2 * Copyright (c) 2007, 2008, 2009, 2010, 2011 Nicira Networks.
3 * Distributed under the terms of the GNU GPL version 2.
5 * Significant portions of this file may be copied from parts of the Linux
6 * kernel, by Linus Torvalds and others.
9 /* Functions for managing the dp interface/device. */
11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
13 #include <linux/init.h>
14 #include <linux/module.h>
15 #include <linux/if_arp.h>
16 #include <linux/if_vlan.h>
19 #include <linux/jhash.h>
20 #include <linux/delay.h>
21 #include <linux/time.h>
22 #include <linux/etherdevice.h>
23 #include <linux/genetlink.h>
24 #include <linux/kernel.h>
25 #include <linux/kthread.h>
26 #include <linux/mutex.h>
27 #include <linux/percpu.h>
28 #include <linux/rcupdate.h>
29 #include <linux/tcp.h>
30 #include <linux/udp.h>
31 #include <linux/version.h>
32 #include <linux/ethtool.h>
33 #include <linux/wait.h>
34 #include <asm/system.h>
35 #include <asm/div64.h>
37 #include <linux/highmem.h>
38 #include <linux/netfilter_bridge.h>
39 #include <linux/netfilter_ipv4.h>
40 #include <linux/inetdevice.h>
41 #include <linux/list.h>
42 #include <linux/rculist.h>
43 #include <linux/dmi.h>
44 #include <net/inet_ecn.h>
45 #include <net/genetlink.h>
47 #include "openvswitch/datapath-protocol.h"
52 #include "loop_counter.h"
55 #include "vport-internal_dev.h"
57 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,18) || \
58 LINUX_VERSION_CODE > KERNEL_VERSION(2,6,38)
59 #error Kernels before 2.6.18 or after 2.6.38 are not supported by this version of Open vSwitch.
62 int (*dp_ioctl_hook)(struct net_device *dev, struct ifreq *rq, int cmd);
63 EXPORT_SYMBOL(dp_ioctl_hook);
68 * Writes to device state (add/remove datapath, port, set operations on vports,
69 * etc.) are protected by RTNL.
71 * Writes to other state (flow table modifications, set miscellaneous datapath
72 * parameters such as drop frags, etc.) are protected by genl_mutex. The RTNL
73 * lock nests inside genl_mutex.
75 * Reads are protected by RCU.
77 * There are a few special cases (mostly stats) that have their own
78 * synchronization but they nest under all of above and don't interact with
82 /* Global list of datapaths to enable dumping them all out.
83 * Protected by genl_mutex.
85 static LIST_HEAD(dps);
87 static struct vport *new_vport(const struct vport_parms *);
88 static int queue_control_packets(struct datapath *, struct sk_buff *,
89 const struct dp_upcall_info *);
91 /* Must be called with rcu_read_lock, genl_mutex, or RTNL lock. */
92 struct datapath *get_dp(int dp_ifindex)
94 struct datapath *dp = NULL;
95 struct net_device *dev;
98 dev = dev_get_by_index_rcu(&init_net, dp_ifindex);
100 struct vport *vport = internal_dev_get_vport(dev);
108 EXPORT_SYMBOL_GPL(get_dp);
110 /* Must be called with genl_mutex. */
111 static struct tbl *get_table_protected(struct datapath *dp)
113 return rcu_dereference_protected(dp->table, lockdep_genl_is_held());
116 /* Must be called with rcu_read_lock or RTNL lock. */
117 static struct vport *get_vport_protected(struct datapath *dp, u16 port_no)
119 return rcu_dereference_rtnl(dp->ports[port_no]);
122 /* Must be called with rcu_read_lock or RTNL lock. */
123 const char *dp_name(const struct datapath *dp)
125 return vport_get_name(rcu_dereference_rtnl(dp->ports[ODPP_LOCAL]));
128 static inline size_t br_nlmsg_size(void)
130 return NLMSG_ALIGN(sizeof(struct ifinfomsg))
131 + nla_total_size(IFNAMSIZ) /* IFLA_IFNAME */
132 + nla_total_size(MAX_ADDR_LEN) /* IFLA_ADDRESS */
133 + nla_total_size(4) /* IFLA_MASTER */
134 + nla_total_size(4) /* IFLA_MTU */
135 + nla_total_size(4) /* IFLA_LINK */
136 + nla_total_size(1); /* IFLA_OPERSTATE */
139 /* Caller must hold RTNL lock. */
140 static int dp_fill_ifinfo(struct sk_buff *skb,
141 const struct vport *port,
142 int event, unsigned int flags)
144 struct datapath *dp = port->dp;
145 int ifindex = vport_get_ifindex(port);
146 int iflink = vport_get_iflink(port);
147 struct ifinfomsg *hdr;
148 struct nlmsghdr *nlh;
156 nlh = nlmsg_put(skb, 0, 0, event, sizeof(*hdr), flags);
160 hdr = nlmsg_data(nlh);
161 hdr->ifi_family = AF_BRIDGE;
163 hdr->ifi_type = ARPHRD_ETHER;
164 hdr->ifi_index = ifindex;
165 hdr->ifi_flags = vport_get_flags(port);
168 NLA_PUT_STRING(skb, IFLA_IFNAME, vport_get_name(port));
169 NLA_PUT_U32(skb, IFLA_MASTER,
170 vport_get_ifindex(get_vport_protected(dp, ODPP_LOCAL)));
171 NLA_PUT_U32(skb, IFLA_MTU, vport_get_mtu(port));
172 #ifdef IFLA_OPERSTATE
173 NLA_PUT_U8(skb, IFLA_OPERSTATE,
174 vport_is_running(port)
175 ? vport_get_operstate(port)
179 NLA_PUT(skb, IFLA_ADDRESS, ETH_ALEN, vport_get_addr(port));
181 if (ifindex != iflink)
182 NLA_PUT_U32(skb, IFLA_LINK,iflink);
184 return nlmsg_end(skb, nlh);
187 nlmsg_cancel(skb, nlh);
191 /* Caller must hold RTNL lock. */
192 static void dp_ifinfo_notify(int event, struct vport *port)
197 skb = nlmsg_new(br_nlmsg_size(), GFP_KERNEL);
201 err = dp_fill_ifinfo(skb, port, event, 0);
203 /* -EMSGSIZE implies BUG in br_nlmsg_size() */
204 WARN_ON(err == -EMSGSIZE);
208 rtnl_notify(skb, &init_net, 0, RTNLGRP_LINK, NULL, GFP_KERNEL);
212 rtnl_set_sk_err(&init_net, RTNLGRP_LINK, err);
215 static void release_dp(struct kobject *kobj)
217 struct datapath *dp = container_of(kobj, struct datapath, ifobj);
221 static struct kobj_type dp_ktype = {
222 .release = release_dp
225 static void destroy_dp_rcu(struct rcu_head *rcu)
227 struct datapath *dp = container_of(rcu, struct datapath, rcu);
229 tbl_destroy((struct tbl __force *)dp->table, flow_free_tbl);
230 free_percpu(dp->stats_percpu);
231 kobject_put(&dp->ifobj);
234 /* Called with RTNL lock and genl_lock. */
235 static struct vport *new_vport(const struct vport_parms *parms)
239 vport = vport_add(parms);
240 if (!IS_ERR(vport)) {
241 struct datapath *dp = parms->dp;
243 rcu_assign_pointer(dp->ports[parms->port_no], vport);
244 list_add(&vport->node, &dp->port_list);
246 dp_ifinfo_notify(RTM_NEWLINK, vport);
252 /* Called with RTNL lock. */
253 int dp_detach_port(struct vport *p)
257 if (p->port_no != ODPP_LOCAL)
259 dp_ifinfo_notify(RTM_DELLINK, p);
261 /* First drop references to device. */
263 rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
265 /* Then destroy it. */
269 /* Must be called with rcu_read_lock. */
270 void dp_process_received_packet(struct vport *p, struct sk_buff *skb)
272 struct datapath *dp = p->dp;
273 struct dp_stats_percpu *stats;
274 int stats_counter_off;
275 struct sw_flow_actions *acts;
276 struct loop_counter *loop;
279 OVS_CB(skb)->vport = p;
281 if (!OVS_CB(skb)->flow) {
282 struct sw_flow_key key;
283 struct tbl_node *flow_node;
286 /* Extract flow from 'skb' into 'key'. */
287 error = flow_extract(skb, p->port_no, &key, &is_frag);
288 if (unlikely(error)) {
293 if (is_frag && dp->drop_frags) {
295 stats_counter_off = offsetof(struct dp_stats_percpu, n_frags);
300 flow_node = tbl_lookup(rcu_dereference(dp->table), &key,
301 flow_hash(&key), flow_cmp);
302 if (unlikely(!flow_node)) {
303 struct dp_upcall_info upcall;
305 upcall.cmd = ODP_PACKET_CMD_MISS;
308 upcall.sample_pool = 0;
309 upcall.actions = NULL;
310 upcall.actions_len = 0;
311 dp_upcall(dp, skb, &upcall);
312 stats_counter_off = offsetof(struct dp_stats_percpu, n_missed);
316 OVS_CB(skb)->flow = flow_cast(flow_node);
319 stats_counter_off = offsetof(struct dp_stats_percpu, n_hit);
320 flow_used(OVS_CB(skb)->flow, skb);
322 acts = rcu_dereference(OVS_CB(skb)->flow->sf_acts);
324 /* Check whether we've looped too much. */
325 loop = loop_get_counter();
326 if (unlikely(++loop->count > MAX_LOOPS))
327 loop->looping = true;
328 if (unlikely(loop->looping)) {
329 loop_suppress(dp, acts);
334 /* Execute actions. */
335 execute_actions(dp, skb, &OVS_CB(skb)->flow->key, acts->actions,
338 /* Check whether sub-actions looped too much. */
339 if (unlikely(loop->looping))
340 loop_suppress(dp, acts);
343 /* Decrement loop counter. */
345 loop->looping = false;
349 /* Update datapath statistics. */
351 stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
353 write_seqcount_begin(&stats->seqlock);
354 (*(u64 *)((u8 *)stats + stats_counter_off))++;
355 write_seqcount_end(&stats->seqlock);
360 static void copy_and_csum_skb(struct sk_buff *skb, void *to)
362 u16 csum_start, csum_offset;
365 get_skb_csum_pointers(skb, &csum_start, &csum_offset);
366 csum_start -= skb_headroom(skb);
368 skb_copy_bits(skb, 0, to, csum_start);
370 csum = skb_copy_and_csum_bits(skb, csum_start, to + csum_start,
371 skb->len - csum_start, 0);
372 *(__sum16 *)(to + csum_start + csum_offset) = csum_fold(csum);
375 static struct genl_family dp_packet_genl_family = {
376 .id = GENL_ID_GENERATE,
377 .hdrsize = sizeof(struct odp_header),
378 .name = ODP_PACKET_FAMILY,
380 .maxattr = ODP_PACKET_ATTR_MAX
383 /* Generic Netlink multicast groups for upcalls.
385 * We really want three unique multicast groups per datapath, but we can't even
386 * get one, because genl_register_mc_group() takes genl_lock, which is also
387 * held during Generic Netlink message processing, so trying to acquire
388 * multicast groups during ODP_DP_NEW processing deadlocks. Instead, we
389 * preallocate a few groups and use them round-robin for datapaths. Collision
390 * isn't fatal--multicast listeners should check that the family is the one
391 * that they want and discard others--but it wastes time and memory to receive
394 #define PACKET_N_MC_GROUPS 16
395 static struct genl_multicast_group packet_mc_groups[PACKET_N_MC_GROUPS];
397 static u32 packet_mc_group(struct datapath *dp, u8 cmd)
400 BUILD_BUG_ON_NOT_POWER_OF_2(PACKET_N_MC_GROUPS);
402 idx = jhash_2words(dp->dp_ifindex, cmd, 0) & (PACKET_N_MC_GROUPS - 1);
403 return packet_mc_groups[idx].id;
406 static int packet_register_mc_groups(void)
410 for (i = 0; i < PACKET_N_MC_GROUPS; i++) {
411 struct genl_multicast_group *group = &packet_mc_groups[i];
414 sprintf(group->name, "packet%d", i);
415 error = genl_register_mc_group(&dp_packet_genl_family, group);
422 int dp_upcall(struct datapath *dp, struct sk_buff *skb, const struct dp_upcall_info *upcall_info)
424 struct dp_stats_percpu *stats;
427 WARN_ON_ONCE(skb_shared(skb));
429 forward_ip_summed(skb);
431 err = vswitch_skb_checksum_setup(skb);
435 /* Break apart GSO packets into their component pieces. Otherwise
436 * userspace may try to stuff a 64kB packet into a 1500-byte MTU. */
437 if (skb_is_gso(skb)) {
438 struct sk_buff *nskb = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
448 err = queue_control_packets(dp, skb, upcall_info);
458 stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
460 write_seqcount_begin(&stats->seqlock);
462 write_seqcount_end(&stats->seqlock);
469 /* Send each packet in the 'skb' list to userspace for 'dp' as directed by
470 * 'upcall_info'. There will be only one packet unless we broke up a GSO
473 static int queue_control_packets(struct datapath *dp, struct sk_buff *skb,
474 const struct dp_upcall_info *upcall_info)
476 u32 group = packet_mc_group(dp, upcall_info->cmd);
477 struct sk_buff *nskb;
481 if (OVS_CB(skb)->vport)
482 port_no = OVS_CB(skb)->vport->port_no;
484 port_no = ODPP_LOCAL;
487 struct odp_header *upcall;
488 struct sk_buff *user_skb; /* to be queued to userspace */
495 err = vlan_deaccel_tag(skb);
499 if (nla_attr_size(skb->len) > USHRT_MAX)
502 len = sizeof(struct odp_header);
503 len += nla_total_size(skb->len);
504 len += nla_total_size(FLOW_BUFSIZE);
505 if (upcall_info->userdata)
506 len += nla_total_size(8);
507 if (upcall_info->sample_pool)
508 len += nla_total_size(4);
509 if (upcall_info->actions_len)
510 len += nla_total_size(upcall_info->actions_len);
512 user_skb = genlmsg_new(len, GFP_ATOMIC);
514 netlink_set_err(INIT_NET_GENL_SOCK, 0, group, -ENOBUFS);
518 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family, 0, upcall_info->cmd);
519 upcall->dp_ifindex = dp->dp_ifindex;
521 nla = nla_nest_start(user_skb, ODP_PACKET_ATTR_KEY);
522 flow_to_nlattrs(upcall_info->key, user_skb);
523 nla_nest_end(user_skb, nla);
525 if (upcall_info->userdata)
526 nla_put_u64(user_skb, ODP_PACKET_ATTR_USERDATA, upcall_info->userdata);
527 if (upcall_info->sample_pool)
528 nla_put_u32(user_skb, ODP_PACKET_ATTR_SAMPLE_POOL, upcall_info->sample_pool);
529 if (upcall_info->actions_len) {
530 const struct nlattr *actions = upcall_info->actions;
531 u32 actions_len = upcall_info->actions_len;
533 nla = nla_nest_start(user_skb, ODP_PACKET_ATTR_ACTIONS);
534 memcpy(__skb_put(user_skb, actions_len), actions, actions_len);
535 nla_nest_end(user_skb, nla);
538 nla = __nla_reserve(user_skb, ODP_PACKET_ATTR_PACKET, skb->len);
539 if (skb->ip_summed == CHECKSUM_PARTIAL)
540 copy_and_csum_skb(skb, nla_data(nla));
542 skb_copy_bits(skb, 0, nla_data(nla), skb->len);
544 err = genlmsg_multicast(user_skb, 0, group, GFP_ATOMIC);
555 while ((skb = nskb) != NULL) {
562 /* Called with genl_mutex. */
563 static int flush_flows(int dp_ifindex)
565 struct tbl *old_table;
566 struct tbl *new_table;
569 dp = get_dp(dp_ifindex);
573 old_table = get_table_protected(dp);
574 new_table = tbl_create(TBL_MIN_BUCKETS);
578 rcu_assign_pointer(dp->table, new_table);
580 tbl_deferred_destroy(old_table, flow_free_tbl);
585 static int validate_actions(const struct nlattr *attr)
587 const struct nlattr *a;
590 nla_for_each_nested(a, attr, rem) {
591 static const u32 action_lens[ODP_ACTION_ATTR_MAX + 1] = {
592 [ODP_ACTION_ATTR_OUTPUT] = 4,
593 [ODP_ACTION_ATTR_CONTROLLER] = 8,
594 [ODP_ACTION_ATTR_SET_DL_TCI] = 2,
595 [ODP_ACTION_ATTR_STRIP_VLAN] = 0,
596 [ODP_ACTION_ATTR_SET_DL_SRC] = ETH_ALEN,
597 [ODP_ACTION_ATTR_SET_DL_DST] = ETH_ALEN,
598 [ODP_ACTION_ATTR_SET_NW_SRC] = 4,
599 [ODP_ACTION_ATTR_SET_NW_DST] = 4,
600 [ODP_ACTION_ATTR_SET_NW_TOS] = 1,
601 [ODP_ACTION_ATTR_SET_TP_SRC] = 2,
602 [ODP_ACTION_ATTR_SET_TP_DST] = 2,
603 [ODP_ACTION_ATTR_SET_TUNNEL] = 8,
604 [ODP_ACTION_ATTR_SET_PRIORITY] = 4,
605 [ODP_ACTION_ATTR_POP_PRIORITY] = 0,
606 [ODP_ACTION_ATTR_DROP_SPOOFED_ARP] = 0,
608 int type = nla_type(a);
610 if (type > ODP_ACTION_ATTR_MAX || nla_len(a) != action_lens[type])
614 case ODP_ACTION_ATTR_UNSPEC:
617 case ODP_ACTION_ATTR_CONTROLLER:
618 case ODP_ACTION_ATTR_STRIP_VLAN:
619 case ODP_ACTION_ATTR_SET_DL_SRC:
620 case ODP_ACTION_ATTR_SET_DL_DST:
621 case ODP_ACTION_ATTR_SET_NW_SRC:
622 case ODP_ACTION_ATTR_SET_NW_DST:
623 case ODP_ACTION_ATTR_SET_TP_SRC:
624 case ODP_ACTION_ATTR_SET_TP_DST:
625 case ODP_ACTION_ATTR_SET_TUNNEL:
626 case ODP_ACTION_ATTR_SET_PRIORITY:
627 case ODP_ACTION_ATTR_POP_PRIORITY:
628 case ODP_ACTION_ATTR_DROP_SPOOFED_ARP:
629 /* No validation needed. */
632 case ODP_ACTION_ATTR_OUTPUT:
633 if (nla_get_u32(a) >= DP_MAX_PORTS)
637 case ODP_ACTION_ATTR_SET_DL_TCI:
638 if (nla_get_be16(a) & htons(VLAN_CFI_MASK))
642 case ODP_ACTION_ATTR_SET_NW_TOS:
643 if (nla_get_u8(a) & INET_ECN_MASK)
657 static void clear_stats(struct sw_flow *flow)
661 flow->packet_count = 0;
662 flow->byte_count = 0;
665 /* Called with genl_mutex. */
666 static int expand_table(struct datapath *dp)
668 struct tbl *old_table = get_table_protected(dp);
669 struct tbl *new_table;
671 new_table = tbl_expand(old_table);
672 if (IS_ERR(new_table))
673 return PTR_ERR(new_table);
675 rcu_assign_pointer(dp->table, new_table);
676 tbl_deferred_destroy(old_table, NULL);
681 static int odp_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
683 struct odp_header *odp_header = info->userhdr;
684 struct nlattr **a = info->attrs;
685 struct sk_buff *packet;
686 struct sw_flow_key key;
694 if (!a[ODP_PACKET_ATTR_PACKET] || !a[ODP_PACKET_ATTR_ACTIONS] ||
695 nla_len(a[ODP_PACKET_ATTR_PACKET]) < ETH_HLEN)
698 err = validate_actions(a[ODP_PACKET_ATTR_ACTIONS]);
702 len = nla_len(a[ODP_PACKET_ATTR_PACKET]);
703 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
707 skb_reserve(packet, NET_IP_ALIGN);
709 memcpy(__skb_put(packet, len), nla_data(a[ODP_PACKET_ATTR_PACKET]), len);
711 skb_reset_mac_header(packet);
712 eth = eth_hdr(packet);
714 /* Normally, setting the skb 'protocol' field would be handled by a
715 * call to eth_type_trans(), but it assumes there's a sending
716 * device, which we may not have. */
717 if (ntohs(eth->h_proto) >= 1536)
718 packet->protocol = eth->h_proto;
720 packet->protocol = htons(ETH_P_802_2);
722 /* Initialize OVS_CB (it came from Netlink so might not be zeroed). */
723 memset(OVS_CB(packet), 0, sizeof(struct ovs_skb_cb));
725 err = flow_extract(packet, -1, &key, &is_frag);
730 dp = get_dp(odp_header->dp_ifindex);
734 err = execute_actions(dp, packet, &key,
735 nla_data(a[ODP_PACKET_ATTR_ACTIONS]),
736 nla_len(a[ODP_PACKET_ATTR_ACTIONS]));
748 static const struct nla_policy packet_policy[ODP_PACKET_ATTR_MAX + 1] = {
749 [ODP_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
750 [ODP_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
753 static struct genl_ops dp_packet_genl_ops[] = {
754 { .cmd = ODP_PACKET_CMD_EXECUTE,
755 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
756 .policy = packet_policy,
757 .doit = odp_packet_cmd_execute
761 static void get_dp_stats(struct datapath *dp, struct odp_stats *stats)
765 stats->n_frags = stats->n_hit = stats->n_missed = stats->n_lost = 0;
766 for_each_possible_cpu(i) {
767 const struct dp_stats_percpu *percpu_stats;
768 struct dp_stats_percpu local_stats;
771 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
774 seqcount = read_seqcount_begin(&percpu_stats->seqlock);
775 local_stats = *percpu_stats;
776 } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount));
778 stats->n_frags += local_stats.n_frags;
779 stats->n_hit += local_stats.n_hit;
780 stats->n_missed += local_stats.n_missed;
781 stats->n_lost += local_stats.n_lost;
785 /* MTU of the dp pseudo-device: ETH_DATA_LEN or the minimum of the ports.
786 * Called with RTNL lock.
788 int dp_min_mtu(const struct datapath *dp)
795 list_for_each_entry (p, &dp->port_list, node) {
798 /* Skip any internal ports, since that's what we're trying to
800 if (is_internal_vport(p))
803 dev_mtu = vport_get_mtu(p);
806 if (!mtu || dev_mtu < mtu)
810 return mtu ? mtu : ETH_DATA_LEN;
813 /* Sets the MTU of all datapath devices to the minimum of the ports
814 * Called with RTNL lock.
816 void set_internal_devs_mtu(const struct datapath *dp)
823 mtu = dp_min_mtu(dp);
825 list_for_each_entry (p, &dp->port_list, node) {
826 if (is_internal_vport(p))
827 vport_set_mtu(p, mtu);
831 static const struct nla_policy flow_policy[ODP_FLOW_ATTR_MAX + 1] = {
832 [ODP_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
833 [ODP_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
834 [ODP_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
837 static struct genl_family dp_flow_genl_family = {
838 .id = GENL_ID_GENERATE,
839 .hdrsize = sizeof(struct odp_header),
840 .name = ODP_FLOW_FAMILY,
842 .maxattr = ODP_FLOW_ATTR_MAX
845 static struct genl_multicast_group dp_flow_multicast_group = {
846 .name = ODP_FLOW_MCGROUP
849 /* Called with genl_lock. */
850 static int odp_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
851 struct sk_buff *skb, u32 pid, u32 seq, u32 flags, u8 cmd)
853 const int skb_orig_len = skb->len;
854 const struct sw_flow_actions *sf_acts;
855 struct odp_flow_stats stats;
856 struct odp_header *odp_header;
862 sf_acts = rcu_dereference_protected(flow->sf_acts,
863 lockdep_genl_is_held());
865 odp_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags, cmd);
869 odp_header->dp_ifindex = dp->dp_ifindex;
871 nla = nla_nest_start(skb, ODP_FLOW_ATTR_KEY);
873 goto nla_put_failure;
874 err = flow_to_nlattrs(&flow->key, skb);
877 nla_nest_end(skb, nla);
879 spin_lock_bh(&flow->lock);
881 stats.n_packets = flow->packet_count;
882 stats.n_bytes = flow->byte_count;
883 tcp_flags = flow->tcp_flags;
884 spin_unlock_bh(&flow->lock);
887 NLA_PUT_U64(skb, ODP_FLOW_ATTR_USED, flow_used_time(used));
890 NLA_PUT(skb, ODP_FLOW_ATTR_STATS, sizeof(struct odp_flow_stats), &stats);
893 NLA_PUT_U8(skb, ODP_FLOW_ATTR_TCP_FLAGS, tcp_flags);
895 /* If ODP_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
896 * this is the first flow to be dumped into 'skb'. This is unusual for
897 * Netlink but individual action lists can be longer than
898 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
899 * The userspace caller can always fetch the actions separately if it
900 * really wants them. (Most userspace callers in fact don't care.)
902 * This can only fail for dump operations because the skb is always
903 * properly sized for single flows.
905 err = nla_put(skb, ODP_FLOW_ATTR_ACTIONS, sf_acts->actions_len,
907 if (err < 0 && skb_orig_len)
910 return genlmsg_end(skb, odp_header);
915 genlmsg_cancel(skb, odp_header);
919 static struct sk_buff *odp_flow_cmd_alloc_info(struct sw_flow *flow)
921 const struct sw_flow_actions *sf_acts;
924 sf_acts = rcu_dereference_protected(flow->sf_acts,
925 lockdep_genl_is_held());
927 len = nla_total_size(FLOW_BUFSIZE); /* ODP_FLOW_ATTR_KEY */
928 len += nla_total_size(sf_acts->actions_len); /* ODP_FLOW_ATTR_ACTIONS */
929 len += nla_total_size(sizeof(struct odp_flow_stats)); /* ODP_FLOW_ATTR_STATS */
930 len += nla_total_size(1); /* ODP_FLOW_ATTR_TCP_FLAGS */
931 len += nla_total_size(8); /* ODP_FLOW_ATTR_USED */
932 return genlmsg_new(NLMSG_ALIGN(sizeof(struct odp_header)) + len, GFP_KERNEL);
935 static struct sk_buff *odp_flow_cmd_build_info(struct sw_flow *flow, struct datapath *dp,
936 u32 pid, u32 seq, u8 cmd)
941 skb = odp_flow_cmd_alloc_info(flow);
943 return ERR_PTR(-ENOMEM);
945 retval = odp_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd);
950 static int odp_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
952 struct nlattr **a = info->attrs;
953 struct odp_header *odp_header = info->userhdr;
954 struct tbl_node *flow_node;
955 struct sw_flow_key key;
956 struct sw_flow *flow;
957 struct sk_buff *reply;
965 if (!a[ODP_FLOW_ATTR_KEY])
967 error = flow_from_nlattrs(&key, a[ODP_FLOW_ATTR_KEY]);
971 /* Validate actions. */
972 if (a[ODP_FLOW_ATTR_ACTIONS]) {
973 error = validate_actions(a[ODP_FLOW_ATTR_ACTIONS]);
976 } else if (info->genlhdr->cmd == ODP_FLOW_CMD_NEW) {
981 dp = get_dp(odp_header->dp_ifindex);
986 hash = flow_hash(&key);
987 table = get_table_protected(dp);
988 flow_node = tbl_lookup(table, &key, hash, flow_cmp);
990 struct sw_flow_actions *acts;
992 /* Bail out if we're not allowed to create a new flow. */
994 if (info->genlhdr->cmd == ODP_FLOW_CMD_SET)
997 /* Expand table, if necessary, to make room. */
998 if (tbl_count(table) >= tbl_n_buckets(table)) {
999 error = expand_table(dp);
1002 table = get_table_protected(dp);
1005 /* Allocate flow. */
1006 flow = flow_alloc();
1008 error = PTR_ERR(flow);
1014 /* Obtain actions. */
1015 acts = flow_actions_alloc(a[ODP_FLOW_ATTR_ACTIONS]);
1016 error = PTR_ERR(acts);
1018 goto error_free_flow;
1019 rcu_assign_pointer(flow->sf_acts, acts);
1021 /* Put flow in bucket. */
1022 error = tbl_insert(table, &flow->tbl_node, hash);
1024 goto error_free_flow;
1026 reply = odp_flow_cmd_build_info(flow, dp, info->snd_pid,
1027 info->snd_seq, ODP_FLOW_CMD_NEW);
1029 /* We found a matching flow. */
1030 struct sw_flow_actions *old_acts;
1032 /* Bail out if we're not allowed to modify an existing flow.
1033 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1034 * because Generic Netlink treats the latter as a dump
1035 * request. We also accept NLM_F_EXCL in case that bug ever
1039 if (info->genlhdr->cmd == ODP_FLOW_CMD_NEW &&
1040 info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1043 /* Update actions. */
1044 flow = flow_cast(flow_node);
1045 old_acts = rcu_dereference_protected(flow->sf_acts,
1046 lockdep_genl_is_held());
1047 if (a[ODP_FLOW_ATTR_ACTIONS] &&
1048 (old_acts->actions_len != nla_len(a[ODP_FLOW_ATTR_ACTIONS]) ||
1049 memcmp(old_acts->actions, nla_data(a[ODP_FLOW_ATTR_ACTIONS]),
1050 old_acts->actions_len))) {
1051 struct sw_flow_actions *new_acts;
1053 new_acts = flow_actions_alloc(a[ODP_FLOW_ATTR_ACTIONS]);
1054 error = PTR_ERR(new_acts);
1055 if (IS_ERR(new_acts))
1058 rcu_assign_pointer(flow->sf_acts, new_acts);
1059 flow_deferred_free_acts(old_acts);
1062 reply = odp_flow_cmd_build_info(flow, dp, info->snd_pid,
1063 info->snd_seq, ODP_FLOW_CMD_NEW);
1066 if (a[ODP_FLOW_ATTR_CLEAR]) {
1067 spin_lock_bh(&flow->lock);
1069 spin_unlock_bh(&flow->lock);
1074 genl_notify(reply, genl_info_net(info), info->snd_pid,
1075 dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1077 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1078 dp_flow_multicast_group.id, PTR_ERR(reply));
1087 static int odp_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1089 struct nlattr **a = info->attrs;
1090 struct odp_header *odp_header = info->userhdr;
1091 struct sw_flow_key key;
1092 struct tbl_node *flow_node;
1093 struct sk_buff *reply;
1094 struct sw_flow *flow;
1095 struct datapath *dp;
1099 if (!a[ODP_FLOW_ATTR_KEY])
1101 err = flow_from_nlattrs(&key, a[ODP_FLOW_ATTR_KEY]);
1105 dp = get_dp(odp_header->dp_ifindex);
1109 table = get_table_protected(dp);
1110 flow_node = tbl_lookup(table, &key, flow_hash(&key), flow_cmp);
1114 flow = flow_cast(flow_node);
1115 reply = odp_flow_cmd_build_info(flow, dp, info->snd_pid, info->snd_seq, ODP_FLOW_CMD_NEW);
1117 return PTR_ERR(reply);
1119 return genlmsg_reply(reply, info);
1122 static int odp_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1124 struct nlattr **a = info->attrs;
1125 struct odp_header *odp_header = info->userhdr;
1126 struct sw_flow_key key;
1127 struct tbl_node *flow_node;
1128 struct sk_buff *reply;
1129 struct sw_flow *flow;
1130 struct datapath *dp;
1134 if (!a[ODP_FLOW_ATTR_KEY])
1135 return flush_flows(odp_header->dp_ifindex);
1136 err = flow_from_nlattrs(&key, a[ODP_FLOW_ATTR_KEY]);
1140 dp = get_dp(odp_header->dp_ifindex);
1144 table = get_table_protected(dp);
1145 flow_node = tbl_lookup(table, &key, flow_hash(&key), flow_cmp);
1148 flow = flow_cast(flow_node);
1150 reply = odp_flow_cmd_alloc_info(flow);
1154 err = tbl_remove(table, flow_node);
1160 err = odp_flow_cmd_fill_info(flow, dp, reply, info->snd_pid,
1161 info->snd_seq, 0, ODP_FLOW_CMD_DEL);
1164 flow_deferred_free(flow);
1166 genl_notify(reply, genl_info_net(info), info->snd_pid,
1167 dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1171 static int odp_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1173 struct odp_header *odp_header = genlmsg_data(nlmsg_data(cb->nlh));
1174 struct datapath *dp;
1176 dp = get_dp(odp_header->dp_ifindex);
1181 struct tbl_node *flow_node;
1182 struct sw_flow *flow;
1185 bucket = cb->args[0];
1187 flow_node = tbl_next(get_table_protected(dp), &bucket, &obj);
1191 flow = flow_cast(flow_node);
1192 if (odp_flow_cmd_fill_info(flow, dp, skb, NETLINK_CB(cb->skb).pid,
1193 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1194 ODP_FLOW_CMD_NEW) < 0)
1197 cb->args[0] = bucket;
1203 static struct genl_ops dp_flow_genl_ops[] = {
1204 { .cmd = ODP_FLOW_CMD_NEW,
1205 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1206 .policy = flow_policy,
1207 .doit = odp_flow_cmd_new_or_set
1209 { .cmd = ODP_FLOW_CMD_DEL,
1210 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1211 .policy = flow_policy,
1212 .doit = odp_flow_cmd_del
1214 { .cmd = ODP_FLOW_CMD_GET,
1215 .flags = 0, /* OK for unprivileged users. */
1216 .policy = flow_policy,
1217 .doit = odp_flow_cmd_get,
1218 .dumpit = odp_flow_cmd_dump
1220 { .cmd = ODP_FLOW_CMD_SET,
1221 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1222 .policy = flow_policy,
1223 .doit = odp_flow_cmd_new_or_set,
1227 static const struct nla_policy datapath_policy[ODP_DP_ATTR_MAX + 1] = {
1228 #ifdef HAVE_NLA_NUL_STRING
1229 [ODP_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1231 [ODP_DP_ATTR_IPV4_FRAGS] = { .type = NLA_U32 },
1232 [ODP_DP_ATTR_SAMPLING] = { .type = NLA_U32 },
1235 static struct genl_family dp_datapath_genl_family = {
1236 .id = GENL_ID_GENERATE,
1237 .hdrsize = sizeof(struct odp_header),
1238 .name = ODP_DATAPATH_FAMILY,
1240 .maxattr = ODP_DP_ATTR_MAX
1243 static struct genl_multicast_group dp_datapath_multicast_group = {
1244 .name = ODP_DATAPATH_MCGROUP
1247 static int odp_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1248 u32 pid, u32 seq, u32 flags, u8 cmd)
1250 struct odp_header *odp_header;
1254 odp_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family,
1259 odp_header->dp_ifindex = dp->dp_ifindex;
1262 err = nla_put_string(skb, ODP_DP_ATTR_NAME, dp_name(dp));
1265 goto nla_put_failure;
1267 nla = nla_reserve(skb, ODP_DP_ATTR_STATS, sizeof(struct odp_stats));
1269 goto nla_put_failure;
1270 get_dp_stats(dp, nla_data(nla));
1272 NLA_PUT_U32(skb, ODP_DP_ATTR_IPV4_FRAGS,
1273 dp->drop_frags ? ODP_DP_FRAG_DROP : ODP_DP_FRAG_ZERO);
1275 if (dp->sflow_probability)
1276 NLA_PUT_U32(skb, ODP_DP_ATTR_SAMPLING, dp->sflow_probability);
1278 nla = nla_nest_start(skb, ODP_DP_ATTR_MCGROUPS);
1280 goto nla_put_failure;
1281 NLA_PUT_U32(skb, ODP_PACKET_CMD_MISS, packet_mc_group(dp, ODP_PACKET_CMD_MISS));
1282 NLA_PUT_U32(skb, ODP_PACKET_CMD_ACTION, packet_mc_group(dp, ODP_PACKET_CMD_ACTION));
1283 NLA_PUT_U32(skb, ODP_PACKET_CMD_SAMPLE, packet_mc_group(dp, ODP_PACKET_CMD_SAMPLE));
1284 nla_nest_end(skb, nla);
1286 return genlmsg_end(skb, odp_header);
1289 genlmsg_cancel(skb, odp_header);
1294 static struct sk_buff *odp_dp_cmd_build_info(struct datapath *dp, u32 pid,
1297 struct sk_buff *skb;
1300 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1302 return ERR_PTR(-ENOMEM);
1304 retval = odp_dp_cmd_fill_info(dp, skb, pid, seq, 0, cmd);
1307 return ERR_PTR(retval);
1312 static int odp_dp_cmd_validate(struct nlattr *a[ODP_DP_ATTR_MAX + 1])
1314 if (a[ODP_DP_ATTR_IPV4_FRAGS]) {
1315 u32 frags = nla_get_u32(a[ODP_DP_ATTR_IPV4_FRAGS]);
1317 if (frags != ODP_DP_FRAG_ZERO && frags != ODP_DP_FRAG_DROP)
1321 return CHECK_NUL_STRING(a[ODP_DP_ATTR_NAME], IFNAMSIZ - 1);
1324 /* Called with genl_mutex and optionally with RTNL lock also. */
1325 static struct datapath *lookup_datapath(struct odp_header *odp_header, struct nlattr *a[ODP_DP_ATTR_MAX + 1])
1327 struct datapath *dp;
1329 if (!a[ODP_DP_ATTR_NAME])
1330 dp = get_dp(odp_header->dp_ifindex);
1332 struct vport *vport;
1335 vport = vport_locate(nla_data(a[ODP_DP_ATTR_NAME]));
1336 dp = vport && vport->port_no == ODPP_LOCAL ? vport->dp : NULL;
1339 return dp ? dp : ERR_PTR(-ENODEV);
1342 /* Called with genl_mutex. */
1343 static void change_datapath(struct datapath *dp, struct nlattr *a[ODP_DP_ATTR_MAX + 1])
1345 if (a[ODP_DP_ATTR_IPV4_FRAGS])
1346 dp->drop_frags = nla_get_u32(a[ODP_DP_ATTR_IPV4_FRAGS]) == ODP_DP_FRAG_DROP;
1347 if (a[ODP_DP_ATTR_SAMPLING])
1348 dp->sflow_probability = nla_get_u32(a[ODP_DP_ATTR_SAMPLING]);
1351 static int odp_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1353 struct nlattr **a = info->attrs;
1354 struct vport_parms parms;
1355 struct sk_buff *reply;
1356 struct datapath *dp;
1357 struct vport *vport;
1361 if (!a[ODP_DP_ATTR_NAME])
1364 err = odp_dp_cmd_validate(a);
1370 if (!try_module_get(THIS_MODULE))
1371 goto err_unlock_rtnl;
1374 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1376 goto err_put_module;
1377 INIT_LIST_HEAD(&dp->port_list);
1379 /* Initialize kobject for bridge. This will be added as
1380 * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
1381 dp->ifobj.kset = NULL;
1382 kobject_init(&dp->ifobj, &dp_ktype);
1384 /* Allocate table. */
1386 rcu_assign_pointer(dp->table, tbl_create(TBL_MIN_BUCKETS));
1390 /* Set up our datapath device. */
1391 parms.name = nla_data(a[ODP_DP_ATTR_NAME]);
1392 parms.type = ODP_VPORT_TYPE_INTERNAL;
1393 parms.options = NULL;
1395 parms.port_no = ODPP_LOCAL;
1396 vport = new_vport(&parms);
1397 if (IS_ERR(vport)) {
1398 err = PTR_ERR(vport);
1402 goto err_destroy_table;
1404 dp->dp_ifindex = vport_get_ifindex(vport);
1407 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1408 if (!dp->stats_percpu) {
1410 goto err_destroy_local_port;
1413 change_datapath(dp, a);
1415 reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_NEW);
1416 err = PTR_ERR(reply);
1418 goto err_destroy_local_port;
1420 list_add_tail(&dp->list_node, &dps);
1421 dp_sysfs_add_dp(dp);
1425 genl_notify(reply, genl_info_net(info), info->snd_pid,
1426 dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1429 err_destroy_local_port:
1430 dp_detach_port(get_vport_protected(dp, ODPP_LOCAL));
1432 tbl_destroy(get_table_protected(dp), NULL);
1436 module_put(THIS_MODULE);
1443 static int odp_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1445 struct vport *vport, *next_vport;
1446 struct sk_buff *reply;
1447 struct datapath *dp;
1450 err = odp_dp_cmd_validate(info->attrs);
1455 dp = lookup_datapath(info->userhdr, info->attrs);
1460 reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_DEL);
1461 err = PTR_ERR(reply);
1465 list_for_each_entry_safe (vport, next_vport, &dp->port_list, node)
1466 if (vport->port_no != ODPP_LOCAL)
1467 dp_detach_port(vport);
1469 dp_sysfs_del_dp(dp);
1470 list_del(&dp->list_node);
1471 dp_detach_port(get_vport_protected(dp, ODPP_LOCAL));
1473 /* rtnl_unlock() will wait until all the references to devices that
1474 * are pending unregistration have been dropped. We do it here to
1475 * ensure that any internal devices (which contain DP pointers) are
1476 * fully destroyed before freeing the datapath.
1480 call_rcu(&dp->rcu, destroy_dp_rcu);
1481 module_put(THIS_MODULE);
1483 genl_notify(reply, genl_info_net(info), info->snd_pid,
1484 dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1494 static int odp_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1496 struct sk_buff *reply;
1497 struct datapath *dp;
1500 err = odp_dp_cmd_validate(info->attrs);
1504 dp = lookup_datapath(info->userhdr, info->attrs);
1508 change_datapath(dp, info->attrs);
1510 reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_NEW);
1511 if (IS_ERR(reply)) {
1512 err = PTR_ERR(reply);
1513 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1514 dp_datapath_multicast_group.id, err);
1518 genl_notify(reply, genl_info_net(info), info->snd_pid,
1519 dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1523 static int odp_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1525 struct sk_buff *reply;
1526 struct datapath *dp;
1529 err = odp_dp_cmd_validate(info->attrs);
1533 dp = lookup_datapath(info->userhdr, info->attrs);
1537 reply = odp_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, ODP_DP_CMD_NEW);
1539 return PTR_ERR(reply);
1541 return genlmsg_reply(reply, info);
1544 static int odp_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1546 struct datapath *dp;
1547 int skip = cb->args[0];
1550 list_for_each_entry (dp, &dps, list_node) {
1553 if (odp_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).pid,
1554 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1555 ODP_DP_CMD_NEW) < 0)
1565 static struct genl_ops dp_datapath_genl_ops[] = {
1566 { .cmd = ODP_DP_CMD_NEW,
1567 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1568 .policy = datapath_policy,
1569 .doit = odp_dp_cmd_new
1571 { .cmd = ODP_DP_CMD_DEL,
1572 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1573 .policy = datapath_policy,
1574 .doit = odp_dp_cmd_del
1576 { .cmd = ODP_DP_CMD_GET,
1577 .flags = 0, /* OK for unprivileged users. */
1578 .policy = datapath_policy,
1579 .doit = odp_dp_cmd_get,
1580 .dumpit = odp_dp_cmd_dump
1582 { .cmd = ODP_DP_CMD_SET,
1583 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1584 .policy = datapath_policy,
1585 .doit = odp_dp_cmd_set,
1589 static const struct nla_policy vport_policy[ODP_VPORT_ATTR_MAX + 1] = {
1590 #ifdef HAVE_NLA_NUL_STRING
1591 [ODP_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1592 [ODP_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1593 [ODP_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1594 [ODP_VPORT_ATTR_STATS] = { .len = sizeof(struct rtnl_link_stats64) },
1595 [ODP_VPORT_ATTR_ADDRESS] = { .len = ETH_ALEN },
1597 [ODP_VPORT_ATTR_STATS] = { .minlen = sizeof(struct rtnl_link_stats64) },
1598 [ODP_VPORT_ATTR_ADDRESS] = { .minlen = ETH_ALEN },
1600 [ODP_VPORT_ATTR_MTU] = { .type = NLA_U32 },
1601 [ODP_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1604 static struct genl_family dp_vport_genl_family = {
1605 .id = GENL_ID_GENERATE,
1606 .hdrsize = sizeof(struct odp_header),
1607 .name = ODP_VPORT_FAMILY,
1609 .maxattr = ODP_VPORT_ATTR_MAX
1612 static struct genl_multicast_group dp_vport_multicast_group = {
1613 .name = ODP_VPORT_MCGROUP
1616 /* Called with RTNL lock or RCU read lock. */
1617 static int odp_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1618 u32 pid, u32 seq, u32 flags, u8 cmd)
1620 struct odp_header *odp_header;
1622 int ifindex, iflink;
1626 odp_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family,
1631 odp_header->dp_ifindex = vport->dp->dp_ifindex;
1633 NLA_PUT_U32(skb, ODP_VPORT_ATTR_PORT_NO, vport->port_no);
1634 NLA_PUT_U32(skb, ODP_VPORT_ATTR_TYPE, vport_get_type(vport));
1635 NLA_PUT_STRING(skb, ODP_VPORT_ATTR_NAME, vport_get_name(vport));
1637 nla = nla_reserve(skb, ODP_VPORT_ATTR_STATS, sizeof(struct rtnl_link_stats64));
1639 goto nla_put_failure;
1640 if (vport_get_stats(vport, nla_data(nla)))
1641 __skb_trim(skb, skb->len - nla->nla_len);
1643 NLA_PUT(skb, ODP_VPORT_ATTR_ADDRESS, ETH_ALEN, vport_get_addr(vport));
1645 mtu = vport_get_mtu(vport);
1647 NLA_PUT_U32(skb, ODP_VPORT_ATTR_MTU, mtu);
1649 err = vport_get_options(vport, skb);
1650 if (err == -EMSGSIZE)
1653 ifindex = vport_get_ifindex(vport);
1655 NLA_PUT_U32(skb, ODP_VPORT_ATTR_IFINDEX, ifindex);
1657 iflink = vport_get_iflink(vport);
1659 NLA_PUT_U32(skb, ODP_VPORT_ATTR_IFLINK, iflink);
1661 return genlmsg_end(skb, odp_header);
1666 genlmsg_cancel(skb, odp_header);
1670 /* Called with RTNL lock or RCU read lock. */
1671 static struct sk_buff *odp_vport_cmd_build_info(struct vport *vport, u32 pid,
1674 struct sk_buff *skb;
1677 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1679 return ERR_PTR(-ENOMEM);
1681 retval = odp_vport_cmd_fill_info(vport, skb, pid, seq, 0, cmd);
1684 return ERR_PTR(retval);
1689 static int odp_vport_cmd_validate(struct nlattr *a[ODP_VPORT_ATTR_MAX + 1])
1691 return CHECK_NUL_STRING(a[ODP_VPORT_ATTR_NAME], IFNAMSIZ - 1);
1694 /* Called with RTNL lock or RCU read lock. */
1695 static struct vport *lookup_vport(struct odp_header *odp_header,
1696 struct nlattr *a[ODP_VPORT_ATTR_MAX + 1])
1698 struct datapath *dp;
1699 struct vport *vport;
1701 if (a[ODP_VPORT_ATTR_NAME]) {
1702 vport = vport_locate(nla_data(a[ODP_VPORT_ATTR_NAME]));
1704 return ERR_PTR(-ENODEV);
1706 } else if (a[ODP_VPORT_ATTR_PORT_NO]) {
1707 u32 port_no = nla_get_u32(a[ODP_VPORT_ATTR_PORT_NO]);
1709 if (port_no >= DP_MAX_PORTS)
1710 return ERR_PTR(-EFBIG);
1712 dp = get_dp(odp_header->dp_ifindex);
1714 return ERR_PTR(-ENODEV);
1716 vport = get_vport_protected(dp, port_no);
1718 return ERR_PTR(-ENOENT);
1721 return ERR_PTR(-EINVAL);
1724 /* Called with RTNL lock. */
1725 static int change_vport(struct vport *vport, struct nlattr *a[ODP_VPORT_ATTR_MAX + 1])
1728 if (a[ODP_VPORT_ATTR_STATS])
1729 err = vport_set_stats(vport, nla_data(a[ODP_VPORT_ATTR_STATS]));
1730 if (!err && a[ODP_VPORT_ATTR_ADDRESS])
1731 err = vport_set_addr(vport, nla_data(a[ODP_VPORT_ATTR_ADDRESS]));
1732 if (!err && a[ODP_VPORT_ATTR_MTU])
1733 err = vport_set_mtu(vport, nla_get_u32(a[ODP_VPORT_ATTR_MTU]));
1737 static int odp_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1739 struct nlattr **a = info->attrs;
1740 struct odp_header *odp_header = info->userhdr;
1741 struct vport_parms parms;
1742 struct sk_buff *reply;
1743 struct vport *vport;
1744 struct datapath *dp;
1749 if (!a[ODP_VPORT_ATTR_NAME] || !a[ODP_VPORT_ATTR_TYPE])
1752 err = odp_vport_cmd_validate(a);
1757 dp = get_dp(odp_header->dp_ifindex);
1762 if (a[ODP_VPORT_ATTR_PORT_NO]) {
1763 port_no = nla_get_u32(a[ODP_VPORT_ATTR_PORT_NO]);
1766 if (port_no >= DP_MAX_PORTS)
1769 vport = get_vport_protected(dp, port_no);
1774 for (port_no = 1; ; port_no++) {
1775 if (port_no >= DP_MAX_PORTS) {
1779 vport = get_vport_protected(dp, port_no);
1785 parms.name = nla_data(a[ODP_VPORT_ATTR_NAME]);
1786 parms.type = nla_get_u32(a[ODP_VPORT_ATTR_TYPE]);
1787 parms.options = a[ODP_VPORT_ATTR_OPTIONS];
1789 parms.port_no = port_no;
1791 vport = new_vport(&parms);
1792 err = PTR_ERR(vport);
1796 set_internal_devs_mtu(dp);
1797 dp_sysfs_add_if(vport);
1799 err = change_vport(vport, a);
1801 reply = odp_vport_cmd_build_info(vport, info->snd_pid,
1802 info->snd_seq, ODP_VPORT_CMD_NEW);
1804 err = PTR_ERR(reply);
1807 dp_detach_port(vport);
1810 genl_notify(reply, genl_info_net(info), info->snd_pid,
1811 dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1820 static int odp_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1822 struct nlattr **a = info->attrs;
1823 struct sk_buff *reply;
1824 struct vport *vport;
1827 err = odp_vport_cmd_validate(a);
1832 vport = lookup_vport(info->userhdr, a);
1833 err = PTR_ERR(vport);
1838 if (a[ODP_VPORT_ATTR_OPTIONS])
1839 err = vport_set_options(vport, a[ODP_VPORT_ATTR_OPTIONS]);
1841 err = change_vport(vport, a);
1843 reply = odp_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1845 if (IS_ERR(reply)) {
1846 err = PTR_ERR(reply);
1847 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1848 dp_vport_multicast_group.id, err);
1852 genl_notify(reply, genl_info_net(info), info->snd_pid,
1853 dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1861 static int odp_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1863 struct nlattr **a = info->attrs;
1864 struct sk_buff *reply;
1865 struct vport *vport;
1868 err = odp_vport_cmd_validate(a);
1873 vport = lookup_vport(info->userhdr, a);
1874 err = PTR_ERR(vport);
1878 if (vport->port_no == ODPP_LOCAL) {
1883 reply = odp_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1885 err = PTR_ERR(reply);
1889 err = dp_detach_port(vport);
1891 genl_notify(reply, genl_info_net(info), info->snd_pid,
1892 dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1900 static int odp_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1902 struct nlattr **a = info->attrs;
1903 struct odp_header *odp_header = info->userhdr;
1904 struct sk_buff *reply;
1905 struct vport *vport;
1908 err = odp_vport_cmd_validate(a);
1913 vport = lookup_vport(odp_header, a);
1914 err = PTR_ERR(vport);
1918 reply = odp_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1920 err = PTR_ERR(reply);
1924 err = genlmsg_reply(reply, info);
1932 static int odp_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1934 struct odp_header *odp_header = genlmsg_data(nlmsg_data(cb->nlh));
1935 struct datapath *dp;
1939 dp = get_dp(odp_header->dp_ifindex);
1944 for (port_no = cb->args[0]; port_no < DP_MAX_PORTS; port_no++) {
1945 struct vport *vport;
1947 vport = get_vport_protected(dp, port_no);
1951 if (odp_vport_cmd_fill_info(vport, skb, NETLINK_CB(cb->skb).pid,
1952 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1953 ODP_VPORT_CMD_NEW) < 0)
1958 cb->args[0] = port_no;
1964 static struct genl_ops dp_vport_genl_ops[] = {
1965 { .cmd = ODP_VPORT_CMD_NEW,
1966 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1967 .policy = vport_policy,
1968 .doit = odp_vport_cmd_new
1970 { .cmd = ODP_VPORT_CMD_DEL,
1971 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1972 .policy = vport_policy,
1973 .doit = odp_vport_cmd_del
1975 { .cmd = ODP_VPORT_CMD_GET,
1976 .flags = 0, /* OK for unprivileged users. */
1977 .policy = vport_policy,
1978 .doit = odp_vport_cmd_get,
1979 .dumpit = odp_vport_cmd_dump
1981 { .cmd = ODP_VPORT_CMD_SET,
1982 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1983 .policy = vport_policy,
1984 .doit = odp_vport_cmd_set,
1988 struct genl_family_and_ops {
1989 struct genl_family *family;
1990 struct genl_ops *ops;
1992 struct genl_multicast_group *group;
1995 static const struct genl_family_and_ops dp_genl_families[] = {
1996 { &dp_datapath_genl_family,
1997 dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1998 &dp_datapath_multicast_group },
1999 { &dp_vport_genl_family,
2000 dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
2001 &dp_vport_multicast_group },
2002 { &dp_flow_genl_family,
2003 dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
2004 &dp_flow_multicast_group },
2005 { &dp_packet_genl_family,
2006 dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
2010 static void dp_unregister_genl(int n_families)
2014 for (i = 0; i < n_families; i++) {
2015 genl_unregister_family(dp_genl_families[i].family);
2019 static int dp_register_genl(void)
2026 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2027 const struct genl_family_and_ops *f = &dp_genl_families[i];
2029 err = genl_register_family_with_ops(f->family, f->ops,
2036 err = genl_register_mc_group(f->family, f->group);
2042 err = packet_register_mc_groups();
2048 dp_unregister_genl(n_registered);
2052 static int __init dp_init(void)
2054 struct sk_buff *dummy_skb;
2057 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
2059 printk("Open vSwitch %s, built "__DATE__" "__TIME__"\n", VERSION BUILDNR);
2067 goto error_flow_exit;
2069 err = register_netdevice_notifier(&dp_device_notifier);
2071 goto error_vport_exit;
2073 err = dp_register_genl();
2075 goto error_unreg_notifier;
2079 error_unreg_notifier:
2080 unregister_netdevice_notifier(&dp_device_notifier);
2089 static void dp_cleanup(void)
2092 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2093 unregister_netdevice_notifier(&dp_device_notifier);
2098 module_init(dp_init);
2099 module_exit(dp_cleanup);
2101 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2102 MODULE_LICENSE("GPL");