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 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,18) || \
57 LINUX_VERSION_CODE > KERNEL_VERSION(2,6,39)
58 #error Kernels before 2.6.18 or after 2.6.39 are not supported by this version of Open vSwitch.
61 int (*dp_ioctl_hook)(struct net_device *dev, struct ifreq *rq, int cmd);
62 EXPORT_SYMBOL(dp_ioctl_hook);
67 * Writes to device state (add/remove datapath, port, set operations on vports,
68 * etc.) are protected by RTNL.
70 * Writes to other state (flow table modifications, set miscellaneous datapath
71 * parameters such as drop frags, etc.) are protected by genl_mutex. The RTNL
72 * lock nests inside genl_mutex.
74 * Reads are protected by RCU.
76 * There are a few special cases (mostly stats) that have their own
77 * synchronization but they nest under all of above and don't interact with
81 /* Global list of datapaths to enable dumping them all out.
82 * Protected by genl_mutex.
84 static LIST_HEAD(dps);
86 static struct vport *new_vport(const struct vport_parms *);
87 static int queue_control_packets(struct datapath *, struct sk_buff *,
88 const struct dp_upcall_info *);
90 /* Must be called with rcu_read_lock, genl_mutex, or RTNL lock. */
91 struct datapath *get_dp(int dp_ifindex)
93 struct datapath *dp = NULL;
94 struct net_device *dev;
97 dev = dev_get_by_index_rcu(&init_net, dp_ifindex);
99 struct vport *vport = internal_dev_get_vport(dev);
107 EXPORT_SYMBOL_GPL(get_dp);
109 /* Must be called with genl_mutex. */
110 static struct tbl *get_table_protected(struct datapath *dp)
112 return rcu_dereference_protected(dp->table, lockdep_genl_is_held());
115 /* Must be called with rcu_read_lock or RTNL lock. */
116 static struct vport *get_vport_protected(struct datapath *dp, u16 port_no)
118 return rcu_dereference_rtnl(dp->ports[port_no]);
121 /* Must be called with rcu_read_lock or RTNL lock. */
122 const char *dp_name(const struct datapath *dp)
124 return vport_get_name(rcu_dereference_rtnl(dp->ports[ODPP_LOCAL]));
127 static inline size_t br_nlmsg_size(void)
129 return NLMSG_ALIGN(sizeof(struct ifinfomsg))
130 + nla_total_size(IFNAMSIZ) /* IFLA_IFNAME */
131 + nla_total_size(MAX_ADDR_LEN) /* IFLA_ADDRESS */
132 + nla_total_size(4) /* IFLA_MASTER */
133 + nla_total_size(4) /* IFLA_MTU */
134 + nla_total_size(4) /* IFLA_LINK */
135 + nla_total_size(1); /* IFLA_OPERSTATE */
138 /* Caller must hold RTNL lock. */
139 static int dp_fill_ifinfo(struct sk_buff *skb,
140 const struct vport *port,
141 int event, unsigned int flags)
143 struct datapath *dp = port->dp;
144 int ifindex = vport_get_ifindex(port);
145 int iflink = vport_get_iflink(port);
146 struct ifinfomsg *hdr;
147 struct nlmsghdr *nlh;
155 nlh = nlmsg_put(skb, 0, 0, event, sizeof(*hdr), flags);
159 hdr = nlmsg_data(nlh);
160 hdr->ifi_family = AF_BRIDGE;
162 hdr->ifi_type = ARPHRD_ETHER;
163 hdr->ifi_index = ifindex;
164 hdr->ifi_flags = vport_get_flags(port);
167 NLA_PUT_STRING(skb, IFLA_IFNAME, vport_get_name(port));
168 NLA_PUT_U32(skb, IFLA_MASTER,
169 vport_get_ifindex(get_vport_protected(dp, ODPP_LOCAL)));
170 NLA_PUT_U32(skb, IFLA_MTU, vport_get_mtu(port));
171 #ifdef IFLA_OPERSTATE
172 NLA_PUT_U8(skb, IFLA_OPERSTATE,
173 vport_is_running(port)
174 ? vport_get_operstate(port)
178 NLA_PUT(skb, IFLA_ADDRESS, ETH_ALEN, vport_get_addr(port));
180 if (ifindex != iflink)
181 NLA_PUT_U32(skb, IFLA_LINK,iflink);
183 return nlmsg_end(skb, nlh);
186 nlmsg_cancel(skb, nlh);
190 /* Caller must hold RTNL lock. */
191 static void dp_ifinfo_notify(int event, struct vport *port)
196 skb = nlmsg_new(br_nlmsg_size(), GFP_KERNEL);
200 err = dp_fill_ifinfo(skb, port, event, 0);
202 /* -EMSGSIZE implies BUG in br_nlmsg_size() */
203 WARN_ON(err == -EMSGSIZE);
207 rtnl_notify(skb, &init_net, 0, RTNLGRP_LINK, NULL, GFP_KERNEL);
211 rtnl_set_sk_err(&init_net, RTNLGRP_LINK, err);
214 static void release_dp(struct kobject *kobj)
216 struct datapath *dp = container_of(kobj, struct datapath, ifobj);
220 static struct kobj_type dp_ktype = {
221 .release = release_dp
224 static void destroy_dp_rcu(struct rcu_head *rcu)
226 struct datapath *dp = container_of(rcu, struct datapath, rcu);
228 tbl_destroy((struct tbl __force *)dp->table, flow_free_tbl);
229 free_percpu(dp->stats_percpu);
230 kobject_put(&dp->ifobj);
233 /* Called with RTNL lock and genl_lock. */
234 static struct vport *new_vport(const struct vport_parms *parms)
238 vport = vport_add(parms);
239 if (!IS_ERR(vport)) {
240 struct datapath *dp = parms->dp;
242 rcu_assign_pointer(dp->ports[parms->port_no], vport);
243 list_add(&vport->node, &dp->port_list);
245 dp_ifinfo_notify(RTM_NEWLINK, vport);
251 /* Called with RTNL lock. */
252 int dp_detach_port(struct vport *p)
256 if (p->port_no != ODPP_LOCAL)
258 dp_ifinfo_notify(RTM_DELLINK, p);
260 /* First drop references to device. */
262 rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
264 /* Then destroy it. */
268 /* Must be called with rcu_read_lock. */
269 void dp_process_received_packet(struct vport *p, struct sk_buff *skb)
271 struct datapath *dp = p->dp;
272 struct dp_stats_percpu *stats;
273 int stats_counter_off;
276 OVS_CB(skb)->vport = p;
278 if (!OVS_CB(skb)->flow) {
279 struct sw_flow_key key;
280 struct tbl_node *flow_node;
284 /* Extract flow from 'skb' into 'key'. */
285 error = flow_extract(skb, p->port_no, &key, &key_len, &is_frag);
286 if (unlikely(error)) {
291 if (is_frag && dp->drop_frags) {
293 stats_counter_off = offsetof(struct dp_stats_percpu, n_frags);
298 flow_node = tbl_lookup(rcu_dereference(dp->table), &key, key_len,
299 flow_hash(&key, key_len), flow_cmp);
300 if (unlikely(!flow_node)) {
301 struct dp_upcall_info upcall;
303 upcall.cmd = ODP_PACKET_CMD_MISS;
306 upcall.sample_pool = 0;
307 upcall.actions = NULL;
308 upcall.actions_len = 0;
309 dp_upcall(dp, skb, &upcall);
310 stats_counter_off = offsetof(struct dp_stats_percpu, n_missed);
314 OVS_CB(skb)->flow = flow_cast(flow_node);
317 stats_counter_off = offsetof(struct dp_stats_percpu, n_hit);
318 flow_used(OVS_CB(skb)->flow, skb);
319 execute_actions(dp, skb);
322 /* Update datapath statistics. */
324 stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
326 write_seqcount_begin(&stats->seqlock);
327 (*(u64 *)((u8 *)stats + stats_counter_off))++;
328 write_seqcount_end(&stats->seqlock);
333 static void copy_and_csum_skb(struct sk_buff *skb, void *to)
335 u16 csum_start, csum_offset;
338 get_skb_csum_pointers(skb, &csum_start, &csum_offset);
339 csum_start -= skb_headroom(skb);
341 skb_copy_bits(skb, 0, to, csum_start);
343 csum = skb_copy_and_csum_bits(skb, csum_start, to + csum_start,
344 skb->len - csum_start, 0);
345 *(__sum16 *)(to + csum_start + csum_offset) = csum_fold(csum);
348 static struct genl_family dp_packet_genl_family = {
349 .id = GENL_ID_GENERATE,
350 .hdrsize = sizeof(struct odp_header),
351 .name = ODP_PACKET_FAMILY,
353 .maxattr = ODP_PACKET_ATTR_MAX
356 /* Generic Netlink multicast groups for upcalls.
358 * We really want three unique multicast groups per datapath, but we can't even
359 * get one, because genl_register_mc_group() takes genl_lock, which is also
360 * held during Generic Netlink message processing, so trying to acquire
361 * multicast groups during ODP_DP_NEW processing deadlocks. Instead, we
362 * preallocate a few groups and use them round-robin for datapaths. Collision
363 * isn't fatal--multicast listeners should check that the family is the one
364 * that they want and discard others--but it wastes time and memory to receive
367 #define PACKET_N_MC_GROUPS 16
368 static struct genl_multicast_group packet_mc_groups[PACKET_N_MC_GROUPS];
370 static u32 packet_mc_group(struct datapath *dp, u8 cmd)
373 BUILD_BUG_ON_NOT_POWER_OF_2(PACKET_N_MC_GROUPS);
375 idx = jhash_2words(dp->dp_ifindex, cmd, 0) & (PACKET_N_MC_GROUPS - 1);
376 return packet_mc_groups[idx].id;
379 static int packet_register_mc_groups(void)
383 for (i = 0; i < PACKET_N_MC_GROUPS; i++) {
384 struct genl_multicast_group *group = &packet_mc_groups[i];
387 sprintf(group->name, "packet%d", i);
388 error = genl_register_mc_group(&dp_packet_genl_family, group);
395 int dp_upcall(struct datapath *dp, struct sk_buff *skb, const struct dp_upcall_info *upcall_info)
397 struct dp_stats_percpu *stats;
400 WARN_ON_ONCE(skb_shared(skb));
402 forward_ip_summed(skb, true);
404 /* Break apart GSO packets into their component pieces. Otherwise
405 * userspace may try to stuff a 64kB packet into a 1500-byte MTU. */
406 if (skb_is_gso(skb)) {
407 struct sk_buff *nskb = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
417 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,
568 int type = nla_type(a);
570 if (type > ODP_ACTION_ATTR_MAX || nla_len(a) != action_lens[type])
574 case ODP_ACTION_ATTR_UNSPEC:
577 case ODP_ACTION_ATTR_CONTROLLER:
578 case ODP_ACTION_ATTR_STRIP_VLAN:
579 case ODP_ACTION_ATTR_SET_DL_SRC:
580 case ODP_ACTION_ATTR_SET_DL_DST:
581 case ODP_ACTION_ATTR_SET_NW_SRC:
582 case ODP_ACTION_ATTR_SET_NW_DST:
583 case ODP_ACTION_ATTR_SET_TP_SRC:
584 case ODP_ACTION_ATTR_SET_TP_DST:
585 case ODP_ACTION_ATTR_SET_TUNNEL:
586 case ODP_ACTION_ATTR_SET_PRIORITY:
587 case ODP_ACTION_ATTR_POP_PRIORITY:
588 /* No validation needed. */
591 case ODP_ACTION_ATTR_OUTPUT:
592 if (nla_get_u32(a) >= DP_MAX_PORTS)
596 case ODP_ACTION_ATTR_SET_DL_TCI:
597 if (nla_get_be16(a) & htons(VLAN_CFI_MASK))
601 case ODP_ACTION_ATTR_SET_NW_TOS:
602 if (nla_get_u8(a) & INET_ECN_MASK)
616 static void clear_stats(struct sw_flow *flow)
620 flow->packet_count = 0;
621 flow->byte_count = 0;
624 /* Called with genl_mutex. */
625 static int expand_table(struct datapath *dp)
627 struct tbl *old_table = get_table_protected(dp);
628 struct tbl *new_table;
630 new_table = tbl_expand(old_table);
631 if (IS_ERR(new_table))
632 return PTR_ERR(new_table);
634 rcu_assign_pointer(dp->table, new_table);
635 tbl_deferred_destroy(old_table, NULL);
640 static int odp_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
642 struct odp_header *odp_header = info->userhdr;
643 struct nlattr **a = info->attrs;
644 struct sw_flow_actions *acts;
645 struct sk_buff *packet;
646 struct sw_flow *flow;
655 if (!a[ODP_PACKET_ATTR_PACKET] || !a[ODP_PACKET_ATTR_KEY] ||
656 !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, &key_len, &is_frag);
693 flow->tbl_node.hash = flow_hash(&flow->key, key_len);
695 err = flow_metadata_from_nlattrs(&flow->key.eth.in_port,
696 &flow->key.eth.tun_id,
697 a[ODP_PACKET_ATTR_KEY]);
701 acts = flow_actions_alloc(a[ODP_PACKET_ATTR_ACTIONS]);
705 rcu_assign_pointer(flow->sf_acts, acts);
707 OVS_CB(packet)->flow = flow;
710 dp = get_dp(odp_header->dp_ifindex);
714 err = execute_actions(dp, packet);
730 static const struct nla_policy packet_policy[ODP_PACKET_ATTR_MAX + 1] = {
731 [ODP_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
732 [ODP_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
733 [ODP_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
736 static struct genl_ops dp_packet_genl_ops[] = {
737 { .cmd = ODP_PACKET_CMD_EXECUTE,
738 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
739 .policy = packet_policy,
740 .doit = odp_packet_cmd_execute
744 static void get_dp_stats(struct datapath *dp, struct odp_stats *stats)
748 stats->n_frags = stats->n_hit = stats->n_missed = stats->n_lost = 0;
749 for_each_possible_cpu(i) {
750 const struct dp_stats_percpu *percpu_stats;
751 struct dp_stats_percpu local_stats;
754 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
757 seqcount = read_seqcount_begin(&percpu_stats->seqlock);
758 local_stats = *percpu_stats;
759 } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount));
761 stats->n_frags += local_stats.n_frags;
762 stats->n_hit += local_stats.n_hit;
763 stats->n_missed += local_stats.n_missed;
764 stats->n_lost += local_stats.n_lost;
768 /* MTU of the dp pseudo-device: ETH_DATA_LEN or the minimum of the ports.
769 * Called with RTNL lock.
771 int dp_min_mtu(const struct datapath *dp)
778 list_for_each_entry (p, &dp->port_list, node) {
781 /* Skip any internal ports, since that's what we're trying to
783 if (is_internal_vport(p))
786 dev_mtu = vport_get_mtu(p);
789 if (!mtu || dev_mtu < mtu)
793 return mtu ? mtu : ETH_DATA_LEN;
796 /* Sets the MTU of all datapath devices to the minimum of the ports
797 * Called with RTNL lock.
799 void set_internal_devs_mtu(const struct datapath *dp)
806 mtu = dp_min_mtu(dp);
808 list_for_each_entry (p, &dp->port_list, node) {
809 if (is_internal_vport(p))
810 vport_set_mtu(p, mtu);
814 static const struct nla_policy flow_policy[ODP_FLOW_ATTR_MAX + 1] = {
815 [ODP_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
816 [ODP_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
817 [ODP_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
820 static struct genl_family dp_flow_genl_family = {
821 .id = GENL_ID_GENERATE,
822 .hdrsize = sizeof(struct odp_header),
823 .name = ODP_FLOW_FAMILY,
825 .maxattr = ODP_FLOW_ATTR_MAX
828 static struct genl_multicast_group dp_flow_multicast_group = {
829 .name = ODP_FLOW_MCGROUP
832 /* Called with genl_lock. */
833 static int odp_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
834 struct sk_buff *skb, u32 pid, u32 seq, u32 flags, u8 cmd)
836 const int skb_orig_len = skb->len;
837 const struct sw_flow_actions *sf_acts;
838 struct odp_flow_stats stats;
839 struct odp_header *odp_header;
845 sf_acts = rcu_dereference_protected(flow->sf_acts,
846 lockdep_genl_is_held());
848 odp_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags, cmd);
852 odp_header->dp_ifindex = dp->dp_ifindex;
854 nla = nla_nest_start(skb, ODP_FLOW_ATTR_KEY);
856 goto nla_put_failure;
857 err = flow_to_nlattrs(&flow->key, skb);
860 nla_nest_end(skb, nla);
862 spin_lock_bh(&flow->lock);
864 stats.n_packets = flow->packet_count;
865 stats.n_bytes = flow->byte_count;
866 tcp_flags = flow->tcp_flags;
867 spin_unlock_bh(&flow->lock);
870 NLA_PUT_U64(skb, ODP_FLOW_ATTR_USED, flow_used_time(used));
873 NLA_PUT(skb, ODP_FLOW_ATTR_STATS, sizeof(struct odp_flow_stats), &stats);
876 NLA_PUT_U8(skb, ODP_FLOW_ATTR_TCP_FLAGS, tcp_flags);
878 /* If ODP_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
879 * this is the first flow to be dumped into 'skb'. This is unusual for
880 * Netlink but individual action lists can be longer than
881 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
882 * The userspace caller can always fetch the actions separately if it
883 * really wants them. (Most userspace callers in fact don't care.)
885 * This can only fail for dump operations because the skb is always
886 * properly sized for single flows.
888 err = nla_put(skb, ODP_FLOW_ATTR_ACTIONS, sf_acts->actions_len,
890 if (err < 0 && skb_orig_len)
893 return genlmsg_end(skb, odp_header);
898 genlmsg_cancel(skb, odp_header);
902 static struct sk_buff *odp_flow_cmd_alloc_info(struct sw_flow *flow)
904 const struct sw_flow_actions *sf_acts;
907 sf_acts = rcu_dereference_protected(flow->sf_acts,
908 lockdep_genl_is_held());
910 len = nla_total_size(FLOW_BUFSIZE); /* ODP_FLOW_ATTR_KEY */
911 len += nla_total_size(sf_acts->actions_len); /* ODP_FLOW_ATTR_ACTIONS */
912 len += nla_total_size(sizeof(struct odp_flow_stats)); /* ODP_FLOW_ATTR_STATS */
913 len += nla_total_size(1); /* ODP_FLOW_ATTR_TCP_FLAGS */
914 len += nla_total_size(8); /* ODP_FLOW_ATTR_USED */
915 return genlmsg_new(NLMSG_ALIGN(sizeof(struct odp_header)) + len, GFP_KERNEL);
918 static struct sk_buff *odp_flow_cmd_build_info(struct sw_flow *flow, struct datapath *dp,
919 u32 pid, u32 seq, u8 cmd)
924 skb = odp_flow_cmd_alloc_info(flow);
926 return ERR_PTR(-ENOMEM);
928 retval = odp_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd);
933 static int odp_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
935 struct nlattr **a = info->attrs;
936 struct odp_header *odp_header = info->userhdr;
937 struct tbl_node *flow_node;
938 struct sw_flow_key key;
939 struct sw_flow *flow;
940 struct sk_buff *reply;
949 if (!a[ODP_FLOW_ATTR_KEY])
951 error = flow_from_nlattrs(&key, &key_len, 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, key_len);
971 table = get_table_protected(dp);
972 flow_node = tbl_lookup(table, &key, key_len, 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;
1084 if (!a[ODP_FLOW_ATTR_KEY])
1086 err = flow_from_nlattrs(&key, &key_len, a[ODP_FLOW_ATTR_KEY]);
1090 dp = get_dp(odp_header->dp_ifindex);
1094 table = get_table_protected(dp);
1095 flow_node = tbl_lookup(table, &key, key_len, flow_hash(&key, key_len),
1100 flow = flow_cast(flow_node);
1101 reply = odp_flow_cmd_build_info(flow, dp, info->snd_pid, info->snd_seq, ODP_FLOW_CMD_NEW);
1103 return PTR_ERR(reply);
1105 return genlmsg_reply(reply, info);
1108 static int odp_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1110 struct nlattr **a = info->attrs;
1111 struct odp_header *odp_header = info->userhdr;
1112 struct sw_flow_key key;
1113 struct tbl_node *flow_node;
1114 struct sk_buff *reply;
1115 struct sw_flow *flow;
1116 struct datapath *dp;
1121 if (!a[ODP_FLOW_ATTR_KEY])
1122 return flush_flows(odp_header->dp_ifindex);
1123 err = flow_from_nlattrs(&key, &key_len, a[ODP_FLOW_ATTR_KEY]);
1127 dp = get_dp(odp_header->dp_ifindex);
1131 table = get_table_protected(dp);
1132 flow_node = tbl_lookup(table, &key, key_len, flow_hash(&key, key_len),
1136 flow = flow_cast(flow_node);
1138 reply = odp_flow_cmd_alloc_info(flow);
1142 err = tbl_remove(table, flow_node);
1148 err = odp_flow_cmd_fill_info(flow, dp, reply, info->snd_pid,
1149 info->snd_seq, 0, ODP_FLOW_CMD_DEL);
1152 flow_deferred_free(flow);
1154 genl_notify(reply, genl_info_net(info), info->snd_pid,
1155 dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1159 static int odp_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1161 struct odp_header *odp_header = genlmsg_data(nlmsg_data(cb->nlh));
1162 struct datapath *dp;
1164 dp = get_dp(odp_header->dp_ifindex);
1169 struct tbl_node *flow_node;
1170 struct sw_flow *flow;
1173 bucket = cb->args[0];
1175 flow_node = tbl_next(get_table_protected(dp), &bucket, &obj);
1179 flow = flow_cast(flow_node);
1180 if (odp_flow_cmd_fill_info(flow, dp, skb, NETLINK_CB(cb->skb).pid,
1181 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1182 ODP_FLOW_CMD_NEW) < 0)
1185 cb->args[0] = bucket;
1191 static struct genl_ops dp_flow_genl_ops[] = {
1192 { .cmd = ODP_FLOW_CMD_NEW,
1193 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1194 .policy = flow_policy,
1195 .doit = odp_flow_cmd_new_or_set
1197 { .cmd = ODP_FLOW_CMD_DEL,
1198 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1199 .policy = flow_policy,
1200 .doit = odp_flow_cmd_del
1202 { .cmd = ODP_FLOW_CMD_GET,
1203 .flags = 0, /* OK for unprivileged users. */
1204 .policy = flow_policy,
1205 .doit = odp_flow_cmd_get,
1206 .dumpit = odp_flow_cmd_dump
1208 { .cmd = ODP_FLOW_CMD_SET,
1209 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1210 .policy = flow_policy,
1211 .doit = odp_flow_cmd_new_or_set,
1215 static const struct nla_policy datapath_policy[ODP_DP_ATTR_MAX + 1] = {
1216 #ifdef HAVE_NLA_NUL_STRING
1217 [ODP_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1219 [ODP_DP_ATTR_IPV4_FRAGS] = { .type = NLA_U32 },
1220 [ODP_DP_ATTR_SAMPLING] = { .type = NLA_U32 },
1223 static struct genl_family dp_datapath_genl_family = {
1224 .id = GENL_ID_GENERATE,
1225 .hdrsize = sizeof(struct odp_header),
1226 .name = ODP_DATAPATH_FAMILY,
1228 .maxattr = ODP_DP_ATTR_MAX
1231 static struct genl_multicast_group dp_datapath_multicast_group = {
1232 .name = ODP_DATAPATH_MCGROUP
1235 static int odp_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1236 u32 pid, u32 seq, u32 flags, u8 cmd)
1238 struct odp_header *odp_header;
1242 odp_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family,
1247 odp_header->dp_ifindex = dp->dp_ifindex;
1250 err = nla_put_string(skb, ODP_DP_ATTR_NAME, dp_name(dp));
1253 goto nla_put_failure;
1255 nla = nla_reserve(skb, ODP_DP_ATTR_STATS, sizeof(struct odp_stats));
1257 goto nla_put_failure;
1258 get_dp_stats(dp, nla_data(nla));
1260 NLA_PUT_U32(skb, ODP_DP_ATTR_IPV4_FRAGS,
1261 dp->drop_frags ? ODP_DP_FRAG_DROP : ODP_DP_FRAG_ZERO);
1263 if (dp->sflow_probability)
1264 NLA_PUT_U32(skb, ODP_DP_ATTR_SAMPLING, dp->sflow_probability);
1266 nla = nla_nest_start(skb, ODP_DP_ATTR_MCGROUPS);
1268 goto nla_put_failure;
1269 NLA_PUT_U32(skb, ODP_PACKET_CMD_MISS, packet_mc_group(dp, ODP_PACKET_CMD_MISS));
1270 NLA_PUT_U32(skb, ODP_PACKET_CMD_ACTION, packet_mc_group(dp, ODP_PACKET_CMD_ACTION));
1271 NLA_PUT_U32(skb, ODP_PACKET_CMD_SAMPLE, packet_mc_group(dp, ODP_PACKET_CMD_SAMPLE));
1272 nla_nest_end(skb, nla);
1274 return genlmsg_end(skb, odp_header);
1277 genlmsg_cancel(skb, odp_header);
1282 static struct sk_buff *odp_dp_cmd_build_info(struct datapath *dp, u32 pid,
1285 struct sk_buff *skb;
1288 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1290 return ERR_PTR(-ENOMEM);
1292 retval = odp_dp_cmd_fill_info(dp, skb, pid, seq, 0, cmd);
1295 return ERR_PTR(retval);
1300 static int odp_dp_cmd_validate(struct nlattr *a[ODP_DP_ATTR_MAX + 1])
1302 if (a[ODP_DP_ATTR_IPV4_FRAGS]) {
1303 u32 frags = nla_get_u32(a[ODP_DP_ATTR_IPV4_FRAGS]);
1305 if (frags != ODP_DP_FRAG_ZERO && frags != ODP_DP_FRAG_DROP)
1309 return CHECK_NUL_STRING(a[ODP_DP_ATTR_NAME], IFNAMSIZ - 1);
1312 /* Called with genl_mutex and optionally with RTNL lock also. */
1313 static struct datapath *lookup_datapath(struct odp_header *odp_header, struct nlattr *a[ODP_DP_ATTR_MAX + 1])
1315 struct datapath *dp;
1317 if (!a[ODP_DP_ATTR_NAME])
1318 dp = get_dp(odp_header->dp_ifindex);
1320 struct vport *vport;
1323 vport = vport_locate(nla_data(a[ODP_DP_ATTR_NAME]));
1324 dp = vport && vport->port_no == ODPP_LOCAL ? vport->dp : NULL;
1327 return dp ? dp : ERR_PTR(-ENODEV);
1330 /* Called with genl_mutex. */
1331 static void change_datapath(struct datapath *dp, struct nlattr *a[ODP_DP_ATTR_MAX + 1])
1333 if (a[ODP_DP_ATTR_IPV4_FRAGS])
1334 dp->drop_frags = nla_get_u32(a[ODP_DP_ATTR_IPV4_FRAGS]) == ODP_DP_FRAG_DROP;
1335 if (a[ODP_DP_ATTR_SAMPLING])
1336 dp->sflow_probability = nla_get_u32(a[ODP_DP_ATTR_SAMPLING]);
1339 static int odp_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1341 struct nlattr **a = info->attrs;
1342 struct vport_parms parms;
1343 struct sk_buff *reply;
1344 struct datapath *dp;
1345 struct vport *vport;
1349 if (!a[ODP_DP_ATTR_NAME])
1352 err = odp_dp_cmd_validate(a);
1358 if (!try_module_get(THIS_MODULE))
1359 goto err_unlock_rtnl;
1362 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1364 goto err_put_module;
1365 INIT_LIST_HEAD(&dp->port_list);
1367 /* Initialize kobject for bridge. This will be added as
1368 * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
1369 dp->ifobj.kset = NULL;
1370 kobject_init(&dp->ifobj, &dp_ktype);
1372 /* Allocate table. */
1374 rcu_assign_pointer(dp->table, tbl_create(TBL_MIN_BUCKETS));
1378 /* Set up our datapath device. */
1379 parms.name = nla_data(a[ODP_DP_ATTR_NAME]);
1380 parms.type = ODP_VPORT_TYPE_INTERNAL;
1381 parms.options = NULL;
1383 parms.port_no = ODPP_LOCAL;
1384 vport = new_vport(&parms);
1385 if (IS_ERR(vport)) {
1386 err = PTR_ERR(vport);
1390 goto err_destroy_table;
1392 dp->dp_ifindex = vport_get_ifindex(vport);
1395 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1396 if (!dp->stats_percpu) {
1398 goto err_destroy_local_port;
1401 change_datapath(dp, a);
1403 reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_NEW);
1404 err = PTR_ERR(reply);
1406 goto err_destroy_local_port;
1408 list_add_tail(&dp->list_node, &dps);
1409 dp_sysfs_add_dp(dp);
1413 genl_notify(reply, genl_info_net(info), info->snd_pid,
1414 dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1417 err_destroy_local_port:
1418 dp_detach_port(get_vport_protected(dp, ODPP_LOCAL));
1420 tbl_destroy(get_table_protected(dp), NULL);
1424 module_put(THIS_MODULE);
1431 static int odp_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1433 struct vport *vport, *next_vport;
1434 struct sk_buff *reply;
1435 struct datapath *dp;
1438 err = odp_dp_cmd_validate(info->attrs);
1443 dp = lookup_datapath(info->userhdr, info->attrs);
1448 reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_DEL);
1449 err = PTR_ERR(reply);
1453 list_for_each_entry_safe (vport, next_vport, &dp->port_list, node)
1454 if (vport->port_no != ODPP_LOCAL)
1455 dp_detach_port(vport);
1457 dp_sysfs_del_dp(dp);
1458 list_del(&dp->list_node);
1459 dp_detach_port(get_vport_protected(dp, ODPP_LOCAL));
1461 /* rtnl_unlock() will wait until all the references to devices that
1462 * are pending unregistration have been dropped. We do it here to
1463 * ensure that any internal devices (which contain DP pointers) are
1464 * fully destroyed before freeing the datapath.
1468 call_rcu(&dp->rcu, destroy_dp_rcu);
1469 module_put(THIS_MODULE);
1471 genl_notify(reply, genl_info_net(info), info->snd_pid,
1472 dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1482 static int odp_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1484 struct sk_buff *reply;
1485 struct datapath *dp;
1488 err = odp_dp_cmd_validate(info->attrs);
1492 dp = lookup_datapath(info->userhdr, info->attrs);
1496 change_datapath(dp, info->attrs);
1498 reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_NEW);
1499 if (IS_ERR(reply)) {
1500 err = PTR_ERR(reply);
1501 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1502 dp_datapath_multicast_group.id, err);
1506 genl_notify(reply, genl_info_net(info), info->snd_pid,
1507 dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1511 static int odp_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1513 struct sk_buff *reply;
1514 struct datapath *dp;
1517 err = odp_dp_cmd_validate(info->attrs);
1521 dp = lookup_datapath(info->userhdr, info->attrs);
1525 reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_NEW);
1527 return PTR_ERR(reply);
1529 return genlmsg_reply(reply, info);
1532 static int odp_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1534 struct datapath *dp;
1535 int skip = cb->args[0];
1538 list_for_each_entry (dp, &dps, list_node) {
1541 if (odp_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).pid,
1542 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1543 ODP_DP_CMD_NEW) < 0)
1553 static struct genl_ops dp_datapath_genl_ops[] = {
1554 { .cmd = ODP_DP_CMD_NEW,
1555 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1556 .policy = datapath_policy,
1557 .doit = odp_dp_cmd_new
1559 { .cmd = ODP_DP_CMD_DEL,
1560 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1561 .policy = datapath_policy,
1562 .doit = odp_dp_cmd_del
1564 { .cmd = ODP_DP_CMD_GET,
1565 .flags = 0, /* OK for unprivileged users. */
1566 .policy = datapath_policy,
1567 .doit = odp_dp_cmd_get,
1568 .dumpit = odp_dp_cmd_dump
1570 { .cmd = ODP_DP_CMD_SET,
1571 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1572 .policy = datapath_policy,
1573 .doit = odp_dp_cmd_set,
1577 static const struct nla_policy vport_policy[ODP_VPORT_ATTR_MAX + 1] = {
1578 #ifdef HAVE_NLA_NUL_STRING
1579 [ODP_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1580 [ODP_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1581 [ODP_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1582 [ODP_VPORT_ATTR_STATS] = { .len = sizeof(struct rtnl_link_stats64) },
1583 [ODP_VPORT_ATTR_ADDRESS] = { .len = ETH_ALEN },
1585 [ODP_VPORT_ATTR_STATS] = { .minlen = sizeof(struct rtnl_link_stats64) },
1586 [ODP_VPORT_ATTR_ADDRESS] = { .minlen = ETH_ALEN },
1588 [ODP_VPORT_ATTR_MTU] = { .type = NLA_U32 },
1589 [ODP_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1592 static struct genl_family dp_vport_genl_family = {
1593 .id = GENL_ID_GENERATE,
1594 .hdrsize = sizeof(struct odp_header),
1595 .name = ODP_VPORT_FAMILY,
1597 .maxattr = ODP_VPORT_ATTR_MAX
1600 static struct genl_multicast_group dp_vport_multicast_group = {
1601 .name = ODP_VPORT_MCGROUP
1604 /* Called with RTNL lock or RCU read lock. */
1605 static int odp_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1606 u32 pid, u32 seq, u32 flags, u8 cmd)
1608 struct odp_header *odp_header;
1610 int ifindex, iflink;
1614 odp_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family,
1619 odp_header->dp_ifindex = vport->dp->dp_ifindex;
1621 NLA_PUT_U32(skb, ODP_VPORT_ATTR_PORT_NO, vport->port_no);
1622 NLA_PUT_U32(skb, ODP_VPORT_ATTR_TYPE, vport_get_type(vport));
1623 NLA_PUT_STRING(skb, ODP_VPORT_ATTR_NAME, vport_get_name(vport));
1625 nla = nla_reserve(skb, ODP_VPORT_ATTR_STATS, sizeof(struct rtnl_link_stats64));
1627 goto nla_put_failure;
1628 if (vport_get_stats(vport, nla_data(nla)))
1629 __skb_trim(skb, skb->len - nla->nla_len);
1631 NLA_PUT(skb, ODP_VPORT_ATTR_ADDRESS, ETH_ALEN, vport_get_addr(vport));
1633 mtu = vport_get_mtu(vport);
1635 NLA_PUT_U32(skb, ODP_VPORT_ATTR_MTU, mtu);
1637 err = vport_get_options(vport, skb);
1638 if (err == -EMSGSIZE)
1641 ifindex = vport_get_ifindex(vport);
1643 NLA_PUT_U32(skb, ODP_VPORT_ATTR_IFINDEX, ifindex);
1645 iflink = vport_get_iflink(vport);
1647 NLA_PUT_U32(skb, ODP_VPORT_ATTR_IFLINK, iflink);
1649 return genlmsg_end(skb, odp_header);
1654 genlmsg_cancel(skb, odp_header);
1658 /* Called with RTNL lock or RCU read lock. */
1659 static struct sk_buff *odp_vport_cmd_build_info(struct vport *vport, u32 pid,
1662 struct sk_buff *skb;
1665 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1667 return ERR_PTR(-ENOMEM);
1669 retval = odp_vport_cmd_fill_info(vport, skb, pid, seq, 0, cmd);
1672 return ERR_PTR(retval);
1677 static int odp_vport_cmd_validate(struct nlattr *a[ODP_VPORT_ATTR_MAX + 1])
1679 return CHECK_NUL_STRING(a[ODP_VPORT_ATTR_NAME], IFNAMSIZ - 1);
1682 /* Called with RTNL lock or RCU read lock. */
1683 static struct vport *lookup_vport(struct odp_header *odp_header,
1684 struct nlattr *a[ODP_VPORT_ATTR_MAX + 1])
1686 struct datapath *dp;
1687 struct vport *vport;
1689 if (a[ODP_VPORT_ATTR_NAME]) {
1690 vport = vport_locate(nla_data(a[ODP_VPORT_ATTR_NAME]));
1692 return ERR_PTR(-ENODEV);
1694 } else if (a[ODP_VPORT_ATTR_PORT_NO]) {
1695 u32 port_no = nla_get_u32(a[ODP_VPORT_ATTR_PORT_NO]);
1697 if (port_no >= DP_MAX_PORTS)
1698 return ERR_PTR(-EFBIG);
1700 dp = get_dp(odp_header->dp_ifindex);
1702 return ERR_PTR(-ENODEV);
1704 vport = get_vport_protected(dp, port_no);
1706 return ERR_PTR(-ENOENT);
1709 return ERR_PTR(-EINVAL);
1712 /* Called with RTNL lock. */
1713 static int change_vport(struct vport *vport, struct nlattr *a[ODP_VPORT_ATTR_MAX + 1])
1716 if (a[ODP_VPORT_ATTR_STATS])
1717 err = vport_set_stats(vport, nla_data(a[ODP_VPORT_ATTR_STATS]));
1718 if (!err && a[ODP_VPORT_ATTR_ADDRESS])
1719 err = vport_set_addr(vport, nla_data(a[ODP_VPORT_ATTR_ADDRESS]));
1720 if (!err && a[ODP_VPORT_ATTR_MTU])
1721 err = vport_set_mtu(vport, nla_get_u32(a[ODP_VPORT_ATTR_MTU]));
1725 static int odp_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1727 struct nlattr **a = info->attrs;
1728 struct odp_header *odp_header = info->userhdr;
1729 struct vport_parms parms;
1730 struct sk_buff *reply;
1731 struct vport *vport;
1732 struct datapath *dp;
1737 if (!a[ODP_VPORT_ATTR_NAME] || !a[ODP_VPORT_ATTR_TYPE])
1740 err = odp_vport_cmd_validate(a);
1745 dp = get_dp(odp_header->dp_ifindex);
1750 if (a[ODP_VPORT_ATTR_PORT_NO]) {
1751 port_no = nla_get_u32(a[ODP_VPORT_ATTR_PORT_NO]);
1754 if (port_no >= DP_MAX_PORTS)
1757 vport = get_vport_protected(dp, port_no);
1762 for (port_no = 1; ; port_no++) {
1763 if (port_no >= DP_MAX_PORTS) {
1767 vport = get_vport_protected(dp, port_no);
1773 parms.name = nla_data(a[ODP_VPORT_ATTR_NAME]);
1774 parms.type = nla_get_u32(a[ODP_VPORT_ATTR_TYPE]);
1775 parms.options = a[ODP_VPORT_ATTR_OPTIONS];
1777 parms.port_no = port_no;
1779 vport = new_vport(&parms);
1780 err = PTR_ERR(vport);
1784 set_internal_devs_mtu(dp);
1785 dp_sysfs_add_if(vport);
1787 err = change_vport(vport, a);
1789 reply = odp_vport_cmd_build_info(vport, info->snd_pid,
1790 info->snd_seq, ODP_VPORT_CMD_NEW);
1792 err = PTR_ERR(reply);
1795 dp_detach_port(vport);
1798 genl_notify(reply, genl_info_net(info), info->snd_pid,
1799 dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1808 static int odp_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1810 struct nlattr **a = info->attrs;
1811 struct sk_buff *reply;
1812 struct vport *vport;
1815 err = odp_vport_cmd_validate(a);
1820 vport = lookup_vport(info->userhdr, a);
1821 err = PTR_ERR(vport);
1826 if (a[ODP_VPORT_ATTR_OPTIONS])
1827 err = vport_set_options(vport, a[ODP_VPORT_ATTR_OPTIONS]);
1829 err = change_vport(vport, a);
1831 reply = odp_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1833 if (IS_ERR(reply)) {
1834 err = PTR_ERR(reply);
1835 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1836 dp_vport_multicast_group.id, err);
1840 genl_notify(reply, genl_info_net(info), info->snd_pid,
1841 dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1849 static int odp_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1851 struct nlattr **a = info->attrs;
1852 struct sk_buff *reply;
1853 struct vport *vport;
1856 err = odp_vport_cmd_validate(a);
1861 vport = lookup_vport(info->userhdr, a);
1862 err = PTR_ERR(vport);
1866 if (vport->port_no == ODPP_LOCAL) {
1871 reply = odp_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1873 err = PTR_ERR(reply);
1877 err = dp_detach_port(vport);
1879 genl_notify(reply, genl_info_net(info), info->snd_pid,
1880 dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1888 static int odp_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1890 struct nlattr **a = info->attrs;
1891 struct odp_header *odp_header = info->userhdr;
1892 struct sk_buff *reply;
1893 struct vport *vport;
1896 err = odp_vport_cmd_validate(a);
1901 vport = lookup_vport(odp_header, a);
1902 err = PTR_ERR(vport);
1906 reply = odp_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1908 err = PTR_ERR(reply);
1914 return genlmsg_reply(reply, info);
1922 static int odp_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1924 struct odp_header *odp_header = genlmsg_data(nlmsg_data(cb->nlh));
1925 struct datapath *dp;
1929 dp = get_dp(odp_header->dp_ifindex);
1934 for (port_no = cb->args[0]; port_no < DP_MAX_PORTS; port_no++) {
1935 struct vport *vport;
1937 vport = get_vport_protected(dp, port_no);
1941 if (odp_vport_cmd_fill_info(vport, skb, NETLINK_CB(cb->skb).pid,
1942 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1943 ODP_VPORT_CMD_NEW) < 0)
1948 cb->args[0] = port_no;
1954 static struct genl_ops dp_vport_genl_ops[] = {
1955 { .cmd = ODP_VPORT_CMD_NEW,
1956 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1957 .policy = vport_policy,
1958 .doit = odp_vport_cmd_new
1960 { .cmd = ODP_VPORT_CMD_DEL,
1961 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1962 .policy = vport_policy,
1963 .doit = odp_vport_cmd_del
1965 { .cmd = ODP_VPORT_CMD_GET,
1966 .flags = 0, /* OK for unprivileged users. */
1967 .policy = vport_policy,
1968 .doit = odp_vport_cmd_get,
1969 .dumpit = odp_vport_cmd_dump
1971 { .cmd = ODP_VPORT_CMD_SET,
1972 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1973 .policy = vport_policy,
1974 .doit = odp_vport_cmd_set,
1978 struct genl_family_and_ops {
1979 struct genl_family *family;
1980 struct genl_ops *ops;
1982 struct genl_multicast_group *group;
1985 static const struct genl_family_and_ops dp_genl_families[] = {
1986 { &dp_datapath_genl_family,
1987 dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1988 &dp_datapath_multicast_group },
1989 { &dp_vport_genl_family,
1990 dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1991 &dp_vport_multicast_group },
1992 { &dp_flow_genl_family,
1993 dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1994 &dp_flow_multicast_group },
1995 { &dp_packet_genl_family,
1996 dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
2000 static void dp_unregister_genl(int n_families)
2004 for (i = 0; i < n_families; i++) {
2005 genl_unregister_family(dp_genl_families[i].family);
2009 static int dp_register_genl(void)
2016 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2017 const struct genl_family_and_ops *f = &dp_genl_families[i];
2019 err = genl_register_family_with_ops(f->family, f->ops,
2026 err = genl_register_mc_group(f->family, f->group);
2032 err = packet_register_mc_groups();
2038 dp_unregister_genl(n_registered);
2042 static int __init dp_init(void)
2044 struct sk_buff *dummy_skb;
2047 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
2049 printk("Open vSwitch %s, built "__DATE__" "__TIME__"\n", VERSION BUILDNR);
2057 goto error_flow_exit;
2059 err = register_netdevice_notifier(&dp_device_notifier);
2061 goto error_vport_exit;
2063 err = dp_register_genl();
2065 goto error_unreg_notifier;
2069 error_unreg_notifier:
2070 unregister_netdevice_notifier(&dp_device_notifier);
2079 static void dp_cleanup(void)
2082 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2083 unregister_netdevice_notifier(&dp_device_notifier);
2088 module_init(dp_init);
2089 module_exit(dp_cleanup);
2091 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2092 MODULE_LICENSE("GPL");