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/openvswitch.h>
43 #include <linux/rculist.h>
44 #include <linux/dmi.h>
45 #include <net/inet_ecn.h>
46 #include <net/genetlink.h>
54 #include "vport-internal_dev.h"
56 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,18) || \
57 LINUX_VERSION_CODE >= KERNEL_VERSION(3,1,0)
58 #error Kernels before 2.6.18 or after 3.0 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_userspace_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 flow_table *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[OVSP_LOCAL]));
127 static int get_dpifindex(struct datapath *dp)
134 local = get_vport_protected(dp, OVSP_LOCAL);
136 ifindex = vport_get_ifindex(local);
145 static inline size_t br_nlmsg_size(void)
147 return NLMSG_ALIGN(sizeof(struct ifinfomsg))
148 + nla_total_size(IFNAMSIZ) /* IFLA_IFNAME */
149 + nla_total_size(MAX_ADDR_LEN) /* IFLA_ADDRESS */
150 + nla_total_size(4) /* IFLA_MASTER */
151 + nla_total_size(4) /* IFLA_MTU */
152 + nla_total_size(1); /* IFLA_OPERSTATE */
155 /* Caller must hold RTNL lock. */
156 static int dp_fill_ifinfo(struct sk_buff *skb,
157 const struct vport *port,
158 int event, unsigned int flags)
160 struct datapath *dp = port->dp;
161 int ifindex = vport_get_ifindex(port);
162 struct ifinfomsg *hdr;
163 struct nlmsghdr *nlh;
168 nlh = nlmsg_put(skb, 0, 0, event, sizeof(*hdr), flags);
172 hdr = nlmsg_data(nlh);
173 hdr->ifi_family = AF_BRIDGE;
175 hdr->ifi_type = ARPHRD_ETHER;
176 hdr->ifi_index = ifindex;
177 hdr->ifi_flags = vport_get_flags(port);
180 NLA_PUT_STRING(skb, IFLA_IFNAME, vport_get_name(port));
181 NLA_PUT_U32(skb, IFLA_MASTER, get_dpifindex(dp));
182 NLA_PUT_U32(skb, IFLA_MTU, vport_get_mtu(port));
183 #ifdef IFLA_OPERSTATE
184 NLA_PUT_U8(skb, IFLA_OPERSTATE,
185 vport_is_running(port)
186 ? vport_get_operstate(port)
190 NLA_PUT(skb, IFLA_ADDRESS, ETH_ALEN, vport_get_addr(port));
192 return nlmsg_end(skb, nlh);
195 nlmsg_cancel(skb, nlh);
199 /* Caller must hold RTNL lock. */
200 static void dp_ifinfo_notify(int event, struct vport *port)
205 skb = nlmsg_new(br_nlmsg_size(), GFP_KERNEL);
209 err = dp_fill_ifinfo(skb, port, event, 0);
211 /* -EMSGSIZE implies BUG in br_nlmsg_size() */
212 WARN_ON(err == -EMSGSIZE);
216 rtnl_notify(skb, &init_net, 0, RTNLGRP_LINK, NULL, GFP_KERNEL);
220 rtnl_set_sk_err(&init_net, RTNLGRP_LINK, err);
223 static void release_dp(struct kobject *kobj)
225 struct datapath *dp = container_of(kobj, struct datapath, ifobj);
229 static struct kobj_type dp_ktype = {
230 .release = release_dp
233 static void destroy_dp_rcu(struct rcu_head *rcu)
235 struct datapath *dp = container_of(rcu, struct datapath, rcu);
237 flow_tbl_destroy(dp->table);
238 free_percpu(dp->stats_percpu);
239 kobject_put(&dp->ifobj);
242 /* Called with RTNL lock and genl_lock. */
243 static struct vport *new_vport(const struct vport_parms *parms)
247 vport = vport_add(parms);
248 if (!IS_ERR(vport)) {
249 struct datapath *dp = parms->dp;
251 rcu_assign_pointer(dp->ports[parms->port_no], vport);
252 list_add(&vport->node, &dp->port_list);
254 dp_ifinfo_notify(RTM_NEWLINK, vport);
260 /* Called with RTNL lock. */
261 void dp_detach_port(struct vport *p)
265 if (p->port_no != OVSP_LOCAL)
267 dp_ifinfo_notify(RTM_DELLINK, p);
269 /* First drop references to device. */
271 rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
273 /* Then destroy it. */
277 /* Must be called with rcu_read_lock. */
278 void dp_process_received_packet(struct vport *p, struct sk_buff *skb)
280 struct datapath *dp = p->dp;
281 struct sw_flow *flow;
282 struct dp_stats_percpu *stats;
286 stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
287 OVS_CB(skb)->vport = p;
289 if (!OVS_CB(skb)->flow) {
290 struct sw_flow_key key;
294 /* Extract flow from 'skb' into 'key'. */
295 error = flow_extract(skb, p->port_no, &key, &key_len, &is_frag);
296 if (unlikely(error)) {
301 if (is_frag && dp->drop_frags) {
303 stats_counter = &stats->n_frags;
308 flow = flow_tbl_lookup(rcu_dereference(dp->table), &key, key_len);
309 if (unlikely(!flow)) {
310 struct dp_upcall_info upcall;
312 upcall.cmd = OVS_PACKET_CMD_MISS;
314 upcall.userdata = NULL;
315 upcall.pid = p->upcall_pid;
316 dp_upcall(dp, skb, &upcall);
318 stats_counter = &stats->n_missed;
322 OVS_CB(skb)->flow = flow;
325 stats_counter = &stats->n_hit;
326 flow_used(OVS_CB(skb)->flow, skb);
327 execute_actions(dp, skb);
330 /* Update datapath statistics. */
332 write_seqcount_begin(&stats->seqlock);
334 write_seqcount_end(&stats->seqlock);
337 static void copy_and_csum_skb(struct sk_buff *skb, void *to)
339 u16 csum_start, csum_offset;
342 get_skb_csum_pointers(skb, &csum_start, &csum_offset);
343 csum_start -= skb_headroom(skb);
345 skb_copy_bits(skb, 0, to, csum_start);
347 csum = skb_copy_and_csum_bits(skb, csum_start, to + csum_start,
348 skb->len - csum_start, 0);
349 *(__sum16 *)(to + csum_start + csum_offset) = csum_fold(csum);
352 static struct genl_family dp_packet_genl_family = {
353 .id = GENL_ID_GENERATE,
354 .hdrsize = sizeof(struct ovs_header),
355 .name = OVS_PACKET_FAMILY,
357 .maxattr = OVS_PACKET_ATTR_MAX
360 int dp_upcall(struct datapath *dp, struct sk_buff *skb,
361 const struct dp_upcall_info *upcall_info)
363 struct sk_buff *segs = NULL;
364 struct dp_stats_percpu *stats;
367 if (upcall_info->pid == 0) {
372 forward_ip_summed(skb, true);
374 /* Break apart GSO packets into their component pieces. Otherwise
375 * userspace may try to stuff a 64kB packet into a 1500-byte MTU. */
376 if (skb_is_gso(skb)) {
377 segs = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
386 err = queue_userspace_packets(dp, skb, upcall_info);
388 struct sk_buff *next;
389 /* Free GSO-segments */
393 } while ((segs = next) != NULL);
402 stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
404 write_seqcount_begin(&stats->seqlock);
406 write_seqcount_end(&stats->seqlock);
411 /* Send each packet in the 'skb' list to userspace for 'dp' as directed by
412 * 'upcall_info'. There will be only one packet unless we broke up a GSO
415 static int queue_userspace_packets(struct datapath *dp, struct sk_buff *skb,
416 const struct dp_upcall_info *upcall_info)
420 dp_ifindex = get_dpifindex(dp);
425 struct ovs_header *upcall;
426 struct sk_buff *user_skb; /* to be queued to userspace */
431 err = vlan_deaccel_tag(skb);
435 if (nla_attr_size(skb->len) > USHRT_MAX)
438 len = sizeof(struct ovs_header);
439 len += nla_total_size(skb->len);
440 len += nla_total_size(FLOW_BUFSIZE);
441 if (upcall_info->cmd == OVS_PACKET_CMD_ACTION)
442 len += nla_total_size(8);
444 user_skb = genlmsg_new(len, GFP_ATOMIC);
448 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
449 0, upcall_info->cmd);
450 upcall->dp_ifindex = dp_ifindex;
452 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
453 flow_to_nlattrs(upcall_info->key, user_skb);
454 nla_nest_end(user_skb, nla);
456 if (upcall_info->userdata)
457 nla_put_u64(user_skb, OVS_PACKET_ATTR_USERDATA,
458 nla_get_u64(upcall_info->userdata));
460 nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
461 if (skb->ip_summed == CHECKSUM_PARTIAL)
462 copy_and_csum_skb(skb, nla_data(nla));
464 skb_copy_bits(skb, 0, nla_data(nla), skb->len);
466 err = genlmsg_unicast(&init_net, user_skb, upcall_info->pid);
470 } while ((skb = skb->next));
475 /* Called with genl_mutex. */
476 static int flush_flows(int dp_ifindex)
478 struct flow_table *old_table;
479 struct flow_table *new_table;
482 dp = get_dp(dp_ifindex);
486 old_table = get_table_protected(dp);
487 new_table = flow_tbl_alloc(TBL_MIN_BUCKETS);
491 rcu_assign_pointer(dp->table, new_table);
493 flow_tbl_deferred_destroy(old_table);
497 static int validate_actions(const struct nlattr *attr, int depth);
499 static int validate_sample(const struct nlattr *attr, int depth)
501 static const struct nla_policy sample_policy[OVS_SAMPLE_ATTR_MAX + 1] =
503 [OVS_SAMPLE_ATTR_PROBABILITY] = {.type = NLA_U32 },
504 [OVS_SAMPLE_ATTR_ACTIONS] = {.type = NLA_UNSPEC },
506 struct nlattr *a[OVS_SAMPLE_ATTR_MAX + 1];
509 error = nla_parse_nested(a, OVS_SAMPLE_ATTR_MAX, attr, sample_policy);
513 if (!a[OVS_SAMPLE_ATTR_PROBABILITY])
515 if (!a[OVS_SAMPLE_ATTR_ACTIONS])
518 return validate_actions(a[OVS_SAMPLE_ATTR_ACTIONS], (depth + 1));
521 static int validate_userspace(const struct nlattr *attr)
523 static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] =
525 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
526 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_U64 },
528 struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
531 error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX, attr, userspace_policy);
535 if (!a[OVS_USERSPACE_ATTR_PID] || !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
541 static int validate_actions(const struct nlattr *attr, int depth)
543 const struct nlattr *a;
546 if (depth >= SAMPLE_ACTION_DEPTH)
549 nla_for_each_nested(a, attr, rem) {
550 /* Expected argument lengths, (u32)-1 for variable length. */
551 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
552 [OVS_ACTION_ATTR_OUTPUT] = 4,
553 [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
554 [OVS_ACTION_ATTR_PUSH_VLAN] = 2,
555 [OVS_ACTION_ATTR_POP_VLAN] = 0,
556 [OVS_ACTION_ATTR_SET_DL_SRC] = ETH_ALEN,
557 [OVS_ACTION_ATTR_SET_DL_DST] = ETH_ALEN,
558 [OVS_ACTION_ATTR_SET_NW_SRC] = 4,
559 [OVS_ACTION_ATTR_SET_NW_DST] = 4,
560 [OVS_ACTION_ATTR_SET_NW_TOS] = 1,
561 [OVS_ACTION_ATTR_SET_TP_SRC] = 2,
562 [OVS_ACTION_ATTR_SET_TP_DST] = 2,
563 [OVS_ACTION_ATTR_SET_TUNNEL] = 8,
564 [OVS_ACTION_ATTR_SET_PRIORITY] = 4,
565 [OVS_ACTION_ATTR_POP_PRIORITY] = 0,
566 [OVS_ACTION_ATTR_SAMPLE] = (u32)-1
568 int type = nla_type(a);
570 if (type > OVS_ACTION_ATTR_MAX ||
571 (action_lens[type] != nla_len(a) &&
572 action_lens[type] != (u32)-1))
576 case OVS_ACTION_ATTR_UNSPEC:
579 case OVS_ACTION_ATTR_POP_VLAN:
580 case OVS_ACTION_ATTR_SET_DL_SRC:
581 case OVS_ACTION_ATTR_SET_DL_DST:
582 case OVS_ACTION_ATTR_SET_NW_SRC:
583 case OVS_ACTION_ATTR_SET_NW_DST:
584 case OVS_ACTION_ATTR_SET_TP_SRC:
585 case OVS_ACTION_ATTR_SET_TP_DST:
586 case OVS_ACTION_ATTR_SET_TUNNEL:
587 case OVS_ACTION_ATTR_SET_PRIORITY:
588 case OVS_ACTION_ATTR_POP_PRIORITY:
589 /* No validation needed. */
592 case OVS_ACTION_ATTR_USERSPACE:
593 err = validate_userspace(a);
598 case OVS_ACTION_ATTR_OUTPUT:
599 if (nla_get_u32(a) >= DP_MAX_PORTS)
603 case OVS_ACTION_ATTR_PUSH_VLAN:
604 if (nla_get_be16(a) & htons(VLAN_CFI_MASK))
608 case OVS_ACTION_ATTR_SET_NW_TOS:
609 if (nla_get_u8(a) & INET_ECN_MASK)
613 case OVS_ACTION_ATTR_SAMPLE:
614 err = validate_sample(a, depth);
629 static void clear_stats(struct sw_flow *flow)
633 flow->packet_count = 0;
634 flow->byte_count = 0;
637 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
639 struct ovs_header *ovs_header = info->userhdr;
640 struct nlattr **a = info->attrs;
641 struct sw_flow_actions *acts;
642 struct sk_buff *packet;
643 struct sw_flow *flow;
652 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
653 !a[OVS_PACKET_ATTR_ACTIONS] ||
654 nla_len(a[OVS_PACKET_ATTR_PACKET]) < ETH_HLEN)
657 err = validate_actions(a[OVS_PACKET_ATTR_ACTIONS], 0);
661 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
662 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
666 skb_reserve(packet, NET_IP_ALIGN);
668 memcpy(__skb_put(packet, len), nla_data(a[OVS_PACKET_ATTR_PACKET]), len);
670 skb_reset_mac_header(packet);
671 eth = eth_hdr(packet);
673 /* Normally, setting the skb 'protocol' field would be handled by a
674 * call to eth_type_trans(), but it assumes there's a sending
675 * device, which we may not have. */
676 if (ntohs(eth->h_proto) >= 1536)
677 packet->protocol = eth->h_proto;
679 packet->protocol = htons(ETH_P_802_2);
681 /* Build an sw_flow for sending this packet. */
687 err = flow_extract(packet, -1, &flow->key, &key_len, &is_frag);
691 err = flow_metadata_from_nlattrs(&flow->key.eth.in_port,
692 &flow->key.eth.tun_id,
693 a[OVS_PACKET_ATTR_KEY]);
697 flow->hash = flow_hash(&flow->key, key_len);
699 acts = flow_actions_alloc(a[OVS_PACKET_ATTR_ACTIONS]);
703 rcu_assign_pointer(flow->sf_acts, acts);
705 OVS_CB(packet)->flow = flow;
708 dp = get_dp(ovs_header->dp_ifindex);
713 if (flow->key.eth.in_port < DP_MAX_PORTS)
714 OVS_CB(packet)->vport = get_vport_protected(dp,
715 flow->key.eth.in_port);
718 err = execute_actions(dp, packet);
735 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
736 [OVS_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
737 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
738 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
741 static struct genl_ops dp_packet_genl_ops[] = {
742 { .cmd = OVS_PACKET_CMD_EXECUTE,
743 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
744 .policy = packet_policy,
745 .doit = ovs_packet_cmd_execute
749 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
752 struct flow_table *table = get_table_protected(dp);
754 stats->n_flows = flow_tbl_count(table);
756 stats->n_frags = stats->n_hit = stats->n_missed = stats->n_lost = 0;
757 for_each_possible_cpu(i) {
758 const struct dp_stats_percpu *percpu_stats;
759 struct dp_stats_percpu local_stats;
762 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
765 seqcount = read_seqcount_begin(&percpu_stats->seqlock);
766 local_stats = *percpu_stats;
767 } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount));
769 stats->n_frags += local_stats.n_frags;
770 stats->n_hit += local_stats.n_hit;
771 stats->n_missed += local_stats.n_missed;
772 stats->n_lost += local_stats.n_lost;
776 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
777 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
778 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
779 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
782 static struct genl_family dp_flow_genl_family = {
783 .id = GENL_ID_GENERATE,
784 .hdrsize = sizeof(struct ovs_header),
785 .name = OVS_FLOW_FAMILY,
787 .maxattr = OVS_FLOW_ATTR_MAX
790 static struct genl_multicast_group dp_flow_multicast_group = {
791 .name = OVS_FLOW_MCGROUP
794 /* Called with genl_lock. */
795 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
796 struct sk_buff *skb, u32 pid, u32 seq, u32 flags, u8 cmd)
798 const int skb_orig_len = skb->len;
799 const struct sw_flow_actions *sf_acts;
800 struct ovs_flow_stats stats;
801 struct ovs_header *ovs_header;
807 sf_acts = rcu_dereference_protected(flow->sf_acts,
808 lockdep_genl_is_held());
810 ovs_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags, cmd);
814 ovs_header->dp_ifindex = get_dpifindex(dp);
816 nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
818 goto nla_put_failure;
819 err = flow_to_nlattrs(&flow->key, skb);
822 nla_nest_end(skb, nla);
824 spin_lock_bh(&flow->lock);
826 stats.n_packets = flow->packet_count;
827 stats.n_bytes = flow->byte_count;
828 tcp_flags = flow->tcp_flags;
829 spin_unlock_bh(&flow->lock);
832 NLA_PUT_U64(skb, OVS_FLOW_ATTR_USED, flow_used_time(used));
835 NLA_PUT(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats);
838 NLA_PUT_U8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags);
840 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
841 * this is the first flow to be dumped into 'skb'. This is unusual for
842 * Netlink but individual action lists can be longer than
843 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
844 * The userspace caller can always fetch the actions separately if it
845 * really wants them. (Most userspace callers in fact don't care.)
847 * This can only fail for dump operations because the skb is always
848 * properly sized for single flows.
850 err = nla_put(skb, OVS_FLOW_ATTR_ACTIONS, sf_acts->actions_len,
852 if (err < 0 && skb_orig_len)
855 return genlmsg_end(skb, ovs_header);
860 genlmsg_cancel(skb, ovs_header);
864 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
866 const struct sw_flow_actions *sf_acts;
869 sf_acts = rcu_dereference_protected(flow->sf_acts,
870 lockdep_genl_is_held());
872 len = nla_total_size(FLOW_BUFSIZE); /* OVS_FLOW_ATTR_KEY */
873 len += nla_total_size(sf_acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
874 len += nla_total_size(sizeof(struct ovs_flow_stats)); /* OVS_FLOW_ATTR_STATS */
875 len += nla_total_size(1); /* OVS_FLOW_ATTR_TCP_FLAGS */
876 len += nla_total_size(8); /* OVS_FLOW_ATTR_USED */
877 return genlmsg_new(NLMSG_ALIGN(sizeof(struct ovs_header)) + len, GFP_KERNEL);
880 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow, struct datapath *dp,
881 u32 pid, u32 seq, u8 cmd)
886 skb = ovs_flow_cmd_alloc_info(flow);
888 return ERR_PTR(-ENOMEM);
890 retval = ovs_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd);
895 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
897 struct nlattr **a = info->attrs;
898 struct ovs_header *ovs_header = info->userhdr;
899 struct sw_flow_key key;
900 struct sw_flow *flow;
901 struct sk_buff *reply;
903 struct flow_table *table;
909 if (!a[OVS_FLOW_ATTR_KEY])
911 error = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
915 /* Validate actions. */
916 if (a[OVS_FLOW_ATTR_ACTIONS]) {
917 error = validate_actions(a[OVS_FLOW_ATTR_ACTIONS], 0);
920 } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
925 dp = get_dp(ovs_header->dp_ifindex);
930 table = get_table_protected(dp);
931 flow = flow_tbl_lookup(table, &key, key_len);
933 struct sw_flow_actions *acts;
935 /* Bail out if we're not allowed to create a new flow. */
937 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
940 /* Expand table, if necessary, to make room. */
941 if (flow_tbl_need_to_expand(table)) {
942 struct flow_table *new_table;
944 new_table = flow_tbl_expand(table);
945 if (!IS_ERR(new_table)) {
946 rcu_assign_pointer(dp->table, new_table);
947 flow_tbl_deferred_destroy(table);
948 table = get_table_protected(dp);
955 error = PTR_ERR(flow);
961 /* Obtain actions. */
962 acts = flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]);
963 error = PTR_ERR(acts);
965 goto error_free_flow;
966 rcu_assign_pointer(flow->sf_acts, acts);
968 /* Put flow in bucket. */
969 flow->hash = flow_hash(&key, key_len);
970 flow_tbl_insert(table, flow);
972 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
973 info->snd_seq, OVS_FLOW_CMD_NEW);
975 /* We found a matching flow. */
976 struct sw_flow_actions *old_acts;
978 /* Bail out if we're not allowed to modify an existing flow.
979 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
980 * because Generic Netlink treats the latter as a dump
981 * request. We also accept NLM_F_EXCL in case that bug ever
985 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
986 info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
989 /* Update actions. */
990 old_acts = rcu_dereference_protected(flow->sf_acts,
991 lockdep_genl_is_held());
992 if (a[OVS_FLOW_ATTR_ACTIONS] &&
993 (old_acts->actions_len != nla_len(a[OVS_FLOW_ATTR_ACTIONS]) ||
994 memcmp(old_acts->actions, nla_data(a[OVS_FLOW_ATTR_ACTIONS]),
995 old_acts->actions_len))) {
996 struct sw_flow_actions *new_acts;
998 new_acts = flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]);
999 error = PTR_ERR(new_acts);
1000 if (IS_ERR(new_acts))
1003 rcu_assign_pointer(flow->sf_acts, new_acts);
1004 flow_deferred_free_acts(old_acts);
1007 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1008 info->snd_seq, OVS_FLOW_CMD_NEW);
1011 if (a[OVS_FLOW_ATTR_CLEAR]) {
1012 spin_lock_bh(&flow->lock);
1014 spin_unlock_bh(&flow->lock);
1019 genl_notify(reply, genl_info_net(info), info->snd_pid,
1020 dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1022 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1023 dp_flow_multicast_group.id, PTR_ERR(reply));
1032 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1034 struct nlattr **a = info->attrs;
1035 struct ovs_header *ovs_header = info->userhdr;
1036 struct sw_flow_key key;
1037 struct sk_buff *reply;
1038 struct sw_flow *flow;
1039 struct datapath *dp;
1040 struct flow_table *table;
1044 if (!a[OVS_FLOW_ATTR_KEY])
1046 err = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1050 dp = get_dp(ovs_header->dp_ifindex);
1054 table = get_table_protected(dp);
1055 flow = flow_tbl_lookup(table, &key, key_len);
1059 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid, info->snd_seq, OVS_FLOW_CMD_NEW);
1061 return PTR_ERR(reply);
1063 return genlmsg_reply(reply, info);
1066 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1068 struct nlattr **a = info->attrs;
1069 struct ovs_header *ovs_header = info->userhdr;
1070 struct sw_flow_key key;
1071 struct sk_buff *reply;
1072 struct sw_flow *flow;
1073 struct datapath *dp;
1074 struct flow_table *table;
1078 if (!a[OVS_FLOW_ATTR_KEY])
1079 return flush_flows(ovs_header->dp_ifindex);
1080 err = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1084 dp = get_dp(ovs_header->dp_ifindex);
1088 table = get_table_protected(dp);
1089 flow = flow_tbl_lookup(table, &key, key_len);
1093 reply = ovs_flow_cmd_alloc_info(flow);
1097 flow_tbl_remove(table, flow);
1099 err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_pid,
1100 info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1103 flow_deferred_free(flow);
1105 genl_notify(reply, genl_info_net(info), info->snd_pid,
1106 dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1110 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1112 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1113 struct datapath *dp;
1115 dp = get_dp(ovs_header->dp_ifindex);
1120 struct sw_flow *flow;
1123 bucket = cb->args[0];
1125 flow = flow_tbl_next(get_table_protected(dp), &bucket, &obj);
1129 if (ovs_flow_cmd_fill_info(flow, dp, skb, NETLINK_CB(cb->skb).pid,
1130 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1131 OVS_FLOW_CMD_NEW) < 0)
1134 cb->args[0] = bucket;
1140 static struct genl_ops dp_flow_genl_ops[] = {
1141 { .cmd = OVS_FLOW_CMD_NEW,
1142 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1143 .policy = flow_policy,
1144 .doit = ovs_flow_cmd_new_or_set
1146 { .cmd = OVS_FLOW_CMD_DEL,
1147 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1148 .policy = flow_policy,
1149 .doit = ovs_flow_cmd_del
1151 { .cmd = OVS_FLOW_CMD_GET,
1152 .flags = 0, /* OK for unprivileged users. */
1153 .policy = flow_policy,
1154 .doit = ovs_flow_cmd_get,
1155 .dumpit = ovs_flow_cmd_dump
1157 { .cmd = OVS_FLOW_CMD_SET,
1158 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1159 .policy = flow_policy,
1160 .doit = ovs_flow_cmd_new_or_set,
1164 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1165 #ifdef HAVE_NLA_NUL_STRING
1166 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1168 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1169 [OVS_DP_ATTR_IPV4_FRAGS] = { .type = NLA_U32 },
1172 static struct genl_family dp_datapath_genl_family = {
1173 .id = GENL_ID_GENERATE,
1174 .hdrsize = sizeof(struct ovs_header),
1175 .name = OVS_DATAPATH_FAMILY,
1177 .maxattr = OVS_DP_ATTR_MAX
1180 static struct genl_multicast_group dp_datapath_multicast_group = {
1181 .name = OVS_DATAPATH_MCGROUP
1184 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1185 u32 pid, u32 seq, u32 flags, u8 cmd)
1187 struct ovs_header *ovs_header;
1191 ovs_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family,
1196 ovs_header->dp_ifindex = get_dpifindex(dp);
1199 err = nla_put_string(skb, OVS_DP_ATTR_NAME, dp_name(dp));
1202 goto nla_put_failure;
1204 nla = nla_reserve(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats));
1206 goto nla_put_failure;
1207 get_dp_stats(dp, nla_data(nla));
1209 NLA_PUT_U32(skb, OVS_DP_ATTR_IPV4_FRAGS,
1210 dp->drop_frags ? OVS_DP_FRAG_DROP : OVS_DP_FRAG_ZERO);
1212 return genlmsg_end(skb, ovs_header);
1215 genlmsg_cancel(skb, ovs_header);
1220 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 pid,
1223 struct sk_buff *skb;
1226 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1228 return ERR_PTR(-ENOMEM);
1230 retval = ovs_dp_cmd_fill_info(dp, skb, pid, seq, 0, cmd);
1233 return ERR_PTR(retval);
1238 static int ovs_dp_cmd_validate(struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1240 if (a[OVS_DP_ATTR_IPV4_FRAGS]) {
1241 u32 frags = nla_get_u32(a[OVS_DP_ATTR_IPV4_FRAGS]);
1243 if (frags != OVS_DP_FRAG_ZERO && frags != OVS_DP_FRAG_DROP)
1247 return CHECK_NUL_STRING(a[OVS_DP_ATTR_NAME], IFNAMSIZ - 1);
1250 /* Called with genl_mutex and optionally with RTNL lock also. */
1251 static struct datapath *lookup_datapath(struct ovs_header *ovs_header, struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1253 struct datapath *dp;
1255 if (!a[OVS_DP_ATTR_NAME])
1256 dp = get_dp(ovs_header->dp_ifindex);
1258 struct vport *vport;
1261 vport = vport_locate(nla_data(a[OVS_DP_ATTR_NAME]));
1262 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1265 return dp ? dp : ERR_PTR(-ENODEV);
1268 /* Called with genl_mutex. */
1269 static void change_datapath(struct datapath *dp, struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1271 if (a[OVS_DP_ATTR_IPV4_FRAGS])
1272 dp->drop_frags = nla_get_u32(a[OVS_DP_ATTR_IPV4_FRAGS]) == OVS_DP_FRAG_DROP;
1275 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1277 struct nlattr **a = info->attrs;
1278 struct vport_parms parms;
1279 struct sk_buff *reply;
1280 struct datapath *dp;
1281 struct vport *vport;
1285 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1288 err = ovs_dp_cmd_validate(a);
1294 if (!try_module_get(THIS_MODULE))
1295 goto err_unlock_rtnl;
1298 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1300 goto err_put_module;
1301 INIT_LIST_HEAD(&dp->port_list);
1303 /* Initialize kobject for bridge. This will be added as
1304 * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
1305 dp->ifobj.kset = NULL;
1306 kobject_init(&dp->ifobj, &dp_ktype);
1308 /* Allocate table. */
1310 rcu_assign_pointer(dp->table, flow_tbl_alloc(TBL_MIN_BUCKETS));
1315 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1316 if (!dp->stats_percpu) {
1318 goto err_destroy_table;
1321 change_datapath(dp, a);
1323 /* Set up our datapath device. */
1324 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1325 parms.type = OVS_VPORT_TYPE_INTERNAL;
1326 parms.options = NULL;
1328 parms.port_no = OVSP_LOCAL;
1329 parms.upcall_pid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1331 vport = new_vport(&parms);
1332 if (IS_ERR(vport)) {
1333 err = PTR_ERR(vport);
1337 goto err_destroy_percpu;
1340 reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_NEW);
1341 err = PTR_ERR(reply);
1343 goto err_destroy_local_port;
1345 list_add_tail(&dp->list_node, &dps);
1346 dp_sysfs_add_dp(dp);
1350 genl_notify(reply, genl_info_net(info), info->snd_pid,
1351 dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1354 err_destroy_local_port:
1355 dp_detach_port(get_vport_protected(dp, OVSP_LOCAL));
1357 free_percpu(dp->stats_percpu);
1359 flow_tbl_destroy(get_table_protected(dp));
1363 module_put(THIS_MODULE);
1370 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1372 struct vport *vport, *next_vport;
1373 struct sk_buff *reply;
1374 struct datapath *dp;
1377 err = ovs_dp_cmd_validate(info->attrs);
1382 dp = lookup_datapath(info->userhdr, info->attrs);
1387 reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_DEL);
1388 err = PTR_ERR(reply);
1392 list_for_each_entry_safe (vport, next_vport, &dp->port_list, node)
1393 if (vport->port_no != OVSP_LOCAL)
1394 dp_detach_port(vport);
1396 dp_sysfs_del_dp(dp);
1397 list_del(&dp->list_node);
1398 dp_detach_port(get_vport_protected(dp, OVSP_LOCAL));
1400 /* rtnl_unlock() will wait until all the references to devices that
1401 * are pending unregistration have been dropped. We do it here to
1402 * ensure that any internal devices (which contain DP pointers) are
1403 * fully destroyed before freeing the datapath.
1407 call_rcu(&dp->rcu, destroy_dp_rcu);
1408 module_put(THIS_MODULE);
1410 genl_notify(reply, genl_info_net(info), info->snd_pid,
1411 dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1421 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1423 struct sk_buff *reply;
1424 struct datapath *dp;
1427 err = ovs_dp_cmd_validate(info->attrs);
1431 dp = lookup_datapath(info->userhdr, info->attrs);
1435 change_datapath(dp, info->attrs);
1437 reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_NEW);
1438 if (IS_ERR(reply)) {
1439 err = PTR_ERR(reply);
1440 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1441 dp_datapath_multicast_group.id, err);
1445 genl_notify(reply, genl_info_net(info), info->snd_pid,
1446 dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1450 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1452 struct sk_buff *reply;
1453 struct datapath *dp;
1456 err = ovs_dp_cmd_validate(info->attrs);
1460 dp = lookup_datapath(info->userhdr, info->attrs);
1464 reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_NEW);
1466 return PTR_ERR(reply);
1468 return genlmsg_reply(reply, info);
1471 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1473 struct datapath *dp;
1474 int skip = cb->args[0];
1477 list_for_each_entry (dp, &dps, list_node) {
1480 if (ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).pid,
1481 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1482 OVS_DP_CMD_NEW) < 0)
1492 static struct genl_ops dp_datapath_genl_ops[] = {
1493 { .cmd = OVS_DP_CMD_NEW,
1494 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1495 .policy = datapath_policy,
1496 .doit = ovs_dp_cmd_new
1498 { .cmd = OVS_DP_CMD_DEL,
1499 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1500 .policy = datapath_policy,
1501 .doit = ovs_dp_cmd_del
1503 { .cmd = OVS_DP_CMD_GET,
1504 .flags = 0, /* OK for unprivileged users. */
1505 .policy = datapath_policy,
1506 .doit = ovs_dp_cmd_get,
1507 .dumpit = ovs_dp_cmd_dump
1509 { .cmd = OVS_DP_CMD_SET,
1510 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1511 .policy = datapath_policy,
1512 .doit = ovs_dp_cmd_set,
1516 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1517 #ifdef HAVE_NLA_NUL_STRING
1518 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1519 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1520 [OVS_VPORT_ATTR_ADDRESS] = { .len = ETH_ALEN },
1522 [OVS_VPORT_ATTR_STATS] = { .minlen = sizeof(struct ovs_vport_stats) },
1523 [OVS_VPORT_ATTR_ADDRESS] = { .minlen = ETH_ALEN },
1525 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1526 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1527 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1528 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1531 static struct genl_family dp_vport_genl_family = {
1532 .id = GENL_ID_GENERATE,
1533 .hdrsize = sizeof(struct ovs_header),
1534 .name = OVS_VPORT_FAMILY,
1536 .maxattr = OVS_VPORT_ATTR_MAX
1539 struct genl_multicast_group dp_vport_multicast_group = {
1540 .name = OVS_VPORT_MCGROUP
1543 /* Called with RTNL lock or RCU read lock. */
1544 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1545 u32 pid, u32 seq, u32 flags, u8 cmd)
1547 struct ovs_header *ovs_header;
1551 ovs_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family,
1556 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1558 NLA_PUT_U32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no);
1559 NLA_PUT_U32(skb, OVS_VPORT_ATTR_TYPE, vport_get_type(vport));
1560 NLA_PUT_STRING(skb, OVS_VPORT_ATTR_NAME, vport_get_name(vport));
1561 NLA_PUT_U32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_pid);
1563 nla = nla_reserve(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats));
1565 goto nla_put_failure;
1567 vport_get_stats(vport, nla_data(nla));
1569 NLA_PUT(skb, OVS_VPORT_ATTR_ADDRESS, ETH_ALEN, vport_get_addr(vport));
1571 err = vport_get_options(vport, skb);
1572 if (err == -EMSGSIZE)
1575 return genlmsg_end(skb, ovs_header);
1580 genlmsg_cancel(skb, ovs_header);
1584 /* Called with RTNL lock or RCU read lock. */
1585 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 pid,
1588 struct sk_buff *skb;
1591 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1593 return ERR_PTR(-ENOMEM);
1595 retval = ovs_vport_cmd_fill_info(vport, skb, pid, seq, 0, cmd);
1598 return ERR_PTR(retval);
1603 static int ovs_vport_cmd_validate(struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1605 return CHECK_NUL_STRING(a[OVS_VPORT_ATTR_NAME], IFNAMSIZ - 1);
1608 /* Called with RTNL lock or RCU read lock. */
1609 static struct vport *lookup_vport(struct ovs_header *ovs_header,
1610 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1612 struct datapath *dp;
1613 struct vport *vport;
1615 if (a[OVS_VPORT_ATTR_NAME]) {
1616 vport = vport_locate(nla_data(a[OVS_VPORT_ATTR_NAME]));
1618 return ERR_PTR(-ENODEV);
1620 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1621 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1623 if (port_no >= DP_MAX_PORTS)
1624 return ERR_PTR(-EFBIG);
1626 dp = get_dp(ovs_header->dp_ifindex);
1628 return ERR_PTR(-ENODEV);
1630 vport = get_vport_protected(dp, port_no);
1632 return ERR_PTR(-ENOENT);
1635 return ERR_PTR(-EINVAL);
1638 /* Called with RTNL lock. */
1639 static int change_vport(struct vport *vport, struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1643 if (a[OVS_VPORT_ATTR_STATS])
1644 vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1646 if (a[OVS_VPORT_ATTR_ADDRESS])
1647 err = vport_set_addr(vport, nla_data(a[OVS_VPORT_ATTR_ADDRESS]));
1652 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1654 struct nlattr **a = info->attrs;
1655 struct ovs_header *ovs_header = info->userhdr;
1656 struct vport_parms parms;
1657 struct sk_buff *reply;
1658 struct vport *vport;
1659 struct datapath *dp;
1664 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1665 !a[OVS_VPORT_ATTR_UPCALL_PID])
1668 err = ovs_vport_cmd_validate(a);
1673 dp = get_dp(ovs_header->dp_ifindex);
1678 if (a[OVS_VPORT_ATTR_PORT_NO]) {
1679 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1682 if (port_no >= DP_MAX_PORTS)
1685 vport = get_vport_protected(dp, port_no);
1690 for (port_no = 1; ; port_no++) {
1691 if (port_no >= DP_MAX_PORTS) {
1695 vport = get_vport_protected(dp, port_no);
1701 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1702 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1703 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1705 parms.port_no = port_no;
1706 parms.upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1708 vport = new_vport(&parms);
1709 err = PTR_ERR(vport);
1713 dp_sysfs_add_if(vport);
1715 err = change_vport(vport, a);
1717 reply = ovs_vport_cmd_build_info(vport, info->snd_pid,
1718 info->snd_seq, OVS_VPORT_CMD_NEW);
1720 err = PTR_ERR(reply);
1723 dp_detach_port(vport);
1726 genl_notify(reply, genl_info_net(info), info->snd_pid,
1727 dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1736 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1738 struct nlattr **a = info->attrs;
1739 struct sk_buff *reply;
1740 struct vport *vport;
1743 err = ovs_vport_cmd_validate(a);
1748 vport = lookup_vport(info->userhdr, a);
1749 err = PTR_ERR(vport);
1754 if (a[OVS_VPORT_ATTR_TYPE] && nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport_get_type(vport))
1756 if (!err && a[OVS_VPORT_ATTR_OPTIONS])
1757 err = vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1759 err = change_vport(vport, a);
1760 if (!err && a[OVS_VPORT_ATTR_UPCALL_PID])
1761 vport->upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1763 reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1765 if (IS_ERR(reply)) {
1766 err = PTR_ERR(reply);
1767 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1768 dp_vport_multicast_group.id, err);
1772 genl_notify(reply, genl_info_net(info), info->snd_pid,
1773 dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1781 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1783 struct nlattr **a = info->attrs;
1784 struct sk_buff *reply;
1785 struct vport *vport;
1788 err = ovs_vport_cmd_validate(a);
1793 vport = lookup_vport(info->userhdr, a);
1794 err = PTR_ERR(vport);
1798 if (vport->port_no == OVSP_LOCAL) {
1803 reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1805 err = PTR_ERR(reply);
1809 dp_detach_port(vport);
1811 genl_notify(reply, genl_info_net(info), info->snd_pid,
1812 dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1820 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1822 struct nlattr **a = info->attrs;
1823 struct ovs_header *ovs_header = info->userhdr;
1824 struct sk_buff *reply;
1825 struct vport *vport;
1828 err = ovs_vport_cmd_validate(a);
1833 vport = lookup_vport(ovs_header, a);
1834 err = PTR_ERR(vport);
1838 reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1840 err = PTR_ERR(reply);
1846 return genlmsg_reply(reply, info);
1854 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1856 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1857 struct datapath *dp;
1861 dp = get_dp(ovs_header->dp_ifindex);
1866 for (port_no = cb->args[0]; port_no < DP_MAX_PORTS; port_no++) {
1867 struct vport *vport;
1869 vport = get_vport_protected(dp, port_no);
1873 if (ovs_vport_cmd_fill_info(vport, skb, NETLINK_CB(cb->skb).pid,
1874 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1875 OVS_VPORT_CMD_NEW) < 0)
1880 cb->args[0] = port_no;
1886 static struct genl_ops dp_vport_genl_ops[] = {
1887 { .cmd = OVS_VPORT_CMD_NEW,
1888 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1889 .policy = vport_policy,
1890 .doit = ovs_vport_cmd_new
1892 { .cmd = OVS_VPORT_CMD_DEL,
1893 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1894 .policy = vport_policy,
1895 .doit = ovs_vport_cmd_del
1897 { .cmd = OVS_VPORT_CMD_GET,
1898 .flags = 0, /* OK for unprivileged users. */
1899 .policy = vport_policy,
1900 .doit = ovs_vport_cmd_get,
1901 .dumpit = ovs_vport_cmd_dump
1903 { .cmd = OVS_VPORT_CMD_SET,
1904 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1905 .policy = vport_policy,
1906 .doit = ovs_vport_cmd_set,
1910 struct genl_family_and_ops {
1911 struct genl_family *family;
1912 struct genl_ops *ops;
1914 struct genl_multicast_group *group;
1917 static const struct genl_family_and_ops dp_genl_families[] = {
1918 { &dp_datapath_genl_family,
1919 dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1920 &dp_datapath_multicast_group },
1921 { &dp_vport_genl_family,
1922 dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1923 &dp_vport_multicast_group },
1924 { &dp_flow_genl_family,
1925 dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1926 &dp_flow_multicast_group },
1927 { &dp_packet_genl_family,
1928 dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1932 static void dp_unregister_genl(int n_families)
1936 for (i = 0; i < n_families; i++)
1937 genl_unregister_family(dp_genl_families[i].family);
1940 static int dp_register_genl(void)
1947 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
1948 const struct genl_family_and_ops *f = &dp_genl_families[i];
1950 err = genl_register_family_with_ops(f->family, f->ops,
1957 err = genl_register_mc_group(f->family, f->group);
1966 dp_unregister_genl(n_registered);
1970 static int __init dp_init(void)
1972 struct sk_buff *dummy_skb;
1975 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
1977 printk("Open vSwitch %s, built "__DATE__" "__TIME__"\n", VERSION BUILDNR);
1985 goto error_tnl_exit;
1989 goto error_flow_exit;
1991 err = register_netdevice_notifier(&dp_device_notifier);
1993 goto error_vport_exit;
1995 err = dp_register_genl();
1997 goto error_unreg_notifier;
2001 error_unreg_notifier:
2002 unregister_netdevice_notifier(&dp_device_notifier);
2013 static void dp_cleanup(void)
2016 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2017 unregister_netdevice_notifier(&dp_device_notifier);
2023 module_init(dp_init);
2024 module_exit(dp_cleanup);
2026 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2027 MODULE_LICENSE("GPL");