
1552 { .cmd = ODP_DP_CMD_GET,
1553 .flags = 0, /* OK for unprivileged users. */
1554 .policy = datapath_policy,
1555 .doit = odp_dp_cmd_get,
1556 .dumpit = odp_dp_cmd_dump
1558 { .cmd = ODP_DP_CMD_SET,
1559 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1560 .policy = datapath_policy,
1561 .doit = odp_dp_cmd_set,
1565 static const struct nla_policy vport_policy[ODP_VPORT_ATTR_MAX + 1] = {
1566 #ifdef HAVE_NLA_NUL_STRING
1567 [ODP_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1568 [ODP_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1569 [ODP_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1570 [ODP_VPORT_ATTR_STATS] = { .len = sizeof(struct rtnl_link_stats64) },
1571 [ODP_VPORT_ATTR_ADDRESS] = { .len = ETH_ALEN },
1573 [ODP_VPORT_ATTR_STATS] = { .minlen = sizeof(struct rtnl_link_stats64) },
1574 [ODP_VPORT_ATTR_ADDRESS] = { .minlen = ETH_ALEN },
1576 [ODP_VPORT_ATTR_MTU] = { .type = NLA_U32 },
1577 [ODP_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1580 static struct genl_family dp_vport_genl_family = {
1581 .id = GENL_ID_GENERATE,
1582 .hdrsize = sizeof(struct odp_header),
1583 .name = ODP_VPORT_FAMILY,
1585 .maxattr = ODP_VPORT_ATTR_MAX
1588 static struct genl_multicast_group dp_vport_multicast_group = {
1589 .name = ODP_VPORT_MCGROUP
1592 /* Called with RTNL lock or RCU read lock. */
1593 static int odp_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1594 u32 pid, u32 seq, u32 flags, u8 cmd)
1596 struct odp_header *odp_header;
1598 int ifindex, iflink;
1602 odp_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family,
1607 odp_header->dp_ifindex = vport->dp->dp_ifindex;
1609 NLA_PUT_U32(skb, ODP_VPORT_ATTR_PORT_NO, vport->port_no);
1610 NLA_PUT_U32(skb, ODP_VPORT_ATTR_TYPE, vport_get_type(vport));
1611 NLA_PUT_STRING(skb, ODP_VPORT_ATTR_NAME, vport_get_name(vport));
1613 nla = nla_reserve(skb, ODP_VPORT_ATTR_STATS, sizeof(struct rtnl_link_stats64));
1615 goto nla_put_failure;
1616 if (vport_get_stats(vport, nla_data(nla)))
1617 __skb_trim(skb, skb->len - nla->nla_len);
1619 NLA_PUT(skb, ODP_VPORT_ATTR_ADDRESS, ETH_ALEN, vport_get_addr(vport));
1621 mtu = vport_get_mtu(vport);
1623 NLA_PUT_U32(skb, ODP_VPORT_ATTR_MTU, mtu);
1625 err = vport_get_options(vport, skb);
1626 if (err == -EMSGSIZE)
1629 ifindex = vport_get_ifindex(vport);
1631 NLA_PUT_U32(skb, ODP_VPORT_ATTR_IFINDEX, ifindex);
1633 iflink = vport_get_iflink(vport);
1635 NLA_PUT_U32(skb, ODP_VPORT_ATTR_IFLINK, iflink);
1637 return genlmsg_end(skb, odp_header);
1642 genlmsg_cancel(skb, odp_header);
1646 /* Called with RTNL lock or RCU read lock. */
1647 static struct sk_buff *odp_vport_cmd_build_info(struct vport *vport, u32 pid,
1650 struct sk_buff *skb;
1653 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1655 return ERR_PTR(-ENOMEM);
1657 retval = odp_vport_cmd_fill_info(vport, skb, pid, seq, 0, cmd);
1660 return ERR_PTR(retval);
1665 static int odp_vport_cmd_validate(struct nlattr *a[ODP_VPORT_ATTR_MAX + 1])
1667 return CHECK_NUL_STRING(a[ODP_VPORT_ATTR_NAME], IFNAMSIZ - 1);
1670 /* Called with RTNL lock or RCU read lock. */
1671 static struct vport *lookup_vport(struct odp_header *odp_header,
1672 struct nlattr *a[ODP_VPORT_ATTR_MAX + 1])
1674 struct datapath *dp;
1675 struct vport *vport;
1677 if (a[ODP_VPORT_ATTR_NAME]) {
1678 vport = vport_locate(nla_data(a[ODP_VPORT_ATTR_NAME]));
1680 return ERR_PTR(-ENODEV);
1682 } else if (a[ODP_VPORT_ATTR_PORT_NO]) {
1683 u32 port_no = nla_get_u32(a[ODP_VPORT_ATTR_PORT_NO]);
1685 if (port_no >= DP_MAX_PORTS)
1686 return ERR_PTR(-EFBIG);
1688 dp = get_dp(odp_header->dp_ifindex);
1690 return ERR_PTR(-ENODEV);
1692 vport = get_vport_protected(dp, port_no);
1694 return ERR_PTR(-ENOENT);
1697 return ERR_PTR(-EINVAL);
1700 /* Called with RTNL lock. */
1701 static int change_vport(struct vport *vport, struct nlattr *a[ODP_VPORT_ATTR_MAX + 1])
1704 if (a[ODP_VPORT_ATTR_STATS])
1705 err = vport_set_stats(vport, nla_data(a[ODP_VPORT_ATTR_STATS]));
1706 if (!err && a[ODP_VPORT_ATTR_ADDRESS])
1707 err = vport_set_addr(vport, nla_data(a[ODP_VPORT_ATTR_ADDRESS]));
1708 if (!err && a[ODP_VPORT_ATTR_MTU])
1709 err = vport_set_mtu(vport, nla_get_u32(a[ODP_VPORT_ATTR_MTU]));
1713 static int odp_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1715 struct nlattr **a = info->attrs;
1716 struct odp_header *odp_header = info->userhdr;
1717 struct vport_parms parms;
1718 struct sk_buff *reply;
1719 struct vport *vport;
1720 struct datapath *dp;
1725 if (!a[ODP_VPORT_ATTR_NAME] || !a[ODP_VPORT_ATTR_TYPE])
1728 err = odp_vport_cmd_validate(a);
1733 dp = get_dp(odp_header->dp_ifindex);
1738 if (a[ODP_VPORT_ATTR_PORT_NO]) {
1739 port_no = nla_get_u32(a[ODP_VPORT_ATTR_PORT_NO]);
1742 if (port_no >= DP_MAX_PORTS)
1745 vport = get_vport_protected(dp, port_no);
1750 for (port_no = 1; ; port_no++) {
1751 if (port_no >= DP_MAX_PORTS) {
1755 vport = get_vport_protected(dp, port_no);
1761 parms.name = nla_data(a[ODP_VPORT_ATTR_NAME]);
1762 parms.type = nla_get_u32(a[ODP_VPORT_ATTR_TYPE]);
1763 parms.options = a[ODP_VPORT_ATTR_OPTIONS];
1765 parms.port_no = port_no;
1767 vport = new_vport(&parms);
1768 err = PTR_ERR(vport);
1772 set_internal_devs_mtu(dp);
1773 dp_sysfs_add_if(vport);
1775 err = change_vport(vport, a);
1777 reply = odp_vport_cmd_build_info(vport, info->snd_pid,
1778 info->snd_seq, ODP_VPORT_CMD_NEW);
1780 err = PTR_ERR(reply);
1783 dp_detach_port(vport);
1786 genl_notify(reply, genl_info_net(info), info->snd_pid,
1787 dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1796 static int odp_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1798 struct nlattr **a = info->attrs;
1799 struct sk_buff *reply;
1800 struct vport *vport;
1803 err = odp_vport_cmd_validate(a);
1808 vport = lookup_vport(info->userhdr, a);
1809 err = PTR_ERR(vport);
1814 if (a[ODP_VPORT_ATTR_OPTIONS])
1815 err = vport_set_options(vport, a[ODP_VPORT_ATTR_OPTIONS]);
1817 err = change_vport(vport, a);
1819 reply = odp_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1821 if (IS_ERR(reply)) {
1822 err = PTR_ERR(reply);
1823 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1824 dp_vport_multicast_group.id, err);
1828 genl_notify(reply, genl_info_net(info), info->snd_pid,
1829 dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1837 static int odp_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1839 struct nlattr **a = info->attrs;
1840 struct sk_buff *reply;
1841 struct vport *vport;
1844 err = odp_vport_cmd_validate(a);
1849 vport = lookup_vport(info->userhdr, a);
1850 err = PTR_ERR(vport);
1854 if (vport->port_no == ODPP_LOCAL) {
1859 reply = odp_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1861 err = PTR_ERR(reply);
1865 err = dp_detach_port(vport);
1867 genl_notify(reply, genl_info_net(info), info->snd_pid,
1868 dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1876 static int odp_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1878 struct nlattr **a = info->attrs;
1879 struct odp_header *odp_header = info->userhdr;
1880 struct sk_buff *reply;
1881 struct vport *vport;
1884 err = odp_vport_cmd_validate(a);
1889 vport = lookup_vport(odp_header, a);
1890 err = PTR_ERR(vport);
1894 reply = odp_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1896 err = PTR_ERR(reply);
1900 err = genlmsg_reply(reply, info);
1908 static int odp_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1910 struct odp_header *odp_header = genlmsg_data(nlmsg_data(cb->nlh));
1911 struct datapath *dp;
1915 dp = get_dp(odp_header->dp_ifindex);
1920 for (port_no = cb->args[0]; port_no < DP_MAX_PORTS; port_no++) {
1921 struct vport *vport;
1923 vport = get_vport_protected(dp, port_no);
1927 if (odp_vport_cmd_fill_info(vport, skb, NETLINK_CB(cb->skb).pid,
1928 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1929 ODP_VPORT_CMD_NEW) < 0)
1934 cb->args[0] = port_no;
1940 static struct genl_ops dp_vport_genl_ops[] = {
1941 { .cmd = ODP_VPORT_CMD_NEW,
1942 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1943 .policy = vport_policy,
1944 .doit = odp_vport_cmd_new
1946 { .cmd = ODP_VPORT_CMD_DEL,
1947 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1948 .policy = vport_policy,
1949 .doit = odp_vport_cmd_del
1951 { .cmd = ODP_VPORT_CMD_GET,
1952 .flags = 0, /* OK for unprivileged users. */
1953 .policy = vport_policy,
1954 .doit = odp_vport_cmd_get,
1955 .dumpit = odp_vport_cmd_dump
1957 { .cmd = ODP_VPORT_CMD_SET,
1958 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1959 .policy = vport_policy,
1960 .doit = odp_vport_cmd_set,
1964 struct genl_family_and_ops {
1965 struct genl_family *family;
1966 struct genl_ops *ops;
1968 struct genl_multicast_group *group;
1971 static const struct genl_family_and_ops dp_genl_families[] = {
1972 { &dp_datapath_genl_family,
1973 dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1974 &dp_datapath_multicast_group },
1975 { &dp_vport_genl_family,
1976 dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1977 &dp_vport_multicast_group },
1978 { &dp_flow_genl_family,
1979 dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1980 &dp_flow_multicast_group },
1981 { &dp_packet_genl_family,
1982 dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1986 static void dp_unregister_genl(int n_families)
1990 for (i = 0; i < n_families; i++) {
1991 genl_unregister_family(dp_genl_families[i].family);
1995 static int dp_register_genl(void)
2002 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2003 const struct genl_family_and_ops *f = &dp_genl_families[i];
2005 err = genl_register_family_with_ops(f->family, f->ops,
2012 err = genl_register_mc_group(f->family, f->group);
2018 err = packet_register_mc_groups();
2024 dp_unregister_genl(n_registered);
2028 static int __init dp_init(void)
2030 struct sk_buff *dummy_skb;
2033 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
2035 printk("Open vSwitch %s, built "__DATE__" "__TIME__"\n", VERSION BUILDNR);
2043 goto error_flow_exit;
2045 err = register_netdevice_notifier(&dp_device_notifier);
2047 goto error_vport_exit;
2049 err = dp_register_genl();
2051 goto error_unreg_notifier;
2055 error_unreg_notifier:
2056 unregister_netdevice_notifier(&dp_device_notifier);
2065 static void dp_cleanup(void)
2068 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2069 unregister_netdevice_notifier(&dp_device_notifier);
2074 module_init(dp_init);
2075 module_exit(dp_cleanup);
2077 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2078 MODULE_LICENSE("GPL");