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"
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 /* Protected by genl_mutex. */
77 static struct datapath __rcu *dps[256];
79 static struct vport *new_vport(const struct vport_parms *);
81 /* Must be called with rcu_read_lock, genl_mutex, or RTNL lock. */
82 struct datapath *get_dp(int dp_idx)
84 if (dp_idx < 0 || dp_idx >= ARRAY_SIZE(dps))
87 return rcu_dereference_check(dps[dp_idx], rcu_read_lock_held() ||
88 lockdep_rtnl_is_held() ||
89 lockdep_genl_is_held());
91 EXPORT_SYMBOL_GPL(get_dp);
93 /* Must be called with genl_mutex. */
94 static struct tbl *get_table_protected(struct datapath *dp)
96 return rcu_dereference_protected(dp->table, lockdep_genl_is_held());
99 /* Must be called with rcu_read_lock or RTNL lock. */
100 static struct vport *get_vport_protected(struct datapath *dp, u16 port_no)
102 return rcu_dereference_rtnl(dp->ports[port_no]);
105 /* Must be called with rcu_read_lock or RTNL lock. */
106 const char *dp_name(const struct datapath *dp)
108 return vport_get_name(rcu_dereference_rtnl(dp->ports[ODPP_LOCAL]));
111 static inline size_t br_nlmsg_size(void)
113 return NLMSG_ALIGN(sizeof(struct ifinfomsg))
114 + nla_total_size(IFNAMSIZ) /* IFLA_IFNAME */
115 + nla_total_size(MAX_ADDR_LEN) /* IFLA_ADDRESS */
116 + nla_total_size(4) /* IFLA_MASTER */
117 + nla_total_size(4) /* IFLA_MTU */
118 + nla_total_size(4) /* IFLA_LINK */
119 + nla_total_size(1); /* IFLA_OPERSTATE */
122 /* Caller must hold RTNL lock. */
123 static int dp_fill_ifinfo(struct sk_buff *skb,
124 const struct vport *port,
125 int event, unsigned int flags)
127 struct datapath *dp = port->dp;
128 int ifindex = vport_get_ifindex(port);
129 int iflink = vport_get_iflink(port);
130 struct ifinfomsg *hdr;
131 struct nlmsghdr *nlh;
139 nlh = nlmsg_put(skb, 0, 0, event, sizeof(*hdr), flags);
143 hdr = nlmsg_data(nlh);
144 hdr->ifi_family = AF_BRIDGE;
146 hdr->ifi_type = ARPHRD_ETHER;
147 hdr->ifi_index = ifindex;
148 hdr->ifi_flags = vport_get_flags(port);
151 NLA_PUT_STRING(skb, IFLA_IFNAME, vport_get_name(port));
152 NLA_PUT_U32(skb, IFLA_MASTER,
153 vport_get_ifindex(get_vport_protected(dp, ODPP_LOCAL)));
154 NLA_PUT_U32(skb, IFLA_MTU, vport_get_mtu(port));
155 #ifdef IFLA_OPERSTATE
156 NLA_PUT_U8(skb, IFLA_OPERSTATE,
157 vport_is_running(port)
158 ? vport_get_operstate(port)
162 NLA_PUT(skb, IFLA_ADDRESS, ETH_ALEN, vport_get_addr(port));
164 if (ifindex != iflink)
165 NLA_PUT_U32(skb, IFLA_LINK,iflink);
167 return nlmsg_end(skb, nlh);
170 nlmsg_cancel(skb, nlh);
174 /* Caller must hold RTNL lock. */
175 static void dp_ifinfo_notify(int event, struct vport *port)
180 skb = nlmsg_new(br_nlmsg_size(), GFP_KERNEL);
184 err = dp_fill_ifinfo(skb, port, event, 0);
186 /* -EMSGSIZE implies BUG in br_nlmsg_size() */
187 WARN_ON(err == -EMSGSIZE);
191 rtnl_notify(skb, &init_net, 0, RTNLGRP_LINK, NULL, GFP_KERNEL);
195 rtnl_set_sk_err(&init_net, RTNLGRP_LINK, err);
198 static void release_dp(struct kobject *kobj)
200 struct datapath *dp = container_of(kobj, struct datapath, ifobj);
204 static struct kobj_type dp_ktype = {
205 .release = release_dp
208 static void destroy_dp_rcu(struct rcu_head *rcu)
210 struct datapath *dp = container_of(rcu, struct datapath, rcu);
212 tbl_destroy((struct tbl __force *)dp->table, flow_free_tbl);
213 free_percpu(dp->stats_percpu);
214 kobject_put(&dp->ifobj);
217 /* Called with RTNL lock and genl_lock. */
218 static struct vport *new_vport(const struct vport_parms *parms)
222 vport = vport_add(parms);
223 if (!IS_ERR(vport)) {
224 struct datapath *dp = parms->dp;
226 rcu_assign_pointer(dp->ports[parms->port_no], vport);
227 list_add(&vport->node, &dp->port_list);
229 dp_ifinfo_notify(RTM_NEWLINK, vport);
235 /* Called with RTNL lock. */
236 int dp_detach_port(struct vport *p)
240 if (p->port_no != ODPP_LOCAL)
242 dp_ifinfo_notify(RTM_DELLINK, p);
244 /* First drop references to device. */
246 rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
248 /* Then destroy it. */
252 /* Must be called with rcu_read_lock. */
253 void dp_process_received_packet(struct vport *p, struct sk_buff *skb)
255 struct datapath *dp = p->dp;
256 struct dp_stats_percpu *stats;
257 int stats_counter_off;
258 struct sw_flow_actions *acts;
259 struct loop_counter *loop;
262 OVS_CB(skb)->vport = p;
264 if (!OVS_CB(skb)->flow) {
265 struct sw_flow_key key;
266 struct tbl_node *flow_node;
269 /* Extract flow from 'skb' into 'key'. */
270 error = flow_extract(skb, p->port_no, &key, &is_frag);
271 if (unlikely(error)) {
276 if (is_frag && dp->drop_frags) {
278 stats_counter_off = offsetof(struct dp_stats_percpu, n_frags);
283 flow_node = tbl_lookup(rcu_dereference(dp->table), &key,
284 flow_hash(&key), flow_cmp);
285 if (unlikely(!flow_node)) {
286 struct dp_upcall_info upcall;
288 upcall.cmd = ODP_PACKET_CMD_MISS;
291 upcall.sample_pool = 0;
292 upcall.actions = NULL;
293 upcall.actions_len = 0;
294 dp_upcall(dp, skb, &upcall);
295 stats_counter_off = offsetof(struct dp_stats_percpu, n_missed);
299 OVS_CB(skb)->flow = flow_cast(flow_node);
302 stats_counter_off = offsetof(struct dp_stats_percpu, n_hit);
303 flow_used(OVS_CB(skb)->flow, skb);
305 acts = rcu_dereference(OVS_CB(skb)->flow->sf_acts);
307 /* Check whether we've looped too much. */
308 loop = loop_get_counter();
309 if (unlikely(++loop->count > MAX_LOOPS))
310 loop->looping = true;
311 if (unlikely(loop->looping)) {
312 loop_suppress(dp, acts);
317 /* Execute actions. */
318 execute_actions(dp, skb, &OVS_CB(skb)->flow->key, acts->actions,
321 /* Check whether sub-actions looped too much. */
322 if (unlikely(loop->looping))
323 loop_suppress(dp, acts);
326 /* Decrement loop counter. */
328 loop->looping = false;
332 /* Update datapath statistics. */
334 stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
336 write_seqcount_begin(&stats->seqlock);
337 (*(u64 *)((u8 *)stats + stats_counter_off))++;
338 write_seqcount_end(&stats->seqlock);
343 static void copy_and_csum_skb(struct sk_buff *skb, void *to)
345 u16 csum_start, csum_offset;
348 get_skb_csum_pointers(skb, &csum_start, &csum_offset);
349 csum_start -= skb_headroom(skb);
350 BUG_ON(csum_start >= skb_headlen(skb));
352 skb_copy_bits(skb, 0, to, csum_start);
354 csum = skb_copy_and_csum_bits(skb, csum_start, to + csum_start,
355 skb->len - csum_start, 0);
356 *(__sum16 *)(to + csum_start + csum_offset) = csum_fold(csum);
359 static struct genl_family dp_packet_genl_family;
360 #define PACKET_N_MC_GROUPS 16
362 static int packet_mc_group(struct datapath *dp, u8 cmd)
364 BUILD_BUG_ON_NOT_POWER_OF_2(PACKET_N_MC_GROUPS);
365 return jhash_2words(dp->dp_idx, cmd, 0) & (PACKET_N_MC_GROUPS - 1);
368 /* Send each packet in the 'skb' list to userspace for 'dp' as directed by
369 * 'upcall_info'. There will be only one packet unless we broke up a GSO
372 static int queue_control_packets(struct datapath *dp, struct sk_buff *skb,
373 const struct dp_upcall_info *upcall_info)
375 u32 group = packet_mc_group(dp, upcall_info->cmd);
376 struct sk_buff *nskb;
380 if (OVS_CB(skb)->vport)
381 port_no = OVS_CB(skb)->vport->port_no;
383 port_no = ODPP_LOCAL;
386 struct odp_header *upcall;
387 struct sk_buff *user_skb; /* to be queued to userspace */
394 len = sizeof(struct odp_header);
395 len += nla_total_size(4); /* ODP_PACKET_ATTR_TYPE. */
396 len += nla_total_size(skb->len);
397 len += nla_total_size(FLOW_BUFSIZE);
398 if (upcall_info->userdata)
399 len += nla_total_size(8);
400 if (upcall_info->sample_pool)
401 len += nla_total_size(4);
402 if (upcall_info->actions_len)
403 len += nla_total_size(upcall_info->actions_len);
405 user_skb = genlmsg_new(len, GFP_ATOMIC);
407 netlink_set_err(INIT_NET_GENL_SOCK, 0, group, -ENOBUFS);
411 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family, 0, upcall_info->cmd);
412 upcall->dp_idx = dp->dp_idx;
414 nla = nla_nest_start(user_skb, ODP_PACKET_ATTR_KEY);
415 flow_to_nlattrs(upcall_info->key, user_skb);
416 nla_nest_end(user_skb, nla);
418 if (upcall_info->userdata)
419 nla_put_u64(user_skb, ODP_PACKET_ATTR_USERDATA, upcall_info->userdata);
420 if (upcall_info->sample_pool)
421 nla_put_u32(user_skb, ODP_PACKET_ATTR_SAMPLE_POOL, upcall_info->sample_pool);
422 if (upcall_info->actions_len) {
423 const struct nlattr *actions = upcall_info->actions;
424 u32 actions_len = upcall_info->actions_len;
426 nla = nla_nest_start(user_skb, ODP_PACKET_ATTR_ACTIONS);
427 memcpy(__skb_put(user_skb, actions_len), actions, actions_len);
428 nla_nest_end(user_skb, nla);
431 nla = __nla_reserve(user_skb, ODP_PACKET_ATTR_PACKET, skb->len);
432 if (skb->ip_summed == CHECKSUM_PARTIAL)
433 copy_and_csum_skb(skb, nla_data(nla));
435 skb_copy_bits(skb, 0, nla_data(nla), skb->len);
437 err = genlmsg_multicast(user_skb, 0, group, GFP_ATOMIC);
448 while ((skb = nskb) != NULL) {
455 /* Generic Netlink multicast groups for upcalls.
457 * We really want three unique multicast groups per datapath, but we can't even
458 * get one, because genl_register_mc_group() takes genl_lock, which is also
459 * held during Generic Netlink message processing, so trying to acquire
460 * multicast groups during ODP_DP_NEW processing deadlocks. Instead, we
461 * preallocate a few groups and use them round-robin for datapaths. Collision
462 * isn't fatal--multicast listeners should check that the family is the one
463 * that they want and discard others--but it wastes time and memory to receive
466 static struct genl_multicast_group packet_mc_groups[PACKET_N_MC_GROUPS];
468 static struct genl_family dp_packet_genl_family = {
469 .id = GENL_ID_GENERATE,
470 .hdrsize = sizeof(struct odp_header),
471 .name = ODP_PACKET_FAMILY,
473 .maxattr = ODP_PACKET_ATTR_MAX
476 static int packet_register_mc_groups(void)
480 for (i = 0; i < PACKET_N_MC_GROUPS; i++) {
481 struct genl_multicast_group *group = &packet_mc_groups[i];
484 sprintf(group->name, "packet%d", i);
485 error = genl_register_mc_group(&dp_packet_genl_family, group);
492 int dp_upcall(struct datapath *dp, struct sk_buff *skb, const struct dp_upcall_info *upcall_info)
494 struct dp_stats_percpu *stats;
497 WARN_ON_ONCE(skb_shared(skb));
499 forward_ip_summed(skb);
501 err = vswitch_skb_checksum_setup(skb);
505 /* Break apart GSO packets into their component pieces. Otherwise
506 * userspace may try to stuff a 64kB packet into a 1500-byte MTU. */
507 if (skb_is_gso(skb)) {
508 struct sk_buff *nskb = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
518 return queue_control_packets(dp, skb, upcall_info);
524 stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
526 write_seqcount_begin(&stats->seqlock);
528 write_seqcount_end(&stats->seqlock);
535 /* Called with genl_mutex. */
536 static int flush_flows(int dp_idx)
538 struct tbl *old_table;
539 struct tbl *new_table;
546 old_table = get_table_protected(dp);
547 new_table = tbl_create(TBL_MIN_BUCKETS);
551 rcu_assign_pointer(dp->table, new_table);
553 tbl_deferred_destroy(old_table, flow_free_tbl);
558 static int validate_actions(const struct nlattr *attr)
560 const struct nlattr *a;
563 nla_for_each_nested(a, attr, rem) {
564 static const u32 action_lens[ODPAT_MAX + 1] = {
566 [ODPAT_CONTROLLER] = 8,
567 [ODPAT_SET_DL_TCI] = 2,
568 [ODPAT_STRIP_VLAN] = 0,
569 [ODPAT_SET_DL_SRC] = ETH_ALEN,
570 [ODPAT_SET_DL_DST] = ETH_ALEN,
571 [ODPAT_SET_NW_SRC] = 4,
572 [ODPAT_SET_NW_DST] = 4,
573 [ODPAT_SET_NW_TOS] = 1,
574 [ODPAT_SET_TP_SRC] = 2,
575 [ODPAT_SET_TP_DST] = 2,
576 [ODPAT_SET_TUNNEL] = 8,
577 [ODPAT_SET_PRIORITY] = 4,
578 [ODPAT_POP_PRIORITY] = 0,
579 [ODPAT_DROP_SPOOFED_ARP] = 0,
581 int type = nla_type(a);
583 if (type > ODPAT_MAX || nla_len(a) != action_lens[type])
590 case ODPAT_CONTROLLER:
591 case ODPAT_STRIP_VLAN:
592 case ODPAT_SET_DL_SRC:
593 case ODPAT_SET_DL_DST:
594 case ODPAT_SET_NW_SRC:
595 case ODPAT_SET_NW_DST:
596 case ODPAT_SET_TP_SRC:
597 case ODPAT_SET_TP_DST:
598 case ODPAT_SET_TUNNEL:
599 case ODPAT_SET_PRIORITY:
600 case ODPAT_POP_PRIORITY:
601 case ODPAT_DROP_SPOOFED_ARP:
602 /* No validation needed. */
606 if (nla_get_u32(a) >= DP_MAX_PORTS)
610 case ODPAT_SET_DL_TCI:
611 if (nla_get_be16(a) & htons(VLAN_CFI_MASK))
615 case ODPAT_SET_NW_TOS:
616 if (nla_get_u8(a) & INET_ECN_MASK)
630 static void clear_stats(struct sw_flow *flow)
634 flow->packet_count = 0;
635 flow->byte_count = 0;
638 /* Called with genl_mutex. */
639 static int expand_table(struct datapath *dp)
641 struct tbl *old_table = get_table_protected(dp);
642 struct tbl *new_table;
644 new_table = tbl_expand(old_table);
645 if (IS_ERR(new_table))
646 return PTR_ERR(new_table);
648 rcu_assign_pointer(dp->table, new_table);
649 tbl_deferred_destroy(old_table, NULL);
654 static int odp_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
656 struct odp_header *odp_header = info->userhdr;
657 struct nlattr **a = info->attrs;
658 struct sk_buff *packet;
659 struct sw_flow_key key;
666 if (!a[ODP_PACKET_ATTR_PACKET] || !a[ODP_PACKET_ATTR_ACTIONS] ||
667 nla_len(a[ODP_PACKET_ATTR_PACKET]) < ETH_HLEN)
670 err = validate_actions(a[ODP_PACKET_ATTR_ACTIONS]);
674 packet = skb_clone(skb, GFP_KERNEL);
678 packet->data = nla_data(a[ODP_PACKET_ATTR_PACKET]);
679 packet->len = nla_len(a[ODP_PACKET_ATTR_PACKET]);
681 skb_reset_mac_header(packet);
682 eth = eth_hdr(packet);
684 /* Normally, setting the skb 'protocol' field would be handled by a
685 * call to eth_type_trans(), but it assumes there's a sending
686 * device, which we may not have. */
687 if (ntohs(eth->h_proto) >= 1536)
688 packet->protocol = eth->h_proto;
690 packet->protocol = htons(ETH_P_802_2);
692 err = flow_extract(packet, -1, &key, &is_frag);
697 dp = get_dp(odp_header->dp_idx);
700 err = execute_actions(dp, packet, &key,
701 nla_data(a[ODP_PACKET_ATTR_ACTIONS]),
702 nla_len(a[ODP_PACKET_ATTR_ACTIONS]));
709 static const struct nla_policy packet_policy[ODP_PACKET_ATTR_MAX + 1] = {
710 [ODP_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
711 [ODP_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
714 static struct genl_ops dp_packet_genl_ops[] = {
715 { .cmd = ODP_PACKET_CMD_EXECUTE,
716 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
717 .policy = packet_policy,
718 .doit = odp_packet_cmd_execute
722 static void get_dp_stats(struct datapath *dp, struct odp_stats *stats)
726 stats->n_frags = stats->n_hit = stats->n_missed = stats->n_lost = 0;
727 for_each_possible_cpu(i) {
728 const struct dp_stats_percpu *percpu_stats;
729 struct dp_stats_percpu local_stats;
732 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
735 seqcount = read_seqcount_begin(&percpu_stats->seqlock);
736 local_stats = *percpu_stats;
737 } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount));
739 stats->n_frags += local_stats.n_frags;
740 stats->n_hit += local_stats.n_hit;
741 stats->n_missed += local_stats.n_missed;
742 stats->n_lost += local_stats.n_lost;
746 /* MTU of the dp pseudo-device: ETH_DATA_LEN or the minimum of the ports.
747 * Called with RTNL lock.
749 int dp_min_mtu(const struct datapath *dp)
756 list_for_each_entry (p, &dp->port_list, node) {
759 /* Skip any internal ports, since that's what we're trying to
761 if (is_internal_vport(p))
764 dev_mtu = vport_get_mtu(p);
765 if (!mtu || dev_mtu < mtu)
769 return mtu ? mtu : ETH_DATA_LEN;
772 /* Sets the MTU of all datapath devices to the minimum of the ports
773 * Called with RTNL lock.
775 void set_internal_devs_mtu(const struct datapath *dp)
782 mtu = dp_min_mtu(dp);
784 list_for_each_entry (p, &dp->port_list, node) {
785 if (is_internal_vport(p))
786 vport_set_mtu(p, mtu);
790 static const struct nla_policy flow_policy[ODP_FLOW_ATTR_MAX + 1] = {
791 [ODP_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
792 [ODP_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
793 [ODP_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
796 static struct genl_family dp_flow_genl_family = {
797 .id = GENL_ID_GENERATE,
798 .hdrsize = sizeof(struct odp_header),
799 .name = ODP_FLOW_FAMILY,
801 .maxattr = ODP_FLOW_ATTR_MAX
804 static struct genl_multicast_group dp_flow_multicast_group = {
805 .name = ODP_FLOW_MCGROUP
808 /* Called with genl_lock. */
809 static int odp_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
810 struct sk_buff *skb, u32 pid, u32 seq, u32 flags, u8 cmd)
812 const int skb_orig_len = skb->len;
813 const struct sw_flow_actions *sf_acts;
814 struct odp_flow_stats stats;
815 struct odp_header *odp_header;
822 sf_acts = rcu_dereference_protected(flow->sf_acts,
823 lockdep_genl_is_held());
825 odp_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags, cmd);
829 odp_header->dp_idx = dp->dp_idx;
831 nla = nla_nest_start(skb, ODP_FLOW_ATTR_KEY);
833 goto nla_put_failure;
834 err = flow_to_nlattrs(&flow->key, skb);
837 nla_nest_end(skb, nla);
839 spin_lock_bh(&flow->lock);
841 stats.n_packets = flow->packet_count;
842 stats.n_bytes = flow->byte_count;
843 tcp_flags = flow->tcp_flags;
844 spin_unlock_bh(&flow->lock);
847 NLA_PUT_MSECS(skb, ODP_FLOW_ATTR_USED, used);
850 NLA_PUT(skb, ODP_FLOW_ATTR_STATS, sizeof(struct odp_flow_stats), &stats);
853 NLA_PUT_U8(skb, ODP_FLOW_ATTR_TCP_FLAGS, tcp_flags);
855 /* If ODP_FLOW_ATTR_ACTIONS doesn't fit, and this is the first flow to
856 * be dumped into 'skb', then expand the skb. This is unusual for
857 * Netlink but individual action lists can be longer than a page and
858 * thus entirely undumpable if we didn't do this. */
859 nla_len = nla_total_size(sf_acts->actions_len);
860 if (nla_len > skb_tailroom(skb) && !skb_orig_len) {
861 int hdr_off = (unsigned char *)odp_header - skb->data;
863 err = pskb_expand_head(skb, 0, nla_len - skb_tailroom(skb), GFP_KERNEL);
867 odp_header = (struct odp_header *)(skb->data + hdr_off);
869 nla = nla_nest_start(skb, ODP_FLOW_ATTR_ACTIONS);
870 memcpy(__skb_put(skb, sf_acts->actions_len), sf_acts->actions, sf_acts->actions_len);
871 nla_nest_end(skb, nla);
873 return genlmsg_end(skb, odp_header);
878 genlmsg_cancel(skb, odp_header);
882 static struct sk_buff *odp_flow_cmd_alloc_info(struct sw_flow *flow)
884 const struct sw_flow_actions *sf_acts;
887 sf_acts = rcu_dereference_protected(flow->sf_acts,
888 lockdep_genl_is_held());
890 len = nla_total_size(FLOW_BUFSIZE); /* ODP_FLOW_ATTR_KEY */
891 len += nla_total_size(sf_acts->actions_len); /* ODP_FLOW_ATTR_ACTIONS */
892 len += nla_total_size(sizeof(struct odp_flow_stats)); /* ODP_FLOW_ATTR_STATS */
893 len += nla_total_size(1); /* ODP_FLOW_ATTR_TCP_FLAGS */
894 len += nla_total_size(8); /* ODP_FLOW_ATTR_USED */
895 return genlmsg_new(NLMSG_ALIGN(sizeof(struct odp_header)) + len, GFP_KERNEL);
898 static struct sk_buff *odp_flow_cmd_build_info(struct sw_flow *flow, struct datapath *dp,
899 u32 pid, u32 seq, u8 cmd)
904 skb = odp_flow_cmd_alloc_info(flow);
906 return ERR_PTR(-ENOMEM);
908 retval = odp_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd);
913 static int odp_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
915 struct nlattr **a = info->attrs;
916 struct odp_header *odp_header = info->userhdr;
917 struct tbl_node *flow_node;
918 struct sw_flow_key key;
919 struct sw_flow *flow;
920 struct sk_buff *reply;
928 if (!a[ODP_FLOW_ATTR_KEY])
930 error = flow_from_nlattrs(&key, a[ODP_FLOW_ATTR_KEY]);
934 /* Validate actions. */
935 if (a[ODP_FLOW_ATTR_ACTIONS]) {
936 error = validate_actions(a[ODP_FLOW_ATTR_ACTIONS]);
939 } else if (info->genlhdr->cmd == ODP_FLOW_CMD_NEW) {
944 dp = get_dp(odp_header->dp_idx);
949 hash = flow_hash(&key);
950 table = get_table_protected(dp);
951 flow_node = tbl_lookup(table, &key, hash, flow_cmp);
953 struct sw_flow_actions *acts;
955 /* Bail out if we're not allowed to create a new flow. */
957 if (info->genlhdr->cmd == ODP_FLOW_CMD_SET)
960 /* Expand table, if necessary, to make room. */
961 if (tbl_count(table) >= tbl_n_buckets(table)) {
962 error = expand_table(dp);
965 table = get_table_protected(dp);
971 error = PTR_ERR(flow);
977 /* Obtain actions. */
978 acts = flow_actions_alloc(a[ODP_FLOW_ATTR_ACTIONS]);
979 error = PTR_ERR(acts);
981 goto error_free_flow;
982 rcu_assign_pointer(flow->sf_acts, acts);
984 /* Put flow in bucket. */
985 error = tbl_insert(table, &flow->tbl_node, hash);
987 goto error_free_flow;
989 reply = odp_flow_cmd_build_info(flow, dp, info->snd_pid,
990 info->snd_seq, ODP_FLOW_CMD_NEW);
992 /* We found a matching flow. */
993 struct sw_flow_actions *old_acts;
995 /* Bail out if we're not allowed to modify an existing flow.
996 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
997 * because Generic Netlink treats the latter as a dump
998 * request. We also accept NLM_F_EXCL in case that bug ever
1002 if (info->genlhdr->cmd == ODP_FLOW_CMD_NEW &&
1003 info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1006 /* Update actions. */
1007 flow = flow_cast(flow_node);
1008 old_acts = rcu_dereference_protected(flow->sf_acts,
1009 lockdep_genl_is_held());
1010 if (a[ODP_FLOW_ATTR_ACTIONS] &&
1011 (old_acts->actions_len != nla_len(a[ODP_FLOW_ATTR_ACTIONS]) ||
1012 memcmp(old_acts->actions, nla_data(a[ODP_FLOW_ATTR_ACTIONS]),
1013 old_acts->actions_len))) {
1014 struct sw_flow_actions *new_acts;
1016 new_acts = flow_actions_alloc(a[ODP_FLOW_ATTR_ACTIONS]);
1017 error = PTR_ERR(new_acts);
1018 if (IS_ERR(new_acts))
1021 rcu_assign_pointer(flow->sf_acts, new_acts);
1022 flow_deferred_free_acts(old_acts);
1025 reply = odp_flow_cmd_build_info(flow, dp, info->snd_pid,
1026 info->snd_seq, ODP_FLOW_CMD_NEW);
1029 if (a[ODP_FLOW_ATTR_CLEAR]) {
1030 spin_lock_bh(&flow->lock);
1032 spin_unlock_bh(&flow->lock);
1037 genl_notify(reply, genl_info_net(info), info->snd_pid,
1038 dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1040 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1041 dp_flow_multicast_group.id, PTR_ERR(reply));
1050 static int odp_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1052 struct nlattr **a = info->attrs;
1053 struct odp_header *odp_header = info->userhdr;
1054 struct sw_flow_key key;
1055 struct tbl_node *flow_node;
1056 struct sk_buff *reply;
1057 struct sw_flow *flow;
1058 struct datapath *dp;
1062 if (!a[ODP_FLOW_ATTR_KEY])
1064 err = flow_from_nlattrs(&key, a[ODP_FLOW_ATTR_KEY]);
1068 dp = get_dp(odp_header->dp_idx);
1072 table = get_table_protected(dp);
1073 flow_node = tbl_lookup(table, &key, flow_hash(&key), flow_cmp);
1077 flow = flow_cast(flow_node);
1078 reply = odp_flow_cmd_build_info(flow, dp, info->snd_pid, info->snd_seq, ODP_FLOW_CMD_NEW);
1080 return PTR_ERR(reply);
1082 return genlmsg_reply(reply, info);
1085 static int odp_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1087 struct nlattr **a = info->attrs;
1088 struct odp_header *odp_header = info->userhdr;
1089 struct sw_flow_key key;
1090 struct tbl_node *flow_node;
1091 struct sk_buff *reply;
1092 struct sw_flow *flow;
1093 struct datapath *dp;
1097 if (!a[ODP_FLOW_ATTR_KEY])
1098 return flush_flows(odp_header->dp_idx);
1099 err = flow_from_nlattrs(&key, a[ODP_FLOW_ATTR_KEY]);
1103 dp = get_dp(odp_header->dp_idx);
1107 table = get_table_protected(dp);
1108 flow_node = tbl_lookup(table, &key, flow_hash(&key), flow_cmp);
1111 flow = flow_cast(flow_node);
1113 reply = odp_flow_cmd_alloc_info(flow);
1117 err = tbl_remove(table, flow_node);
1123 err = odp_flow_cmd_fill_info(flow, dp, reply, info->snd_pid,
1124 info->snd_seq, 0, ODP_FLOW_CMD_DEL);
1127 flow_deferred_free(flow);
1129 genl_notify(reply, genl_info_net(info), info->snd_pid,
1130 dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1134 static int odp_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1136 struct odp_header *odp_header = genlmsg_data(nlmsg_data(cb->nlh));
1137 struct datapath *dp;
1139 dp = get_dp(odp_header->dp_idx);
1144 struct tbl_node *flow_node;
1145 struct sw_flow *flow;
1148 bucket = cb->args[0];
1150 flow_node = tbl_next(get_table_protected(dp), &bucket, &obj);
1154 flow = flow_cast(flow_node);
1155 if (odp_flow_cmd_fill_info(flow, dp, skb, NETLINK_CB(cb->skb).pid,
1156 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1157 ODP_FLOW_CMD_NEW) < 0)
1160 cb->args[0] = bucket;
1166 static struct genl_ops dp_flow_genl_ops[] = {
1167 { .cmd = ODP_FLOW_CMD_NEW,
1168 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1169 .policy = flow_policy,
1170 .doit = odp_flow_cmd_new_or_set
1172 { .cmd = ODP_FLOW_CMD_DEL,
1173 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1174 .policy = flow_policy,
1175 .doit = odp_flow_cmd_del
1177 { .cmd = ODP_FLOW_CMD_GET,
1178 .flags = 0, /* OK for unprivileged users. */
1179 .policy = flow_policy,
1180 .doit = odp_flow_cmd_get,
1181 .dumpit = odp_flow_cmd_dump
1183 { .cmd = ODP_FLOW_CMD_SET,
1184 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1185 .policy = flow_policy,
1186 .doit = odp_flow_cmd_new_or_set,
1190 static const struct nla_policy datapath_policy[ODP_DP_ATTR_MAX + 1] = {
1191 #ifdef HAVE_NLA_NUL_STRING
1192 [ODP_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1194 [ODP_DP_ATTR_IPV4_FRAGS] = { .type = NLA_U32 },
1195 [ODP_DP_ATTR_SAMPLING] = { .type = NLA_U32 },
1198 static struct genl_family dp_datapath_genl_family = {
1199 .id = GENL_ID_GENERATE,
1200 .hdrsize = sizeof(struct odp_header),
1201 .name = ODP_DATAPATH_FAMILY,
1203 .maxattr = ODP_DP_ATTR_MAX
1206 static struct genl_multicast_group dp_datapath_multicast_group = {
1207 .name = ODP_DATAPATH_MCGROUP
1210 static int odp_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1211 u32 pid, u32 seq, u32 flags, u8 cmd)
1213 struct odp_header *odp_header;
1217 odp_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family,
1222 odp_header->dp_idx = dp->dp_idx;
1225 err = nla_put_string(skb, ODP_DP_ATTR_NAME, dp_name(dp));
1228 goto nla_put_failure;
1230 nla = nla_reserve(skb, ODP_DP_ATTR_STATS, sizeof(struct odp_stats));
1232 goto nla_put_failure;
1233 get_dp_stats(dp, nla_data(nla));
1235 NLA_PUT_U32(skb, ODP_DP_ATTR_IPV4_FRAGS,
1236 dp->drop_frags ? ODP_DP_FRAG_DROP : ODP_DP_FRAG_ZERO);
1238 if (dp->sflow_probability)
1239 NLA_PUT_U32(skb, ODP_DP_ATTR_SAMPLING, dp->sflow_probability);
1241 nla = nla_nest_start(skb, ODP_DP_ATTR_MCGROUPS);
1243 goto nla_put_failure;
1244 NLA_PUT_U32(skb, ODP_PACKET_CMD_MISS, packet_mc_group(dp, ODP_PACKET_CMD_MISS));
1245 NLA_PUT_U32(skb, ODP_PACKET_CMD_ACTION, packet_mc_group(dp, ODP_PACKET_CMD_ACTION));
1246 NLA_PUT_U32(skb, ODP_PACKET_CMD_SAMPLE, packet_mc_group(dp, ODP_PACKET_CMD_SAMPLE));
1247 nla_nest_end(skb, nla);
1249 return genlmsg_end(skb, odp_header);
1252 genlmsg_cancel(skb, odp_header);
1257 static struct sk_buff *odp_dp_cmd_build_info(struct datapath *dp, u32 pid,
1260 struct sk_buff *skb;
1263 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1265 return ERR_PTR(-ENOMEM);
1267 retval = odp_dp_cmd_fill_info(dp, skb, pid, seq, 0, cmd);
1270 return ERR_PTR(retval);
1275 static int odp_dp_cmd_validate(struct nlattr *a[ODP_DP_ATTR_MAX + 1])
1277 if (a[ODP_DP_ATTR_IPV4_FRAGS]) {
1278 u32 frags = nla_get_u32(a[ODP_DP_ATTR_IPV4_FRAGS]);
1280 if (frags != ODP_DP_FRAG_ZERO && frags != ODP_DP_FRAG_DROP)
1284 return VERIFY_NUL_STRING(a[ODP_DP_ATTR_NAME], IFNAMSIZ - 1);
1287 /* Called with genl_mutex and optionally with RTNL lock also. */
1288 static struct datapath *lookup_datapath(struct odp_header *odp_header, struct nlattr *a[ODP_DP_ATTR_MAX + 1])
1290 if (!a[ODP_DP_ATTR_NAME]) {
1291 struct datapath *dp = get_dp(odp_header->dp_idx);
1293 return ERR_PTR(-ENODEV);
1296 struct vport *vport;
1300 vport = vport_locate(nla_data(a[ODP_DP_ATTR_NAME]));
1301 dp_idx = vport && vport->port_no == ODPP_LOCAL ? vport->dp->dp_idx : -1;
1305 return ERR_PTR(-ENODEV);
1310 /* Called with genl_mutex. */
1311 static void change_datapath(struct datapath *dp, struct nlattr *a[ODP_DP_ATTR_MAX + 1])
1313 if (a[ODP_DP_ATTR_IPV4_FRAGS])
1314 dp->drop_frags = nla_get_u32(a[ODP_DP_ATTR_IPV4_FRAGS]) == ODP_DP_FRAG_DROP;
1315 if (a[ODP_DP_ATTR_SAMPLING])
1316 dp->sflow_probability = nla_get_u32(a[ODP_DP_ATTR_SAMPLING]);
1319 static int odp_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1321 struct nlattr **a = info->attrs;
1322 struct odp_header *odp_header = info->userhdr;
1323 struct vport_parms parms;
1324 struct sk_buff *reply;
1325 struct datapath *dp;
1326 struct vport *vport;
1331 if (!a[ODP_DP_ATTR_NAME])
1334 err = odp_dp_cmd_validate(a);
1340 if (!try_module_get(THIS_MODULE))
1341 goto err_unlock_rtnl;
1343 dp_idx = odp_header->dp_idx;
1346 for (dp_idx = 0; dp_idx < ARRAY_SIZE(dps); dp_idx++) {
1352 } else if (dp_idx < ARRAY_SIZE(dps))
1353 err = get_dp(dp_idx) ? -EBUSY : 0;
1357 goto err_put_module;
1360 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1362 goto err_put_module;
1363 INIT_LIST_HEAD(&dp->port_list);
1364 dp->dp_idx = dp_idx;
1366 /* Initialize kobject for bridge. This will be added as
1367 * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
1368 dp->ifobj.kset = NULL;
1369 kobject_init(&dp->ifobj, &dp_ktype);
1371 /* Allocate table. */
1373 rcu_assign_pointer(dp->table, tbl_create(TBL_MIN_BUCKETS));
1377 /* Set up our datapath device. */
1378 parms.name = nla_data(a[ODP_DP_ATTR_NAME]);
1379 parms.type = ODP_VPORT_TYPE_INTERNAL;
1380 parms.options = NULL;
1382 parms.port_no = ODPP_LOCAL;
1383 vport = new_vport(&parms);
1384 if (IS_ERR(vport)) {
1385 err = PTR_ERR(vport);
1389 goto err_destroy_table;
1393 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1394 if (!dp->stats_percpu) {
1396 goto err_destroy_local_port;
1399 change_datapath(dp, a);
1401 reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_NEW);
1402 err = PTR_ERR(reply);
1404 goto err_destroy_local_port;
1406 rcu_assign_pointer(dps[dp_idx], dp);
1407 dp_sysfs_add_dp(dp);
1411 genl_notify(reply, genl_info_net(info), info->snd_pid,
1412 dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1415 err_destroy_local_port:
1416 dp_detach_port(get_vport_protected(dp, ODPP_LOCAL));
1418 tbl_destroy(get_table_protected(dp), NULL);
1422 module_put(THIS_MODULE);
1429 static int odp_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1431 struct vport *vport, *next_vport;
1432 struct sk_buff *reply;
1433 struct datapath *dp;
1436 err = odp_dp_cmd_validate(info->attrs);
1441 dp = lookup_datapath(info->userhdr, info->attrs);
1446 reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_DEL);
1447 err = PTR_ERR(reply);
1451 list_for_each_entry_safe (vport, next_vport, &dp->port_list, node)
1452 if (vport->port_no != ODPP_LOCAL)
1453 dp_detach_port(vport);
1455 dp_sysfs_del_dp(dp);
1456 rcu_assign_pointer(dps[dp->dp_idx], NULL);
1457 dp_detach_port(get_vport_protected(dp, ODPP_LOCAL));
1459 call_rcu(&dp->rcu, destroy_dp_rcu);
1460 module_put(THIS_MODULE);
1462 genl_notify(reply, genl_info_net(info), info->snd_pid,
1463 dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1472 static int odp_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1474 struct sk_buff *reply;
1475 struct datapath *dp;
1478 err = odp_dp_cmd_validate(info->attrs);
1482 dp = lookup_datapath(info->userhdr, info->attrs);
1486 change_datapath(dp, info->attrs);
1488 reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_NEW);
1489 if (IS_ERR(reply)) {
1490 err = PTR_ERR(reply);
1491 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1492 dp_datapath_multicast_group.id, err);
1496 genl_notify(reply, genl_info_net(info), info->snd_pid,
1497 dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1501 static int odp_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1503 struct sk_buff *reply;
1504 struct datapath *dp;
1507 err = odp_dp_cmd_validate(info->attrs);
1511 dp = lookup_datapath(info->userhdr, info->attrs);
1515 reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_NEW);
1517 return PTR_ERR(reply);
1519 return genlmsg_reply(reply, info);
1522 static int odp_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1526 for (dp_idx = cb->args[0]; dp_idx < ARRAY_SIZE(dps); dp_idx++) {
1527 struct datapath *dp = get_dp(dp_idx);
1530 if (odp_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).pid,
1531 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1532 ODP_DP_CMD_NEW) < 0)
1536 cb->args[0] = dp_idx;
1540 static struct genl_ops dp_datapath_genl_ops[] = {
1541 { .cmd = ODP_DP_CMD_NEW,
1542 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1543 .policy = datapath_policy,
1544 .doit = odp_dp_cmd_new
1546 { .cmd = ODP_DP_CMD_DEL,
1547 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1548 .policy = datapath_policy,
1549 .doit = odp_dp_cmd_del
1551 { .cmd = ODP_DP_CMD_GET,
1552 .flags = 0, /* OK for unprivileged users. */
1553 .policy = datapath_policy,
1554 .doit = odp_dp_cmd_get,
1555 .dumpit = odp_dp_cmd_dump
1557 { .cmd = ODP_DP_CMD_SET,
1558 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1559 .policy = datapath_policy,
1560 .doit = odp_dp_cmd_set,
1564 static const struct nla_policy vport_policy[ODP_VPORT_ATTR_MAX + 1] = {
1565 #ifdef HAVE_NLA_NUL_STRING
1566 [ODP_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1567 [ODP_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1568 [ODP_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1569 [ODP_VPORT_ATTR_STATS] = { .len = sizeof(struct rtnl_link_stats64) },
1570 [ODP_VPORT_ATTR_ADDRESS] = { .len = ETH_ALEN },
1572 [ODP_VPORT_ATTR_STATS] = { .minlen = sizeof(struct rtnl_link_stats64) },
1573 [ODP_VPORT_ATTR_ADDRESS] = { .minlen = ETH_ALEN },
1575 [ODP_VPORT_ATTR_MTU] = { .type = NLA_U32 },
1576 [ODP_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1579 static struct genl_family dp_vport_genl_family = {
1580 .id = GENL_ID_GENERATE,
1581 .hdrsize = sizeof(struct odp_header),
1582 .name = ODP_VPORT_FAMILY,
1584 .maxattr = ODP_VPORT_ATTR_MAX
1587 static struct genl_multicast_group dp_vport_multicast_group = {
1588 .name = ODP_VPORT_MCGROUP
1591 /* Called with RTNL lock or RCU read lock. */
1592 static int odp_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1593 u32 pid, u32 seq, u32 flags, u8 cmd)
1595 struct odp_header *odp_header;
1597 int ifindex, iflink;
1600 odp_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family,
1605 odp_header->dp_idx = vport->dp->dp_idx;
1607 NLA_PUT_U32(skb, ODP_VPORT_ATTR_PORT_NO, vport->port_no);
1608 NLA_PUT_U32(skb, ODP_VPORT_ATTR_TYPE, vport_get_type(vport));
1609 NLA_PUT_STRING(skb, ODP_VPORT_ATTR_NAME, vport_get_name(vport));
1611 nla = nla_reserve(skb, ODP_VPORT_ATTR_STATS, sizeof(struct rtnl_link_stats64));
1613 goto nla_put_failure;
1614 if (vport_get_stats(vport, nla_data(nla)))
1615 __skb_trim(skb, skb->len - nla->nla_len);
1617 NLA_PUT(skb, ODP_VPORT_ATTR_ADDRESS, ETH_ALEN, vport_get_addr(vport));
1619 NLA_PUT_U32(skb, ODP_VPORT_ATTR_MTU, vport_get_mtu(vport));
1621 err = vport_get_options(vport, skb);
1622 if (err == -EMSGSIZE)
1625 ifindex = vport_get_ifindex(vport);
1627 NLA_PUT_U32(skb, ODP_VPORT_ATTR_IFINDEX, ifindex);
1629 iflink = vport_get_iflink(vport);
1631 NLA_PUT_U32(skb, ODP_VPORT_ATTR_IFLINK, iflink);
1633 return genlmsg_end(skb, odp_header);
1638 genlmsg_cancel(skb, odp_header);
1642 /* Called with RTNL lock or RCU read lock. */
1643 static struct sk_buff *odp_vport_cmd_build_info(struct vport *vport, u32 pid,
1646 struct sk_buff *skb;
1649 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1651 return ERR_PTR(-ENOMEM);
1653 retval = odp_vport_cmd_fill_info(vport, skb, pid, seq, 0, cmd);
1656 return ERR_PTR(retval);
1661 static int odp_vport_cmd_validate(struct nlattr *a[ODP_VPORT_ATTR_MAX + 1])
1663 return VERIFY_NUL_STRING(a[ODP_VPORT_ATTR_NAME], IFNAMSIZ - 1);
1666 /* Called with RTNL lock or RCU read lock. */
1667 static struct vport *lookup_vport(struct odp_header *odp_header,
1668 struct nlattr *a[ODP_VPORT_ATTR_MAX + 1])
1670 struct datapath *dp;
1671 struct vport *vport;
1673 if (a[ODP_VPORT_ATTR_NAME]) {
1674 vport = vport_locate(nla_data(a[ODP_VPORT_ATTR_NAME]));
1676 return ERR_PTR(-ENODEV);
1678 } else if (a[ODP_VPORT_ATTR_PORT_NO]) {
1679 u32 port_no = nla_get_u32(a[ODP_VPORT_ATTR_PORT_NO]);
1681 if (port_no >= DP_MAX_PORTS)
1682 return ERR_PTR(-EFBIG);
1684 dp = get_dp(odp_header->dp_idx);
1686 return ERR_PTR(-ENODEV);
1688 vport = get_vport_protected(dp, port_no);
1690 return ERR_PTR(-ENOENT);
1693 return ERR_PTR(-EINVAL);
1696 /* Called with RTNL lock. */
1697 static int change_vport(struct vport *vport, struct nlattr *a[ODP_VPORT_ATTR_MAX + 1])
1700 if (a[ODP_VPORT_ATTR_STATS])
1701 err = vport_set_stats(vport, nla_data(a[ODP_VPORT_ATTR_STATS]));
1702 if (!err && a[ODP_VPORT_ATTR_ADDRESS])
1703 err = vport_set_addr(vport, nla_data(a[ODP_VPORT_ATTR_ADDRESS]));
1704 if (!err && a[ODP_VPORT_ATTR_MTU])
1705 err = vport_set_mtu(vport, nla_get_u32(a[ODP_VPORT_ATTR_MTU]));
1709 static int odp_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1711 struct nlattr **a = info->attrs;
1712 struct odp_header *odp_header = info->userhdr;
1713 struct vport_parms parms;
1714 struct sk_buff *reply;
1715 struct vport *vport;
1716 struct datapath *dp;
1721 if (!a[ODP_VPORT_ATTR_NAME] || !a[ODP_VPORT_ATTR_TYPE])
1724 err = odp_vport_cmd_validate(a);
1729 dp = get_dp(odp_header->dp_idx);
1734 if (a[ODP_VPORT_ATTR_PORT_NO]) {
1735 port_no = nla_get_u32(a[ODP_VPORT_ATTR_PORT_NO]);
1738 if (port_no >= DP_MAX_PORTS)
1741 vport = get_vport_protected(dp, port_no);
1746 for (port_no = 1; ; port_no++) {
1747 if (port_no >= DP_MAX_PORTS) {
1751 vport = get_vport_protected(dp, port_no);
1757 parms.name = nla_data(a[ODP_VPORT_ATTR_NAME]);
1758 parms.type = nla_get_u32(a[ODP_VPORT_ATTR_TYPE]);
1759 parms.options = a[ODP_VPORT_ATTR_OPTIONS];
1761 parms.port_no = port_no;
1763 vport = new_vport(&parms);
1764 err = PTR_ERR(vport);
1768 set_internal_devs_mtu(dp);
1769 dp_sysfs_add_if(vport);
1771 err = change_vport(vport, a);
1773 reply = odp_vport_cmd_build_info(vport, info->snd_pid,
1774 info->snd_seq, ODP_VPORT_CMD_NEW);
1776 err = PTR_ERR(reply);
1779 dp_detach_port(vport);
1782 genl_notify(reply, genl_info_net(info), info->snd_pid,
1783 dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1792 static int odp_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1794 struct nlattr **a = info->attrs;
1795 struct sk_buff *reply;
1796 struct vport *vport;
1799 err = odp_vport_cmd_validate(a);
1804 vport = lookup_vport(info->userhdr, a);
1805 err = PTR_ERR(vport);
1810 if (a[ODP_VPORT_ATTR_OPTIONS])
1811 err = vport_set_options(vport, a[ODP_VPORT_ATTR_OPTIONS]);
1813 err = change_vport(vport, a);
1815 reply = odp_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1817 if (IS_ERR(reply)) {
1818 err = PTR_ERR(reply);
1819 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1820 dp_vport_multicast_group.id, err);
1824 genl_notify(reply, genl_info_net(info), info->snd_pid,
1825 dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1833 static int odp_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1835 struct nlattr **a = info->attrs;
1836 struct sk_buff *reply;
1837 struct vport *vport;
1840 err = odp_vport_cmd_validate(a);
1845 vport = lookup_vport(info->userhdr, a);
1846 err = PTR_ERR(vport);
1850 if (vport->port_no == ODPP_LOCAL) {
1855 reply = odp_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1857 err = PTR_ERR(reply);
1861 err = dp_detach_port(vport);
1863 genl_notify(reply, genl_info_net(info), info->snd_pid,
1864 dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1872 static int odp_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1874 struct nlattr **a = info->attrs;
1875 struct odp_header *odp_header = info->userhdr;
1876 struct sk_buff *reply;
1877 struct vport *vport;
1880 err = odp_vport_cmd_validate(a);
1885 vport = lookup_vport(odp_header, a);
1886 err = PTR_ERR(vport);
1890 reply = odp_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1892 err = PTR_ERR(reply);
1896 err = genlmsg_reply(reply, info);
1904 static int odp_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1906 struct odp_header *odp_header = genlmsg_data(nlmsg_data(cb->nlh));
1907 struct datapath *dp;
1911 dp = get_dp(odp_header->dp_idx);
1916 for (port_no = cb->args[0]; port_no < DP_MAX_PORTS; port_no++) {
1917 struct vport *vport;
1919 vport = get_vport_protected(dp, port_no);
1923 if (odp_vport_cmd_fill_info(vport, skb, NETLINK_CB(cb->skb).pid,
1924 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1925 ODP_VPORT_CMD_NEW) < 0)
1930 cb->args[0] = port_no;
1936 static struct genl_ops dp_vport_genl_ops[] = {
1937 { .cmd = ODP_VPORT_CMD_NEW,
1938 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1939 .policy = vport_policy,
1940 .doit = odp_vport_cmd_new
1942 { .cmd = ODP_VPORT_CMD_DEL,
1943 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1944 .policy = vport_policy,
1945 .doit = odp_vport_cmd_del
1947 { .cmd = ODP_VPORT_CMD_GET,
1948 .flags = 0, /* OK for unprivileged users. */
1949 .policy = vport_policy,
1950 .doit = odp_vport_cmd_get,
1951 .dumpit = odp_vport_cmd_dump
1953 { .cmd = ODP_VPORT_CMD_SET,
1954 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1955 .policy = vport_policy,
1956 .doit = odp_vport_cmd_set,
1960 struct genl_family_and_ops {
1961 struct genl_family *family;
1962 struct genl_ops *ops;
1964 struct genl_multicast_group *group;
1967 static const struct genl_family_and_ops dp_genl_families[] = {
1968 { &dp_datapath_genl_family,
1969 dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1970 &dp_datapath_multicast_group },
1971 { &dp_vport_genl_family,
1972 dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1973 &dp_vport_multicast_group },
1974 { &dp_flow_genl_family,
1975 dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1976 &dp_flow_multicast_group },
1977 { &dp_packet_genl_family,
1978 dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1982 static void dp_unregister_genl(int n_families)
1986 for (i = 0; i < n_families; i++) {
1987 genl_unregister_family(dp_genl_families[i].family);
1991 static int dp_register_genl(void)
1998 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
1999 const struct genl_family_and_ops *f = &dp_genl_families[i];
2001 err = genl_register_family_with_ops(f->family, f->ops,
2008 err = genl_register_mc_group(f->family, f->group);
2014 err = packet_register_mc_groups();
2020 dp_unregister_genl(n_registered);
2024 static int __init dp_init(void)
2026 struct sk_buff *dummy_skb;
2029 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
2031 printk("Open vSwitch %s, built "__DATE__" "__TIME__"\n", VERSION BUILDNR);
2039 goto error_flow_exit;
2041 err = register_netdevice_notifier(&dp_device_notifier);
2043 goto error_vport_exit;
2045 err = dp_register_genl();
2047 goto error_unreg_notifier;
2051 error_unreg_notifier:
2052 unregister_netdevice_notifier(&dp_device_notifier);
2061 static void dp_cleanup(void)
2064 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2065 unregister_netdevice_notifier(&dp_device_notifier);
2070 module_init(dp_init);
2071 module_exit(dp_cleanup);
2073 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2074 MODULE_LICENSE("GPL");